Skip to contents

Add the configured derived covariates to the covariate grid

Source: R/derivoce.R
add_derivoce_covariates.Rd

Runs each covariates.derivoce step in order, passing its fields straight through to the derivoce function it names. Steps see the columns the earlier ones produced, so they chain: a current_speed step followed by a horizontal_gradient step on speed reproduces the original pipeline's uv_grad.

Usage

add_derivoce_covariates(env_dat, config, bathy = NULL)

Arguments

env_dat

covariate data from fetch_covariates()

config

a config list, as returned by load_config()

bathy

optional static bathymetry SpatRaster from datamatch::fetch_bathymetry(), needed only by steps reading a depth column

Value

env_dat with one column per derived covariate added

Details

This runs on the covariate grid, before stations are matched to it, because a gradient or a front is a property of the field and cannot be recovered from scattered station points. Everything downstream then treats the results as ordinary covariate columns: attach_covariates() matches them to stations, covariate_grid() carries them onto the projection grid, and they are picked up as predictors automatically. Contrast add_derived_covariates(), which adds jday to a table of observations and needs one date per row rather than a grid.

Three consequences are worth stating plainly, since each one costs data:

  • Lags, integrals, and temporal gradients are undefined for the first time step(s) of the record, so stations in those months are dropped when covariates are matched, and the earliest month's projection has nothing to predict on.

  • A step reading a point's surroundings is undefined on the edge of the study area, where the point has no complete neighbourhood. That border is lost from both the training stations and the projected maps, so a study area drawn to just contain the stations will lose the outermost ones.

  • A neighbourhood step computed from a variable that was upsampled from a coarser product measures the source grid rather than the ocean. Those steps warn when their input is one of the upsampled variables recorded by fetch_covariates().

See also

derivoce_covariates() for the step types and the columns they produce

Examples

if (FALSE) { # \dontrun{
env_dat <- fetch_covariates(config)
env_dat <- add_derivoce_covariates(env_dat, config)
} # }