Dengue transmission risk mapping using urban heat and greenness
Urban heat islands and patchy vegetation create the thermal and habitat conditions that Aedes mosquitoes exploit. Satellite-derived land surface temperature, greenness and rainfall anomalies can map where those conditions converge weekly, giving health authorities a spatial head start before case counts rise.
Sensors
- ECOSTRESS (ISS-mounted): Thermal infrared radiometer delivering land surface temperature at approximately 70 m spatial resolution. Revisit is irregular, roughly 3 to 5 days at mid-latitudes, because the ISS orbit precesses rather than following a fixed sun-synchronous ground track. Acquires at multiple local times, which matters for capturing peak afternoon heat in urban canyons.
- Sentinel-2 MSI: Multispectral imager at 10 m (visible and near-infrared) and 20 m (red-edge, SWIR) resolution. The red and near-infrared bands produce NDVI at 10 m, sufficient to resolve individual tree crowns and small garden plots. Five-day revisit at the equator with both satellites combined, though cloud cover in humid tropical cities routinely reduces effective clear-sky frequency to once every two to three weeks.
- MODIS Land Surface Temperature (Terra/Aqua): Daily LST composites at 1 km resolution from the MOD11A1 and MYD11A1 products. Coarser than ECOSTRESS but consistent daily coverage since 2000, making it the backbone of long-term thermal climatology and anomaly detection. Useful for establishing the baseline against which ECOSTRESS anomalies are assessed.
- GPM IMERG: Global Precipitation Measurement Integrated Multi-satellitE Retrievals, providing near-real-time precipitation estimates at 0.1-degree resolution and 30-minute temporal steps. Rainfall accumulation over the preceding 10 to 14 days is the primary driver of container-water availability and therefore larval habitat creation. IMERG latency for the early run is approximately 4 hours.
Why temperature and greenness are the wrong variables to ignore
Aedes aegypti and Aedes albopictus are container breeders. They do not need rivers or swamps. They need small volumes of standing water, warm air to accelerate larval development, and enough shade to keep adult mosquitoes active through the day. In a city, all three conditions are set by the urban fabric itself.
Land surface temperature governs the extrinsic incubation period of dengue virus inside the mosquito. Laboratory studies show that the period shortens sharply above roughly 25 degrees Celsius and that vector competence peaks in the 28 to 32 degree range. Urban heat islands routinely push surface temperatures 3 to 8 degrees above surrounding rural land, compressing incubation and extending the transmission season. Satellite LST does not measure air temperature directly, but the correlation between daytime LST and near-surface air temperature in dense urban areas is well established in the published literature, and LST anomalies are a practical proxy for identifying the hottest urban micro-zones.
What ECOSTRESS adds that MODIS cannot
At 1 km, MODIS LST averages across entire city blocks. A single pixel might contain a hospital car park, a market roof, a pocket park and a row of shophouses. The thermal signal is blurred into meaninglessness at the neighbourhood scale where vector control teams actually operate.
ECOSTRESS at 70 m resolves individual blocks and, in many cases, individual buildings. That is the scale at which impervious surface fraction, shaded courtyards and rooftop water storage vary. The irregular revisit is a genuine constraint: you cannot guarantee a cloud-free ECOSTRESS acquisition on any given day in a humid tropical city. The practical approach is to fuse ECOSTRESS acquisitions, when available, with the MODIS daily thermal climatology to produce a downscaled LST estimate, using Sentinel-2 impervious surface fraction and NDVI as the spatial disaggregation covariates. This is a published method class, not a proprietary trick.
Rainfall anomalies close the habitat loop
Temperature and greenness define suitability. Rainfall determines whether suitable habitat is actually wet. GPM IMERG 10-day accumulation anomalies, expressed as departure from the long-term mean for the same calendar period, identify weeks when container-filling rainfall has exceeded normal levels. When an above-normal rainfall anomaly coincides with elevated LST and moderate NDVI, the three conditions for productive larval habitat are simultaneously present.
The combination produces a weekly risk surface: a gridded map at roughly 70 to 100 m resolution where each cell carries a score reflecting how far current thermal, vegetative and rainfall conditions depart from the long-term baseline. High-scoring cells are not confirmed breeding sites. They are places where conditions are favourable enough to warrant ground inspection.
The honest gap between suitability and incidence
This is the point where the model has to be honest with the people buying it. Vector suitability is not case incidence. A neighbourhood can score high on every environmental indicator and still have low transmission if the local population has recent immunity, if vector control is effective, or if the dominant serotype is not circulating that season. Conversely, a neighbourhood with moderate environmental scores can produce a cluster if a single viraemic traveller arrives and the local Aedes population is dense.
