NoahSnowCover
The snow-cover diagnostics of WRF's unified Noah land-surface model: SUBROUTINE SNFRAC (the fraction of the cell the pack covers), SUBROUTINE ALCALC (the albedo and emissivity that cover implies, including the ageing of fresh snow) and SUBROUTINE CSNOW (the thermal conductivity of the pack), all of phys/module_sf_noahlsm.F (WRF 4.8, sf_surface_physics = 2). These are the algebraic quantities that turn a snow water equivalent into what the radiation and the soil actually see. SNFRAC is the Anderson/Koren depletion curve sncovr = 1 - (exp(-salp rsnow) - rsnow exp(-salp)) with rsnow = sneqv/snup, saturating at 1 once the pack exceeds the vegetation type's snup; SFLX then CAPS the result at 0.98, so a fully buried cell still reports 2 % bare ground and sncovr = 1 is unreachable in practice -- which matters, because the 0.98 value is what selects the pure-snow conductivity branch at module_sf_noahlsm.F:627. ALBEDO ageing: fresh snow starts at snoalb1 = snoalb + lvcoef (0.85 - snoalb) and decays as snoalb1 snacca^((snotime1/86400)^snaccb), floored at the snow-free albedo; snotime1 is the time since the last snowfall, an accumulator that WRF resets to zero whenever it is snowing, so this component exposes its rate dsnotime1_dt rather than owning it. The cell albedo is then the cover-weighted blend alb + sncovr (snoalb2 - alb), clipped so it can never exceed the aged snow albedo itself. Emissivity blends the background against a fixed snow value the same way. CSNOW is Yen's exponential fit sncond = 2 unit 0.328 10^(2.25 dsnow), where dsnow is the pack DENSITY in g/cm^3 and unit = 0.11631 converts Yen's units. NOT TRANSCRIBED: the whole UA_PHYS canopy-burial branch of SNFRAC (fvb, gama, fbur, fgsn, and the leaf-area reduction it applies), which is off in every reference run and is a separate parameterization; and the snow-conductivity BLEND that SFLX applies around this routine, which belongs to the snow-pack energy balance rather than to the cover diagnostics. THE NO-PACK CUTOFF. SFLX treats any pack of sneqv <= 1e-7 m as no pack at all (module_sf_noahlsm.F:503-507): it zeroes sneqv and snowh, sets sndens = 0 and sncond = 1.0 instead of calling CSNOW, and then, finding sneqv == 0, skips SNFRAC and ALCALC and reports sncovr = 0, albedo = alb, emissi = embrd. This component applies the same cutoff to its sneqv input, so a snow-free cell gives finite, WRF-equal diagnostics rather than 0/0. One ordering detail cannot be represented by an input-state component and is stated rather than hidden: SFLX applies the cutoff to the pack ENTERING the step and only then adds the step's new snowfall, so a step that snows onto a <= 1e-7 m pack diagnoses the cover of the new snow alone; in the dt -> 0 limit this component computes, that increment vanishes and the two coincide. The component is algebraic in the consumer's state and owns nothing.
Description
The snow-cover diagnostics of WRF’s unified Noah land-surface model: SUBROUTINE SNFRAC (the fraction of the cell the pack covers), SUBROUTINE ALCALC (the albedo and emissivity that cover implies, including the ageing of fresh snow) and SUBROUTINE CSNOW (the thermal conductivity of the pack), all of phys/module_sf_noahlsm.F (WRF 4.8, sf_surface_physics = 2). These are the algebraic quantities that turn a snow water equivalent into what the radiation and the soil actually see. SNFRAC is the Anderson/Koren depletion curve sncovr = 1 - (exp(-salp rsnow) - rsnow exp(-salp)) with rsnow = sneqv/snup, saturating at 1 once the pack exceeds the vegetation type’s snup; SFLX then CAPS the result at 0.98, so a fully buried cell still reports 2 % bare ground and sncovr = 1 is unreachable in practice – which matters, because the 0.98 value is what selects the pure-snow conductivity branch at module_sf_noahlsm.F:627. ALBEDO ageing: fresh snow starts at snoalb1 = snoalb + lvcoef (0.85 - snoalb) and decays as snoalb1 snacca^((snotime1/86400)^snaccb), floored at the snow-free albedo; snotime1 is the time since the last snowfall, an accumulator that WRF resets to zero whenever it is snowing, so this component exposes its rate dsnotime1_dt rather than owning it. The cell albedo is then the cover-weighted blend alb + sncovr (snoalb2 - alb), clipped so it can never exceed the aged snow albedo itself. Emissivity blends the background against a fixed snow value the same way. CSNOW is Yen’s exponential fit sncond = 2 unit 0.328 10^(2.25 dsnow), where dsnow is the pack DENSITY in g/cm^3 and unit = 0.11631 converts Yen’s units. NOT TRANSCRIBED: the whole UA_PHYS canopy-burial branch of SNFRAC (fvb, gama, fbur, fgsn, and the leaf-area reduction it applies), which is off in every reference run and is a separate parameterization; and the snow-conductivity BLEND that SFLX applies around this routine, which belongs to the snow-pack energy balance rather than to the cover diagnostics. THE NO-PACK CUTOFF. SFLX treats any pack of sneqv <= 1e-7 m as no pack at all (module_sf_noahlsm.F:503-507): it zeroes sneqv and snowh, sets sndens = 0 and sncond = 1.0 instead of calling CSNOW, and then, finding sneqv == 0, skips SNFRAC and ALCALC and reports sncovr = 0, albedo = alb, emissi = embrd. This component applies the same cutoff to its sneqv input, so a snow-free cell gives finite, WRF-equal diagnostics rather than 0/0. One ordering detail cannot be represented by an input-state component and is stated rather than hidden: SFLX applies the cutoff to the pack ENTERING the step and only then adds the step’s new snowfall, so a step that snows onto a <= 1e-7 m pack diagnoses the cover of the new snow alone; in the dt -> 0 limit this component computes, that increment vanishes and the two coincide. The component is algebraic in the consumer’s state and owns nothing.
