Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions NEWS.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
# fireSense_spreadPredict (development version)

* A ledger row fitted with the split climate covariate (`fireSense_dataPrepFit`'s `spreadClimateSplit`: parameters `<var>_yr` and `<var>_dev`) is predicted with the same two terms: `spreadProbOneELF()` makes them from the year's climate with `fireSenseUtils::splitSpreadClimate()`, the median being over the ELF's own burnable pixels (`elfPixels`: every row with one fitted ELF, the ELF's own pixels of `rasterToMatchLargeELF` with several, not the blend zone), and only its flammable pixels (`fireSenseUtils::climateSplitRefPixels()`, the choice the fit makes). `ELFblendWeights()` reads the cells' ELF labels with `fireSenseUtils::elfCellLabels()`. A row without split terms predicts as before. Needs fireSenseUtils with `splitSpreadClimate()` (Depends-on PredictiveEcology/fireSenseUtils); the `reqdPkgs` floor is set after that merges.

- With `paramSets`: the cache of a parameter set's surface no longer reads all the values of `flammableRTM` and `rasterToMatchLargeELF` every replicate (a new raster object or a change at 1000 sampled cells is a change) and returns the stored surface without copying it (a changed stored surface is recomputed). On a 20-million-cell raster a replicate that reuses a surface takes 0.35 s instead of 0.9 s, most of it SpaDES. `paramSets` stops if the ledger rows have the simulated ELF twice, or if a file of ledger rows has none for it.

