Developing the Role of Earth Observation in Spatio-Temporal Mosquito Modelling to Identify Malaria Hot-Spots

Abstract

Anopheles mosquitoes are the vectors of human malaria, a disease responsible for a significant burden of global disease and over half a million deaths in 2020. Here, methods using a time series of cost-free Earth Observation (EO) data, 45,844 in situ mosquito monitoring captures, and the cloud processing platform Google Earth Engine are developed to identify the biogeographical variables driving the abundance and distribution of three malaria vectors—Anopheles gambiae s.l., An. funestus, and An. paludis—in two highly endemic areas in the Democratic Republic of the Congo. EO-derived topographical and time series land surface temperature and rainfall data sets are analysed using Random Forests (RFs) to identify their relative importance in relation to the abundance of the three mosquito species, and they show how spatial and temporal distributions vary by site, by mosquito species, and by month. The observed relationships differed between species and study areas, with the overall number of biogeographical variables identified as important in relation to species abundance, being 30 for An. gambiae s.l. and An. funestus and 26 for An. paludis. Results indicate rainfall and land surface temperature to consistently be the variables of highest importance, with higher rainfall resulting in greater mosquito abundance through the creation of pools acting as mosquito larval habitats; however, proportional coverage of forest and grassland, as well as proximity to forests, are also consistently identified as important. Predictive application of the RF models generated monthly abundance maps for each species, identifying both spatial and temporal hot-spots of high abundance and, by proxy, increased malaria infection risk. Results indicate greater temporal variability in An. gambiae s.l. and An. paludis abundances in response to seasonal rainfall, whereas An. funestus is generally more temporally stable, with maximum predicted abundances of 122 for An. gambiae s.l., 283 for An. funestus, and 120 for An. paludis. Model validation produced R2 values of 0.717 for An. gambiae s.l., 0.861 for An. funestus, and 0.448 for An. paludis. Monthly abundance values were extracted for 248,089 individual buildings, demonstrating how species abundance, and therefore biting pressure, varies spatially and seasonally on a building-to-building basis. These methods advance previous broader regional mosquito mapping and can provide a crucial tool for designing bespoke control programs and for improving the targeting of resource-constrained disease control activities to reduce malaria transmission and subsequent mortality in endemic regions, in line with the WHO’s ‘High Burden to High Impact’ initiative. The developed method was designed to be widely applicable to other areas, where suitable in situ mosquito monitoring data are available. Training materials were also made freely available in multiple languages, enabling wider uptake and implementation of the methods by users without requiring prior expertise in EO.

Three machine-learning models to predict dengue vectors Breteau Index based on meteorology and biotope in Fujian China

“…With the incorporation of local new data, machine learning models can learn and improve autonomously. RF has been applied to monitor the intensity and distribution of adult mosquitoes to identify variables that in uence the distribution and abundance of three mosquito malaria vectors and their relative importance [13]. Salim et al used dengue outbreak viable based on historical data, combined with meteorology viable, and constructed a Support Vector Machine (SVM) model, which is of high accuracy to predict the strike of dengue while in uenced by unbalanced data [14].…”

Mapping small mammal optimal habitats using satellite-derived proxy variables and species distribution models

“…While Earth observation (EO) data and derived products have been applied for SDM, for example [30][31][32], integration of remotely sensed data in SDM remains rare in practice [33]; further opportunities exist to develop SDMs for predictive and explanatory purposes through a close integration of SDM and EO [34]. The broad-scale coverage offered by satellite sensors along with regular revisit periods and cost-free data availability enables characterisation of landscape features and environmental processes underlying species distributions to be quantified and included within SDMs.…”