Visceral leishmaniasis sandfly habitat mapping from land surface data
Phlebotomine sandflies, vectors of visceral leishmaniasis, occupy micro-climatic niches detectable through MODIS land surface temperature, Sentinel-2 bare-soil fraction and SMAP soil moisture. Combined, these layers produce habitat suitability surfaces used to prioritise indoor residual spraying campaigns across South Asia and East Africa.
Sensors
- MODIS Terra and Aqua (LST): Provides land surface temperature at 1 km spatial resolution with twice-daily overpasses (Terra ~10:30 local, Aqua ~13:30 local). The diurnal temperature range derived from MOD11A1 and MYD11A1 products is a published proxy for soil thermal inertia, which correlates with the sandy, low-moisture substrates favoured by Phlebotomus argentipes in Bihar and P. orientalis in East Africa.
- Sentinel-2 MSI: Ten-metre multispectral imagery with a 5-day revisit at the equator (combined Sentinel-2A and 2B). Bare-soil fraction is extracted via spectral unmixing using the shortwave-infrared bands (Band 11 at 1610 nm, Band 12 at 2190 nm) and the red-edge bands. Seasonal composites isolate periods when vegetation is absent and soil surface conditions are most diagnostic.
- SMAP (NASA Soil Moisture Active Passive): L-band (1.4 GHz) passive radiometer retrieving volumetric soil moisture at 36 km effective resolution with a 2-3 day global revisit. The coarse footprint is the principal constraint in heterogeneous agricultural landscapes where sandfly breeding patches can be smaller than a single pixel. SMAP L4 model-assimilated products improve spatial detail to 9 km but remain coarse for village-scale targeting.
- Sentinel-1 SAR (C-band): Interferometric Wide swath mode at 10 m resolution, 6-day revisit per satellite. C-band backscatter is sensitive to near-surface soil moisture in vegetated areas where SMAP is most degraded. Used as a spatial disaggregation layer to downscale SMAP estimates, a method documented in peer-reviewed literature using the change-detection approach.
- MODIS NDVI (MOD13A2): 500 m or 1 km vegetation index at 16-day compositing intervals. Seasonal NDVI trajectories characterise the low-to-moderate vegetation cover associated with sandfly resting sites and help separate productive agricultural land from fallow sandy ground in the habitat model.
What the soil tells you that entomology surveys cannot
Field-trapping Phlebotomine sandflies is expensive, slow and geographically sparse. A single CDC light-trap survey covers perhaps a few dozen sites per district. Satellite data covers every square kilometre simultaneously, every few days, for free. The trade-off is that you are measuring habitat proxies, not insects directly. Getting that distinction wrong is the fastest way to produce a map that misleads a spraying team.
The proxies are well-grounded. P. argentipes, the dominant vector in the Indian subcontinent, breeds in sandy alluvial soils with moderate organic content, low surface moisture during the dry season and proximity to cattle. P. orientalis in Sudan and Ethiopia associates with black-cotton soils and Acacia-Balanites woodland. These niches leave fingerprints: elevated diurnal LST range (sandy soils heat and cool fast), low NDVI during the pre-monsoon period and specific bare-soil spectral signatures in Sentinel-2 shortwave infrared. WHO-linked modelling studies published in PLOS Neglected Tropical Diseases and similar journals have used these variables to produce district-level suitability surfaces that correlate with recorded kala-azar case distributions.
Building the suitability surface: inputs, weights and honest uncertainty
The standard workflow stacks three or four seasonal composites. MODIS LST diurnal range is computed per pixel as the mean difference between daytime and night-time retrievals across the dry season, typically January to April in Bihar. Sentinel-2 bare-soil fraction is extracted using a constrained linear spectral unmixing model with end-members derived from field spectra or high-confidence pure pixels. SMAP soil moisture provides a seasonality signal: the model is looking for moderate moisture in the resting season and relative dryness at the surface during breeding.
The suitability index is usually produced through MaxEnt or a comparable species-distribution modelling framework, trained on georeferenced entomological presence records. MaxEnt is well-documented and widely used for vector habitat work; its outputs are continuous probability surfaces, not binary risk zones, and they should be presented that way. The honest uncertainty comes from three sources. First, entomological training data are often biased toward accessible locations near roads. Second, SMAP at 36 km cannot resolve the field-to-field moisture variation that matters at the village scale. Third, LST at 1 km conflates multiple land-cover types within a single pixel in fragmented agricultural mosaics.
The 36-kilometre problem and what partially solves it
SMAP's 36 km footprint is not a minor caveat. A single pixel in the Gangetic plain of Bihar covers roughly 1,300 square kilometres and may contain dozens of villages, multiple soil types and irrigated fields sitting alongside sandy fallow land. Assigning a single volumetric moisture value to that area and using it to predict sandfly abundance at the household level is statistically unsound.
Two partial remedies exist in the published literature. The first is statistical downscaling using Sentinel-1 SAR backscatter. The change-detection method, which relates temporal changes in C-band backscatter to changes in SMAP retrievals, can produce spatially disaggregated moisture estimates at 1 km or finer, though with increased uncertainty and a requirement for cloud-free Sentinel-1 acquisitions. The second is to treat SMAP as a seasonal regime indicator rather than a pixel-level predictor: it tells you which months the landscape is in a moisture state compatible with sandfly activity, and Sentinel-2 and MODIS carry the spatial detail within that temporal window. Neither approach fully resolves the constraint. Any programme using this analysis for village-level spraying prioritisation should validate against local entomological trap data before acting on the map.
