6 Mins Read

30 Sep 2026

What is Earth Observation Data? Types, Sources

What is Earth observation data?

Earth observation data is the imagery and measurements of the Earth’s surface, oceans, and atmosphere collected by satellites, aircraft, and drones. It includes optical and radar imagery, elevation, land cover, sea surface temperature, and atmospheric gases, delivered as georeferenced rasters with a timestamp. Open programs such as Copernicus and Landsat supply most of it free.

What Earth observation data is

Earth observation data is the output of remote sensing: any measurement of the planet made from above, processed into a georeferenced product with a known date.

The defining property is that it is systematic. A satellite images the same ground on a fixed orbit and schedule, so Earth observation data is a time series over every place, whether or not anyone asked for it.

Four families cover most of it. Optical imagery records reflected sunlight in visible and infrared bands, and radar imagery records microwave backscatter through cloud.

Thermal imagery records emitted heat. Atmospheric instruments record gases, aerosols, and clouds.

Derived products sit on top. Elevation models, land cover maps, vegetation indices, sea surface temperature, and soil moisture are all Earth observation data one processing step removed from the sensor.

Producers split into public programs, led by the European Union’s Copernicus and the United States’ Landsat and NASA missions, and commercial operators such as Planet, Maxar, ICEYE, and Airbus.

What Earth observation data does

Earth observation data replaces sampling with coverage. It measures every field, forest, coastline, and city on a schedule, and the archive shows how each one has changed.

Governments report crop area, forest loss, and urban expansion from it under climate and trade agreements.

Insurers quantify exposure before an event and damage after it, across a whole portfolio at once.

Energy companies site wind and solar plants on irradiance, terrain, and land cover, then monitor construction and vegetation encroachment.

Supply chain teams watch ports, mines, and factories for activity and disruption without a person on site.

Agencies detect wildfires within hours of ignition, map floods within a day, and track methane leaks from orbit.

Each use shares one requirement: joining the observation to the assets, boundaries, or policies it concerns, which is where most of the engineering effort goes.

How Earth observation data works

Earth observation data arrives as scenes, each covering a fixed footprint on one date, and is organized by processing level.

Level	What it contains	Ready for
0	Raw sensor counts	Reprocessing only
1	Georeferenced radiance or amplitude	Visualization
2	Surface reflectance or calibrated backscatter, with quality masks	Analysis
3	Composites and mosaics over time	Time series, mapping
4	Model output such as biomass or soil moisture	Decision products

Analysis Ready Data (ARD) is the industry term for Level 2 products on a consistent grid with masks applied, so a user can compute on them without preprocessing.

Formats are raster. Cloud Optimized GeoTIFF holds most imagery, NetCDF and Zarr hold multidimensional atmospheric and ocean data, and point clouds hold lidar.

Discovery runs through STAC, the SpatioTemporal Asset Catalog, a JSON standard that indexes scenes by footprint, date, and asset URL. Most open archives on AWS, Microsoft Planetary Computer, and the Copernicus Data Space expose a STAC API.

Volume shapes everything. Copernicus alone distributes tens of terabytes a day, so Earth observation data lives in cloud object storage and is processed where it sits.

The output of an Earth observation workflow is a table: a value per asset, per field, or per boundary, per date.

Earth observation data in Wherobots

WherobotsDB reads Earth observation data where it is stored and joins it to vector data in the same SQL query, which is the step most workflows spend their effort on.

The STAC reader queries any STAC API by area and date and returns one row per scene, with each band asset already registered as an out-db raster. Pixels are read only when a function touches them.

RS_ functions do the raster work. RS_MapAlgebra computes indices, RS_Clip cuts to a boundary, RS_ZonalStats summarizes per polygon, and RS_Value samples at a point, each in parallel across every scene.

Havasu, the Apache Iceberg table format extended with native raster and geometry types, stores the resulting time series with ACID guarantees and spatial filter pushdown.

RasterFlow supplies the imagery preparation and model layer. It builds cloud-filtered Sentinel-2 and NAIP mosaics for an area of interest and runs built-in or custom PyTorch models over them.

The Wherobots Data Hub offers ready tables, including ESA WorldCover land cover and Overture Maps buildings, so a join has a partner on day one.

Nothing moves. Data stays in the customer’s S3 bucket or Iceberg catalog, and compute comes to it.

The Wherobots MCP Server exposes the same catalog and functions to Claude Code and other agents. WherobotsDB is built by the original creators of Apache Sedona, 100% code compatible across all spatial functions.

-- Dominant land cover class and pixel count per administrative area (ESA WorldCover)
SELECT a.name,
       RS_ZonalStats(lc.rast, a.geometry, 1, 'mode')  AS dominant_class,
       RS_ZonalStats(lc.rast, a.geometry, 1, 'count') AS pixels
FROM wherobots.esa.esa_world_cover lc
JOIN admin_areas a ON RS_Intersects(lc.rast, a.geometry);