A Copernicus fetch over a decade is millions of grid points, which is why
run_taupatch() summarises the covariates rather than returning them. But a
summary cannot be mapped, and a map is the fastest way to see that a covariate
is wrong — a field of zeros, a land mask in the wrong place, a month that
failed to download.
Arguments
- env_dat
covariate data from
fetch_covariates()- max_cells
the most rows to keep across all time steps
Details
This keeps a regular subsample: every nth location, the same ones in every
time step, chosen so the whole object stays under max_cells. A map drawn
from it is coarser than the data and shows the same thing, which is what it is
for. Nothing modelled is thinned; this is a copy kept for looking at.
Locations rather than rows, because dropping rows at random would give a different set of points each month and a map that flickers between them.