deepguess/wrf-250m-severe-weather-overlays

WRF 250m Severe Weather Overlays

18 high-resolution severe weather simulations, ready to plot.

250-meter WRF model output covering tornadoes, wildfires, hurricanes, blizzards, heat waves, and flooding events across the US. Each event includes surface weather fields at 1-minute temporal resolution on an 800x800 grid — just load with numpy and start making maps.

What's in the box

Each event folder contains:

• coords.npz — latitude/longitude arrays (xlat, xlong, both 800x800 float32)
• t_0000.npz through t_NNNN.npz — one file per timestep, each containing 16 weather fields
• metadata.json — event info, timestamps, field descriptions

Quick start

import numpy as np

# Load coordinates
coords = np.load("carr_fire/coords.npz")
lat, lon = coords["xlat"], coords["xlong"]

# Load a single timestep
data = np.load("carr_fire/t_0200.npz")

# Plot wind speed
import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(10, 10))
c = ax.pcolormesh(lon, lat, data["wind_speed_10m"], cmap="YlOrRd", vmin=0, vmax=30)
ax.set_title("Carr Fire — 10m Wind Speed (m/s)")
plt.colorbar(c)
plt.savefig("carr_fire_wind.png", dpi=150)

Make an animation

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
import json

with open("carr_fire/metadata.json") as f:
    meta = json.load(f)

coords = np.load("carr_fire/coords.npz")
lat, lon = coords["xlat"], coords["xlong"]

fig, ax = plt.subplots(figsize=(10, 10))
data0 = np.load("carr_fire/t_0000.npz")
mesh = ax.pcolormesh(lon, lat, data0["refc"], cmap="turbo", vmin=-10, vmax=70)
title = ax.set_title("")

def update(frame):
    data = np.load(f"carr_fire/t_{frame:04d}.npz")
    mesh.set_array(data["refc"].ravel())
    title.set_text(f"Carr Fire Reflectivity — {meta['times'][frame]}")
    return mesh, title

anim = FuncAnimation(fig, update, frames=range(0, 421, 5), interval=100)
anim.save("carr_fire_refc.mp4", dpi=100)

Fields

Every .npz timestep file contains these 16 fields, all float32 at 800x800:

Field	Description	Units	Notes
t2m	2-meter temperature	degC
u10	10-meter U-wind component	m/s
v10	10-meter V-wind component	m/s
wind_speed_10m	10-meter wind speed	m/s	Derived: sqrt(u10^2 + v10^2)
wind_direction_10m	10-meter wind direction	degrees	Meteorological convention (wind FROM)
surface_pressure	Surface pressure	hPa
pblh	Planetary boundary layer height	m
hfx	Surface sensible heat flux	W/m2
lh	Surface latent heat flux	W/m2
rain_rate	Precipitation rate	mm/hr	Derived from accumulated RAINNC
refc	Composite reflectivity	dBZ	Column-max of 3D reflectivity
wspd10max	Max 10m wind gust	m/s	Running max since simulation start
w_up_max	Max updraft speed	m/s	Running max since simulation start
w_dn_max	Max downdraft speed	m/s	Running max since simulation start
updraft_helicity	Max updraft helicity 2-5 km	m2/s2	Running max since simulation start
hail_max	Max hail diameter	mm	Running max since simulation start

Note on "running max" fields: wspd10max, w_up_max, w_dn_max, updraft_helicity, and hail_max are accumulated maxima that increase monotonically over the simulation. They show the worst conditions experienced at each grid point up to that time — useful for swath maps showing the full storm impact.

