Dust storm detection and health-warning forecasting

Mineral dust plumes from the Sahara, Arabian Peninsula and Gobi routinely push PM10 past WHO thresholds within hours of lofting. Satellites detect the plume; models translate it into surface concentrations and 72-hour health warnings.

Sensors

Sentinel-5P TROPOMI: Absorbing Aerosol Index (AAI) at 3.5 × 5.5 km nadir resolution, daily global coverage. AAI distinguishes UV-absorbing aerosols (dust, smoke) from non-absorbing types; values above roughly 1–2 are considered significant dust loading, though the threshold is scene-dependent.
MODIS Deep Blue AOD (Terra and Aqua): Aerosol Optical Depth retrieval specifically designed for bright desert surfaces where the standard Dark Target algorithm fails. 10 km resolution at nadir, twice-daily combined revisit. Provides columnar aerosol load but not surface concentration directly.
Meteosat SEVIRI (MSG): Spinning Enhanced Visible and InfraRed Imager aboard Meteosat Second Generation. 15-minute repeat over Africa, Europe and the Middle East at roughly 3 km resolution in the thermal infrared. The Dust RGB composite (BTD 12.0–10.8 µm, BTD 10.8–8.7 µm, 10.8 µm brightness temperature) identifies lofted mineral dust as a characteristic pink-magenta signature, day and night.
VIIRS (Suomi-NPP / NOAA-20): Deep Blue AOD product comparable to MODIS, plus a 375 m resolution I-band useful for resolving plume boundaries. Daily global coverage; NOAA-20 and Suomi-NPP together provide two overpasses per day.
Sentinel-3 OLCI / SLSTR: 300 m ocean and land colour instrument provides aerosol retrievals over ocean and coastal zones. SLSTR dual-view geometry aids aerosol characterisation. Revisit roughly every two days at mid-latitudes, daily near the poles.

What the satellite actually sees, and what it does not

Every satellite aerosol product measures a column-integrated quantity: the total scattering or absorption through the full atmospheric depth above the surface. TROPOMI's Absorbing Aerosol Index is a unitless measure derived from the ratio of observed to modelled UV radiance at 340 and 380 nm. A high AAI tells you that UV-absorbing particles are present somewhere in the column. It does not tell you how high they are, how thick the layer is in grams per square metre, or what the PM10 concentration is at street level.

That last step requires a chemical transport model. Systems such as ECMWF's CAMS (Copernicus Atmosphere Monitoring Service) ingest the satellite retrievals as constraints and then integrate meteorological fields, emission inventories and particle settling physics to estimate surface mass concentration. The satellite provides the spatial fingerprint and the timing; the model provides the health-relevant number. Treating a satellite AOD or AAI value as a direct PM10 reading is one of the more common and consequential errors in applied air-quality work.

Three source regions, three detection challenges

The Sahara is the world's largest dust source by emission flux, contributing an estimated 400–700 Tg per year to the global aerosol budget (a range reflecting genuine uncertainty in emission models, not sloppy accounting). Plumes regularly cross the Mediterranean within 24–48 hours of lofting and reach the Caribbean within five to seven days. SEVIRI's 15-minute cadence makes it the operational workhorse for early detection over North Africa and Europe; the Dust RGB composite is interpretable in near-real time without algorithm processing.

The Arabian Peninsula presents a harder retrieval environment. Surface albedo in the 8–12 µm thermal infrared window is highly variable across sand, gravel and rocky desert, which complicates the brightness-temperature-difference (BTD) dust signal. The BTD method exploits the fact that mineral dust has a negative BTD at 10.8–8.7 µm (dust is colder in the 10.8 µm window than in the 8.7 µm window), but over hot, bright surfaces this contrast can be suppressed or reversed. False-alarm rates over the Rub' al Khali and similar terrain are non-trivial.

The Gobi and Taklimakan deserts in Central Asia feed the East Asian dust pathway, which affects China, the Korean Peninsula and Japan. Spring episodes can push PM10 above 1,000 µg/m³ in downwind cities, far above the WHO 24-hour guideline of 45 µg/m³ set in the 2021 guidelines. MODIS Deep Blue and VIIRS perform better here than over the Arabian Peninsula because surface albedo in the visible is lower over Gobi gravel plains than over bright sand seas.

The false-alarm problem over bright surfaces

Arid surfaces reflect strongly in the shortwave, which compresses the contrast between dusty and clear-sky scenes in visible and near-infrared bands. The Deep Blue algorithm was designed specifically to work over these surfaces by using blue-channel reflectance (around 412 nm) where the surface is relatively dark even over bright desert. It works well but carries higher retrieval uncertainty than over ocean or vegetated land, typically ±0.03 + 30% of AOD rather than the ±0.05 + 15% uncertainty quoted for Dark Target over ocean.

Thin cirrus clouds and the Saharan Air Layer, a warm, dry, dust-laden layer that sits between roughly 1.5 and 5 km altitude, can both produce AAI signals that resemble fresh surface-lofted dust. Operationally, analysts cross-reference multiple products: AAI for absorbing character, BTD for lofted mineral dust specifically, AOD for column loading, and back-trajectory analysis to confirm a plausible source region. No single product is sufficient.

From detection to a 72-hour health warning

The operational chain has four steps. First, satellite products flag a plume and characterise its initial AOD or AAI footprint. Second, a chemical transport model (CAMS, NAAPS, or a national equivalent) assimilates the satellite data and runs a forward simulation, typically to 72 hours, producing gridded PM10 and PM2.5 fields at hourly intervals. Third, those fields are intersected with population grids and health-vulnerability indices (age distribution, prevalence of asthma or COPD, access to healthcare) to identify populations at elevated risk. Fourth, an alert is issued through whatever public-health communication channel the client operates.

