Malaria vector habitat mapping from land surface and water data
Sentinel-2, Landsat and MODIS land surface temperature data can locate Anopheles breeding habitat weeks before peak transmission, giving vector-control teams a spatial head start. Cloud cover during the wet season is the hard limit.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with both satellites. Red-edge and SWIR bands allow precise NDVI and NDWI calculation for separating open water from waterlogged vegetation and distinguishing emergent aquatic plants from dry ground.
- Landsat 8/9 OLI: 30 m multispectral resolution, 16-day revisit per satellite (8-day combined). The archive stretches to 1984, enabling long-term surface water seasonality analysis via the JRC Global Surface Water dataset, which classifies permanent, seasonal and ephemeral water bodies globally.
- MODIS Land Surface Temperature (MOD11A1/MOD11A2): 1 km spatial resolution, daily revisit. LST products derived from thermal infrared bands 31 and 32 provide the temperature screening layer: Anopheles larval development stalls below roughly 16°C, so sub-threshold grid cells can be masked from habitat models without field visits.
- JRC Global Surface Water (Landsat archive): 30 m, derived from the full Landsat archive since 1984. Provides monthly water occurrence, seasonality, recurrence and transition layers that characterise whether a water body is newly flooded, perennially present or episodic. Episodic and newly flooded surfaces are the highest-risk class for Anopheles gambiae sensu lato.
Why shallow, warm and slow is the problem
Anopheles mosquitoes, the genus responsible for Plasmodium transmission, are selective breeders. Larvae require shallow water with low turbulence, sufficient sunlight to warm the surface above the developmental threshold of roughly 16°C, and enough aquatic vegetation or organic matter to provide shelter and food. Deep, fast-moving rivers are largely irrelevant. The habitats that matter are rice paddies, seasonal pools, roadside ditches, brick-pit ponds, and the shallow margins of larger water bodies. These are precisely the features that multispectral satellites at 10 to 30 m resolution can resolve.
The transmission cycle has a pronounced seasonal pulse in most endemic regions. Rainfall creates new water surfaces, temperatures rise, and larval populations explode within two to three weeks. Vector-control teams that receive spatial predictions before that pulse can target larviciding or environmental management where it is most needed. Predictions that arrive after peak emergence are operationally useless. The satellite data pipeline therefore has to be fast as well as accurate.
What NDVI seasonality reveals about breeding risk
Vegetation indices do two jobs in a malaria habitat model. First, NDVI derived from Sentinel-2 red and near-infrared bands tracks the phenological cycle of irrigated agriculture and seasonal wetlands. A sharp NDVI increase following the first rains indicates newly inundated or waterlogged ground, which is a leading indicator of larval habitat creation. Second, NDWI (Normalised Difference Water Index, using green and near-infrared bands) and MNDWI (using green and SWIR) discriminate open water from wet soil and from submerged vegetation, each of which carries different larval density expectations.
Time-series analysis across multiple Sentinel-2 acquisitions can characterise whether a pixel is consistently wet, intermittently flooded, or dry. Consistently flooded pixels often support predator communities that suppress larval survival. Intermittently flooded pixels, particularly those that dry and refill, tend to produce the highest adult emergence rates. That distinction, invisible to a single-date image, becomes clear across a 60-day stack. Landsat's longer archive allows the same analysis to be extended back decades, which matters when assessing whether a newly irrigated scheme has created persistent new habitat.
Combining water extent, temperature and species distribution models
The standard analytical workflow layers three inputs: surface water extent (from JRC Global Surface Water or a freshly computed NDWI mask), land surface temperature from MODIS MOD11, and NDVI seasonality metrics. These become predictor variables in a species distribution model, typically MaxEnt or a random forest classifier, calibrated against entomological field survey data recording where adult Anopheles were caught and at what densities. The model learns which combinations of water permanence, temperature and vegetation structure correspond to high larval productivity in a given landscape.
Calibration data quality is the limiting factor. A model trained on 50 well-georeferenced trap locations will outperform one trained on 500 imprecisely recorded catches. Satellite inputs are only as useful as the ground truth they are matched against. Published studies from the Sahel and East Africa have demonstrated reasonable predictive accuracy at 1 to 2 km spatial scales using this approach, though performance degrades in landscapes with high micro-habitat heterogeneity, such as dense urban areas where small containers dominate over natural water bodies.
The wet-season cloud problem is not solvable from optical data alone
The principal limitation of this entire approach is cloud cover, and it is worth being direct about the scale of the problem. In West and Central Africa, cloud fraction over the Guinea coast exceeds 80 percent for months at a time during the rainy season. This is exactly when new breeding habitats are forming and when a spatial prediction would have the highest operational value. Sentinel-2 and Landsat are both passive optical sensors; they see nothing through cloud.
