Evaluate the spread objective function without DEoptim — runSpreadWithoutDEoptim • fireSense

In "debug" mode, runs fireSenseUtils::.objfunSpreadFit() once per parameter set, with plots. Otherwise calibrates the SNLL threshold: evaluates iterThresh random parameter sets, each with a random candidate threshold, in forked processes, and returns pickThreshold() of the results.

runSpreadWithoutDEoptim(
  iterThresh,
  lower,
  upper,
  fireSense_spreadFormula,
  flammableRTM,
  annualDTx1000,
  nonAnnualDTx1000,
  fireBufferedListDT,
  mutuallyExclusive = list(youngAge = "vegPC"),
  doObjFunAssertions = getOption("fireSenseUtils.assertions", TRUE),
  historicalFires,
  covMinMax,
  objfunFireReps,
  maxFireSpread,
  weighted = TRUE,
  tests = c("snll_fs", "adtest"),
  formulaToFit,
  pars = NULL,
  plot.it = TRUE,
  mode = "fit",
  seed = NULL,
  escapeSizeHa = NULL,
  sizeLik = "kde",
  sizeLikDf = 5,
  adWeight = "auto",
  link = NULL,
  jumpTries = 0,
  jumpMeanDist = 0,
  yearAreaWeight = 0,
  areaDistWeight = 0
)

Arguments

iterThresh: integer; number of random parameter sets (and candidate thresholds).
lower, upper: numeric; bounds the random parameter sets are drawn between.
fireSense_spreadFormula: character; passed to FS_formula.
flammableRTM: SpatRaster; passed to landscape.
annualDTx1000, nonAnnualDTx1000, fireBufferedListDT, historicalFires: elements of the list made by fireSenseUtils::covsX1000AndSetDF().
mutuallyExclusive: named list of mutually exclusive covariates.
doObjFunAssertions: logical; passed to doAssertions.
covMinMax: data.table of covariate min and max, or NULL for no rescaling.
objfunFireReps: integer; passed to Nreps, the replicates per fire.
maxFireSpread: numeric; upper limit on mean spread probability.
weighted: logical; weight the SNLL by log fire size. Also used in the rough threshold estimate.
tests: character; objective function tests, e.g. "SNLL_FS", "adTest".
formulaToFit: character; the spread formula.
pars: optional numeric vector, or list of them, to evaluate instead of random sets.
plot.it: passed to .objfunSpreadFit() in "debug" mode.
mode: character; if it includes "debug", the debug branch runs.
seed: integer or NULL; set.seed() value. NULL draws one at random.
escapeSizeHa: passed to .objfunSpreadFit(): fit escaped fires only (NULL: any fire over 1 pixel).
sizeLik, sizeLikDf, adWeight, link, jumpTries, jumpMeanDist, yearAreaWeight, areaDistWeight: passed to .objfunSpreadFit(), so the threshold is calibrated on the objective the fit uses. The defaults are that function's.

Value

the calibrated threshold (numeric, or NA if every trial failed); NULL in "debug" mode.