Reference
- Anderson, E. A. (1976). A point energy and mass balance model of a snow cover. NOAA Tech. Rep. NWS 19.
- Koren, V., et al. (1999). A parameterization of snowpack and frozen ground intended for NCEP weather and climate models. J. Geophys. Res., 104(D16), 19569-19585.
- Yen, Y.-C. (1981). Review of thermal properties of snow, ice and sea ice. CRREL Report 81-10.
- Livneh, B., et al. (2010). Noah LSM snow model diagnostics and enhancements. J. Hydrometeor., 11, 721-738 (the snotime1 albedo ageing).
- wrf-model/WRF v4.8.0 phys/module_sf_noahlsm.F SUBROUTINE SNFRAC lines 2817-2919, ALCALC lines 891-1006, CSNOW lines 1148-1187, and the SNCOVR cap at line 588 (cold-season reference run instrumented in ctessum-claude/WRF branch earthsciml-instrumented-noahcold).
Parameters
| Name | Units | Default | Description |
|---|---|---|---|
sneqv | m | 0.01 | Snow water equivalent of the pack, WRF’s SNEQV, as a depth of liquid water. |
snowh | m | 0.05 | Physical depth of the pack, WRF’s SNOWH. Together with sneqv it gives the density. |
snup | m | 0.04 | Snow water equivalent at which the vegetation type is considered fully covered, VEGPARM.TBL SNUPTBL. |
alb | 1 | 0.17 | Snow-free background albedo of the cell, WRF’s ALB from the greenness interpolation. |
snoalb | 1 | 0.7 | Maximum (deep, fresh) snow albedo of the cell, WRF’s SNOALB. |
embrd | 1 | 0.985 | Snow-free background emissivity of the cell. |
snotime1 | s | 0.0 | Time elapsed since the last snowfall, WRF’s SNOTIME1. A consumer-supplied accumulator: this component reports its rate and does not own it. |
snowing | 1 | 0.0 | Indicator (1/0) for WRF’s SNOWNG. While it is one the pack is refreshed, so the albedo returns to its fresh value and the age accumulator is held at zero. |
emissi_s | 1 | 0.95 | Emissivity of a fully snow-covered surface, the module PARAMETER EMISSI_S of module_sf_noahlsm.F. |
Observed
| Name | Units | Default | Description |
|---|---|---|---|
rsnow | 1 | Pack depth relative to the full-cover threshold, sneqv/snup, capped at one where the pack is deeper. | |
sncovr_raw | 1 | The depletion curve itself, 1 - (exp(-salp rsnow) - rsnow exp(-salp)) below the threshold and exactly 1 at or above it. | |
sncovr | 1 | Snow-cover fraction as the rest of SFLX sees it: the depletion curve CAPPED AT 0.98, and ZERO where sneqv <= 1e-7 m. The cap is not in SNFRAC; SFLX applies it immediately after the call, and it is what makes a completely buried cell still report 2 % bare ground. The zero is SFLX’s ‘IF (SNEQV <= 1.E-7) SNEQV = 0’ cutoff (module_sf_noahlsm.F:503) followed by its ‘IF (SNEQV == 0.0) SNCOVR = 0’ branch (:563), which skips SNFRAC and ALCALC altogether; with the cover zero the blends below return albedo = alb and emissi = embrd, exactly what that branch assigns. | |
snoalb1 | 1 | Albedo of fresh snow on this cell, snoalb + lvcoef (0.85 - snoalb): the tabulated maximum pulled part of the way towards 0.85 by the ageing coefficient. | |
snoalb2 | 1 | Aged snow albedo, snoalb1 snacca^((snotime1/86400)^snaccb) when it is not snowing and snoalb1 itself when it is, floored at the snow-free albedo alb. | |
albedo | 1 | Cell albedo, the cover-weighted blend alb + sncovr (snoalb2 - alb), clipped so it never exceeds the aged snow albedo. | |
emissi | 1 | Cell emissivity, embrd + sncovr (emissi_s - embrd). | |
sndens | g/cm^3 | Bulk density of the pack in the units CSNOW expects, sneqv/snowh expressed in g/cm^3 (numerically the ratio itself, since a water-equivalent depth over a physical depth times 1000 kg/m^3 is that ratio in g/cm^3). ZERO where sneqv <= 1e-7 m, SFLX’s own assignment (module_sf_noahlsm.F:503-507), which is also what keeps a snow-free cell (sneqv = snowh = 0) from forming 0/0. | |
sncond | W/(m*K) | Thermal conductivity of the pack, Yen’s 2 unit 0.328 10^(2.25 sndens); 1.0 W/(m K) where sneqv <= 1e-7 m, the value SFLX assigns in place of calling CSNOW (module_sf_noahlsm.F:506). | |
dsnotime1_dt | 1 | Rate of the snow-age accumulator: one while no snow is falling, and zero while it is, because WRF resets snotime1 to zero on a snowing step. The reset itself is a projection of the state, not a rate, so a consumer integrating this must apply it separately – the indicator snowing is what tells it to. |
Observed expressions
$$ rsnow = \mathrm{min}\left(\frac{sneqv}{snup}, 1.0\right) $$Pack depth relative to the full-cover threshold, sneqv/snup, capped at one where the pack is deeper.