From suitability surface to spraying calendar
The operational output for a national vector-control programme is not a single annual map. It is a rolling suitability surface updated monthly or quarterly, combined with a phenological trigger that flags when conditions cross the threshold associated with peak sandfly activity. In Bihar, that window broadly corresponds to the post-monsoon period from October to December, when surface moisture has dropped enough for sandy soil to be accessible but the landscape has not yet desiccated completely.
A district health officer needs two things from this analysis: a ranked list of sub-districts by habitat suitability and a recommended spray timing window. The satellite-derived surface provides the first. The phenological trigger, built from MODIS LST and NDVI time series, provides the second. Combined, they allow indoor residual spraying resources, always limited, to be concentrated where the habitat evidence is strongest and the season is right. The WHO's kala-azar elimination programme in South-East Asia has used spatially explicit prioritisation of this type, though the specific satellite inputs vary by implementing partner.
Limits the map cannot resolve
Cloud cover during the monsoon interrupts Sentinel-2 optical acquisitions for weeks at a time across much of South Asia and East Africa. SAR from Sentinel-1 penetrates cloud but loses sensitivity to soil moisture under dense vegetation. The result is that the period immediately after monsoon withdrawal, arguably the most important for predicting the coming transmission season, is also the period with the patchiest satellite coverage.
Animal reservoir density, primarily domestic dogs in some foci and cattle in others, is not directly observable from orbit. Human settlement density can be approximated from Sentinel-2 or nighttime light data, but household-level risk factors such as mud-floor construction and cattle-shed proximity require field survey. The satellite surface is a first-pass filter, not a substitute for ground truth. Programmes that have treated modelled suitability as confirmed risk have misdirected resources. Satellize builds these analyses with explicit confidence intervals and recommends a field-validation protocol before any spraying calendar is finalised. Our crop-estimation work in Tonga used a comparable validation-first approach, and the principle transfers directly to vector habitat work.
What a well-specified analysis actually delivers
A properly scoped satellite habitat analysis for a national leishmaniasis programme produces four concrete outputs. A seasonal suitability raster at the finest defensible resolution, typically 100 to 500 m depending on the downscaling approach chosen. A ranked district or sub-district table with confidence bands. A phenological spray-timing calendar derived from the LST and NDVI time series. And a change-detection layer comparing current-year habitat conditions against a multi-year baseline, flagging areas where suitability is expanding or contracting.
The change-detection layer is often the most operationally useful. It answers the question a programme manager actually has: not where is the habitat in absolute terms, but where has it changed since last year? Expansion into previously low-suitability areas may indicate irrigation expansion, land-use change or climate shift. Contraction may reflect genuine improvement or simply a wet year suppressing bare-soil signals. Either way, the anomaly map is where limited field-verification resources should go first.
Typical figures
| LST spatial resolution (MODIS) | 1 km (MOD11A1/MYD11A1 daily products) |
| LST revisit | Twice daily per satellite; four observations per day with Terra and Aqua combined |
| Bare-soil fraction resolution (Sentinel-2) | 10 m (visible/NIR bands); 20 m (SWIR bands used for soil discrimination) |
| Sentinel-2 revisit | 5 days at equator (2A + 2B combined); cloud-free composites typically monthly in monsoon-affected regions |
| SMAP soil moisture resolution | 36 km (passive L2); 9 km (L4 model-assimilated); downscaled to ~1 km via Sentinel-1 SAR change-detection |
| SMAP revisit | 2 to 3 days global |
| Sentinel-1 SAR resolution | 10 m (IW mode GRD); 6-day revisit per satellite |
| Suitability model output resolution | 100 to 500 m depending on downscaling approach; not finer than the limiting input layer |
| MODIS archive depth | Terra from February 2000; Aqua from July 2002; enables 20+ year baseline |
| Sentinel-2 archive depth | From 2015 (2A); 2017 (2B); systematic global coverage from 2017 |
Analytics Satellize can run
| Seasonal habitat suitability raster | MaxEnt or BRT species-distribution model trained on georeferenced entomological records, with MODIS LST diurnal range, Sentinel-2 bare-soil fraction and downscaled SMAP as predictors | GeoTIFF suitability surface (0 to 1 probability) at 100-500 m, updated quarterly, with per-pixel confidence interval layer |
| Sub-district spraying priority ranking | Zonal statistics aggregating suitability surface to administrative boundaries; weighted by population exposure layer derived from settlement density | CSV and GIS polygon layer with ranked priority scores and confidence bands, compatible with DHIS2 import |
| Phenological spray-timing calendar | MODIS LST and NDVI time-series harmonic analysis to detect seasonal transitions into high-suitability moisture and temperature regimes | District-level calendar report (PDF and JSON) flagging optimal spraying windows 4 to 6 weeks in advance |
| Year-on-year habitat change detection | Pixel-level comparison of current-season suitability composite against multi-year median baseline; anomalies flagged at two standard deviations | Change-detection GeoTIFF and summary report highlighting expansion or contraction zones for field verification prioritisation |
| SMAP downscaling layer | Sentinel-1 C-band backscatter change-detection disaggregation of SMAP L2 passive retrievals following published statistical downscaling approaches | Monthly 1 km soil moisture GeoTIFF with uncertainty estimate, replacing raw 36 km SMAP in the suitability model for heterogeneous landscapes |
| Bare-soil fraction seasonal composite | Constrained linear spectral unmixing of Sentinel-2 SWIR and red-edge bands using dry-season image stacks to minimise vegetation masking | Annual and seasonal bare-soil fraction GeoTIFF at 20 m, used as a persistent habitat structural layer |
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.