India · Built & Lit

Why alternative data?

Many questions in Indian public policy — economic activity, investment, urbanisation, the functioning of public services — need granular, timely data. India's official statistical system, hampered by limited state capability, releases statistics that are coarse (state-level at best), slow (long lags), and narrow in scope (whole domains such as physical capital stock simply aren't covered).

Satellites help fill these gaps today. The two layers below give a district-level read on the Indian economy from space — one capturing activity in motion, the other capturing capital that has accumulated.

Building volume stock

Satellite imagery and deep learning together yield the 3-D structure of the built environment: footprints, heights, and the volume of every building, every year. We use Google's Open Buildings 2.5D Temporal — Sentinel-2 inputs, ~4 m effective resolution, annual snapshots 2016–2023.

Building volume measures the physical capital stock of a district — the accumulated investment in factories, housing, and urban infrastructure. Unlike flows that fluctuate from month to month, the stock grows monotonically as the place builds out.

Note: the building-volume series is the raw Open Buildings 2.5D output — it has not been put through a cleaning pipeline. The 2022 snapshot in particular looks problematic and should be treated with caution.

Nighttime lights flow

Satellites photograph the Earth at night, and more economic activity goes with more light ( Henderson, Storeygard & Weil, 2012). We use the NOAA VIIRS DNB monthly composite — 500 m pixels, global, since 2014; cleaned with the PSTT2021 pipeline via NighttimeLights.jl.

NTL captures the flow of economic activity — factories running, vehicles moving, homes lit. When activity picks up or slows, the lights move with it. Prosperous places are brighter, and as a place grows more prosperous, it gets brighter.

Note: the shift to LED lighting can lower the radiance a satellite picks up even as activity rises — LEDs emit less in VIIRS's sensitive band. Falling NTL in a district isn't necessarily falling activity.

Together, the two series let you see how activity and capital co-move across India's districts.

Loading data…

Reproduce / adapt

The entire pipeline runs in two notebooks on Google Colab — no setup, no installs. All you need is a Google account (for Colab + Drive) and a free Google Earth Engine project. Click a button, follow the cells, get the same CSVs the dashboard above is built from.

Building volume

Annual built-up volume per district from Google Open Buildings 2.5D Temporal. Pure Python, single Colab kernel.

▶ Open in Colab

Nighttime lights

Monthly VIIRS NTL per district, cleaned with NighttimeLights.jl. Python (download) then Julia (clean), bridged by Drive.

▶ Open in Colab

Want this for your country instead — or at a different administrative level, say cities rather than districts? Swap the boundary GeoJSON URL at the top of each notebook with one for your geography. Everything downstream — the GEE reductions, the cleaning, the per-region aggregation — just works.