$$ sncovr_{raw} = \mathrm{ifelse}\left(\mathrm{<}\left(sneqv, snup\right), 1.0 - \left(e^{\left(0.0 - par.salp\right) \cdot rsnow} - rsnow \cdot e^{0.0 - par.salp}\right), 1.0\right) $$The depletion curve itself, 1 - (exp(-salp rsnow) - rsnow exp(-salp)) below the threshold and exactly 1 at or above it.
$$ sncovr = \mathrm{ifelse}\left(\mathrm{<=}\left(sneqv, 1 \times 10^{-7}\right), 0.0, \mathrm{min}\left(sncovr_{raw}, 0.98\right)\right) $$Snow-cover fraction as the rest of SFLX sees it: the depletion curve CAPPED AT 0.98, and ZERO where sneqv <= 1e-7 m. The cap is not in SNFRAC; SFLX applies it immediately after the call, and it is what makes a completely buried cell still report 2 % bare ground. The zero is SFLX’s ‘IF (SNEQV <= 1.E-7) SNEQV = 0’ cutoff (module_sf_noahlsm.F:503) followed by its ‘IF (SNEQV == 0.0) SNCOVR = 0’ branch (:563), which skips SNFRAC and ALCALC altogether; with the cover zero the blends below return albedo = alb and emissi = embrd, exactly what that branch assigns.
$$ snoalb1 = snoalb + par.lvcoef \cdot \left(0.85 - snoalb\right) $$Albedo of fresh snow on this cell, snoalb + lvcoef (0.85 - snoalb): the tabulated maximum pulled part of the way towards 0.85 by the ageing coefficient.
$$ snoalb2 = \mathrm{max}\left(\mathrm{ifelse}\left(snowing = 1.0, snoalb1, snoalb1 \cdot 0.94^{\frac{snotime1}{86400.0}^{0.58}}\right), alb\right) $$Aged snow albedo, snoalb1 snacca^((snotime1/86400)^snaccb) when it is not snowing and snoalb1 itself when it is, floored at the snow-free albedo alb.
$$ albedo = \mathrm{min}\left(alb + sncovr \cdot \left(snoalb2 - alb\right), snoalb2\right) $$Cell albedo, the cover-weighted blend alb + sncovr (snoalb2 - alb), clipped so it never exceeds the aged snow albedo.
$$ emissi = embrd + sncovr \cdot \left(emissi_{s} - embrd\right) $$Cell emissivity, embrd + sncovr (emissi_s - embrd).
$$ sndens = \mathrm{ifelse}\left(\mathrm{<=}\left(sneqv, 1 \times 10^{-7}\right), 0.0, \frac{sneqv \cdot par.rho_{water\,cgs}}{snowh}\right) $$Bulk density of the pack in the units CSNOW expects, sneqv/snowh expressed in g/cm^3 (numerically the ratio itself, since a water-equivalent depth over a physical depth times 1000 kg/m^3 is that ratio in g/cm^3). ZERO where sneqv <= 1e-7 m, SFLX’s own assignment (module_sf_noahlsm.F:503-507), which is also what keeps a snow-free cell (sneqv = snowh = 0) from forming 0/0.
$$ sncond = \mathrm{ifelse}\left(\mathrm{<=}\left(sneqv, 1 \times 10^{-7}\right), 1.0, 2.0 \cdot 0.11631 \cdot 0.328 \cdot 10.0^{2.25 \cdot sndens}\right) $$Thermal conductivity of the pack, Yen’s 2 unit 0.328 10^(2.25 sndens); 1.0 W/(m K) where sneqv <= 1e-7 m, the value SFLX assigns in place of calling CSNOW (module_sf_noahlsm.F:506).
$$ dsnotime1_{dt} = \mathrm{ifelse}\left(snowing = 1.0, 0.0, 1.0\right) $$Rate of the snow-age accumulator: one while no snow is falling, and zero while it is, because WRF resets snotime1 to zero on a snowing step. The reset itself is a projection of the state, not a rate, so a consumer integrating this must apply it separately – the indicator snowing is what tells it to.
Raw .esm
Raw .esm JSON (this component)
{
"subsystems": {
"par": {
"ref": "./noah_parameters.esm"
}
},
"variables": {
"sneqv": {
"type": "parameter",
"units": "m",
"default": 0.01,
"description": "Snow water equivalent of the pack, WRF's SNEQV, as a depth of liquid water."
},
"snowh": {
"type": "parameter",
"units": "m",
"default": 0.05,
"description": "Physical depth of the pack, WRF's SNOWH. Together with sneqv it gives the density."
