Sensors
- Sentinel-1 SAR (C-band, 5.405 GHz): Interferometric Wide Swath mode delivers 10 m ground range resolution with a 250 km swath, revisiting most of sub-Saharan Africa every 6–12 days. C-band backscatter drops sharply over open water, making inundation mapping reliable even under cloud cover, which is ubiquitous during the wet seasons that drive RVF risk.
- MODIS Terra/Aqua (bands 1–7): 250 m red and near-infrared bands underpin NDVI at daily revisit, enabling anomaly detection against multi-year baselines. The 500 m land surface temperature product adds an independent indicator of evaporative cooling from standing water. Latency for standard products is roughly 1–2 days from overpass.
- GPM IMERG (Global Precipitation Measurement): Half-hourly, 0.1-degree (~11 km) global rainfall estimates from the GPM constellation, with a Final Run latency of about 3.5 months and an Early Run latency under 6 hours. IMERG provides the rainfall forcing that precedes inundation, allowing risk scoring to begin before flooding is visible in SAR imagery.
- Sentinel-2 MSI: 10 m visible and near-infrared bands at 5-day revisit (two-satellite constellation) resolve individual dambo boundaries and distinguish open water from wet soil or dense vegetation far more precisely than MODIS. Cloud cover is the binding constraint in equatorial wet seasons; Sentinel-2 is most useful for post-storm confirmation when skies clear.
Why a flooding event is also a disease event
Rift Valley fever virus circulates silently in drought-resistant Aedes eggs that can remain viable in soil for years. When rainfall is anomalously heavy and low-lying dambos, pans and seasonal wetlands flood, those eggs hatch in synchrony. The resulting mosquito population surge is large enough to amplify virus through livestock herds within days, and human exposure follows.
The epidemiological link between rainfall anomalies and RVF outbreaks is well established. The 1997–98 outbreak across East Africa, associated with El Niño-driven flooding, killed hundreds of people and caused enormous livestock losses across Kenya, Somalia and Tanzania. The 2000 outbreak on the Arabian Peninsula followed a similar pattern. In each case, the environmental signal preceded confirmed cases by weeks. That gap is the window satellite analytics is designed to exploit.
What the sensors actually measure, and what they do not
Sentinel-1 detects inundation through a change in C-band backscatter: open water returns very little energy to the sensor, so flooded ground appears as a dark anomaly against the surrounding landscape. The method works under cloud and at night, which matters enormously in tropical wet seasons. At 10 m resolution it can resolve individual dambo pools large enough to support breeding populations. Smaller ephemeral puddles, perhaps a few metres across, fall below the detection floor.
MODIS NDVI anomalies capture a different signal: the rapid green flush that follows flooding and saturated soils. Comparing current NDVI against a multi-year baseline for the same calendar week isolates the anomalous vegetation response. This is a proxy, not a direct vector count. A strong positive NDVI anomaly in a historically RVF-prone landscape is a flag, not a confirmation.
GPM IMERG provides the temporal trigger. Rainfall accumulation thresholds derived from historical outbreak data, typically several hundred millimetres over a few weeks in susceptible zones, can initiate a monitoring protocol before any flooding is visible in SAR. The combination of IMERG onset, Sentinel-1 inundation extent and MODIS green-up anomaly gives three independent lines of evidence pointing at the same risk window.
The calibration problem the model cannot solve on its own
Satellite data identifies where and when the environmental conditions for vector amplification are present. It does not measure mosquito abundance, and it cannot detect whether RVF virus is circulating in the local Aedes population. Both of those require ground-truth entomological surveillance: light-trap catches, oviposition indices and serological screening of sentinel livestock.
This is not a minor caveat. Two landscapes with identical inundation and NDVI anomaly scores can carry very different vector densities depending on soil type, vegetation structure, historical land use and local Aedes species composition. The satellite model needs to be calibrated against historical trap data from the specific region before its risk thresholds mean anything actionable. Regions with sparse entomological records, which includes most of the areas at highest RVF risk, require investment in ground surveys before the satellite layer can be trusted for operational alerts.
Viral introduction is a separate problem entirely. Even a perfectly calibrated vector-abundance model cannot predict whether infected animals or wind-dispersed infected mosquitoes will arrive in a newly flooded area. The satellite component is a necessary but not sufficient part of an early-warning system.
Lead time, latency and the operational alert window
The 2–4 week lead time cited in RVF early-warning literature refers to the gap between detectable environmental anomaly and first confirmed human or livestock cases. In practice, the usable window depends on how quickly satellite products are processed and delivered. Sentinel-1 Level-1 data is typically available within 1–3 hours of acquisition through the Copernicus Data Space. MODIS near-real-time products through NASA FIRMS and LANCE run at latencies under 3 hours. GPM IMERG Early Run is available within 6 hours.
