Rooftop solar potential mapping from DSM and irradiance data
Stereo and SAR-derived surface models, combined with satellite irradiance records, let planners estimate per-building photovoltaic yield before a surveyor sets foot on site. The method has real limits: rooftop obstructions and cloud seasonality both introduce uncertainty that honest analysis must quantify.
Sensors
- TanDEM-X (TerraSAR-X add-on): Bistatic X-band SAR interferometry producing the global TanDEM-X DEM at 12 m posting, with commercial high-resolution spotlight acquisitions down to 1.5 m pixel spacing. Penetrates cloud cover, enabling DSM generation in persistently overcast cities, but X-band reflects off the first solid surface encountered, so a flat HVAC unit and a bare roof return nearly identical signals.
- Airbus Pléiades Neo tri-stereo: 50 cm panchromatic imagery collected in three along-track passes within a single overpass, enabling photogrammetric DSMs at 25–50 cm ground sampling. Roof ridges, dormer windows and parapet walls resolve clearly. Dependent on cloud-free conditions; a single tasking window may need two or three attempts in tropical cities.
- CAMS McClear / Copernicus Atmosphere Monitoring Service solar irradiance: Provides clear-sky and all-sky surface solar irradiance at up to 1-minute temporal resolution and roughly 3–5 km spatial resolution, derived from Meteosat SEVIRI. Archive extends to 2004. Uncertainty in monthly global horizontal irradiance is typically 3–8% depending on aerosol loading; higher in dust-prone or persistently cloudy regions.
- Sentinel-2 MSI: 10 m multispectral imagery with a 5-day revisit at mid-latitudes (10-day per satellite). Used here not for DSM generation but to build seasonal cloud-fraction composites that identify which months contribute reliable irradiance observations and which must be interpolated or flagged as uncertain.
- Sentinel-1 SAR (C-band): 6-day repeat C-band SAR at 10 m IW mode resolution. Useful as a cross-check on rooftop change between DSM acquisition and analysis date, particularly for detecting new rooftop structures that would invalidate an older surface model.
What a surface model actually sees on a roof
A digital surface model records the first reflective surface a sensor encounters. For a pitched residential roof that is usually the tiles or membrane. For a flat commercial roof it is whatever sits highest: a parapet, a water tank, an HVAC condenser, or a forest of ventilation stacks. The problem is that at 12 m TanDEM-X posting, a 2 m condenser unit and the clear deck beside it average into a single elevation value that understates usable area. Pléiades Neo at 25–50 cm DSM resolution resolves individual obstructions more faithfully, but even there, a low parapet casting a shadow onto the deck can be misclassified as usable surface if the shadow-analysis step uses only the DSM without a companion orthoimage.
This is why rooftop solar mapping pipelines should always combine a DSM with a co-registered high-resolution orthoimage. The image identifies obstruction footprints; the DSM supplies the height field for slope, aspect and horizon-angle calculations. Treating either product as sufficient on its own produces yield estimates that look precise but carry unquantified systematic errors.
From surface model to tilt, aspect and shading geometry
Once a clean DSM is in hand, the analytical steps are well-established. Slope and aspect are computed per pixel using standard neighbourhood gradient operators. Pixels steeper than roughly 60 degrees are excluded as structurally unsuitable for panel mounting. The remaining roof facets are grouped into aspect classes (south-facing in the northern hemisphere being the premium orientation) and slope classes that determine the optimal panel tilt offset from roof pitch.
Horizon angles are computed by ray-casting from each candidate pixel outward in 36 or 72 azimuth directions, sampling the DSM to find the elevation angle at which the horizon blocks direct sunlight. This captures shading from adjacent buildings and rooftop superstructures. The computation is straightforward but expensive at city scale: a 100 km² city at 50 cm resolution involves roughly 400 million pixels, and horizon-angle ray-casting scales with both pixel count and ray-casting distance. Practical pipelines tile the city and run in parallel, typically producing a full shading model within hours on cloud infrastructure.
The output is a per-pixel annual shading factor, expressed as the fraction of potential beam irradiance actually received after horizon obstructions. Combined with the irradiance record, this gives an unshaded and shaded irradiance value for every roof facet.
Irradiance data: what CAMS gives you and where it falls short
CAMS McClear provides physically modelled clear-sky irradiance based on Meteosat observations of aerosol optical depth, water vapour and ozone. The all-sky product folds in cloud cover derived from the same geostationary sensor. Monthly mean global horizontal irradiance uncertainty is typically cited at 3–8% under moderate aerosol conditions, rising to 10–15% in regions with high dust variability or persistent convective cloud.
The spatial resolution of roughly 3–5 km is the most significant limitation for urban applications. A city centre with dense high-albedo surfaces and urban heat-island effects can have a meaningfully different aerosol environment from its suburbs, but CAMS will not resolve that gradient. For grid-capacity planning at district scale, this is usually acceptable. For individual building underwriting, analysts should note the mismatch between the 50 cm DSM pixel and the 3–5 km irradiance cell explicitly in any yield certificate.
Seasonal cloud correction matters more than it is usually given credit for. A 10-year CAMS archive allows the analyst to characterise not just mean annual irradiance but interannual variability, which is the figure a project financier actually needs. A site with high mean irradiance but a coefficient of variation above 15% across years carries a different revenue risk profile from one with lower mean but stable delivery.
Turning geometry and irradiance into a yield estimate
Photovoltaic yield is the product of usable area, panel efficiency, irradiance received and a system performance ratio that accounts for inverter losses, wiring, soiling and temperature derating. The satellite-derived inputs supply the first three. Panel efficiency and performance ratio come from manufacturer datasheets and published empirical ranges, typically 0.75–0.85 for well-maintained grid-tied systems using current silicon technology.
