Six sources, one answer.
PSE, the Planetary Sensing Engine, is an open source data fusion engine that combines Earth observation data from six sources into one model behind a single API. Every value can be traced back to where it came from.
Earth data arrives in pieces.
Temperature comes from ECMWF, sunlight from the Global Solar Atlas, imagery from Copernicus, infrastructure from OpenStreetMap and social data from the World Bank. Each has its own format, grid and time step.
PSE does the alignment once, so every system built on it starts from the same clean picture.
What it reads.
- ERA5
- ECMWF reanalysis at 0.25 degrees, about 28 km, daily with a five day lag, through the Copernicus Climate Data Store.
- Open-Meteo
- 24 hourly weather variables, routed between archive and forecast at a five day cutoff.
- Global Solar Atlas
- World Bank solar resource at about 1 km: GHI, DNI, diffuse and tilted irradiance and PV potential.
- Sentinel-2
- 10 to 60 metre imagery every five days, turned into vegetation and water indices and a four class land cover.
- OpenStreetMap
- Substations, power lines, roads, settlements and waterways through the Overpass API.
- World Bank
- Electricity access, GDP per person, renewable share, CO₂ and population, yearly by country.
How it works.
- Align in space
- Everything is regridded to one grid, from 10 metre pixels to country figures.
- Align in time
- Hourly, daily, monthly, yearly and static sources are put on one time axis for the query.
- Weigh
- Where sources overlap, each gets a quality score of 0.7 times reliability plus 0.3 times recency, and they are blended by it.
- Flag
- When sources disagree beyond a set spread, the disagreement is flagged in the response.
- Trace
- Every fused dataset records its sources, the steps applied, the quality scores and the time of fusion.
- Serve
- A FastAPI service answers point, area and explicit fusion queries with xarray datasets, stored in Zarr and PostGIS.
FLUX, built on PSE.
Renewable energy siting for Indonesia, Kenya and Vietnam, with each country's grid, tariffs, exclusion zones and financial terms.
- Solar yield
- pvlib from sun position to hourly output, five panel types, with P50, P90 and P10.
- Wind yield
- Hub height extrapolation, Weibull fits and power curves for four turbine models.
- Grid and cost
- Nearest substation, connection distance, voltage level and cost, then LCOE, IRR, NPV and payback.
- Climate risk
- Six hazards, from flood and cyclone to drought and sea level rise, weighed into one score.
In figures.
- 6
- Data connectors
- Apache 2.0
- Open source licence
Open engine, private products on top.
PSE was released in March 2026 under Apache 2.0, about 12,500 lines of Python with more than 200 passing tests. FLUX, our renewable energy model, is built on it and runs for Indonesia, Kenya and Vietnam. The data layer is open, and what we build on top of it is our own.
Who it serves.
- Climate and Earth observation researchers working in the Pangeo ecosystem.
- Teams siting solar and wind projects.
- Developers building their own systems on open Earth data.