The satellite model predicts where the physical environment is permissive. It does not predict who gets sick. To be operationally useful, the risk surface must be calibrated against historical case data from the local epidemiological surveillance system. Without that ground truth, the output is a hypothesis map, not an operational alert. Health authorities that have geo-coded case records going back two or more dengue seasons can fit a logistic or Poisson regression linking environmental scores to observed incidence, producing a calibrated probability surface. Those that do not have geo-coded records are working with a weaker product, and they should know that before commissioning the analysis.
Cloud cover is the other persistent limit. In cities near the equator, Sentinel-2 may deliver fewer than ten cloud-free acquisitions per month during the wet season, which is precisely when dengue risk is highest. Gap-filling with Landsat 8 and 9 data (also 10 to 30 m, 16-day revisit) helps, but does not eliminate the problem.
Building the weekly product: data flow and outputs
The operational pipeline ingests four streams: ECOSTRESS LST when a cloud-free acquisition is available, MODIS daily LST for gap-filling and anomaly baseline, Sentinel-2 NDVI composited over the preceding 30 days to smooth cloud gaps, and GPM IMERG 10-day precipitation anomaly. Each stream is resampled to a common 100 m grid. A risk score is computed per cell, typically as a weighted combination of normalised anomalies, with weights derived from the calibration regression if case data exist or from published literature values if they do not.
The deliverable is a GIS layer updated weekly, with an accompanying summary table ranking administrative sub-units by mean risk score. Vector control teams can overlay this with their own field-inspection records to prioritise which blocks to visit first. The layer is not a substitute for entomological surveillance; it is a triage tool that makes limited field capacity go further.
Satellize runs this pipeline on open constellations and can add commercial high-resolution tasking where a client needs sub-50 m thermal data for a specific district. The Tonga crop-estimation programme demonstrated the same fusion logic, combining open and commercial imagery into a single calibrated output, and the same architecture applies here.
What a health authority needs before starting
Three things determine whether the output is worth having. First, a city boundary and administrative subdivision shapefile at the neighbourhood or ward level, so outputs align with how the health authority reports and acts. Second, at least one dengue season of geo-coded case data, ideally at the sub-district level, for calibration. Third, a point of contact in the vector control programme who understands that the satellite product will generate inspection hypotheses, not confirmed outbreaks.
If the case data do not exist or cannot be shared, the uncalibrated suitability surface is still useful for long-term urban planning, identifying which neighbourhoods to prioritise for drainage infrastructure or tree-planting programmes. But the health authority should not use it as a real-time alert system. That distinction matters, and it should be written into any service agreement from the start.
Typical figures
| Primary LST resolution | ~70 m (ECOSTRESS); 1 km (MODIS MOD11A1/MYD11A1) |
| Vegetation index resolution | 10 m NDVI from Sentinel-2 MSI bands B04 and B08 |
| ECOSTRESS revisit | Irregular, approximately 3 to 5 days at tropical latitudes; not guaranteed cloud-free |
| Sentinel-2 revisit | 5 days (both satellites); effective cloud-free revisit in humid tropics typically 10 to 30 days |
| Precipitation product | GPM IMERG Early Run at 0.1° / 30-minute; ~4-hour latency |
| Output grid resolution | 100 m (fused product); degraded to 250 m where ECOSTRESS is unavailable |
| Risk surface update cadence | Weekly, subject to input data availability |
| MODIS LST archive depth | Terra from March 2000; Aqua from July 2002 |
| Sentinel-2 archive depth | From June 2015 (Sentinel-2A); global systematic coverage from 2017 |
| Delivery format | GeoTIFF risk layer, GeoPackage administrative summary, optional WMS tile feed |
Analytics Satellize can run
| Weekly urban vector suitability surface | Weighted anomaly fusion of downscaled LST, NDVI and GPM IMERG 10-day precipitation departure from long-term mean | GeoTIFF layer at 100 m, updated weekly, with per-cell risk score 0 to 1 |
| Calibrated dengue risk probability surface | Logistic regression of environmental suitability scores against geo-coded historical case data, following published statistical vector-habitat modelling approaches | Calibrated probability GeoTIFF with confidence intervals; requires client-supplied case records |
| Administrative-unit risk ranking table | Zonal statistics aggregating cell-level scores to ward or sub-district boundaries | Weekly CSV and PDF summary ranking neighbourhoods by mean and peak risk score |
| Seasonal thermal anomaly trend report | MODIS LST time-series analysis against 20-year baseline to characterise urban heat island intensification and its dengue-season implications | Annual PDF report with maps and trend charts for urban planning and public health strategy teams |
| Impervious surface and greenness change detection | Sentinel-2 NDVI and NDBI (Normalised Difference Built-up Index) bi-annual differencing to track urban expansion into previously vegetated areas | Change-detection GeoTIFF and summary table flagging newly urbanised cells for prospective surveillance |
| Rainfall-triggered habitat alert | GPM IMERG threshold exceedance: alert triggered when 10-day accumulation exceeds the 75th percentile of the historical distribution for that calendar week in a high-suitability zone | Email or API alert with map attachment within 6 hours of IMERG early-run availability |
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.