21 · Satellite II: Earth Engine

Thursday, Nov 5, 2026

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

Objectives

By the end of this session you can:

  • Say when to move the computation to the data instead of the data to your machine.
  • Find a dataset in the catalog and read its documentation.
  • Filter an image collection by date, place, and cloud cover.
  • Reduce a collection over your own polygons and export a table.
  • Bring that table back into R and treat it as any other data frame.
  • Say when Earth Engine is the wrong tool.

What we cover

  • The inversion. Every session so far has fetched data and computed locally. Some questions make that impossible: a decade of a continent is more pixels than your disk holds. You send the computation instead, and only the answer comes back. The same idea runs BigQuery and cloud-hosted archives, so it outlives this one tool.
  • The catalog: Landsat, Sentinel, MODIS, and the derived products (land cover, night lights, gridded weather). What each documentation page tells you, and what it does not.
  • Filtering a collection: date range, bounding box, cloud cover. How many images survive, which is the count check.
  • Server-side thinking, and the mistake everyone makes first. A loop that works in R hangs in Earth Engine, because the work happens on Google’s machines and only what you ask for comes back. Map and reduce instead of iterate.
  • The payoff: reduceRegions(). Zonal statistics over your own polygons, at continental scale. Mean rainfall per county, night lights per district, greenness per field boundary. This is the shape most economics questions actually want.
  • Export to a table, load it in R, and the rest of the semester’s workflow applies unchanged.
  • When not to use it. Small areas, one scene, or anything you need byte-reproducible: the archive is versioned but the service is not yours, so a result you cannot re-run is a result you cannot defend. Cache the exported table and commit it.
  • Access is free for research, and the account setup takes a few days, so we start it in the session before.

Verification habit. Reduce one polygon you can check by hand, and compare it with the same polygon computed locally.