deepguess/wrf-250m-severe-weather

WRF 250m Severe Weather Simulations — Raw WRF Output

2 TB of raw WRF-ARW v4.7.1 output from 18 high-impact weather events at 250m resolution. Three nested domains (3km / 1km / 250m), 81 vertical levels, 216 variables per wrfout file, 1-minute surface output via auxhist2. This is the full simulation output — nothing removed, nothing derived.

If you want pre-extracted surface fields in a simpler format, see wrf-250m-severe-weather-netcdf4 (16 fields, ~150 GB).

This dataset is uploading now. Files appear as events process. Check back for updates.

What's Here

Every file is a standard WRF netCDF file. Open with xarray, netCDF4, wrf-python, Panoply, NCL, or any netCDF tool.

File Types

File Pattern	Description	Grid	Temporal	Typical Size
wrfout_d01_*	Full 3D output, 3km domain	~200x200 x 79 levels	15-min intervals	~75 MB
wrfout_d02_*	Full 3D output, 1km domain	~600x600 x 79 levels	15-min intervals	~300 MB
wrfout_d03_*	Full 3D output, 250m domain	800x800 x 79 levels	15-min intervals	~1.0 GB
auxhist2_d01_*	Surface diagnostics, 3km	~200x200	15-min intervals	~8 MB
auxhist2_d02_*	Surface diagnostics, 1km	~600x600	5-min intervals	~40 MB
auxhist2_d03_*	Surface diagnostics, 250m	800x800	1-min intervals	~165 MB

What's in wrfout (216 variables)

The full WRF state — everything needed to restart the simulation or derive any diagnostic:

3D fields (79 levels, 800x800):

• T — perturbation potential temperature (add 300K for full theta)
• U, V, W — wind components (U/V on staggered grids, destagger before use)
• P, PB — perturbation + base pressure (add for total pressure)
• PH, PHB — perturbation + base geopotential (add, divide by 9.81 for height AGL)
• QVAPOR, QCLOUD, QRAIN, QICE, QSNOW, QGRAUP — moisture species mixing ratios
• CLDFRA — cloud fraction
• REFL_10CM — simulated radar reflectivity at each level (column-max for composite)
• THM — moist potential temperature
• P_HYD — hydrostatic pressure

Surface fields (800x800):

• T2, Q2, U10, V10, PSFC — standard 2m/10m diagnostics
• HFX, LH, GRDFLX — surface heat fluxes
• PBLH — planetary boundary layer height
• RAINNC, RAINC, SNOWNC — accumulated precipitation
• REFD_MAX — running max reflectivity (accumulated, not instantaneous)
• WSPD10MAX, W_UP_MAX, W_DN_MAX — running max wind/updraft/downdraft
• UP_HELI_MAX — running max 2-5km updraft helicity
• HAIL_MAX2D — running max hail diameter
• SWDOWN, GLW, OLR — radiation fields
• SNOW, SNOWH, SNOWC — snow fields
• SST, TSK, TMN — surface/skin/deep soil temperature
• SMOIS, SH2O, TSLB — soil moisture/temperature (4 layers)
• XLAT, XLONG — latitude/longitude arrays
• HGT — terrain height
• LANDMASK, LU_INDEX, IVGTYP, ISLTYP — land use
• Plus ~60 more (radiation budget, map factors, soil, etc.)

What's in auxhist2 (surface diagnostics, 1-min resolution)

The subset written at 1-minute intervals for the 250m domain:

• T2, U10, V10, PSFC — standard surface met
• HFX, LH — surface fluxes
• PBLH — PBL height
• RAINNC — accumulated precip
• REFL_10CM — full 3D reflectivity (79 levels)
• W — full 3D vertical velocity (80 levels, staggered)
• REFD_MAX, WSPD10MAX, W_UP_MAX, W_DN_MAX — running maxima
• UP_HELI_MAX — updraft helicity
• HAIL_MAX2D, HAIL_MAXK1 — hail
• UST — friction velocity

