Grid operators and water utilities face a fundamental forecasting problem: conventional meteorological stations are too sparse and too slow to warn them that a specific district is about to overwhelm its substation or exhaust its reservoir. When a heatwave bakes an urban basin, demand does not rise uniformly — it spikes hardest where surface temperatures are highest, vegetation has already died back, and building stock retains heat overnight. Satellite land surface temperature (LST) data resolves that spatial heterogeneity at 30–100m, giving demand planners a physical signal that weather-station interpolation simply cannot replicate.
A constellation of thermal and multispectral satellites produces hourly LST mosaics, NDVI drought indices and urban albedo maps that feed directly into load-forecasting models. The satellite stack adds two decisive advantages: it sees every rooftop simultaneously rather than sampling, and it captures the nocturnal heat retention that drives overnight air-conditioning demand — the period most dangerous for transformer failure. Fused with smart-meter telemetry and historical demand curves, the resulting ML inference layer can issue 6–48 hour ahead demand forecasts with district-level granularity, hours before a commercial weather service flags anything abnormal.
The operational outcome is the ability to pre-position spinning reserves, pre-cool reservoirs, redirect water flows across distribution zones, and defer non-essential industrial loads before the crisis rather than during it. Nations that have experienced rolling blackouts during extreme heat events — events that now occur with measurable regularity — know that the cost of under-forecasting is measured in lives and economic damage running into hundreds of millions. Sovereign control of that forecasting chain means the data arrives without API rate limits, embargo clauses or commercial prioritisation toward a vendor's premium clients.
Frequently asked
What does 'heat-driven demand forecasting' actually mean in operational terms?
It means using satellite-observed land surface temperature, soil moisture, and urban morphology data — rather than sparse weather-station networks alone — to predict how much electricity, water, and emergency cooling capacity a city or region will need in the next 6–72 hours during a heatwave. The satellite layer fills spatial gaps between ground sensors, captures the urban heat island effect at neighbourhood scale, and updates faster than most numerical weather prediction cycles. Grid operators and water utilities ingest the resulting demand curves into their scheduling and dispatch systems.
Why can't we just buy this as a service from a commercial vendor like Planet or Spire?
You can, and some do — but the moment you rely on a foreign commercial feed for national grid dispatch decisions, you inherit that vendor's uptime risk, pricing power, export-licence constraints, and data-handling jurisdiction. During geopolitical stress or a vendor outage, the feed can be throttled or cut entirely at exactly the moment demand surges are most dangerous. A sovereign constellation means the data pipeline is under national control, the algorithm training data stays onshore, and the capability cannot be sanctioned away.
How many satellites are needed for a useful sovereign capability?
A constellation of 8–16 LEO microsatellites in complementary sun-synchronous and inclined orbits can achieve sub-2-hour revisit over a continental territory with adequate swath width. Below 8 satellites, revisit gaps become operationally significant for intra-day demand ramp forecasting. ESA's Φ-sat programme and NOAA's satellite constellation studies suggest 12 satellites as a practical starting point for regional coverage, with constellation growth driven by national demand density maps.
What spatial resolution is actually needed for demand forecasting — do we need 30 cm imagery?
No. For heat-driven demand forecasting, thermal resolution of 100 m to 1 km is operationally sufficient for city-scale and grid-zone aggregations; sub-100 m resolution adds cost without proportionate forecast skill improvement. The bottleneck is revisit frequency and calibration accuracy, not pixel size. High-resolution optical imagery from providers like Planet or BlackSky is useful for urban morphology basemaps, but the time-critical thermal layer does not need sub-metre resolution.
How does satellite data improve on numerical weather prediction models that utilities already use?
NWP models such as ECMWF's IFS or NOAA's GFS operate on grid cells of 9–25 km and assimilate surface observations that are sparse in many regions. They systematically underestimate urban heat islands — which can add 4–8 °C to demand-relevant temperatures — because the urban canopy is under-resolved. Satellite LST at 100 m–1 km injects direct observation of actual surface temperatures into the forecast chain, correcting the urban bias and improving demand forecast RMSE by up to 18% according to NOAA studies.
Is this capability relevant only to wealthy countries with sophisticated grids?
Emphatically not — it may be more critical for lower-income countries. Nations in South Asia, Sub-Saharan Africa, and the Middle East face the fastest-growing heat exposure (WHO projects 3.5 billion people at dangerous heat risk by 2050), often have the sparsest ground-station networks, and have the least grid headroom to absorb unforecast demand spikes. The WMO's Systematic Observation Financing Facility explicitly identifies satellite data as the primary tool to close observation gaps in data-sparse nations.
What happens to the forecast when satellites are unavailable due to clouds or maintenance?
A resilient operational architecture layers satellite LST with microwave sounder data (which penetrates cloud), NWP model output, and historical climatological demand patterns to produce a degraded-mode forecast when the primary thermal feed is unavailable. Nations should design their sovereign system with an explicit data-gap protocol — tested regularly — so that grid operators know the forecast confidence level and its source at all times. Dependency on a single satellite pass without a fallback is an unacceptable operational design.
How do we handle the transition from buying commercial data today to operating our own constellation?
The standard approach is a 'bridge-and-build' strategy: maintain commercial data service agreements with vendors like Spire or Planet during the 3–5 year constellation development and launch phase, use that period to build sovereign ground-segment infrastructure and train national algorithm teams on live data, and migrate dispatch decisions progressively to the national feed as it proves out. The commercial agreements should include data-format and algorithm-documentation clauses so that transition does not require a cold-start model rebuild.