},
"snup": {
"type": "parameter",
"units": "m",
"default": 0.04,
"description": "Snow water equivalent at which the vegetation type is considered fully covered, VEGPARM.TBL SNUPTBL."
},
"alb": {
"type": "parameter",
"units": "1",
"default": 0.17,
"description": "Snow-free background albedo of the cell, WRF's ALB from the greenness interpolation."
},
"snoalb": {
"type": "parameter",
"units": "1",
"default": 0.7,
"description": "Maximum (deep, fresh) snow albedo of the cell, WRF's SNOALB."
},
"embrd": {
"type": "parameter",
"units": "1",
"default": 0.985,
"description": "Snow-free background emissivity of the cell."
},
"snotime1": {
"type": "parameter",
"units": "s",
"default": 0.0,
"description": "Time elapsed since the last snowfall, WRF's SNOTIME1. A consumer-supplied accumulator: this component reports its rate and does not own it."
},
"snowing": {
"type": "parameter",
"units": "1",
"default": 0.0,
"description": "Indicator (1/0) for WRF's SNOWNG. While it is one the pack is refreshed, so the albedo returns to its fresh value and the age accumulator is held at zero."
},
"emissi_s": {
"type": "parameter",
"units": "1",
"default": 0.95,
"description": "Emissivity of a fully snow-covered surface, the module PARAMETER EMISSI_S of module_sf_noahlsm.F."
},
"rsnow": {
"type": "unknown",
"units": "1",
"description": "Pack depth relative to the full-cover threshold, sneqv/snup, capped at one where the pack is deeper."
},
"sncovr_raw": {
"type": "unknown",
"units": "1",
"description": "The depletion curve itself, 1 - (exp(-salp rsnow) - rsnow exp(-salp)) below the threshold and exactly 1 at or above it."
},
"sncovr": {
"type": "unknown",
"units": "1",
"description": "Snow-cover fraction as the rest of SFLX sees it: the depletion curve CAPPED AT 0.98, and ZERO where sneqv <= 1e-7 m. The cap is not in SNFRAC; SFLX applies it immediately after the call, and it is what makes a completely buried cell still report 2 % bare ground. The zero is SFLX's 'IF (SNEQV <= 1.E-7) SNEQV = 0' cutoff (module_sf_noahlsm.F:503) followed by its 'IF (SNEQV == 0.0) SNCOVR = 0' branch (:563), which skips SNFRAC and ALCALC altogether; with the cover zero the blends below return albedo = alb and emissi = embrd, exactly what that branch assigns."
},
"snoalb1": {
"type": "unknown",
"units": "1",
"description": "Albedo of fresh snow on this cell, snoalb + lvcoef (0.85 - snoalb): the tabulated maximum pulled part of the way towards 0.85 by the ageing coefficient."
},
"snoalb2": {
"type": "unknown",
"units": "1",
"description": "Aged snow albedo, snoalb1 snacca^((snotime1/86400)^snaccb) when it is not snowing and snoalb1 itself when it is, floored at the snow-free albedo alb."
},
"albedo": {
"type": "unknown",
"units": "1",
"description": "Cell albedo, the cover-weighted blend alb + sncovr (snoalb2 - alb), clipped so it never exceeds the aged snow albedo."
},
"emissi": {
"type": "unknown",
"units": "1",
"description": "Cell emissivity, embrd + sncovr (emissi_s - embrd)."
},
"sndens": {
"type": "unknown",
"units": "g/cm^3",
"description": "Bulk density of the pack in the units CSNOW expects, sneqv/snowh expressed in g/cm^3 (numerically the ratio itself, since a water-equivalent depth over a physical depth times 1000 kg/m^3 is that ratio in g/cm^3). ZERO where sneqv <= 1e-7 m, SFLX's own assignment (module_sf_noahlsm.F:503-507), which is also what keeps a snow-free cell (sneqv = snowh = 0) from forming 0/0."
},
"sncond": {
"type": "unknown",
"units": "W/(m*K)",
"description": "Thermal conductivity of the pack, Yen's 2 unit 0.328 10^(2.25 sndens); 1.0 W/(m K) where sneqv <= 1e-7 m, the value SFLX assigns in place of calling CSNOW (module_sf_noahlsm.F:506)."
},
"dsnotime1_dt": {
"type": "unknown",
"units": "1",
"description": "Rate of the snow-age accumulator: one while no snow is falling, and zero while it is, because WRF resets snotime1 to zero on a snowing step. The reset itself is a projection of the state, not a rate, so a consumer integrating this must apply it separately -- the indicator `snowing` is what tells it to."
