Solar irradiance climatology and site screening

Geostationary satellite retrievals of GHI and DNI give solar developers a statistically stable irradiance climatology before a single ground instrument is deployed. The method works, but dusty skies and bright desert surfaces introduce biases that must be corrected explicitly.

Sensors

SEVIRI on Meteosat-9/10/11 (EUMETSAT): 12-channel spinning enhanced visible and infrared imager; full-disc image every 15 minutes (rapid-scan mode every 5 minutes over Europe and North Africa); 3 km nadir resolution in the high-resolution visible channel, 9 km in thermal channels. The primary European and African source for Heliosat-based GHI retrieval.
ABI on GOES-16/18 (NOAA): 16-channel Advanced Baseline Imager; full-disc scan every 10 minutes, continental US every 5 minutes, mesoscale sectors every 60 seconds; 0.5 km visible, 2 km thermal. Feeds NOAA's operational GHI product and underpins REST2-based retrievals for the Americas.
MODIS Terra/Aqua (NASA): 36-band imager at 250 m to 1 km resolution; two overpasses per day per location. Used primarily for aerosol optical depth (AOD) retrieval at 550 nm via the Dark Target and Deep Blue algorithms, which feed aerosol correction in both Heliosat and REST2. Not a primary irradiance source because twice-daily sampling cannot resolve cloud transients.
CAMS McClear / CAMS radiation service (Copernicus Atmosphere Monitoring Service): Not an imaging sensor but a reanalysis-driven aerosol and clear-sky irradiance climatology. McClear provides speciated aerosol optical depth and Linke turbidity at up to 1-minute temporal resolution for any location globally, covering 2004 to near-real-time. Used as the aerosol forcing input when sensor-derived AOD is unavailable or noisy.

Why geostationary imagers, not ground stations, anchor a bankable resource study

A single pyranometer gives you a precise number for one point. It tells you almost nothing about spatial variability across a 50 km² candidate zone, and a 12-month campaign captures perhaps one seasonal cycle, missing the interannual swing that ENSO and regional circulation patterns impose on cloud cover. Geostationary archives now extend to the early 2000s for SEVIRI and the mid-2000s for the GOES series, giving developers 15 to 20 years of sub-hourly cloud observations to draw on before a bankable P50/P90 estimate can be constructed.

The economic logic is straightforward. A ground campaign costs money and time; satellite screening costs a fraction of that and can cover dozens of candidate sites in parallel. The satellite result is not a replacement for a ground campaign at the shortlisted site, but it is a rigorous filter that stops developers wasting measurement budgets on sites that look promising on a map but sit under persistent low cloud or dust plumes.

How Heliosat and REST2 turn reflected light into irradiance

Heliosat, first published by Cano et al. in 1986 and refined repeatedly since, works from the visible channel of a geostationary imager. The core idea is that a pixel's apparent reflectance relative to a cloud-free reference tells you the cloud index, and the cloud index scales the clear-sky GHI down to an all-sky GHI. The reference surface reflectance, the so-called ground albedo map, is built from the darkest observed reflectances over a rolling window. Modern implementations such as HelioClim-3 and the CAMS radiation service use SEVIRI at 15-minute cadence and report mean bias errors of roughly 5 to 8 W/m² for monthly GHI over mid-latitude sites, rising in regions where the ground albedo assumption breaks down.

REST2 (Reference Evaluation of Solar Transmittance, version 2) takes a different path. It is a broadband clear-sky model that requires explicit aerosol, water vapour, and ozone inputs rather than deriving cloud index from reflectance. When driven by CAMS McClear aerosol climatology and combined with a cloud-fraction product from ABI or SEVIRI, REST2 can produce DNI estimates with mean bias errors in the 10 to 20 W/m² range at monthly timescales over clean-sky sites. DNI is harder than GHI because it is acutely sensitive to aerosol optical depth: a change of 0.1 in AOD at 550 nm can shift DNI by 40 to 60 W/m² under a high sun.

Both methods share a common weakness: they assume the atmosphere is plane-parallel and horizontally homogeneous within a pixel. Broken cumulus fields at sub-pixel scale, common in coastal and semi-arid zones, cause systematic underestimation of GHI because the cloud index is averaged across cloud and gap. This is not a solvable problem with current geostationary resolution; it is a known source of scatter in hourly data that largely cancels at monthly averages.

Dust and desert albedo: where the physics gets uncomfortable

The highest DNI values on Earth sit in the Atacama, the Sahara, the Arabian Peninsula and parts of the Thar Desert. These are also the environments most hostile to satellite irradiance retrieval. Mineral dust aerosol is spectrally complex: it absorbs in the ultraviolet and blue, scatters strongly in the visible, and its single-scattering albedo varies with mineralogy. MODIS Dark Target AOD retrieval degrades over bright desert surfaces precisely because the algorithm assumes a dark surface; the Deep Blue algorithm was developed specifically to address this, but uncertainties over highly reflective terrain still reach 0.1 to 0.2 in AOD, propagating directly into DNI errors.

High surface albedo creates a second problem for Heliosat. The cloud-free reference reflectance over sand and salt flats can approach the reflectance of thin cloud, compressing the dynamic range of the cloud index and making thin cirrus effectively invisible. Studies comparing HelioClim-3 against ground measurements in the Sahara have reported mean biases of 15 to 25 W/m² in monthly GHI, roughly two to three times the bias seen over vegetated European sites. Correcting for this requires site-specific albedo characterisation, ideally from a high-resolution optical sensor such as Sentinel-2 or Landsat-8/9, and explicit dust event flagging using aerosol index products from TROPOMI or the CAMS reanalysis.

