How we measure roofs from free satellites and open data, then turn them into a ranked, costed map of where Emissiv’s coating matters most. The public map starts with 12 enriched UK city caches, 4,744 roofs, then scans new UK areas live with Sentinel-2 albedo, model-only material and transparent provenance.
We measure how reflective and how hot every roof is, then turn that into where the cooling coating is worth the most.
Emissiv’s product is a radiative-cooling coating. The first number that decides where it matters most is a roof’s reflectance: dark roofs absorb more sun, run hotter, and have more to gain. The rest of the system adds the context needed to act: what the roof is made of, how it is shaped, what building sits underneath, whether it cools, and how certain each answer is.
The output is a drop-in superset of the Google Solar API: the same per-roof geometry schema, plus reflectance, material and thermal, the layers it doesn’t carry.
A radiative-cooling surface does two things: it reflects sunlight and it radiates heat to the sky. The first is governed by albedo (solar reflectance, 0 to 1) and that is exactly what a satellite can read. A dark roof near 0.10 albedo absorbs about 90% of the sun that hits it. That is both the problem (it bakes) and the opportunity (the coating has the most to give there). Find the dark roofs, and you have found the customers.
The map also keeps a warm road-heat trace on at all times. Its numbers are a live modelled asphalt temperature from Open-Meteo irradiance, air temperature and wind at the map centre, plus assumed asphalt optics. It is not measured road temperature and not an input to the roof calculations. It is there to make the urban heat setting legible while the roof values stay provenance-led.
No imagery bought, nobody waited on. Sentinel-2 and the open-data layers keep variable cost low; OS NGD is the licensed authoritative layer for material and building records.
| Layer | Source | What it gives | Cost |
|---|---|---|---|
| Imagery | Sentinel-2 L2A (ESA), 10 m (visible/NIR; SWIR bands 20 m, resampled to 10 m), 5-day revisit | the pixels we read albedo from, plus SCL masks for cloud, shadow, snow and no-data | free |
| Cross-checks | Sentinel-1, Landsat | independent albedo sanity checks, plus area-scale land-surface-temperature evidence | free |
| Footprints + material | OS NGD (Ordnance Survey) | authoritative per-building outline + roof material | licensed, quota-aware |
| Building record | OS NGD | UPRN, building use, construction year, floors | licensed, quota-aware |
| Geometry | OS NGD + EA LiDAR, 1 m | OS height and flat/pitched shape, LiDAR pitch angle, orientation and fine form | licensed + free |
| Energy + cooling | EPC / SBEM (GOV.UK), joined by UPRN | A-G band, floor area, and per-building cooling demand + efficiency | free |
| Backtest evidence | Private pilot case packs, public aggregate summaries | field-logger surface gaps, indoor comparison, energy claim status and provenance | private raw, tracked aggregates |
| Civic heat risk | Cached cloud-masked Landsat LST, ONS Census 2021, IoD 2025 | area heat exposure, local sensitivity and adaptive-capacity screen | free, OGL/Crown copyright |
| Planning + solar + grid | planning.data.gov.uk, PVGIS, National Grid | listed/conservation/flood, horizontal GHI, live carbon | free |
Each step is simple physics or an authoritative lookup. Nothing is a black box.
Sentinel-2 sees in narrow spectral bands. A narrowband-to-broadband (NTB) physics formula converts them to one solar reflectance, averaging only clear pixels inside the roof footprint of the least-cloudy recent scene. The Scene Classification Layer removes cloud, cloud shadow, snow and no-data pixels. If a roof is fully masked, we skip it rather than invent a low albedo.
Where OS NGD has it (91.95% of the current served roofs) we use its authoritative roof material; elsewhere a model trained on real labels is the fallback and is shown with an uncertainty hedge. The 10 m spectral model is not the product, its leave-one-city-out macro-F1 is still weak at about 0.2. LiDAR geometry is a sanity constraint, not a learned material feature.
Pure physics for an opaque surface.
Absorbing is only half the story. A roof also radiates heat back to the sky, set by thermal emittance (most roofs ~0.90, bare metal ~0.45). We solve the standard ASTM E1980 energy balance for the surface temperature, which gives a modelled peak temperature and a proper SRI (hot black roof = 0, cool white roof = 100). The payoff: a reflective-looking metal roof can run ~10°C hotter than its albedo suggests, because it cannot radiate.