}
},
"equations": [
{
"lhs": "rsnow",
"rhs": {
"op": "min",
"args": [
{
"op": "/",
"args": [
"sneqv",
"snup"
]
},
1.0
]
}
},
{
"lhs": "sncovr_raw",
"rhs": {
"op": "ifelse",
"args": [
{
"op": "<",
"args": [
"sneqv",
"snup"
]
},
{
"op": "-",
"args": [
1.0,
{
"op": "-",
"args": [
{
"op": "exp",
"args": [
{
"op": "*",
"args": [
{
"op": "-",
"args": [
0.0,
"par.salp"
]
},
"rsnow"
]
}
]
},
{
"op": "*",
"args": [
"rsnow",
{
"op": "exp",
"args": [
{
"op": "-",
"args": [
0.0,
"par.salp"
]
}
]
}
]
}
]
}
]
},
1.0
]
}
},
{
"lhs": "sncovr",
"rhs": {
"op": "ifelse",
"args": [
{
"op": "<=",
"args": [
"sneqv",
1e-07
]
},
0.0,
{
"op": "min",
"args": [
"sncovr_raw",
0.98
]
}
]
}
},
{
"lhs": "snoalb1",
"rhs": {
"op": "+",
"args": [
"snoalb",
{
"op": "*",
"args": [
"par.lvcoef",
{
"op": "-",
"args": [
0.85,
"snoalb"
]
}
]
}
]
}
},
{
"lhs": "snoalb2",
"rhs": {
"op": "max",
"args": [
{
"op": "ifelse",
"args": [
{
"op": "==",
"args": [
"snowing",
1.0
]
},
"snoalb1",
{
"op": "*",
"args": [
"snoalb1",
{
"op": "^",
"args": [
0.94,
{
"op": "^",
"args": [
{
"op": "/",
"args": [
"snotime1",
86400.0
]
},
0.58
]
}
]
}
]
}
]
},
"alb"
]
}
},
{
"lhs": "albedo",
"rhs": {
"op": "min",
"args": [
{
"op": "+",
"args": [
"alb",
{
"op": "*",
"args": [
"sncovr",
{
"op": "-",
"args": [
"snoalb2",
"alb"
]
}
]
}
]
},
"snoalb2"
]
}
},
{
"lhs": "emissi",
"rhs": {
"op": "+",
"args": [
"embrd",
{
"op": "*",
"args": [
"sncovr",
{
"op": "-",
"args": [
"emissi_s",
"embrd"
]
}
]
}
]
}
},
{
"lhs": "sndens",
"rhs": {
"op": "ifelse",
"args": [
{
"op": "<=",
"args": [
"sneqv",
1e-07
]
},
0.0,
{
"op": "/",
"args": [
{
"op": "*",
"args": [
"sneqv",
"par.rho_water_cgs"
]
},
"snowh"
]
}
]
}
},
{
"lhs": "sncond",
"rhs": {
"op": "ifelse",
"args": [
{
"op": "<=",
"args": [
"sneqv",
1e-07
]
},
1.0,
{
"op": "*",
"args": [
{
"op": "*",
"args": [
2.0,
0.11631
]
},
{
"op": "*",
"args": [
0.328,
{
"op": "^",
"args": [
10.0,
{
"op": "*",
"args": [
2.25,
"sndens"
]
}
]
}
]
}
]
}
]
}
},
{
"lhs": "dsnotime1_dt",
"rhs": {
"op": "ifelse",
"args": [
{
"op": "==",
"args": [
"snowing",
1.0
]
},
0.0,
1.0
]
}
}
],
"tests": [
{
"id": "cold_caseA_step101",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 101 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0003061296883970499,
"snowh": 0.003943329211324453,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.020270196083057135,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17516890175307298,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9842905571522402,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.07763229291570732,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step1761",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1761 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0034124997910112143,
"snowh": 0.038037508726119995,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2052709015400025,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.222344081313787,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9778155298147908,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.08971407185618176,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step1824",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1824 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.00343441148288548,
"snowh": 0.036144647747278214,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.20645170152417336,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.22264518530763913,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9777742017984534,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.09501853516316583,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step1849",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1849 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.003413871396332979,
"snowh": 0.03516868129372597,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.20534486366955518,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.22236294165669068,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9778129411391985,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.09707135072448943,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step1874",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1874 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.003390752011910081,
"snowh": 0.034208543598651886,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.20409732784970258,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.22204482002485904,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9778566049107394,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.09912003421538136,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step1899",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1899 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.003366285702213645,
"snowh": 0.033275604248046875,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2027751380939995,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.221707661639519,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9779028815711032,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.10116377383159025,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step1924",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1924 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0033420040272176266,
"snowh": 0.03238306939601898,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2014609171200188,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.22137253529350392,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9779488793239925,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.10320220070529869,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step2089",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2089 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0033869065809994936,
"snowh": 0.030685460194945335,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 1e-06,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2038896504724078,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.22199185986112066,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9778638736219156,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.11037496454425459,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step2114",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2114 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.003592143068090081,
"snowh": 0.032493770122528076,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 1020.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.21490396307194037,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2243710364377851,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.977478372523371,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.11054866993095966,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step26",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 26 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 4.9339892029820476e-06,
"snowh": 6.265289266593754e-05,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.00032981949957555656,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17008410575994135,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9849884706179115,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.07875117960259617,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step2847",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2847 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0036658388562500477,
"snowh": 0.03318041190505028,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 45000.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.21882447332697563,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.22194086634736868,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9773411546083614,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.11048201772601168,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step2880",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2880 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0036658267490565777,
"snowh": 0.03317866101861,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 46980.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.21882383072902312,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.22184521934109994,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9773411770992989,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.11048748311453833,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step328",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 328 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.000598000013269484,
"snowh": 0.007400842849165201,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.03923464557835188,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.18000483634046205,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9836268011486163,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.08080160941911331,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step364",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 364 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0006727515137754381,
"snowh": 0.008181992918252945,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.0440357188457946,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.18122911001507497,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9834587635155759,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.08222342899807383,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step397",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 397 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.000754952838178724,
"snowh": 0.009015229530632496,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.049289188207118695,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.18256874469281867,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9832748920127781,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.08374194307700665,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step51",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 51 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 8.478196832584217e-05,
"snowh": 0.001084580784663558,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.005653097633327997,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17144154167452946,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9848021558070801,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.078170265898473,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step519",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 519 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0010642694542184472,
