Repository navigation
Description
With Rothermel, RothermelAndrews2018 and Farsite, the width of a wind-driven fire changes about 2× with spatialIncrement, while the head does not move. The Balbi models don't have this problem.
Flat uniform fuel, 5 m/s wind, from a point, head run ~2.5 km, perimeterResolution=30, relax=1 (script below, run from the repo root):
| model | width at spatialIncrement 10 / 3 / 1 m | ratio |
|---|---|---|
| Farsite | 915 / 610 / 407 | 2.24 |
| Rothermel | 1064 / 681 / 448 | 2.38 |
| RothermelAndrews2018 | 817 / 581 / 399 | 2.05 |
| BalbiNov2011 | 5867 / 5892 / 5898 | 0.99 |
| Balbi2015 | 1027 / 975 / 946 | 1.09 |
| Balbi2020 | 379 / 352 / 339 | 1.12 |
- In the Rothermel family, a flank gets no wind (
getNormalWindis wind · normal), so its rate is the no-wind rate, a few % of the head's. The scheme does not resolve such a peaked rate: the width keeps shrinking asspatialIncrement / perimeterResolutiongoes to 0 (Farsite at 6.5 m/s: L/B 2.7 at 10/30, 8.6 at 0.3/30; the same ratio gives the same width at 100, 60, 30 and 10 m). - The converged answer is not realistic either: about 30 for the per-normal model, against 4.7 for Farsite's own ellipse at that wind. At the usual 3/30 settings, the discretisation error widens fires to L/B ≈ 3 at any wind speed, which looks plausible.
So fixing the numerics alone would make these fires 3–10× narrower than now. The flank needs a rate from the model first, for example the normal speed of Farsite's ellipse (Richards 1990), and then a test that the width no longer depends on spatialIncrement. Would you take a PR along those lines for Farsite, or is there a reason the Rothermel models should stay per-normal?
Found while fitting a real case (Gonfaron, Var, 2021), whose observed L/B (~3) matched only because of the 3/30 ratio.
Side note: normalScheme=splines stalls the front (~200 m in 1 h at any wind).
repro_flank_width_models.py
"""Does the fire's width depend on spatialIncrement / perimeterResolution for every propagation model?
Flat, uniform fuel (the Mediterranean shrub row of tests/python/test_moisture_invariants.py, and
SH5 for Farsite), steady wind along x, from a point. For each model the duration is set so the head
travels about 3 km; then L and B are measured at three spatialIncrement values, perimeterResolution 30.
"""
import os, re, sys
import numpy as np, pyforefire as pyff
sys.path.insert(0, "tests/python") # run from the repository root
from test_moisture_invariants import balbi_table, andrews_table, FUEL_INDEX
L, N, WIND = 16000.0, 160, 5.0
MODELS = {
"Farsite": ("STDfarsiteFuelsTable", 145),
"Rothermel": (balbi_table(0.06), FUEL_INDEX),
"RothermelAndrews2018": (andrews_table(0.06), FUEL_INDEX),
"BalbiNov2011": (balbi_table(0.06), FUEL_INDEX),
"Balbi2015": (balbi_table(0.06), FUEL_INDEX),
"Balbi2020": (balbi_table(0.06), FUEL_INDEX),
}
def run(model, t, si, pr=30.0):
table, idx = MODELS[model]
with pyff.ForeFireRunner() as ff:
ff["propagationModel"] = model; ff["fuelsTable"] = table
ff["minSpeed"] = 0; ff["relax"] = 1.0; ff["spatialIncrement"] = si; ff["perimeterResolution"] = pr
ff["windReductionFactor"] = 1.0; ff["propagationSpeedAdjustmentFactor"] = 1.0
ff.execute(f"FireDomain[sw=(0,0,0);ne=({L},{L},0);t=0]")
ff.addLayer("propagation", model, "propagationModel")
ff.addIndexLayer("table", "fuel", 0, 0, 0, L, L, 0, np.full((1, 1, N, N), idx, np.int32))
ff.addScalarLayer("data", "altitude", 0, 0, 0, L, L, 0, np.zeros((1, 1, 2, 2)))
ff.addScalarLayer("data", "windU", 0, 0, 0, L, L, 0, np.full((1, 1, 2, 2), WIND))
ff.addScalarLayer("data", "windV", 0, 0, 0, L, L, 0, np.zeros((1, 1, 2, 2)))
ff.addScalarLayer("data", "temperature", 0, 0, 0, L, L, 0, np.full((1, 1, 2, 2), 300.0))
ff.addScalarLayer("data", "moisture", 0, 0, 0, L, L, 0, np.full((1, 1, 2, 2), 1.0))
for name, m in (("ones", .05), ("tens", .06), ("hundreds", .07), ("liveh", .3), ("livew", .75)):
ff[f"moistures.{name}"] = m
x0 = 3000.0
ff.execute(f"startFire[loc=({x0},{L/2},0);t=0]")
ff.execute(f"goTo[t={t}]")
xy = np.array([[float(c) for c in m.split(",")[:2]] for m in re.findall(r"loc=\(([^)]*)\)", ff.execute("print[]"))])
return xy[:, 0].max() - x0, np.ptp(xy[:, 0]), np.ptp(xy[:, 1])
print(f"wind {WIND} m/s along x, perimeterResolution 30, relax 1")
print(f"{'model':22s} {'t (s)':>6s} | " + " | ".join(f"si {si:<4} L / B / L/B" for si in (10, 3, 1)) + " | B(si 10)/B(si 1)")
for model in MODELS:
head, _, _ = run(model, 120.0, 3.0)
t = float(np.clip(120.0 * 3000.0 / max(head, 1.0), 60.0, 4 * 3600.0))
cells, bs = [], []
for si in (10, 3, 1):
_, l, b = run(model, t, si)
cells.append(f"{l:5.0f} / {b:5.0f} / {l / max(b, 1):4.1f}"); bs.append(b)
print(f"{model:22s} {t:6.0f} | " + " | ".join(cells) + f" | {bs[0] / max(bs[2], 1):4.2f}", flush=True)Drafted with Claude Opus 5.5.
Activity