For a dark roof, a 95%-reflective coating raises albedo by ΔA. The solar heat it stops absorbing is robust physics that applies to every roof; a slice of it converts to money and carbon for cooled buildings. How firmly we ground that cooling demand follows a three-tier ladder: the building’s own EPC/SBEM cooling figure where it exists, a published BEES benchmark for its building type where the use implies cooling, and otherwise no pounds, only the physics.
Everything down to heat rejected is firm: ΔA and area are measured per roof, and (0.95 − albedo) × G × Area is physics that holds for every roof, cooled or not. G is this roof’s own annual sun, not a flat 1000: the per-city horizontal irradiation from PVGIS, multiplied by a sky-view factor from the 1 m LiDAR DSM. That sky-view factor accounts for overshadowing by taller neighbours and also damps the longwave term because a shaded roof sees less cold sky. Pitch and azimuth do not yet adjust solar incidence, so we call this horizontal-equivalent sun. The cost line is a modelled overlay: it is capped by the building’s own EPC cooling demand where present, capped by a BEES benchmark where building use implies cooling, and zero where there is no cooling signal. The cooling cut has two physical terms: a daylight reflectance term and a 24/7 emittance term for the heat the coating radiates to the sky. The modelled factors are f = Uroof/ho, the small share of surface cooling that reaches the conditioned space, and L ≈ 440 kWh/m²/yr, the UK all-sky annual longwave sky-cooling energy, our most uncertain input. This is a cooling-season figure, not a net-annual claim.
The three-tier cooling ladder. Only about 38% of roofs carry an EPC, so a measured cooling figure is the exception, not the rule. To avoid a blank for every other commercial roof, the cooling demand walks the same ladder as material does. Tier A · measured: the building’s own EPC/SBEM cooling electricity (best, used wherever present). Tier B · estimated: no EPC cooling, but OS NGD building use implies cooling (office, retail, commercial, health and the like), we apply a published UK cooling-intensity benchmark for that use class (BEES 2014-15, the BEIS Building Energy Efficiency Survey: e.g. offices ≈ 18, health ≈ 10, retail ≈ 8 kWh/m²/yr of cooling electricity) through the identical sol-air formula and cap, and the card badges it “estimated”, a distinct, lower-confidence tier, never shown as a certificate-backed figure. Tier C · physics only: no cooling signal, so we report the heat rejected and the peak-temperature cut and claim no pounds. A non-domestic EPC that was assessed and found to have no air-conditioning is a measured “no cooling” signal and stays in Tier C rather than being overwritten by a benchmark.
OS NGD gives each building its UPRN, which we use to join the EPC energy band, the use and the age. So a hot dark roof becomes “a retail building, built 1985, rated D”, which tells you whether it actually carries a cooling load. A physics target becomes a qualified commercial target.
When the heat intelligence control is requested, the map now looks for three ingredients before it shows a civic heat-risk score: a cached cloud-masked Landsat land-surface-temperature composite for the current city, an ONS Census 2021 sensitivity record for the local LSOA, and IoD 2025 adaptive-capacity context. Landsat gives measured surface exposure at 100 m. ONS adds who may be more exposed or less able to adapt, age, disability, health, tenure, overcrowding and housing form. IoD adds deprivation context where it exists. The board is deliberately called a screen: it is useful for local-authority triage, not a mortality model, medical claim or emergency response forecast.
Backtesting means replaying a real site through the same product logic and comparing the model output with measured field data. The first case pack is Earls Court: an aged TPO roof, a partial coating area, field loggers on roof surfaces and paired indoor sensors. The raw bundle stays private; the app reads a small aggregate JSON with the metrics, caveats and provenance.
In hot roof periods, the untreated reference roof averaged 20.6°C warmer than the radiative-cooling coated surface. This supports surface-temperature direction and scale for that site.
The paired indoor workbook showed the uncoated room 1.11°C warmer on average. Useful, but site specific because occupancy, HVAC, orientation and sensor placement are not controlled.
No HVAC, BMS or meter series was present, so the product must not claim measured energy savings. Backtests calibrate confidence envelopes; they do not overwrite the physics.
In Explore, select a roof and request Backtest evidence. The card shows measured, observed and modelled labels, plus a projected road-vicinity ribbon from basemap roads where road geometry is actually available.
The ideal customer is a large, dark, hot, commercial roof. The engine ranks every one of them across a city or the country, for free, with no site visits. Cold outreach becomes a targeted map.