"snowh": 0.012081330642104149,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.06881516109885122,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.1875478677452953,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9825914826822462,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.08809207245299581,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step679",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 679 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0012426043394953012,
"snowh": 0.013881165534257889,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.07990083335811693,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.19037471415158533,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.98220348399459,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.08951729135633246,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseA_step76",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 76 of case A, snow accumulating onto initially bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.00020789193513337523,
"snowh": 0.0026703651528805494,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.01380810719384784,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17352106909787973,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9845167303558036,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.0778515009112726,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step1",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.06000000238418579,
"snowh": 0.29999998211860657,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 1e-06,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.41989998834197856,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.20000001986822602,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step100",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 100 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.06030664220452309,
"snowh": 0.30376335978507996,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19853165387397087,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step1612",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1612 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.06072096899151802,
"snowh": 0.30848824977874756,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 60600.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.3994403900738926,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19683397677365913,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step1637",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1637 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.060493938624858856,
"snowh": 0.3044896721839905,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 62100.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.3991555724036031,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19867320356493975,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step1816",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1816 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.054353997111320496,
"snowh": 0.25674816966056824,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 72840.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.397200458703163,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.2117015953160917,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step1841",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1841 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.05315843224525452,
"snowh": 0.24898476898670197,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 74340.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.39693807829004757,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.21350073926968552,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step1866",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1866 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.05196269601583481,
"snowh": 0.24135594069957733,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 75840.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.3966780848371751,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.21529487057790106,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step2122",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2122 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.045533597469329834,
"snowh": 0.19721251726150513,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 91200.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.39413905695702656,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.23088594021115305,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step26",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 26 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0600040927529335,
"snowh": 0.30001741647720337,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.20000203140699563,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step2880",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2880 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.045533597469329834,
"snowh": 0.19706536829471588,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 136680.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.38759639158736625,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.2310583430429746,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step435",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 435 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.060732778161764145,
"snowh": 0.3098119795322418,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19603108392745472,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step463",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 463 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.060771744698286057,
"snowh": 0.310437947511673,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19576132745812103,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step492",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 492 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.06078648194670677,
"snowh": 0.3108033537864685,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19557859079110138,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step51",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 51 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.06008230149745941,
"snowh": 0.3009808361530304,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19962168444100983,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step517",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 517 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.06078018620610237,
"snowh": 0.31090250611305237,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19549596741900635,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step561",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 561 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0608341321349144,
"snowh": 0.31172868609428406,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4199000000357628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.1951508951489534,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseB_step658",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 658 of case B, a 60 kg/m2 pack under warm advection: SNFRAC saturates and the cover sits at the 0.98 cap; the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.060759976506233215,
"snowh": 0.3112076222896576,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 3360.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.98,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.4159988740090341,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9507000002861022,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19523935840390388,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005000000353902578,
"snowh": 0.02499999850988388,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 1e-06,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2867568586410364,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.24312299680713545,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9749635201505887,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.20000002607703476,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1564",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1564 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.00440554553642869,
"snowh": 0.02148446999490261,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 70380.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.25718852393208325,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2297402015556053,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9759984122883805,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.205057212836713,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step169",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 169 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005403401795774698,
"snowh": 0.03003288432955742,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.30620513880906053,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2480823116369123,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.974282830066498,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.17991617909162547,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1785",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1785 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0037659858353435993,
"snowh": 0.01713281124830246,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 83640.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2241231849202019,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.221539677886829,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9771556996267998,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.21981131880717505,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1810",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1810 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0036482352297753096,