The per-building aggregation step requires building footprints. These can come from a municipal cadastre, from OpenStreetMap, or from a building-footprint extraction layer derived from the same Pléiades Neo imagery used for the DSM. Each footprint clips the yield raster to produce a single number: estimated annual kilowatt-hours per building. That number, combined with local grid tariffs and connection costs, is what a green-finance underwriter or a municipal permitting office actually needs.
Honest reporting should accompany every yield estimate with an uncertainty band. A reasonable first-order decomposition: DSM-derived area uncertainty ±10–20% for flat roofs with significant obstruction density, irradiance uncertainty ±5–10%, performance ratio uncertainty ±5%. Combined in quadrature, the yield estimate for a single building carries roughly ±15–25% uncertainty at one standard deviation. Portfolio-level estimates for hundreds of buildings narrow considerably as systematic biases partially cancel.
Where the method earns its keep, and where it does not
Municipal solar permitting offices benefit most when they use rooftop potential maps to prioritise outreach rather than to certify individual installations. A map showing that a particular industrial district contains 40 hectares of south-facing flat roofing with low obstruction density and high irradiance is a reliable planning input. A map claiming that a specific building at a specific address will generate exactly 18,400 kWh per year is making a precision claim the data do not support without a ground-truth survey.
Utility grid-capacity planning is a natural fit. A distribution network operator needs to know where behind-the-meter generation is likely to concentrate, not the precise output of each panel. The satellite-derived layer, aggregated to feeder or substation zones, gives planners a defensible spatial prior for where curtailment risk and voltage-rise problems are likely to emerge as rooftop adoption grows.
Satellize runs this pipeline using CAMS irradiance, Sentinel-2 cloud composites and commercial DSM tasking on client licence, producing GIS-ready yield layers and uncertainty maps. The Tonga crop-estimation programme demonstrated the same principle in a different domain: satellite-derived physical parameters, combined with honest uncertainty accounting, produce outputs that are genuinely useful for government decisions rather than decorative.
Keeping the model current
A rooftop solar potential map has a shelf life. Cities change. A DSM acquired in 2021 will not capture a rooftop plant room added in 2023, and it will not capture the new eight-storey building next door that now shades a previously open roof in the afternoon. Sentinel-1 SAR, with its 6-day repeat and all-weather capability, provides a practical change-detection layer: coherence change between two SAR acquisitions flags rooftops where the surface has been modified, triggering a targeted re-tasking of optical or SAR DSM collection over those specific parcels rather than a full-city re-acquisition.
The irradiance record, by contrast, improves with age. Each additional year of CAMS data narrows the uncertainty on interannual variability. A yield estimate backed by a 15-year irradiance record is materially more bankable than one backed by three years, and the archive cost is zero.
Typical figures
| DSM spatial resolution (TanDEM-X spotlight) | 1.5–3 m pixel spacing; 12 m in standard global DEM product |
| DSM spatial resolution (Pléiades Neo tri-stereo) | 25–50 cm ground sampling distance under good collection geometry |
| Irradiance product spatial resolution (CAMS McClear) | Approximately 3–5 km; temporal resolution up to 1 minute |
| Sentinel-2 cloud-fraction composite revisit | 5-day combined constellation revisit; monthly composites typically achievable in most climates |
| Irradiance archive depth (CAMS) | 2004 to near-present; roughly 20-year record available |
| Irradiance uncertainty (monthly mean GHI) | 3–8% in moderate aerosol conditions; 10–15% in high-dust or persistent-cloud regions |
| Minimum resolvable rooftop obstruction | ~1 m footprint at Pléiades Neo resolution; ~5–10 m at TanDEM-X spotlight; sub-metre obstructions not reliably detected |
| Per-building yield estimate uncertainty (1 sigma) | ±15–25% for individual buildings; narrows to ±8–12% at portfolio scale of hundreds of buildings |
| Delivery formats | GeoTIFF yield rasters, GeoPackage per-building attribute tables, PDF uncertainty summary, WMS/WMTS tile service |
| DSM acquisition latency (commercial tasking) | Typically 2–6 weeks from order to delivered DSM, subject to cloud clearance windows |
Analytics Satellize can run
| Per-building annual yield estimate | DSM slope/aspect extraction, horizon-angle ray-casting, CAMS irradiance integration, performance-ratio modelling | GeoPackage attribute table with estimated kWh/year, usable area (m²), dominant aspect class and uncertainty band per building footprint |
| City-wide rooftop suitability classification | Slope, aspect and shading thresholds applied to DSM raster; obstruction-density index from orthoimage segmentation | GeoTIFF suitability raster (5-class: excellent / good / moderate / poor / unsuitable) with accompanying PDF methodology note |
| Seasonal irradiance variability report | CAMS 15-year monthly GHI time series; coefficient of variation and P90/P50 ratio computed per irradiance cell | PDF bankability report with monthly irradiance charts, interannual variability statistics and P90 annual yield estimate per zone |
| Grid-feeder aggregated solar capacity potential | Spatial join of per-building yield estimates to utility feeder polygons; cumulative capacity and diversity factor estimation | GIS layer and CSV table of estimated behind-the-meter generation potential per feeder zone, for distribution network planning |
| Rooftop change-detection alert (DSM currency monitoring) | Sentinel-1 SAR coherence change detection between baseline and monitoring epoch; flagging of modified rooftop pixels | Monthly GeoJSON alert layer of changed rooftop parcels requiring DSM re-acquisition or yield model update |
| Cloud-correction quality flag layer | Sentinel-2 monthly cloud-fraction composites used to identify irradiance months with >40% cloud fraction, flagged as interpolated in CAMS record | GeoTIFF quality mask indicating data-confidence level per irradiance cell per season, delivered alongside yield rasters |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.