- New parameters `paramSets` (path to an `.rds` or `.csv` of spread parameter vectors for the simulated ELF; default `NA`, behaviour unchanged) and `repsPerSet`. The sets replace the simulated ELF's ledger rows (neighbour ELFs keep theirs); replicate r uses set `ceiling(r / repsPerSet)`, calls `set.seed(k)` with k its number within the set (common random numbers across sets), and the surface is recomputed only when the set changes. The link, lower asymptote, `gamma` and covariate centres come from the ledger row; a set lacking a coefficient of the row stops. The set index is output in `fireSense_SpreadParamSet`. New input `.ELFind`, as in `fireSense_dataPrepPredict`.
Expand Down
23 changes: 17 additions & 6 deletions fireSense_spreadPredict.R
Original file line number Diff line number Diff line change
Expand Up @@ -244,7 +244,7 @@ spreadPredictRun <- function(sim) {
if (NROW(sa) == 1L) {
params1 <- paramSetForRep(sa$params[[1]], rep)
lk1 <- ledgerLinkArgs(sa, 1L)
pred <- do.call(spreadProbOneELF, c(list(covs, params = params1, covMinMax = ledgerCovMinMax(sa, 1L, sim$covMinMax_spread),
pred <- do.call(spreadProbOneELF, c(list(covs, params = params1, flammableRTM = sim$flammableRTM, covMinMax = ledgerCovMinMax(sa, 1L, sim$covMinMax_spread),
yr = time(sim),
maxFireSpread = P(sim)$maxFireSpread, lowerSpreadProb = P(sim)$lowerSpreadProb,
covCentre = ledgerCovCentre(sa, 1L), climateRef = ledgerClimateRef(sa, 1L)), lk1))
Expand Down Expand Up @@ -279,8 +279,10 @@ spreadPredictRun <- function(sim) {
if (!length(these)) next
parsI <- paramSetForRep(sa$params[[i]], rep)
p <- do.call(spreadProbOneELF, c(list(covs[these], params = parsI, covMinMax = sa$covMinMax_spread[[i]],
yr = time(sim), maxFireSpread = P(sim)$maxFireSpread,
yr = time(sim), maxFireSpread = P(sim)$maxFireSpread, flammableRTM = sim$flammableRTM,
lowerSpreadProb = P(sim)$lowerSpreadProb,
elfPixels = if (any(fireSenseUtils::isClimateSplitCol(names(parsI))))
which(fireSenseUtils::elfCellLabels(sim$rasterToMatchLargeELF[[1]]) %in% ids[i]),
covCentre = ledgerCovCentre(sa, i), climateRef = ledgerClimateRef(sa, i)), lks[[i]]))
rows <- these[match(p$pixelID, covs$pixelID[these])]
acc[rows] <- acc[rows] + w[rows, i] * p$spreadProb
Expand Down Expand Up @@ -443,9 +445,7 @@ ELFblendWeights <- function(elfRas, template, pixelID, ids, width) {
stop("fireSense_spreadPredict: sim$rasterToMatchLargeELF is not on the grid of sim$flammableRTM",
call. = FALSE)
r <- elfRas[[1]]
v <- terra::values(r, mat = FALSE)
lv <- terra::levels(r)[[1]]
lab <- if (is.data.frame(lv) && NCOL(lv) >= 2) as.character(lv[[2]][match(v, lv[[1]])]) else as.character(v)
lab <- fireSenseUtils::elfCellLabels(r)
w <- vapply(ids, function(id) {
core <- r; terra::values(core) <- ifelse(lab %in% id, 1, NA)
if (all(is.na(terra::values(core, mat = FALSE)))) return(numeric(length(pixelID)))
Expand All @@ -467,6 +467,10 @@ ELFblendWeights <- function(elfRas, template, pixelID, ids, width) {
#' @param params one-row `data.frame`: the ELF's parameter set for this replicate (`paramSetForRep()`).
#' @param covMinMax the ELF's `covMinMax_spread` (`ledgerCovMinMax()`): the ranges the fit rescaled its
#' covariates with. Every fitted covariate must be a column of it, or it is left unscaled.
#' @param elfPixels integer; the `pixelID`s of the ELF's own burnable pixels, the pixels the year's median of a split
#' climate covariate (`<var>_yr`, `<var>_dev`; `fireSenseUtils::splitSpreadClimate()`) is taken over, as in the fit.
#' `NULL` (one fitted ELF) is every row of `covs` that is flammable. Not used by a fit without split terms.
#' @param flammableRTM `sim$flammableRTM`; with `elfPixels` it limits the median to the ELF's flammable pixels (`fireSenseUtils::climateSplitRefPixels()`, the fit's choice).
#' @param yr,maxFireSpread,lowerSpreadProb as for `fireSenseUtils::spreadProbFromIntegerCovs()` and
#' `fireSenseUtils::logisticAll()`.
#' @param covCentre the ELF's `covCentre_spread` (`ledgerCovCentre()`): the covariate means the fit subtracted
Expand All @@ -490,7 +494,7 @@ ELFblendWeights <- function(elfRas, template, pixelID, ids, width) {
#' `ledgerRichardsNu()`): with `"powerLawSmooth"` and `"richards"`, the lower asymptote, and `nu` of `"richards"`.
#' @return `data.table` with `pixelID` and `spreadProb`, from the one parameter set in `params`; with a `link`
#' also `linearPredictor`, the `x` that `spreadProb` is the link of.
spreadProbOneELF <- function(covs, params, covMinMax, yr, maxFireSpread, lowerSpreadProb,
spreadProbOneELF <- function(covs, params, covMinMax, yr, maxFireSpread, lowerSpreadProb, elfPixels = NULL, flammableRTM = NULL,
covCentre = NULL, climateRef = NULL, powerLawGamma = NULL,
powerLawShape = fireSenseUtils::spreadPowerLawShape,
link = if (!is.null(powerLawGamma)) "powerLaw",
Expand Down Expand Up @@ -522,6 +526,13 @@ spreadProbOneELF <- function(covs, params, covMinMax, yr, maxFireSpread, lowerSp
## the intercept, if the fit has one, is a coefficient with no covariate: it multiplies a column of 1s
hasIntercept <- fireSenseUtils::spreadInterceptTxt %in% names(params)
fitted <- setdiff(names(params), c(unlist(fireSenseUtils::logisticParamNames), fireSenseUtils::spreadInterceptTxt))
## a fit made with fireSense_dataPrepFit's spreadClimateSplit has the climate as `<var>_yr` and `<var>_dev`: the
## year's median of the variable over this ELF's burnable pixels, and the departure from it, made as the fit made them
splitVars <- unique(fireSenseUtils::climateSplitBase(fitted[fireSenseUtils::isClimateSplitCol(fitted)]))
if (length(splitVars) && all(splitVars %in% names(covs))) {
ref <- covs[covs$pixelID %in% fireSenseUtils::climateSplitRefPixels(covs$pixelID, flammableRTM, elfPixels), splitVars, with = FALSE]
fireSenseUtils::splitSpreadClimate(covs, splitVars, ref)
}
noCov <- setdiff(fitted, names(covs))
if (length(noCov)) {
stop(
Expand Down
78 changes: 78 additions & 0 deletions tests/testthat/test-climateSplit.R
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
## A fit made with fireSense_dataPrepFit's spreadClimateSplit has the climate as two terms, `MDC_yr` (the year's
## median over the ELF's burnable pixels) and `MDC_dev` (MDC minus it). Prediction builds them from the year's
## climate with fireSenseUtils::splitSpreadClimate(), as the fit did, over the ELF's own pixels.
##
## Rescaling: MDC_yr with c(0, 200), MDC_dev with c(-100, 100) (fireSenseUtils::climateSplitRanges()).

