About this project

An independent, open-source test of whether satellite-derived features carry information about tract-level reported crime that last year's counts and census demographics do not. The honest answer so far is: mostly no.

Design

• Unit: 2020 census tracts with at least 100 residents (about 2,190), target years 2019 to 2025.
• Outcomes: felony complaints and shooting incidents from NYC Open Data, counted per tract-year.
• Baseline: log lagged counts of both outcomes, log population and income, and 2023 ACS age, poverty, renter and degree shares (held constant across years).
• Satellite features, previous year: Sentinel-2 vegetation, Landsat surface temperature, VIIRS night lights, AlphaEarth embeddings (8 components), Sentinel-1 radar backscatter and its change.
• Model: Poisson regression with a population offset and year and county effects.
• Validation: spatial-block cross-validation (whole clusters of neighbouring tracts held out) and a forward test on 2024 and 2025.

Limits

• Reported crime reflects reporting and enforcement behaviour as well as offending.
• Everything is annual and tract-level; crime concentrates at finer places and times.
• Census covariates are static (2023); demographic change over the period is not modelled.
• The forward test has two years, and shootings fell sharply in 2025.
• Results are associations, not causes.

Responsible use

Not for targeting individuals or demographic groups, or for resource allocation without independent review of reporting bias. See the model card in the repository.

Data and required notices

This product uses the Census Bureau Data API but is not endorsed or certified by the Census Bureau. Contains modified Copernicus Sentinel data 2018-2025. The AlphaEarth Foundations Satellite Embedding dataset is produced by Google and Google DeepMind. Landsat data courtesy of the U.S. Geological Survey; VIIRS Black Marble courtesy of NASA LAADS DAAC. NYC data courtesy of the NYPD, Department of City Planning and Department of Buildings through NYC Open Data, aggregated and modified here; the City does not endorse this project. Map tiles © OpenStreetMap contributors.

2025 shootings come from a separate NYC Open Data dataset than 2018 to 2024. Full source table and terms review: DATA_SOURCES.md in the repository.

Code and reproduction

All scripts, the model card and the full results write-up are in the GitHub repository. This dashboard is built with Observable Framework, MapLibre GL JS and Observable Plot from aggregated tract-year files; no incident records are published.