The 72-hour window is meaningful because mineral dust at Saharan lofting altitudes of 2–5 km travels roughly 800–1,200 km per day under typical trade-wind conditions. That gives downwind health authorities one to three days of lead time, enough to pre-position medication, issue advisory messages and, in some systems, trigger automatic air-quality index escalations. Beyond 72 hours, transport model uncertainty grows substantially; 120-hour dust forecasts carry wide confidence intervals and should be used only for contingency planning, not public alerts.

Honest limits and what they mean for a health-warning system

Cloud cover is a genuine problem. Thick convective cloud over the Sahel during the monsoon transition season can obscure dust underneath, and the BTD method cannot see through cloud. SEVIRI's 15-minute repeat helps by catching cloud-free windows, but a persistent cloud deck can create a detection gap of six hours or more. Thermal infrared dust detection is also limited to lofted dust; dust within the planetary boundary layer, below roughly 1–2 km, is often invisible to the BTD method because the temperature contrast with the surface is too small.

Surface PM10 estimates derived from satellite AOD carry an additional uncertainty layer: the aerosol vertical profile. If dust is concentrated at 4 km altitude and not mixing down to the surface, a high AOD does not translate to a high surface concentration. Conversely, a shallow, dense dust layer near the surface can produce a moderate AOD while delivering a severe PM10 episode at ground level. Radiosonde profiles, lidar ceilometers and surface monitoring networks remain essential ground-truth for any operationally credible health-warning system.

Satellize runs dust-detection analytics on TROPOMI, MODIS Deep Blue and SEVIRI inputs, combining them with CAMS forecast fields to produce gridded 72-hour PM10 exceedance probability maps. The methodology is the same class used in published operational systems; the honest caveat is that surface concentration uncertainty is model-dependent and should always be presented with a confidence range rather than a point estimate.

What a government health ministry actually needs from this chain

The most useful deliverable is not a pretty dust-plume image. It is a time-stamped, population-weighted exceedance probability: the fraction of a district's population likely to experience PM10 above a defined threshold (say, 50 µg/m³ or 150 µg/m³) in each six-hour window over the next three days. That framing lets a health ministry make a tiered decision: issue a general advisory, activate hospital surge protocols, or do nothing beyond monitoring.

Calibration against historical ground-station data is critical. A ministry that runs a network of even ten to twenty PM10 monitors can use two to three years of co-located satellite-model output to build a bias-correction function specific to its geography. That step alone can reduce systematic error in surface concentration estimates by 20–40%, based on published validation studies for the CAMS dust product over North Africa and the Middle East. Without it, the satellite chain is a qualitative early-warning tool. With it, it becomes a quantitative one.

Typical figures

AAI spatial resolution (TROPOMI)	3.5 × 5.5 km at nadir (upgraded from 7 × 3.5 km in August 2019)
AOD spatial resolution (MODIS Deep Blue)	10 km at nadir
SEVIRI repeat cycle	15 minutes over Africa, Europe and Middle East (rapid-scan mode: 5 minutes over Europe)
SEVIRI spatial resolution	3 km at sub-satellite point in thermal infrared; degrades to ~5 km at 50° latitude
Typical AAI detection threshold	AAI > ~1–2 (scene-dependent; values below 1 are unreliable over bright surfaces)
MODIS Deep Blue AOD uncertainty	±0.03 + 30% of retrieved AOD over bright land surfaces
Transport forecast horizon	72 hours operationally reliable; 120 hours for contingency planning only
Satellite data latency (TROPOMI L2)	Near-real-time product: ~3 hours after overpass; offline product: ~1–2 days
Archive depth	TROPOMI from October 2017; MODIS from 2000; SEVIRI from 2004
WHO PM10 24-hour guideline (2021)	45 µg/m³; Saharan episodes routinely exceed 200–500 µg/m³ at receptor sites

Analytics Satellize can run

Real-time dust plume detection layer	Multi-sensor fusion of TROPOMI AAI, MODIS Deep Blue AOD and SEVIRI Dust RGB; thresholding and spatial continuity filtering	GeoTIFF or GeoJSON polygon layer, updated every 15 minutes (SEVIRI) and at each TROPOMI/MODIS overpass; delivered via API or SFTP
72-hour PM10 exceedance probability grid	CAMS ensemble dust forecast, bias-corrected against client ground-station network where available; population-weighted exceedance probability at 0.1° grid	Gridded NetCDF with six-hourly time steps; summary PDF report for non-specialist health officials
Population-at-risk time series	Intersection of PM10 exceedance grid with WorldPop or national census population raster; optional vulnerability weighting by age or COPD prevalence	District-level CSV with estimated exposed population per threshold tier per six-hour window; updated at each forecast cycle
Historical dust episode catalogue	Retrospective processing of MODIS Deep Blue and TROPOMI archive; episode identification by spatial extent, peak AOD and duration; source-region attribution via HYSPLIT back-trajectory	Structured database of episodes by date, source region, peak AOD, receptor zone and estimated PM10; exportable as GeoPackage or CSV
Automated health-alert trigger	Rule-based threshold logic applied to forecast PM10 fields; configurable thresholds by district and population tier; optional integration with national early-warning systems	Structured JSON alert payload for ingestion by client communication platforms; audit log of all triggers and lead times
Satellite-model bias characterisation report	Co-location of MODIS/TROPOMI retrievals with client PM10 monitor time series; regression and error decomposition following published CAMS validation methodology	One-off technical report with bias-correction coefficients and uncertainty bounds; recommended as a prerequisite before operational deployment

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.