Events

Event	Key	Date	Category	Size	wrfout d03	auxhist2 d03	auxhist2 Status
Carr Fire	carr_fire	2018-07-23	Wildfire	145 GB	29 files	421 files	Full (166 MB/file)
Hurricane Michael	michael	2018-10-10	Hurricane	68 GB	35 files	524 files	Empty shells
Camp Fire	camp_fire	2018-11-08	Wildfire	143 GB	29 files	419 files	Full (167 MB/file)
Nashville EF3	nashville	2020-03-03	Tornado	69 GB	42 files	618 files	Empty shells
Death Valley Record	death_valley	2020-08-16	Heat	127 GB	24 files	358 files	Full (164 MB/file)
LA Fires	la_fires	2020-09-06	Wildfire	142 GB	29 files	421 files	Full (120 MB/file)
SF Bay Fires	sfbay	2020-09-06	Wildfire	139 GB	29 files	421 files	Full (115 MB/file)
PNW Windstorm	pnw_windstorm	2020-09-08	Wind	123 GB	24 files	360 files	Full (168 MB/file)
Texas Freeze	tx_freeze	2021-02-16	Winter	31 GB	22 files	421 files	Empty shells
Seattle Heat Dome	seattle_heat_dome	2021-06-28	Heat	33 GB	21 files	301 files	Empty shells
Mayfield EF4	mayfield	2021-12-11	Tornado	63 GB	22 files	421 files	Empty shells
CA Atmospheric River	ca_atmos_river	2021-12-30	Flooding	144 GB	28 files	421 files	Full (166 MB/file)
Buffalo Blizzard	buffalo_blizzard	2022-12-23	Winter	27 GB	14 files	393 files	Empty shells
CA Pineapple Express	ca_pineapple_express	2023-01-04	Flooding	157 GB	29 files	421 files	Full (163 MB/file)
Pineapple Express	pineapple_express	2024-02-04	Flooding	146 GB	29 files	421 files	Full (118 MB/file)
Denver Hailstorm	denver_hail	2024-05-31	Hail	146 GB	29 files	421 files	Full (165 MB/file)
LA Fires Peak 2025	la_fires_peak	2025-01-07	Wildfire	140 GB	28 files	421 files	Full (119 MB/file)
Enderlin EF5	enderlin_ef5	2025-06-21	Tornado	126 GB	24 files	360 files	Full (166 MB/file)

"Empty shells" = auxhist2 files exist but contain only timestamps (0 MB data). These events were run before we added auxhist2 output fields. wrfout files are complete for all events.

Quirks and Gotchas

Windows Filenames

These were produced on Windows. WRF writes colons in filenames (wrfout_d03_2018-07-23_21:00:00) but NTFS doesn't allow colons. The files use underscores instead of colons in the time portion:

wrfout_d03_2018-07-23_21_00_00    # underscore format (what's on disk)

This is standard for WRF-on-Windows. Most tools handle it fine.

Staggered Grids

WRF uses Arakawa C-grid staggering:

• U is on the west-east staggered grid: shape (1, 79, 800, 801) — destagger with 0.5 * (U[:,:,:,:-1] + U[:,:,:,1:])
• V is on the south-north staggered grid: shape (1, 79, 801, 800) — destagger with 0.5 * (V[:,:,:-1,:] + V[:,:,1:,:])
• W is on the bottom-top staggered grid: shape (1, 80, 800, 800) — destagger with 0.5 * (W[:,:-1,:,:] + W[:,1:,:,:])
• PH, PHB are also vertically staggered (80 levels)

Perturbation Variables

WRF stores several fields as base + perturbation:

pressure_pa = ds['P'][:] + ds['PB'][:]           # Total pressure (Pa)
theta = ds['T'][:] + 300.0                        # Total potential temperature (K)
temperature_k = theta * (pressure_pa / 100000.0) ** 0.2854  # Temperature (K)
geopotential = ds['PH'][:] + ds['PHB'][:]        # Total geopotential (m2/s2)
height_m = geopotential / 9.81                    # Height (m)

Running Max Fields (Accumulated)

REFD_MAX, WSPD10MAX, W_UP_MAX, W_DN_MAX, UP_HELI_MAX, HAIL_MAX2D are running maxima since simulation start — they only increase over time. They are NOT instantaneous values. For instantaneous reflectivity, use REFL_10CM (3D, column-max it yourself).

auxhist2 Has No Coordinates

auxhist2 files do NOT contain XLAT/XLONG arrays. Read coordinates from the corresponding wrfout file:

import xarray as xr
coords = xr.open_dataset("wrfout_d03_2018-07-23_21_00_00")
aux = xr.open_dataset("auxhist2_d03_2018-07-23_21_01_00")
lat = coords["XLAT"].values  # (1, 800, 800)
lon = coords["XLONG"].values

Projection

All domains use Lambert Conformal Conic projection. Parameters vary per event (each is centered on its area of interest). Read from any wrfout file's global attributes:

ds = xr.open_dataset("wrfout_d03_2018-07-23_21_00_00")
print(ds.attrs["CEN_LAT"])     # Center latitude
print(ds.attrs["CEN_LON"])     # Center longitude
print(ds.attrs["TRUELAT1"])    # True latitude 1
print(ds.attrs["TRUELAT2"])    # True latitude 2
print(ds.attrs["STAND_LON"])   # Standard longitude
print(ds.attrs["MAP_PROJ"])    # 1 = Lambert Conformal

Full per-event WRF configuration is in wrf_config.json.

