The increasing spread of multi-drug resistant malaria in African highlands has highlighted the importance of malaria suppression through vector control. Its historical success has meant that larval control has been proposed as part of an integrated malaria vector control program. Due to high operation costs, larval control activities would benefit greatly if the locations of mosquito habitats could be identified quickly and easily, allowing for focal habitat source suppression. Several mosquito habitat models have been developed to predict the location of mosquito habitats. However, to what extent these models can be generalised across time and space to predict the distribution of dynamic mosquito habitats remains largely unexplored. This study used mosquito habitat data collected in six different time periods and four different modelling approaches to establish 24 mosquito habitat models. We systematically tested the generality of these 24 mosquito habitat models. We found that although habitat–environment relationships change temporally, a modest level of performance was attained when validating the models using data collected from different time periods. We also describe flexible approaches to the predictive modelling of mosquito habitats, that provide novel modelling architecture for future research efforts.
Highlights: ► There are temporal variabilities in mosquito habitat–environment relationships. ► Habitat models developed using data collected at wet seasons have good performance in predicting dry season habitats. ► Models developed using different modelling approaches have different performance. ► In our study, the flexibility of modelling approaches and spatial autocorrelation have limited impact on the accuracy of habitat models.