BalbiNov2011 rate of spread rises with live fuel moisture, at the shipped fuel values

[HugoFara]

BalbiNov2011 computes xsi = (Ml - Md)·(Sd/Sl)·(Deltah/DeltaH), the share of combustion energy spent vaporising the moisture the live fuel carries in excess of the dead fuel. For fuel 1 of tests/runff/fuels.csv, xsi crosses 1 near Ml ≈ 0.8. Past that, (1 - xsi) is negative, so flame temperature T drops below ambient — and R00 ∝ T⁴ turns that back into a positive, growing number.

Measured on dev, standard fuel, still air, flat ground, Md = 0.1:

Ml	ROS (m/s)
0.1	0.0227
0.5	9.06e-4	falling correctly
0.8	2.34e-8	minimum
1.0	8.16e-5	shipped default
1.5	0.0175	rising
2.0	0.1596	7× the bone-dry rate

Soaking-wet live fuel spreads seven times faster than bone-dry live fuel, and fuels.csv ships Ml = 1.0, inside the broken region — this is the default, not an edge case.

Isolated to this one model

Same conditions, whole family:

Model	Ml=0.1	Ml=1.0	Ml=2.0	Monotonic?
Balbi2015	0.02245	0.004263	0.0001431	yes
BalbiNov2011Curv	0.02245	0.004264	0.0001431	yes
BalbiUnsteady	0.02245	0.004263	0.0001431	yes
BalbiNov2011	0.02273	8.161e-05	0.1596	NO — inverts
Balbi2020	0.01428	0.01428	0.01428	does not respond to Ml
BalbiNov2011TMdMl	0.02273	0.02273	0.02273	does not respond to Ml
Rothermel	0.02931	0.02931	0.02931	does not respond to Ml

The three siblings that do respond to live moisture agree to four significant figures across the whole range. BalbiNov2011 is the only one that inverts, and the difference is one coefficient: it uses Sd/Sl where the monotonic Balbi2015 uses Sigmal/Sigmad. That is the first thing to check against Santoni et al. (2011).

Separately worth a look: BalbiNov2011TMdMl is named for Md/Ml yet does not move with Ml at all here.

Relationship to #158

That PR clamps xsi at zero, fixing the other end — dead fuel wetter than live. It deliberately does not clamp at 1, because that would change results at the default fuel values, moving pinned rates of spread and possibly runff output. That is a physics decision, which is why it is here as its own issue. The numbers above are identical before and after #158.

Options: clamp xsi at 1; revisit the Sd/Sl coefficient; or establish that Ml = 1.0 is out of range for this model and validate inputs against that. Related to the closed #35 and #99.

---

Drafted by Claude Opus 5 from a codebase audit. Reviewed by a maintainer before filing.

EDIT: rewrote for human readability.