# India, district by district, from open satellite data

> geolith turns one open raster into a map and one number for every district, so a rainfall or greenness question has a public answer for all 787 of them.

## The datasets

Each one is a public raster, averaged over the Survey of India district boundaries, with the source printed on every page.

### Rainfall by district

Rain on a 0.05° grid, per district, against the normal for the same five days.

- Source: Rainfall estimates from satellite and rain-gauge data.
- Resolution: 0.05° grid (about 5.5 km)
- Updates: Every 5 days (pentads)
- Licence: CC BY 4.0
- Page: https://geolith.app/data/in/rainfall

### NDVI by district

Greenness from Copernicus, per district, against the 2019–2025 average.

- Source: Normalised Difference Vegetation Index, from Sentinel-3 OLCI.
- Resolution: 300 m pixels
- Updates: Every 10 days (dekads)
- Licence: Copernicus Land Monitoring Service data policy
- Page: https://geolith.app/data/in/ndvi

## What one district page gives

Open any district: a map, the number for this period beside its usual, and the last twelve periods to see the season move. Nothing needs a key.

- Three views of the same map: the value, the difference from usual, and the raw 300 m or 0.05° cells the number came from.
- Twelve periods on one control, so a dry spell or a late green-up is visible without leaving the page.
- An embed for any site, at no cost and with no key, and a share card that carries the sentence and the place.

## From a raster to one number per district

The engine reads the raster’s grid, lays the district boundary over it, and weights each cell by the share of it that falls inside the district. The district’s number is the area-weighted mean of the cells it contains, so a boundary that splits a cell does not lose it.

- In: one raster, such as CHIRPS rainfall, and one boundary file.
- Out: a map layer, and a table with one row of numbers per district.
- A city district and a desert district are read the same way, at the raster’s own resolution.

## Checked before it publishes

A build’s numbers are compared with GDAL's zonal statistics on the same raster and the same boundaries, for all 787 districts. The comparison is the gate: numbers that disagree are not published.

## Boundaries and district codes

Districts follow the Survey of India outline, so India is drawn whole and never cut at a claim line. The district shapes come from Overture Maps and OpenStreetMap; the India boundaries are the DataMeet India community's. Each district also carries its Local Government Directory code, so a row joins to government data without matching names. Source: Local Government Directory, Government of India.

## What you can do today

- **Read it, and embed it.** Every page and every embed is free: no key, no account, any site may frame one. https://geolith.app/data/in/rainfall
- **Files and the Datasets API.** Data files (CSV, Parquet, GeoParquet) and the Datasets API come with the Scale plan. A key is needed for both. https://geolith.apphttps://docs.geolith.app/datasets/status
- **Self-host the engine.** The engine is the geolith dataset command: it builds both outputs on one machine, from your own raster and boundaries. https://geolith.apphttps://docs.geolith.app/datasets/overview

## Sources and licences

Rainfall: CHIRPS v3.0, Climate Hazards Center, UC Santa Barbara. CC BY 4.0. DOI 10.15780/G2JQ0P
Contains modified Copernicus Land Monitoring Service information 2026
Boundaries: Overture Maps Foundation, © OpenStreetMap contributors, ODbL. India boundaries by DataMeet India community (CC BY 4.0)
Source: Local Government Directory, Government of India

## Questions

### What is a pentad, and what is a dekad?

A pentad is five days; a dekad is ten. Rainfall arrives in pentads and greenness in dekads, because that is how their sources publish: CHIRPS v3 sums rainfall over five days, and Copernicus composites vegetation over ten.

### What does preliminary mean?

The period has closed but its source has not finished revising it. Rainfall is preliminary for a few days after the five-day period ends, then final; greenness settles in about the same time. A page says which it is, and a share card marks a preliminary period.

### How often does it update?

A new rainfall period every five days and a new greenness period every ten, each a few days after the period ends while the source catches up. The page shows the day the data was built.

### Can I get the numbers as a file?

Data files (CSV, Parquet, GeoParquet) and the Datasets API come with the Scale plan. Every page shows its numbers and its source for free; the files and the API need a Scale key. The engine that makes them can also be run on your own machine.

### Which countries?

India today, for every district. The engine reads any raster against any boundary file, so another country is a build, not a rewrite.

### Why does a district say it is less green than usual?

The comparison is against the same ten days across 2019 to 2025, built from Sentinel-3 OLCI. A district can be green in absolute terms and still be below its own average for that time of year, which is what the sentence reports.