The bottleneck is usually not data latency but analysis latency: converting raw imagery into a calibrated risk map, reviewing it against historical outbreak locations and issuing a graded alert to veterinary and public health authorities. Automating that pipeline, and agreeing alert thresholds with national health ministries in advance, determines whether the satellite lead time translates into actual preparedness.
Building a risk layer that health ministries can actually use
A practical RVF risk product combines three spatial layers: a historical susceptibility map derived from past outbreak locations and landscape classification, a near-real-time anomaly score from the current season's IMERG, SAR and NDVI data, and an administrative overlay that assigns risk scores to districts or livestock zones where veterinary response capacity is known.
Sentinel-2 at 10 m adds value for priority sub-districts once a seasonal alert is triggered, resolving which specific dambos are inundated and estimating their surface area. Livestock density data from published global datasets such as the FAO Gridded Livestock of the World series can weight the risk score by the number of amplifying hosts present.
Satellize runs this class of multi-sensor anomaly analysis on open Copernicus and NASA constellations, with the same analytical architecture used in its Tonga crop-estimation programme applied here to phenological and hydrological anomaly detection rather than yield forecasting. The deliverable for a health ministry is a district-level risk bulletin, updated on each Sentinel-1 acquisition cycle, with confidence intervals that are explicit about where ground-truth calibration is thin.
Honest limits and the regions where uncertainty is highest
Dense forest canopy attenuates C-band SAR and can mask shallow inundation beneath it. In woodland savanna with moderate canopy cover, inundation extent is likely underestimated. Persistent cloud in equatorial wet seasons limits Sentinel-2 optical confirmation to intermittent windows. MODIS at 250–500 m cannot resolve the fine-scale heterogeneity of dambo hydrology, particularly in fragmented agricultural landscapes.
The Arabian Peninsula presents a different challenge: sparse historical outbreak data, limited entomological baseline records and highly variable terrain mean that the environmental thresholds derived from East African outbreaks may not transfer directly. Any deployment in a new region requires a validation phase before operational alerts are issued. Acknowledging that openly is part of designing a system that health authorities will trust when it matters.
Typical figures
| SAR spatial resolution (Sentinel-1 IW mode) | 10 m ground range, 250 km swath |
| SAR revisit (Sentinel-1 A+B over Africa) | 6–12 days depending on latitude and orbit overlap |
| NDVI anomaly resolution (MODIS) | 250 m (bands 1–2); daily revisit, 1–2 day product latency |
| Rainfall forcing resolution (GPM IMERG) | ~11 km (0.1°); half-hourly; Early Run latency < 6 hours |
| Optical confirmation resolution (Sentinel-2) | 10 m (NIR/red); 5-day revisit; cloud-dependent availability |
| Minimum detectable inundation patch (Sentinel-1) | Approximately 0.01 ha open water; sub-canopy inundation not reliably detected |
| NDVI anomaly baseline depth (MODIS archive) | Terra operational since 2000; Aqua since 2002; > 20-year climatology available |
| Sentinel-1 archive depth over Africa | From 2014 (Sentinel-1A launch); systematic African coverage from 2015 |
| Alert latency (end-to-end pipeline target) | < 24 hours from SAR acquisition to district risk bulletin, subject to processing configuration |
| Geographic coverage | Sub-Saharan Africa and Arabian Peninsula; global Sentinel-1 and MODIS coverage available |
Analytics Satellize can run
| Seasonal inundation extent map | Sentinel-1 C-band backscatter change detection against dry-season reference; thresholding and morphological filtering to remove speckle artefacts | GeoTIFF or vector polygon layer of flooded area per acquisition, delivered to GIS or web dashboard |
| NDVI anomaly score by district | Z-score of current 8-day MODIS NDVI composite against 20-year calendar-week baseline; spatially aggregated to administrative boundaries | District-level anomaly table and raster, updated every 8 days |
| Rainfall accumulation alert | GPM IMERG rolling 30-day accumulation compared against historically derived outbreak-precursor thresholds for each agro-ecological zone | Automated threshold-crossing alert with zone identifier, issued within 6 hours of IMERG Early Run availability |
| Composite RVF environmental risk score | Weighted combination of inundation anomaly, NDVI anomaly and rainfall accumulation; weights calibrated against historical outbreak records where available; Bayesian uncertainty propagation where ground truth is sparse | Weekly district-level risk bulletin (PDF and GIS layer) with confidence tier (calibrated / uncalibrated) |
| Priority dambo delineation | Sentinel-2 10 m NDWI and NIR-based water-body extraction triggered when composite risk score exceeds defined threshold; intersection with livestock density grids | Named dambo polygons with estimated surface area and livestock-exposure weighting, for veterinary field team dispatch |
| Multi-season trend and anomaly archive | Time-series analysis of Sentinel-1 and MODIS archives from 2015 and 2000 respectively; identification of historically recurrent high-risk zones and inter-annual variability | Static reference report and historical risk raster stack for use in national contingency planning |
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.