splitParams <- function(yr = 1.5, dev = 0.8, fuel = 1, name = "fuelA") {
d <- data.frame(maxAsymptote = 0.25, hillSlope1 = 2, inflectionPoint1 = 1, MDC_yr = yr, MDC_dev = dev, x = fuel)
names(d)[6] <- name
d
}
cmmSplit <- function(fuel = "fuelA") {
d <- data.table::data.table(MDC_yr = c(0, 200), MDC_dev = c(-100, 100), f = c(0, 8))
data.table::setnames(d, "f", fuel)
d
}

test_that("a split fit predicts from the year's median over the burnable pixels and each pixel's departure from it", {
covs <- toyCovariates()
ins <- toyInputs(covs = covs, params = splitParams(), formula = "~ MDC_yr + MDC_dev + fuelA - 1")
ins$covMinMax_spread <- cmmSplit()
v <- predVals(toyRun(ins))
o <- order(covs$pixelID)
mdc <- covs$MDC[o]; fuel <- covs$fuelA[o]
med <- stats::median(covs$MDC) # 100: over the 7 burnable pixels of the covariate table
x <- 1.5 * (med / 200) + 0.8 * ((mdc - med + 100) / 200) + 1 * (fuel / 8)
expect_equal(v[cells <- c(1, 2, 3, 4, 6, 7, 8)], handLogistic3(x), tolerance = 1e-7)
})

test_that("a non-flammable row in the covariate table is left out of the median, as the fit's burnable pixels are", {
covs <- toyCovariates()
extra <- data.table::data.table(pixelID = 5L, MDC = 5000, youngAge = 0, fuelA = 0) # cell 5 of the toy flammableRTM is 0
withExtra <- data.table::rbindlist(list(covs, extra))
run <- function(cv) {
ins <- toyInputs(covs = cv, params = splitParams(), formula = "~ MDC_yr + MDC_dev + fuelA - 1")
ins$covMinMax_spread <- cmmSplit()
predVals(toyRun(ins))
}
a <- run(covs); b <- run(withExtra)
cellsB <- c(1, 2, 3, 4, 6, 7, 8)
expect_equal(b[cellsB], a[cellsB], tolerance = 1e-12) # median 100 with and without the extra row
})

test_that("the between term is the same for every pixel: only the within term moves with the pixel", {
covs <- toyCovariates()
both <- function(yr, dev) {
ins <- toyInputs(covs = data.table::copy(covs), params = splitParams(yr, dev, fuel = 0), formula = "~ MDC_yr + MDC_dev + fuelA - 1")
ins$covMinMax_spread <- cmmSplit()
predVals(toyRun(ins))
}
yrOnly <- both(1.5, 0)
expect_equal(yrOnly[c(1, 2, 3, 4, 6, 7, 8)], rep(handLogistic3(1.5 * 0.5), 7), tolerance = 1e-7) # median 100 / 200 everywhere
devOnly <- both(0, 0.8)
expect_gt(length(unique(round(devOnly[c(1, 2, 3, 4, 6, 7, 8)], 8))), 1L)
})

test_that("with several ELFs each takes the median over its own pixels", {
ins <- multiInputs()
covs <- data.table::data.table(pixelID = 1:10, MDC = c(10, 20, 30, 40, 50, 100, 110, 120, 130, 140), fuelA = 2, fuelB = 3)
ins$fireSense_SpreadCovariates <- covs
d <- function(fuel) { x <- data.frame(maxAsymptote = 0.25, hillSlope1 = 2, inflectionPoint1 = 1, MDC_yr = 1.5, MDC_dev = 0.8, f = 1)
names(x)[6] <- fuel; x }
ins$studyAreaWithSpreadParams$params <- list(d("fuelA"), d("fuelB"))
ins$studyAreaWithSpreadParams$covMinMax_spread <- list(cmmSplit("fuelA"), cmmSplit("fuelB"))
ins$fireSense_spreadFormula <- "~ MDC_yr + MDC_dev + fuelA - 1"
v <- predVals(toyRun(ins, params = list(ELFblendWidth = 20000)))
## pixel 1 is only A's (25 km from B): A's median of pixels 1:5 is 30
xA <- 1.5 * (30 / 200) + 0.8 * ((10 - 30 + 100) / 200) + 1 * (2 / 8)
expect_equal(v[1], handLogistic3(xA), tolerance = 1e-7)
## pixel 10 is only B's: B's median of pixels 6:10 is 120
xB <- 1.5 * (120 / 200) + 0.8 * ((140 - 120 + 100) / 200) + 1 * (3 / 8)
expect_equal(v[10], handLogistic3(xB, upper = 0.25), tolerance = 1e-7)
})

test_that("a fit without split terms predicts as before: no median is taken", {
expect_identical(predVals(toyRun()), predVals(toyRun(toyInputs())))
expect_false(any(fireSenseUtils::isClimateSplitCol(names(toyParams()))))
})
Loading