Aerosol optical depth correction for solar irradiance estimation
Uncorrected satellite irradiance products overestimate solar yield in dust-prone and polluted regions. AOD retrievals from MODIS, VIIRS, and Sentinel-5P TROPOMI quantify column aerosol loading so that irradiance estimates reflect what actually reaches a panel.
Sensors
- MODIS Terra/Aqua (MOD04/MYD04): Dark-target and deep-blue AOD retrievals at 10 km resolution, twice-daily combined overpass frequency. The standard global aerosol record stretches back to 2000, making it the primary source for long-term AOD climatologies.
- VIIRS (AERDB/AERDT products): Successor to MODIS on Suomi-NPP and NOAA-20, providing 6 km AOD at nadir via the deep-blue algorithm. Consistent cross-calibration with MODIS allows the two records to be merged for trend analysis.
- Sentinel-5P TROPOMI: UV-visible spectrometer retrieving AOD and absorbing aerosol index at roughly 5.5 km x 3.5 km pixel size with daily global coverage. Particularly useful for absorbing aerosols such as carbonaceous smoke that the dark-target algorithm underweights.
- CAMS Radiation Service (Copernicus Atmosphere Monitoring Service): Not a sensor but the primary irradiance product this page addresses. CAMS-RAD delivers surface solar irradiance at 15-minute intervals from 2004 to near-real-time, ingesting MODIS AOD climatologies to drive its radiative transfer. Bias in the AOD input propagates directly into GHI estimates.
Why aerosols are a financial problem, not just a physics one
Solar project finance rests on a P90 yield estimate: the irradiance level exceeded 90 percent of years. A systematic positive bias in that estimate shifts the P90 upward, inflating projected revenue and potentially tipping a marginal project into bankability. In dust-prone regions, that bias is not small. Published comparisons of CAMS Radiation Service output against ground pyranometers in the Sahel, Arabian Peninsula, and northern India show mean biases of 5 to 15 percent in months of heavy aerosol loading, with episodic dust events pushing instantaneous errors above 30 percent. Over a 25-year project life, a persistent 8 percent overestimate in annual yield is the difference between a project that services its debt and one that does not.
The mechanism is straightforward. Aerosol particles scatter and absorb incoming solar radiation before it reaches the surface. The fraction lost depends on the aerosol optical depth at each wavelength and on the solar zenith angle. Most satellite-derived irradiance products apply a monthly or climatological AOD correction rather than a scene-by-scene one, because the AOD record is spatially coarse and temporally irregular. Where the climatological AOD is lower than the actual aerosol burden during high-dust months, the irradiance product is optimistic.
What AOD retrievals actually measure, and where they struggle
AOD is the column-integrated extinction of the atmosphere at a reference wavelength, typically 550 nm. A value of 0.1 is a clean maritime atmosphere; values above 1.0 are common during Saharan dust outbreaks or biomass-burning episodes. MODIS dark-target retrievals work well over dark vegetated surfaces but lose accuracy over bright desert soils, exactly the terrain most relevant to utility-scale solar development. The deep-blue algorithm, applied to both MODIS and VIIRS, extends retrieval capability over bright surfaces by using shorter wavelengths where the surface is relatively dark, but uncertainty over sand dunes remains higher than over vegetated land, typically ±0.05 plus 15 to 20 percent of the retrieved value.
TROPOMI adds value for absorbing aerosols and for identifying aerosol type, which matters because the radiative effect of mineral dust differs from that of black carbon at the same AOD. Cloud contamination is the most persistent limitation across all three sensors: pixels flagged as partly cloudy are excluded, and in persistently hazy regions the cloud mask can incorrectly flag thick aerosol layers as cloud, introducing gaps. Retrieval quality also degrades at high solar zenith angles, meaning high-latitude winter retrievals carry wider uncertainty bounds.
The 10 km problem: what coarse AOD means for site assessment
A utility-scale solar farm occupies perhaps 5 to 20 km². A single MODIS AOD pixel at 10 km covers a comparable area, which sounds adequate. In practice, the mismatch is more awkward. AOD pixels are retrieved only where the surface and atmosphere meet the algorithm's quality criteria; a site on the edge of a salt flat or near a persistent cloud field may have substantial data gaps in the time series. Spatial interpolation fills those gaps but adds uncertainty.
More importantly, AOD varies on sub-pixel scales near industrial point sources, coastal gradients, or orographic channels that funnel dust. A 10 km pixel average can mask a local aerosol enhancement that a ground-based sun photometer would resolve. The AERONET network of sun photometers provides point-source validation at high accuracy (AOD uncertainty below 0.01 in cloud-free conditions) and is the reference against which satellite retrievals are validated, but AERONET stations are sparse in many project regions. Where no nearby AERONET station exists, satellite AOD carries its full retrieval uncertainty without an independent check. Buyers should treat site-level AOD corrections derived solely from satellite data as a probabilistic estimate with explicit uncertainty bounds, not a deterministic correction.
