Skip to contents

Project a fitted model to monthly habitat suitability maps

Source: R/project.R
project_patch_model.Rd

Predicts patch probability across the covariate grid for every configured month and year. One pooled model produces month-specific maps because day-of-year is a predictor and the covariates themselves are monthly — the same arrangement as original/buildZoopModel.R, without the raster//1000/manual CRS-string handling it needed to read biomod2's output back off disk.

Usage

project_patch_model(model, env_dat, config, bathy = NULL)

Arguments

model

a fitted model from fit_patch_model()

env_dat

covariate data from fetch_covariates()

config

a config list, as returned by load_config()

bathy

optional static bathymetry SpatRaster from fetch_bathymetry(), attached to every month since it does not vary in time

Value

a tibble with year, month, n_cells (predicted), n_grid (cells available), resolution in degrees, and the geotiff/png paths written. The suitability table's path is on it as a table attribute.

What it projects onto

Every cell of the covariate grid for that month, not the station locations. Stations are used to fit the model and play no part here. The grid's resolution is the one the covariates were joined onto, which covariates.grid decides and any covariates.prejoin resampling can change, so projecting finer means changing those rather than anything in this block.

Cells missing any predictor are dropped, and the count is reported per month along with the predictor most often responsible. That is usually a neighbourhood step, which is undefined on the study-area border, or a lag, which is undefined in the first month of the record.

Getting the numbers out

Each month is a GeoTIFF, which carries its own coordinates and is what a GIS wants. Alongside them goes one suitability.csv for the whole run, in long form: species, year, month, longitude, latitude, probability. That is the one to read into anything that is not a GIS, and it is written a month at a time rather than accumulated, since a decade of a real grid is tens of millions of rows. Set projection.write_csv to false to skip it.

Setting projection.write_grd to true also writes one multi-layer raster holding every month, layer names carrying the dates. See write_suitability_stack().