Topographic based models for predicting malaria vector breeding habitat in western Kenya

To enhance malaria control, topographic variables derived from remotely sensed Digital Elevation Models (DEMs) were used to model the breeding sites of malaria vectors. We further compared the predictive strength of two different DEMs and evaluated the predictability of various habitats types inhabited by Anopheles larvae. Using GIS techniques, topographic variables were extracted from two DEMs: Shuttle Radar Topography Mission 3 (SRTM3, 90-m resolution) and, the Advanced Spaceborne Thermal Emission Reflection Radiometer Global DEM (ASTER, 30-m resolution). We used data on breeding sites from an extensive field survey conducted on an island in western Kenya in 2006. Topographic variables were extracted for 826 breeding sites and 4520 negative points that were randomly assigned. Logistic regression modelling was applied to characterize topographic features of the malaria vector breeding sites and predict their locations. Model accuracy was evaluated using the area under the receiver operating characteristics curve (AUC). Multivariate models based on SRTM had a smaller AIC (3959.3) than did models based on the ASTER (3965.7). AUC results from the training data set indicated higher accuracy for the SRTM model (0.758), than for the ASTER model (0.756). For the test site, the SRTM model also showed greater predictive strength (AUC. 0.829), than did the ASTER model (AUC, 0.816). There was variation in the predictability of habitat types. Predictive potentials for both DEMs were fairly similar and were sufficiently accurate to predict vector habitats suggesting that topographic predictive models could be used by vector control managers in Africa to complement malaria control strategies.