Translating climatology into a site-ranking product

A screening exercise typically produces three spatial layers: long-term mean annual GHI, long-term mean annual DNI (relevant for concentrating solar power), and a variability metric such as the coefficient of variation of monthly GHI across years. DNI is the decisive variable for CSP; GHI drives flat-panel PV yield. The two do not always rank sites the same way, which matters when a developer is choosing technology alongside location.

Revisit rate sets the temporal floor. SEVIRI at 15 minutes captures the diurnal cycle well enough to compute hourly clearness index distributions, which feed directly into P90 yield models. ABI at 5 to 10 minutes is better still for sites in the Americas. Neither sensor resolves cloud shadows at sub-kilometre scale; that limitation is covered by a sibling page on cloud-shadow nowcasting and is outside the scope of a climatology product.

Satellize runs this screening workflow on SEVIRI, ABI and CAMS McClear inputs, with TROPOMI aerosol index as a dust-event flag, delivering ranked candidate-site reports as GIS layers and tabular summaries. The same aerosol correction pipeline underpins the crop-estimation work the team runs for the Kingdom of Tonga, where atmospheric correction quality directly affects surface reflectance accuracy.

Honest limits before you commission a study

Satellite-derived GHI at monthly timescales is reliable to within 5 to 10 percent at most mid-latitude sites with low aerosol loading. DNI uncertainty is larger, typically 10 to 15 percent at monthly scale even in good conditions, and can exceed 20 percent in persistently dusty regions. Hourly data carries substantially higher scatter; root-mean-square errors of 80 to 120 W/m² for hourly GHI are normal and should be expected in any honest validation table.

Archive depth is a real asset but not unlimited. SEVIRI data before 2004 is sparse and not uniformly reprocessed. GOES-East continuity across satellite generations introduces calibration discontinuities that require intercalibration corrections before multi-decade trend analysis is valid. Any study claiming sub-percent trend detection from raw geostationary archives without explicit intercalibration should be read with scepticism.

Finally, satellite climatology cannot replace a ground measurement campaign for project finance. Lenders and independent engineers will require at minimum 12 months of on-site pyranometer data, correlated against the satellite record to establish a site-specific correction factor. The satellite product shortens the campaign needed and increases confidence in the long-term mean, but it does not eliminate the requirement.

Typical figures

Spatial resolution (GHI/DNI grid)	3 to 9 km (SEVIRI); 2 to 4 km (ABI); output typically resampled to 1 to 5 km grid
Temporal sampling	15 min (SEVIRI standard), 5 min (SEVIRI rapid scan); 5 to 10 min (ABI); hourly aggregates standard in delivered products
Archive depth	SEVIRI from 2004; GOES-East usable from 2006 (GOES-13 era); CAMS McClear from 2004
GHI mean bias error (monthly, low-aerosol sites)	Approximately 5 to 8 W/m² (Heliosat/HelioClim-3 validation literature)
DNI mean bias error (monthly, low-aerosol sites)	Approximately 10 to 20 W/m²; 20 to 40 W/m² in dusty or high-albedo terrain
Aerosol optical depth input	MODIS Dark Target / Deep Blue (550 nm); CAMS McClear reanalysis; TROPOMI aerosol index for dust flagging
Geographic coverage	Global; SEVIRI covers Europe, Africa, Middle East; ABI covers Americas; gaps at high latitudes above ~65°
Retrieval methods	Heliosat (cloud-index from visible reflectance); REST2 (broadband clear-sky with explicit aerosol forcing)
Delivery formats	GeoTIFF raster layers, NetCDF time series, tabular CSV per candidate site, PDF screening report
Latency (climatology product)	Retrospective; monthly update possible for near-real-time monitoring applications

Analytics Satellize can run

Long-term mean annual GHI and DNI maps	Heliosat cloud-index retrieval on SEVIRI or ABI archive; REST2 clear-sky model with CAMS McClear aerosol forcing	GeoTIFF raster layers at 1 to 5 km resolution; one layer per variable per candidate zone
Interannual variability and P90 yield estimate	Year-by-year monthly GHI/DNI time series; empirical P50/P90 distribution from 15 to 20-year archive	Tabular summary per candidate site with P50, P90, coefficient of variation and year-of-record range
Dust-event frequency and AOD climatology	MODIS Deep Blue AOD time series; TROPOMI aerosol index dust-flag composites; CAMS reanalysis dust load	Monthly dust-day frequency map and site-level AOD percentile table for DNI loss estimation
Surface albedo characterisation for Heliosat bias correction	Sentinel-2 or Landsat-8/9 multi-temporal compositing to derive site-specific broadband albedo; applied as correction to cloud-index reference surface	Corrected GHI layer with bias-correction metadata; quantified improvement in mean bias for high-albedo terrain
Multi-site ranked screening report	Composite scoring across GHI, DNI, variability, dust frequency and terrain slope (from SRTM or Copernicus DEM)	PDF report ranking up to 20 candidate sites with per-site data tables and spatial maps; ready for independent engineer review
Satellite-to-ground correlation analysis	Regression of satellite-derived hourly GHI against client pyranometer record; derivation of site-specific bias and RMSE	Correction coefficients and validation statistics in CSV; adjusted long-term mean for use in yield models

Sources

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.