Building a correction: from AOD climatology to adjusted GHI
The practical workflow runs in three stages. First, construct a multi-year monthly AOD climatology for the project location by stacking MODIS MOD04 and VIIRS AERDB retrievals from the available archive (2000 to present for MODIS, 2012 onwards for VIIRS on Suomi-NPP). Quality-flag filtering, spatial averaging within a representative radius, and gap-filling via temporal interpolation produce a 12-month AOD profile with interannual variability statistics.
Second, compare that climatology against the AOD assumptions embedded in the irradiance product being used. CAMS Radiation Service documentation specifies the aerosol input used in each product version; differences between the embedded climatology and the site-specific one drive the correction factor. Third, apply a radiative transfer adjustment, using a model such as the simple parameterisation in the Bird clear-sky model or a full spectral code, to translate the AOD difference into a GHI correction at each hour of the typical meteorological year. The correction is largest at low solar elevation angles, when the optical path through the aerosol layer is longest, and during months of peak dust activity. The output is an adjusted GHI time series with an associated uncertainty band that reflects both the AOD retrieval error and the radiative transfer parameterisation uncertainty.
Honest limits and what ground data still has to do
Satellite AOD correction reduces systematic bias but does not eliminate it. The residual uncertainty in corrected GHI from satellite methods alone is typically cited in the literature at 3 to 6 percent annually for well-instrumented regions, rising to 8 to 12 percent in data-sparse, high-aerosol environments. For project finance purposes, lenders and independent engineers generally require at least one year of on-site pyranometer data to anchor the long-term satellite record. Satellite AOD correction is most powerful as a tool for extending and adjusting that ground record, not replacing it.
Satellize applies this correction workflow as part of irradiance analytics on open constellations including MODIS, VIIRS, and TROPOMI, and has applied comparable multi-source atmospheric correction methods in its Tonga crop-estimation programme. The same radiative transfer principles that govern aerosol-irradiance interaction in agricultural yield modelling apply directly to solar resource assessment. Corrected irradiance outputs are delivered as gridded GeoTIFF layers with accompanying uncertainty rasters, formatted for direct ingestion into yield-modelling tools.
Typical figures
| AOD spatial resolution (MODIS MOD04) | 10 km at nadir |
| AOD spatial resolution (VIIRS AERDB/AERDT) | 6 km at nadir |
| AOD spatial resolution (TROPOMI) | approximately 5.5 km × 3.5 km per pixel |
| Combined MODIS Terra + Aqua overpass frequency | Twice daily at most latitudes |
| TROPOMI revisit | Daily global coverage |
| AOD retrieval uncertainty (dark-target, vegetated surface) | ±0.05 + 15% of retrieved value (1-sigma, MODIS validated against AERONET) |
| AOD retrieval uncertainty (deep-blue, bright surface) | ±0.05 + 15–20% of retrieved value; higher over sand |
| Archive depth (MODIS AOD) | 2000 to near-present |
| CAMS Radiation Service irradiance archive | 2004 to near-real-time, 15-minute resolution |
| Corrected GHI residual uncertainty (well-instrumented regions) | 3–6% annually; 8–12% in high-aerosol, data-sparse regions |
Analytics Satellize can run
| Site AOD climatology (monthly, with interannual variability) | Multi-year stack of MODIS MOD04 and VIIRS AERDB quality-filtered retrievals, gap-filled by temporal interpolation | 12-month AOD profile table with standard deviation and percentile bands, delivered as CSV and GeoTIFF |
| Corrected GHI typical meteorological year | Radiative transfer adjustment (Bird clear-sky parameterisation or equivalent) applied to CAMS-RAD output using site AOD climatology versus embedded AOD assumption | Hourly GHI time series with uncertainty raster, GeoTIFF and CSV formats |
| Dust-event frequency and severity climatology | AOD threshold exceedance analysis (AOD > 0.5 at 550 nm) across MODIS and TROPOMI archive, stratified by month and wind direction | Monthly exceedance probability table and mapped dust-event frequency layer |
| Aerosol type classification for radiative weighting | TROPOMI absorbing aerosol index combined with MODIS Ångström exponent to distinguish mineral dust from carbonaceous aerosol | Seasonal aerosol-type breakdown report informing wavelength-specific irradiance corrections |
| Bias assessment against CAMS Radiation Service embedded AOD | Direct comparison of site-specific satellite AOD climatology against documented CAMS-RAD aerosol inputs, with radiative transfer sensitivity runs | Bias report quantifying GHI overestimation by month, with confidence intervals |
| Uncertainty-bounded P90 yield adjustment | Monte Carlo propagation of AOD retrieval uncertainty and radiative transfer parameterisation error through corrected GHI distribution | Adjusted P50/P90 GHI estimates with explicit uncertainty bounds, formatted for independent engineer review |
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.