Partial mitigations exist. Synthetic aperture radar, particularly Sentinel-1 C-band SAR, can detect surface water through cloud and is covered in a sibling page on flood-driven disease risk. MODIS LST, despite its coarse 1 km resolution, has a daily revisit that increases the probability of cloud-free thermal observations within any given week. Gap-filling methods such as harmonic time-series fitting can interpolate NDVI through short cloudy gaps, but they cannot reconstruct the actual state of a newly flooded field that was never observed. The honest position is that satellite-based habitat mapping is most reliable in the dry-to-wet transition period and in drier endemic zones such as the Sahel and southern Africa, where cloud cover is less persistent.
From prediction layer to vector-control operation
A habitat suitability map is not itself an operational product. It becomes one when combined with population data, road-network accessibility and the logistical constraints of a national malaria programme. Prioritisation layers that rank predicted high-risk habitat cells by the number of people within a given radius, and by the time required for a spray team to reach them, convert a scientific output into a field schedule.
Satellize runs this kind of analytics workflow on open Sentinel and Landsat data, with commercial tasking added where finer resolution or faster revisit is needed. The Tonga crop-estimation programme demonstrates the same pattern of combining open-archive time series with in-country calibration data to produce operationally specific outputs rather than generic risk maps. National malaria programmes or regional health authorities wanting to scope a habitat-mapping engagement should start by sharing existing entomological survey records, which determine whether calibration data are sufficient to support a model rather than a simple index map.
What the numbers actually look like
At 10 m resolution, Sentinel-2 can resolve a rice paddy field or a roadside ditch of roughly 30 m width, though a single ditch pixel will be spectrally mixed with surrounding soil or vegetation. Reliable water body detection at 10 m requires a minimum surface area of around 100 to 200 square metres under good atmospheric conditions. MODIS LST at 1 km cannot resolve individual habitat patches but provides the temperature screening layer that eliminates large areas from consideration at low computational cost.
JRC Global Surface Water records monthly water occurrence back to March 1984 at 30 m resolution, giving nearly 40 years of seasonality data. This archive depth is genuinely unusual and allows analysts to distinguish whether a habitat is a structural feature of the landscape or a consequence of recent land-use change such as irrigation expansion. Revisit latency for Sentinel-2 analysis-ready data is typically two to five days after acquisition via the Copernicus Data Space, which is fast enough to support fortnightly prediction updates during the transmission season.
Typical figures
| Finest spatial resolution (optical) | 10 m (Sentinel-2 MSI visible/NIR bands) |
| Landsat water mapping resolution | 30 m (OLI, used in JRC Global Surface Water) |
| LST resolution (MODIS MOD11) | 1 km; daily revisit increases cloud-gap probability |
| Revisit (Sentinel-2 combined) | ~5 days at equator; longer in high-latitude regions |
| Revisit (Landsat 8+9 combined) | ~8 days at equator |
| Minimum detectable water body (Sentinel-2) | ~100–200 m² under clear sky; larger under haze |
| Temperature screening threshold | ~16°C LST lower limit for Anopheles larval development |
| Archive depth (JRC Global Surface Water) | March 1984 to present (Landsat archive) |
| Principal data gap | Cloud cover >80% in humid tropics during wet season; optical sensors blind |
| Analysis-ready data latency | 2–5 days post-acquisition via Copernicus Data Space |
Analytics Satellize can run
| Seasonal habitat suitability map | NDWI/MNDWI water masking from Sentinel-2 time series, combined with MODIS LST threshold filter and JRC water seasonality class | GeoTIFF or vector polygon layer, updated fortnightly during transmission season, classified by habitat persistence class |
| Species distribution model output | MaxEnt or random forest trained on entomological field survey records with satellite-derived predictor stack (NDVI seasonality metrics, water occurrence, LST mean and variance) | Gridded probability surface (GeoTIFF) at 30–100 m resolution with confidence intervals, delivered as GIS layer |
| New habitat emergence alert | Change detection on NDWI time series; pixels crossing water threshold for the first time in a rolling 30-day window flagged as newly inundated | Weekly alert shapefile or API feed, with pixel count and estimated area per administrative unit |
| Prioritised intervention schedule | Overlay of habitat suitability surface with gridded population (WorldPop or equivalent) and road-network accessibility analysis | Ranked list of target sites with estimated population at risk and travel-time from nearest health post, in PDF report and CSV |
| Multi-year habitat trend analysis | Harmonic regression on Landsat NDVI and JRC water occurrence time series to detect land-use-driven habitat expansion or contraction since 1984 | Annual trend map and summary statistics table for programme-area administrative units |
| Temperature suitability calendar | MODIS MOD11 LST climatology aggregated by dekad; days per month above 16°C and below 35°C (upper larval tolerance) mapped per grid cell | Monthly raster stack and tabular summary by district, formatted for import into malaria transmission models |
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.