25 · Spatial II: rasters

Thursday, Nov 19, 2026

Materials for this session are not published yet. They appear here before class.

Objectives

By the end of this session you can:

• Do raster algebra and zonal extraction with terra and exactextractr.
• Defend a weighting choice: population vs area.
• Build the sparse-weight-matrix pipeline for fast zonal statistics, and benchmark it.
• Hand-check one unit end to end.

What we cover

• The workhorse of climate economics: gridded weather aggregated onto counties.
• The personal opener: geocode a place you care about — home town, birthplace — find its grid cell, and pull its century of temperatures from the stack.
• A raster is a georeferenced matrix: matrices from session 2 return for ten minutes — create, index, multiply — because extraction is just subsetting one, and the fast method below is multiplying one.
• Alignment, raster algebra, zonal statistics.
• Fast zonal at scale: precompute the cell-to-polygon weights once as a sparse matrix, and any new layer’s zonal statistics become one sparse multiply — the trick behind 40-year daily county panels.
• Benchmarked against the canned function, not asserted — and the efficiency kit for jobs this size: system.time() from week two returns, mclapply() for the embarrassingly parallel step, fread() for the giant file.

Verification habit. Hand-check one unit; weighted-vs-unweighted differs in a predictable direction — confirm it; benchmark claims are measured, not asserted.