A novel method for mapping village-scale outdoor resting microhabitats of the primary African malaria vector, Anopheles gambiae

Abstract

Background: Knowledge of Anopheles resting habitats is needed to advance outdoor malaria vector control. This study presents a technique to map locations of resting habitats using high-resolution satellite imagery (world view 2) and probabilistic Dempster-Shafer (D-S) modelling, focused on a rural village in southern Mali, West Africa where field sampling was conducted to determine outdoor habitat preferences of Anopheles gambiae, the main vector in the study area.

Methods: A combination of supervised and manual image classification was used to derive an accurate land-cover map from the satellite image that provided classes (i.e., photosynthetically active vegetation, water bodies, wetlands, and buildings) suitable for habitat assessment. Linear fuzzy functions were applied to the different image classes to scale resting habitat covariates into a common data range (0-1) with fuzzy breakpoints parameterized experimentally through comparison with mosquito outdoor resting data. Fuzzy layers were entered into a Dempster-Shafer (D-S) weight-of-evidence model that produced pixel-based probability of resting habitat locations.

Results: The D-S model provided a highly detailed suitability map of resting locations. The results indicated a significant difference (p < 0.001) between D-S values at locations positive for An. gambiae and a set of randomly sampled points. Further, a negative binomial regression indicated that although the D-S estimates did not predict abundance (p > 0.05) subsequent analysis suggested that the D-S modelling approach may provide a reasonable estimate locations of low-to-medium An. gambiae density. These results suggest that that D-S modelling performed well in identifying presence points and specifically resting habitats.

Conclusion: The use of a D-S modelling framework for predicting the outdoor resting habitat locations provided novel information on this little-known aspect of anopheline ecology. The technique used here may be applied more broadly at different geographic scales using Google Earth, Landsat or other remotely-sensed imagery to assess the malaria vector resting habitats where outdoor control measures can reduce the burden of the disease in Africa and elsewhere.

Keywords: Anopheles; Dempster-Schafer modeling; Mali; Resting habitats; Species distribution modeling.

Fig. 1

Individual points mark the locations of the sampling locations within the study site and the colour detonates the area which each location is associated with

Fig. 2

Detailed images showing sampling areas and exact locations where drop nets were placed. The numbers refer to the ID given to each sampling location and correspond with the ID values found in Table 1

Fig. 3

Changes in D-S model values with changing distance-from-vegetation breakpoint values (no count values). The distance at which the D-S value rose above zero for all the affected sites were averaged together for the final D-break point value. The field site ID correlates with the ID values found in Table 1

Fig. 4

Resting catch data. The size of the yellow circles is scaled to reflect the number of mosquitoes caught in each location

Fig. 5

Land cover map based on WorldView 2 imagery covering the study site. The map shows the different land cover classes obtained from supervised classification and segmentation followed by manual correction to correctly delineate wetland sites

Fig. 6

Final Dempster-Shafer (D-S) model of outdoor resting sites (belief output) for An. gambiae. Each pixel value provides a probability estimate of An. gambiae presence

Fig. 7

Comparision of mosquito presence/absence with the final D-S model

Fig. 8

Mosquito abundances at the field sites compared to final D-S model

Fig. 9

Mean of D-S values for locations where An. gambiae specimens were collected versus the D-S mean for 50 randomly sampled points. Error bars show the standard deviation