Bounds should be wide enough that they do not influence the fitted value, except to constrain sign. Covariate coefficients get +/- upperAndLower, except drought-index terms (name matches droughtIndexPattern, e.g. CMD, CMD_sm, cumMDC – lower bound 0, since drought should not increase spread probability), fuel biomass terms (fuelTerms: lower bound 0, upper upperAndLowerFuel) and youngAge (upper bound 0, to prevent self-propagating fires). Fuel is bounded below by 0 because, with the intercept and centred covariates, fits put the intercept near the spread-probability ceiling and gave fuel biomass negative coefficients, so low-biomass (recently burned) stands were the most flammable. That positive feedback produced runaway fires 3-6 times the observed area in NRV runs (ELFs 5.4 and 14.3). With fuel >= 0, more fuel can only raise spread probability and the intercept must come down. Every other term, including any other annual covariate and any interaction (a:b, whatever its covariates' bounds are), is symmetric. The two remaining logistic parameters get fixed bounds. hillSlope1, the logistic slope, and inflectionPoint1, its Richards exponent, are not among them: both are fixed at 1, not fitted. hillSlope1 is not identifiable together with the covariate coefficients – with the linear predictor x = covariates %*% beta, it enters the link only as hillSlope1 * x, so scaling every coefficient by k and dividing hillSlope1 by k leaves every prediction unchanged. inflectionPoint1 is not a location; its fitted values were bimodal and spanned the bounds at no cost in the objective, and the coefficients explained half of its variance (fireSenseUtils::fixedLogisticPars; fixLogisticPars() reinserts both before the objective evaluates par).

estimateSpreadParams(
  fireSense_spreadFormula,
  anyAnnualCovariates,
  whichBound,
  upperAndLower,
  fuelTerms = character(),
  upperAndLowerFuel = upperAndLower,
  upperTailBounds = NULL,
  fireSpreadSDBounds = NULL,
  yearSpreadSDBounds = NULL,
  maxAsymptoteBounds = maxAsymptoteBoundsOf()
)

Arguments

fireSense_spreadFormula

character; the spread formula.

anyAnnualCovariates

list of annual covariate data.tables; kept for call-site compatibility but no longer used – bound sign is now decided by term name, not by annual table membership.

whichBound

"upper" or "lower".

upperAndLower

numeric; absolute bound for covariate coefficients.

fuelTerms, upperAndLowerFuel

character names of the fuel biomass terms, and their upper bound (lower is 0).

upperTailBounds

numeric; if not NULL, the bounds of upperTail1, which is added after maxAsymptote (link "logistic3pUpper").

fireSpreadSDBounds, yearSpreadSDBounds

numeric(2) or NULL; the bounds of the per-fire and per-year random effects' sds, added last, fireSpreadSD before yearSpreadSD (randomEffectBounds()). NULL leaves one out.

maxAsymptoteBounds

numeric(2); the bounds of maxAsymptote (maxAsymptoteBoundsOf()).

Value

named numeric vector: maxAsymptote (bounded by maxAsymptoteBounds, by default fireSenseUtils::spreadProbFloor and fireSenseUtils::spreadProbCeiling), upperTail1 (with upperTailBounds), then fireSenseUtils::spreadInterceptTxt if the formula has an intercept (~ 1 + ...), then formula terms. No hillSlope1 or inflectionPoint1 – both are fixed at 1, not fitted.