crossvalid: Cross validation routine for (species distribution)...
In janhoo/crecs: quantitative species distribution modelling

Description Usage Arguments Examples

View source: R/func.model.R

repeated k-fold cross-validation. Calculate model preformance metrics, disribution, standard deviation and occurence maps.

crossvalid(Ncv, Kfold, data, method, responsetype, response, predictors,
  secondary = NULL, strata = NULL, enviStack = NULL, enviPix = NULL,
  seed, aggregated = NULL, pseudoabsence = NULL, gbm.trees = 2000,
  maxargs = c("outputformat=logistic", "defaultprevalence=0.5"),
  rast = TRUE, flat = FALSE, ...)

`Ncv`	Number of cv repititions.
`Kfold`	Number of folds. Usually take 10 or 5.
`data`	`SpatialPointsDataFrame` containing response and predictors
`method`	SDM methos used: "gam", "rf", "gbm", "max", or "gbm.step". See details for details
`responsetype`	Type of the response: "count", "continuous", or "presence". Presence denotes bimodal responses [0\|1]
`response`	Column name of response in data argument
`predictors`	Column names of predictors to use in data argument
`secondary`	Column name in data. Use this to calculate model performance metrics from instead of response. Default is NULL.
`strata`	criterion for fold generation. #FIXME
`enviStack`	`RasterStack` of predictors. Used to calculate SD map
`enviPix`	`SpatialPixelsDataFrame` of predictors. enviPix<-as(enviStack,"SpatialPixelsDataFrame"). Only for performance.
`seed`	Integer. You probably want reproduceability. Note that Maxent's pseudoabsence generation can't be seeded this way–so expect those results to vary
`aggregated`	logical or `NULL` Default is `NULL`. This is for ensemble calculations and added to metrics in case of not being `NULL`
`pseudoabsence`	logical or `NULL` Default is `NULL`. Same as above
`gbm.trees`	gbm.trees param for dismo::gbm
`maxargs`	argument to pass tp maxent
`rast`	return `list` of metric and raster
`flat`	return model performance metrics only (as `data.frame`)
`...`	ellipsis is used to pass arguments to subsequent functions like `threshold.def` (see `metrics`) or `moranii()` & `makefn.free.model()` (moParams=c("response","residuals"),moRange=list(c(0,0.5),c(0.5,1)),check.names=TRUE)

data<-get.environ(species,deutschebucht)
cv<-crossvalid(Ncv=1,Kfold=5,data=data,method="rf",responsetype = "continuous", response = "species1", predictors = c("mgs","mud","depth"),enviStack = deutschebucht,seed=23, check.names = FALSE)
 
# aggregated results
cv$metric.agg

# plot result maps
par(mfrow=c(2,2))
plot(cv$rmean,main="cv prediction")
plot(cv$sd,main="standart deviation")
plot(raster(deutschebucht,layer=match("species1",names(deutschebucht))),main="true distribution")
plot(cv$rbin,main="prob. of occurrence")