The forecast is only as good as what you can measure
A chiller's workload is driven by conditions at the building — not by the regional weather forecast. A rooftop under summer sun runs far hotter than the air temperature a forecast reports, and that difference is exactly what makes cooling demand hard to predict without instrumenting the site. Earth observation measures it directly. CoolStorAge combines three complementary European data sources:
- Meteosat SEVIRI (EUMETSAT) — land surface temperature at 15-minute cadence, the highest-frequency thermal signal available without on-site sensors.
- Copernicus ERA5-Land — the surface energy balance: skin temperature, radiation and heat fluxes, including the surface-to-air temperature delta that captures urban-heat effects.
- Copernicus Sentinel-2 — 10–20 m imagery characterising each site's surroundings (vegetation, built-up density, water), the slowly-varying context that makes a model transferable between sites.
What the research showed
We validated the approach on measured data from a commercial building in Slovenia during the main part of the 2026 cooling season.
The forecast works. The day-ahead cooling-demand model met its accuracy objective against calorimeter ground truth — roughly 3.5× more accurate than naive baselines.
EO substitutes for missing instrumentation. At a fully metered site, satellite data adds little — the information is already being measured on the roof. But where on-site metering is absent, EO recovers about 63% of the accuracy gap that installing instrumentation would close. That is the finding that matters commercially: it means a new site can be onboarded, and its savings estimated, with nothing installed.
The optimisation pays. Run in shadow mode against the pilot's existing static schedule, the optimiser delivered cooling-electricity savings above the 20–30% target range.
What this enables
The EO result changes what a thermal-storage service can be. Because the forecast works without site instrumentation, satellite data enables services that were previously gated on hardware installation:
Analytics Pre-contract savings assessments
Estimate a prospective customer's savings before anything is installed.
Impact Investment-decision support
Help OEMs size new ice-storage systems on forecast demand, not rules of thumb.
Onboarding Zero-CAPEX onboarding
Bring existing sites into the service with no on-site hardware.
Continuity Continuity through sensor outages
Keep forecasting and optimising when on-site metering fails.
About the activity
CoolStorAge was developed under an ESA ACCESS BASS Kick-Start activity (2026), co-funded by ESA and Abelium. It builds on the EO ingestion and forecasting foundations of our earlier ESA activity EO x Grid, extending them from the supply side (solar production) to the demand side (cooling load).
Similar projects
- EO x Grid - Enhancing grid stability and renewable energy integration through accurate space-based PV forecasting. Funded under the European Space Agency's (ESA) ARTES 4.0 Downstream Applications program.