Total dark-roof area is the addressable market; per building you get the energy, £ and CO₂, costed against the live grid, with listed and conservation buildings flagged so the targetable market is the real one.
The same satellites revisit every few days, so reflectance change can be measured over time and impact verified remotely, the evidence funders and carbon markets need.
Free data in, a scalable pipeline, so one estate or the whole country costs roughly the same to survey.
Google owns geometry-for-solar in the UK with its Solar API. Nobody holds material, reflectance and thermal at national scale. We match Google’s geometry fields and add the layers that decide a cooling sale. The barrier is not the data, which is open. It is the physics pipeline and the honest, validated joins that turn it into a decision.
Every number is labelled measured, modelled or estimated, and regenerable from a script. That discipline is what makes it credible to the people writing cheques.
What each source is, and exactly how it enters the calculation. The point of leading with measured physics is that each step can be checked against its reference. Grouped below; tap a heading to open it, swipe each row sideways.
ESA satellite (10 m) + scene classification mask + narrowband-to-broadband albedo physics.
Feeds: the measured solar reflectance R every roof leads with. Cloud, shadow, snow and no-data pixels are excluded before the NTB calculation.
Ordnance Survey authoritative building data (~97% of GB).
Feeds: roof material, flat/pitched shape, height, UPRN, building use/age/floors. Shown without a hedge.
Environment Agency surface-elevation model.
Feeds: roof pitch / aspect / form / roughness, and the surrounding building heights used for shading.
Sky-view-factor algorithm from a DSM.
Feeds: the per-roof sky-view factor (16 rays to 100 m) that damps both sun-in and heat-radiated-out. Base height = footprint median.
EU solar-radiation database (live API).
Feeds: the per-city annual horizontal sun (GHI), 914 (Glasgow) to 1158 (Cardiff) kWh/m²/yr, replacing a flat national average.
Free live weather API.
Feeds: the “now” surface temperature: live irradiance, air temperature and wind (wind drives convective cooling).
Road vectors from the map baselayer plus live irradiance, air temperature and wind.
Feeds: the always-on modelled road heat context visual and its asphalt-now metric. It is not measured road temperature and not used in roof scoring.
Private pilot bundle reduced to a public-safe aggregate backtest summary.
Feeds: the selected-roof Backtest evidence card: measured roof surface, observed indoor comparison, energy not measured, with raw traces kept private.
Free 100 m land-surface-temperature raster from USGS via Planetary Computer.
Feeds: the area-scale heat-exposure layer. Cloud, shadow, snow and fill bits are masked before the city composite is cached. It validates surface heat, not one roof’s temperature.
Official England and Wales small-area census tables via Nomis bulk data.
Feeds: the heat-sensitivity side of the civic risk board: older residents, young children, disability, bad health, tenure, overcrowding, flats and heating. Crown copyright, OGL v3.0.
Official local deprivation rank for English small areas.
Feeds: the adaptive-capacity context in the heat-risk screen. It is a place-level planning signal, not a person-level health measure.
Solar Reflectance Index standard.
Feeds: the SRI and the modelled surface temperature, combining reflectance and emittance (not reflectance alone).
Building heat-transfer reference.
Feeds: the cooling-load split: a daylight reflectance term (ΔR·G) + a 24/7 emittance term (Δε·L). Constants: longwave ≈63 W/m², hₒ≈17.
Peer-reviewed cool-roof study.
Feeds: the insulation factor f ≈ U/hₒ (~1–3%): only a small share of surface heat reaches the room below. Tiered by building age.
Spectra of Emissiv’s coating.
Feeds: the 95% reflectance + 95% emittance the coated-roof maths uses, modelled opaque. Measured 300–1400 nm; full spectrum to come.
GOV.UK Energy Performance certificates.
Feeds: energy band, floor area, and the per-building cooling demand (kWh/m²) + cooling efficiency (SSEER). England & Wales only.
UK building-energy benchmarks (BEIS BEES 2014-15, Tables A.1 + B.3).
Feeds: the use-class cooling estimate (tier B) where a building has no EPC cooling figure, badged “estimated from type”.
Live UK grid-carbon API.
Feeds: converts cooling-electricity saved into CO₂ avoided at the current grid mix.
Open planning register.
Feeds: listed / conservation / flood flags → coating eligibility.