Domain Nesting

Three one-way nested domains:

• d01 (3km): Parent domain, ~200x200 grid, provides boundary conditions
• d02 (1km): Intermediate nest, ~600x600, 4:1 ratio from d01
• d03 (250m): Inner nest, 800x800, 4:1 ratio from d02

d03 is where the action is. d01/d02 are included for boundary context and coarser diagnostics.

Quick Start

import xarray as xr
import numpy as np

# Open a wrfout file
ds = xr.open_dataset("carr_fire/wrfout_d03_2018-07-23_21_00_00")

# Compute actual temperature from perturbation theta
P = ds["P"].values + ds["PB"].values              # Total pressure (Pa)
theta = ds["T"].values + 300.0                     # Potential temperature (K)
T = theta * (P / 100000.0) ** 0.2854              # Temperature (K)

# Surface fields
t2m = ds["T2"].values - 273.15                     # 2m temp in Celsius
psfc = ds["PSFC"].values / 100.0                   # Surface pressure in hPa

# Composite reflectivity from 3D field
refc = np.max(ds["REFL_10CM"].values, axis=1)      # Max over vertical levels

# Wind (destagger U and V first)
U = 0.5 * (ds["U"].values[:,:,:,:-1] + ds["U"].values[:,:,:,1:])
V = 0.5 * (ds["V"].values[:,:,:-1,:] + ds["V"].values[:,:,1:,:])
wind_speed = np.sqrt(U**2 + V**2)

# Coordinates
lat = ds["XLAT"].values[0]   # (800, 800)
lon = ds["XLONG"].values[0]  # (800, 800)

Using wrf-python

from wrf import getvar, ALL_TIMES
import netCDF4 as nc

ncfile = nc.Dataset("carr_fire/wrfout_d03_2018-07-23_21_00_00")
# wrf-python handles destaggering and derived variables automatically
T2 = getvar(ncfile, "T2")
slp = getvar(ncfile, "slp")        # Sea-level pressure
ua = getvar(ncfile, "ua")          # Destaggered U-wind (m/s)
va = getvar(ncfile, "va")          # Destaggered V-wind (m/s)
z = getvar(ncfile, "z")            # Height AGL (m)
dbz = getvar(ncfile, "dbz")        # Reflectivity (dBZ)
cape = getvar(ncfile, "cape_2d")   # CAPE/CIN

WRF Configuration

All simulations use WRF-ARW v4.7.1 with these key settings:

Setting	Value
Microphysics	Thompson (MP_PHYSICS=8)
Radiation (LW/SW)	RRTMG (RA_LW/SW_PHYSICS=4)
Surface layer	Revised MM5 (SF_SFCLAY_PHYSICS=1)
Land surface	Thermal diffusion (SF_SURFACE_PHYSICS=1)
PBL	None (LES mode, BL_PBL_PHYSICS=0)
Cumulus	None (CU_PHYSICS=0)
Diffusion	2D Smagorinsky (DIFF_OPT=2, KM_OPT=5)
Vertical levels	81 (hybrid, DZBOT=50m)
Adaptive timestep	1-2s
Damping	w-Rayleigh (DAMP_OPT=3, DAMPCOEF=0.2)

No PBL scheme — at 250m, the simulation resolves turbulent eddies directly (implicit LES). This is correct for dx < ~1km.

Full per-event namelist parameters are in wrf_config.json.

File Organization

wrf_config.json
carr_fire/
  wrfout_d01_2018-07-23_21_00_00
  wrfout_d01_2018-07-23_21_15_00
  ...
  wrfout_d02_2018-07-23_21_00_00
  ...
  wrfout_d03_2018-07-23_21_00_00
  ...
  auxhist2_d01_2018-07-23_21_00_00
  ...
  auxhist2_d03_2018-07-23_21_00_00
  auxhist2_d03_2018-07-23_21_01_00    # 1-minute intervals
  auxhist2_d03_2018-07-23_21_02_00
  ...
michael/
  ...

Citation

@dataset{wrf_250m_severe_raw_2025,
  title={WRF 250m Severe Weather Simulations: Raw WRF Output},
  author={Fahrenheit Research},
  year={2025},
  url={https://huggingface.co/datasets/deepguess/wrf-250m-severe-weather},
  license={CC-BY-4.0}
}

License

CC-BY-4.0