"snowh": 0.016465021297335625,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 85140.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.21788963257664729,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2200513816750471,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9773738740479959,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.22157488678043827,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1838",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1838 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0035192957147955894,
"snowh": 0.015743093565106392,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 86820.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2110107330869384,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.21841227462963173,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9776146356285391,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.22354537246823802,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1863",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1863 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0034089668188244104,
"snowh": 0.015130738727748394,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 88320.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.20508036098903404,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2170009325418542,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9778221987368004,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.22530075234134853,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1888",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1888 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0033047660253942013,
"snowh": 0.01455513946712017,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 89820.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.19944155250189521,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.215659654907512,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9780195571145142,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.22705148465781333,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step1978",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1978 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.002997202565893531,
"snowh": 0.012845908291637897,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 95220.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.1825810320054171,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.21164106249393397,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9786096755730825,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.23331962971115228,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step2003",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2003 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0029336686711758375,
"snowh": 0.012480990961194038,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 96720.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.17905735152491842,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.21079494030456836,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9787330044403066,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.23505094109206573,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step2104",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2104 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.002814299426972866,
"snowh": 0.011629151180386543,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 102780.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.17239880166574628,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.20911556019091374,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.978966053780629,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.24200385594201143,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step2119",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2119 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.002814315492287278,
"snowh": 0.011586623266339302,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 103680.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.17239970117168413,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.20909203174029878,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9789660222979083,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.24289350120351708,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step2141",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2141 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.002814929699525237,
"snowh": 0.011589097790420055,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 105000.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.1724340902973429,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.20906518563715193,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9789648186780183,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.24289463687606117,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step2166",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2166 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0028158514760434628,
"snowh": 0.01159284170717001,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 106500.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.1724856975671265,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2090377399302569,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9789630124228376,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.24289570643424754,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step26",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 26 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005003620404750109,
"snowh": 0.02504994161427021,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2869335735641815,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.24316806253392842,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9749573351257507,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.19974579109866086,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step268",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 268 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005462722387164831,
"snowh": 0.03075668215751648,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.3090239097234353,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.24880109821503754,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.974184173044172,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.17761091262119633,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step2880",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 2880 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.0028400917071849108,
"snowh": 0.011691033840179443,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 149340.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.17384175685001746,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.20830986607690224,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9789155503285378,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.24292904682425617,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step380",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 380 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005304833874106407,
"snowh": 0.02874927408993244,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.30149827075390057,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.24688206029126308,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9744475705157608,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.18452061980927115,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step405",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 405 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005202005617320538,
"snowh": 0.027913276106119156,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 840.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2965569687945533,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.245092358805429,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9746205161550239,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.1863631340720497,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step434",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 434 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005062823183834553,
"snowh": 0.026858793571591377,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 2580.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2898179234936723,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.24291310103277428,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9748563828369575,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.18849778827081867,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step459",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 459 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.004942660219967365,
"snowh": 0.02596842683851719,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 4080.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2839524246268551,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2411435046633389,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9750616753812028,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.1903334480265086,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step56",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 56 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.005102005787193775,
"snowh": 0.02628016471862793,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 1.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.29172103106936875,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.24438886418919054,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.974789774044584,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.1941390338188336,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step708",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 708 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.00436509819701314,
"snowh": 0.02129247412085533,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 19020.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.2551364065212315,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.2323063784736019,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9760702364271161,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.2050066221631651,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseC_step733",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 733 of case C, a thin pack melting while the soil thaws and refreezes (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 0.004366421140730381,