Events

Event	Date	Category	Location	Timesteps	Temporal Res
Carr Fire	2018-07-23	Wildfire	Redding, CA	421	1 min
Hurricane Michael	2018-10-10	Hurricane	Panama City, FL	~23	15 min
Camp Fire	2018-11-08	Wildfire	Paradise, CA	418	1 min
Nashville EF3 Tornado	2020-03-03	Tornado	Nashville, TN	~28	15 min
Death Valley Record Heat	2020-08-16	Heat	Death Valley, CA	357	1 min
LA Fires	2020-09-06	Wildfire	Los Angeles, CA	421	1 min
SF Bay Area Fires	2020-09-06	Wildfire	San Francisco, CA	421	1 min
PNW Windstorm	2020-09-08	Wind	Pacific Northwest	360	1 min
Texas Freeze	2021-02-16	Winter	Texas	~22	15 min
Seattle Heat Dome	2021-06-28	Heat	Seattle, WA	~10	15 min
Mayfield EF4 Tornado	2021-12-11	Tornado	Mayfield, KY	~22	15 min
CA Atmospheric River	2021-12-30	Flooding	Northern California	421	1 min
Buffalo Blizzard	2022-12-23	Winter	Buffalo, NY	~14	15 min
CA Pineapple Express 2023	2023-01-04	Flooding	California	421	1 min
Pineapple Express	2024-02-04	Flooding	Southern California	418	1 min
Denver Hailstorm	2024-05-31	Hail	Denver, CO	421	1 min
LA Fires Peak 2025	2025-01-07	Wildfire	Los Angeles, CA	421	1 min
Enderlin EF5 Tornado	2025-06-21	Tornado	Enderlin, ND	~360	1 min

12 events have 1-minute temporal resolution from WRF auxiliary history output. 6 events have 15-minute resolution from standard WRF output files.

Grid details

• Model: WRF-ARW v4, d03 (innermost nest)
• Horizontal resolution: 250 meters
• Grid size: 800 x 800 points (200 km x 200 km)
• Projection: Lambert Conformal Conic (varies per event)
• Coordinates: Each event has its own coords.npz with the exact lat/lon for every grid point

File sizes

• Individual timestep .npz: ~20-25 MB (16 fields, float32, 800x800, numpy compressed)
• Full event (421 timesteps): ~8-10 GB
• Total dataset: ~128 GB

Coordinate system

The WRF model uses a Lambert Conformal Conic projection centered on each event location. The coords.npz file contains the exact latitude and longitude of every grid point. Use these for plotting — don't assume a regular lat/lon grid.

coords = np.load("carr_fire/coords.npz")
lat = coords["xlat"]   # shape (800, 800), float32
lon = coords["xlong"]  # shape (800, 800), float32

# These are NOT regularly spaced in lat/lon
# Use pcolormesh, not imshow, for correct geographic placement

For web map overlays (Leaflet, Mapbox), you'll need to reproject from Lambert Conformal to Web Mercator. The metadata.json for each event includes the projection parameters if you need them.

Source

Produced by Fahrenheit Research using WRF-ARW at 250m resolution. Simulations were run on the d03 innermost nest with 1-minute auxiliary history output enabled for surface fields.

Raw WRF netCDF files were processed into .npz format to make them accessible without WRF-specific tools. All fields are on the native d03 mass grid (no interpolation).

WRF model configuration

Full WRF namelist-equivalent settings for every event are in wrf_config.json. Key settings shared across all 18 simulations:

Setting	Value	Description
WRF version	4.7.1
Grid (d03)	800 x 800	Innermost nest
Horizontal resolution	250m	DX = DY = 250m
Vertical levels	80	Hybrid sigma-pressure
Timestep	1.0s	Adaptive (max 2.0s)
Microphysics	Thompson (8)	MP_PHYSICS = 8
PBL	LES / none (0)	BL_PBL_PHYSICS = 0 (250m resolves turbulence)
Surface layer	Revised MM5 (1)	SF_SFCLAY_PHYSICS = 1
Land surface	Noah (2)	SF_SURFACE_PHYSICS = 2 (most events)
Radiation (LW/SW)	RRTMG (4)	1-minute radiation timestep
Cumulus	None (0)	CU_PHYSICS = 0 (250m resolves convection)
Diffusion	2nd order (2)	DIFF_OPT = 2, KM_OPT = 5 (3D TKE)
Nesting ratio	4:1	d01 3km → d02 1km → d03 250m
Projection	Lambert Conformal	TRUELAT1 = 30, TRUELAT2 = 60 (most events)
Damping	Rayleigh (3)	W-damping enabled

Minor variations exist between events (see wrf_config.json for per-event details):

• Land surface model: Noah (2) for most events, thermal diffusion (1) for Michael, Nashville, Buffalo Blizzard
• Camp Fire uses a slightly larger adaptive timestep (1.975s)
• Buffalo Blizzard has 81 vertical levels instead of 80

License

CC-BY-4.0 — free to use for any purpose with attribution.

Citation

@dataset{fahrenheit_wrf_overlays_2026,
  author = {Fahrenheit Research},
  title = {WRF 250m Severe Weather Overlays},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/deepguess/wrf-250m-severe-weather-overlays}
}