prediction_masks: generate masks from samples

In obrl-soil/dsmartr: dsmartr: An R implementation of the DSMART algorithm

generate masks from samples

Description

Generates a rasterlayer for each dsmart iteration highlighting areas where one or more covariates has a value that is out of range of the sample used to model soil distribution.

Usage

prediction_masks(samples = NULL, covariates = NULL, tolerance = 0L, cpus = 1)

Arguments

samples	List; POINT_sfc objects created by dsmartr::iterate() when option write_samples = TRUE
covariates	RasterStack or RasterBrick; environmental covariate datasets.
tolerance	Integer; the number of out-of-range covariates allowed. Defaults to 0, that is, if any covariate is out of range on a given pixel, that pixel will be masked.
cpus	Integer; number of processors to use in parallel.

Details

For each model iteration this function generates a raster that can be used to mask dsmartr predictions where covariate values are out of range of the sample of cells used in that iteration. The masks can also be combined to produce an overall mask, applicable to final products. This is experimental; the idea is to avoid making predictions in locations where the model won't perform well.

Value

a list of rasters

Examples

## Not run: 
# run dsmartr::iterate() with the example data, then:
sample_list <- lapply(list.files(file.path(getwd(), 'iterations', 'samples'),
pattern = '\\.gpkg', full.names = TRUE), read_sf)
masks <- prediction_masks(samples = sample_list,
                          covariates = heronvale_covariates,
                          cpus = max(1, (parallel::detectCores() - 1)))

# less strict: up to 3 out of range covariates are allowed
masks <- prediction_masks(samples = sample_list,
                          covariates = heronvale_covariates,
                          tolerance = 3,
                          cpus = max(1, (parallel::detectCores() - 1)))

# a 'final product' mask, where outputs are masked only if the mode of that
# cell is 'do not predict'.
all_masks <- stack(masks)
modal_mask <- calc(all_masks, function(cell) ifelse(raster::modal(cell) == 1,
NA, 0))
# the above could be applied to the most-likely soils map after running
# dsmartr::most_likely:
masked_m1 <- most_likely_soil[[1]][[1]] + modal_mask

## End(Not run)

obrl-soil/dsmartr documentation built on Feb. 1, 2024, 10:57 p.m.