"snowh": 0.021298784762620926,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 20520.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.25520360903512096,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.23220048247659297,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9760678843381686,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.20500799408956938,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseD_step1",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 1 of case D, a 0.05 kg/m2 trace pack disappearing into bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 5.0000002374872565e-05,
"snowh": 0.0002500000118743628,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 1e-06,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.0033375665622663586,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17085108121572395,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9848831994276912,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.2,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseD_step26",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 26 of case D, a 0.05 kg/m2 trace pack disappearing into bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 4.9963087803917006e-05,
"snowh": 0.0002498149697203189,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 1500.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.0033351063534242664,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17084212299693996,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9848832855350358,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.20000037571748935,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseD_step392",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 392 of case D, a 0.05 kg/m2 trace pack disappearing into bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 4.588760566548444e-05,
"snowh": 0.00021633169671986252,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 23460.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.003063456438819556,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17074390639968834,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9848927932859329,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.21211688514218424,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseD_step417",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 417 of case D, a 0.05 kg/m2 trace pack disappearing into bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 4.2226442019455135e-05,
"snowh": 0.00019741823780350387,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 24960.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.002819363128271135,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17068339593630347,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.984901336555294,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.21389331851760254,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseD_step51",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 51 of case D, a 0.05 kg/m2 trace pack disappearing into bare frozen ground (partial cover throughout); the pack is on the sub-freezing SNOPAC branch.",
"parameter_overrides": {
"sneqv": 4.995149720343761e-05,
"snowh": 0.0002497566747479141,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 3000.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.0033343338849722848,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17083782448374887,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9848833125714427,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.20000064964772196,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseD_step521",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 521 of case D, a 0.05 kg/m2 trace pack disappearing into bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 1.2073031030013226e-05,
"snowh": 5.456982398754917e-05,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 31200.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.0008068568553057931,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.170194196768761,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9849717743036369,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.22124005811035405,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "cold_caseD_step553",
"description": "Reference: WRF 4.8 em_scm_xy cold-season single-column run, data/eqwefic/noahcold_scm, dumped by ctessum-claude/WRF branch earthsciml-instrumented-noahcold and replayed through the real64 (-fdefault-real-8) kernels/noah_driver at ESM_DT = 1e-6 s. Inputs are the dumped SFLX arguments (real32 values widened). SNFRAC and ALCALC are called early in SFLX, before the NOPAC/SNOPAC branch -- but NOT before the new snowfall is added to the pack, so on a snowing step the cover SFLX diagnoses is that of the pack INCLUDING this step's accumulation and is therefore dt-dependent after all. Measured over the 67 pack steps, sncovr moves by up to 2.4e-5 and the albedo by 5.8e-6 between ESM_DT = 0.01 s and 1e-6 s, so the reference here is taken at 1e-6 s, where it is the input-state limit this component computes. A residual of that increment survives even there on the thinnest packs -- case A's 5e-6 m pack moves sncovr by 2.4e-9 relative over a 1e-6 s step -- which is what sets the 1e-8 bound below; it bounds the REFERENCE, not the component. A second dt dependence is easier to miss still: ALCALC advances the snow-age accumulator snotime1 BEFORE it uses it, so the age that sets the albedo is the POST-update one. This component takes that post-update age as its `snotime1` parameter and the test supplies the replay's snotime1_out, which makes the albedo exact rather than dt-sensitive. sndens is referenced against the density SNOPAC leaves at the END of the step, which at this step size is the input-state ratio to well within the bound below. NOTE which background albedo and emissivity are the right ones: the dump's `albbrd` and `embrd` are the values the DRIVER carried in, and REDPRM overwrites both by greenness before SFLX uses them, so the test supplies the greenness-selected values (albedomin and emissmax on this shdfac >= shdmax column) -- exactly what soil_veg_parameters.esm computes. Using the dumped `embrd` instead moves the emissivity by 7 %. Replay floors from the cold-season notes bound what can honestly be claimed: sncovr 2.2e-5, albedo 9.8e-8. Step 553 of case D, a 0.05 kg/m2 trace pack disappearing into bare frozen ground (partial cover throughout); the pack is on the melting SNOPAC branch.",
"parameter_overrides": {
"sneqv": 1.2517150480562123e-06,
"snowh": 5.6008470892265905e-06,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 33120.000001,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 8.368240136436e-05,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.1700201010513326,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9849970854198699,
"tolerance": {
"rel": 1e-08
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.22348673836685715,
"tolerance": {
"rel": 1e-06
}
}
]
},
{
"id": "nopack_zero",
"description": "Reference: the dumped SFLX arguments of cold-season case A step 101 (data/eqwefic/dumps/noahcold_wrf/caseA/esm_dump_noah_101.json) with sneqv = 0 m, snowh = 0 m and prcp = 0, replayed through the real64 kernels/noah_driver at ESM_DT = 1e-6 s (data/eqwefic/noah_pr58_fixes/r64_snow_zero.json, written by run_scenario.py there). A snow-free cell. SFLX's 'IF (SNEQV <= 1.E-7)' block (module_sf_noahlsm.F:503-507) zeroes the pack and assigns SNDENS = 0 and SNCOND = 1.0, then its 'IF (SNEQV == 0.0)' branch reports SNCOVR = 0, ALBEDO = ALB and EMISSI = EMBRD without calling SNFRAC or ALCALC: the replay returns sncovr = 0, sneqv_out = snowh_out = 0, and the albedo and emissivity asserted below. sndens and sncond are not published on this path (SNOPAC never runs), so those two expected values are SFLX's literal assignments. Before the cutoff was transcribed this state gave sndens = sncond = NaN.",
"parameter_overrides": {
"sneqv": 0.0,
"snowh": 0.0,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.0,
"tolerance": {
"abs": 0.0,
"rel": 0.0
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17000000178813934,
"tolerance": {
"rel": 1e-12
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9850000143051147,
"tolerance": {
"rel": 1e-12
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.0,
"tolerance": {
"abs": 0.0,
"rel": 0.0
}
},
{
"variable": "sncond",
"time": 0.0,
"expected": 1.0,
"tolerance": {
"abs": 0.0,
"rel": 0.0
}
}
]
},
{
"id": "nopack_below_cutoff",
"description": "Reference: the dumped SFLX arguments of cold-season case A step 101 (data/eqwefic/dumps/noahcold_wrf/caseA/esm_dump_noah_101.json) with sneqv = 5e-08 m, snowh = 5e-07 m and prcp = 0, replayed through the real64 kernels/noah_driver at ESM_DT = 1e-6 s (data/eqwefic/noah_pr58_fixes/r64_snow_tiny.json, written by run_scenario.py there). A pack BELOW the 1e-7 m cutoff, which the depletion curve alone would give a cover of ~3e-6. SFLX's 'IF (SNEQV <= 1.E-7)' block (module_sf_noahlsm.F:503-507) zeroes the pack and assigns SNDENS = 0 and SNCOND = 1.0, then its 'IF (SNEQV == 0.0)' branch reports SNCOVR = 0, ALBEDO = ALB and EMISSI = EMBRD without calling SNFRAC or ALCALC: the replay returns sncovr = 0, sneqv_out = snowh_out = 0, and the albedo and emissivity asserted below. sndens and sncond are not published on this path (SNOPAC never runs), so those two expected values are SFLX's literal assignments. Before the cutoff was transcribed this state gave sndens = sncond = NaN.",
"parameter_overrides": {
"sneqv": 5e-08,
"snowh": 5e-07,
"snup": 0.03999999910593033,
"alb": 0.17000000178813934,
"snoalb": 0.0,
"embrd": 0.9850000143051147,
"snotime1": 0.0,
"snowing": 0.0
},
"time_span": {
"start": 0.0,
"end": 1.0
},
"assertions": [
{
"variable": "sncovr",
"time": 0.0,
"expected": 0.0,
"tolerance": {
"abs": 0.0,
"rel": 0.0
}
},
{
"variable": "albedo",
"time": 0.0,
"expected": 0.17000000178813934,
"tolerance": {
"rel": 1e-12
}
},
{
"variable": "emissi",
"time": 0.0,
"expected": 0.9850000143051147,
"tolerance": {
"rel": 1e-12
}
},
{
"variable": "sndens",
"time": 0.0,
"expected": 0.0,
"tolerance": {
"abs": 0.0,
"rel": 0.0
}
},
{
"variable": "sncond",
"time": 0.0,
"expected": 1.0,
"tolerance": {
"abs": 0.0,
"rel": 0.0
}
}
]
}
]
}