Forecasting the risk of pathogen spillover from reservoir populations of wild or domestic animals is essential for the effective deployment of interventions such as wildlife vaccination or culling. Due to the sporadic nature of spillover events and limited availability of data, developing and validating robust, spatially explicit, predictions is challenging. Recent efforts have begun to make progress in this direction by capitalizing on machine learning methodologies. An important weakness of existing approaches, however, is that they generally rely on combining human and reservoir infection data during the training process and thus conflate risk attributable to the prevalence of the pathogen in the reservoir population with the risk attributed to the realized rate of spillover into the human population. Because effective planning of interventions requires that these components of risk be disentangled, we developed a multi-layer machine learning framework that separates these processes. Our approach begins by training models to predict the geographic range of the primary reservoir and the subset of this range in which the pathogen occurs. The spillover risk predicted by the product of these reservoir specific models is then fit to data on realized patterns of historical spillover into the human population. The result is a geographically specific spillover risk forecast that can be easily decomposed and used to guide effective intervention. Applying our method to Lassa virus, a zoonotic pathogen that regularly spills over into the human population across West Africa, results in a model that explains a modest but statistically significant portion of geographic variation in historical patterns of spillover. When combined with a mechanistic mathematical model of infection dynamics, our spillover risk model predicts that 897,700 humans are infected by Lassa virus each year across West Africa, with Nigeria accounting for more than half of these human infections.

Robust classification approaches are required for accurate classification of complex land-use/land-cover categories of desert landscapes using remotely sensed data. Machine-learning ensemble classifiers have proved to be powerful for the classification of remotely sensed data. However, they have not been evaluated for classifying land-cover categories in desert regions. In this study, the performance of two machine-learning ensemble classifiers – random forests (RF) and boosted artificial neural networks – is explored in the context of classification of land use/land cover of desert landscapes. The evaluation is based on the accuracy of classification of remotely sensed data, with and without integration of ancillary data. Landsat-5 Thematic Mapper data captured for a desert landscape in the north-western coastal desert of Egypt are used with ancillary variables derived from a digital terrain model to classify 13 different land-use/land-cover categories. Results show that the two ensemble methods produce accurate land-cover classifications, with and without integrating spectral data with ancillary data. In general, the overall accuracy exceeded 85% and the kappa coefficient (κ) attained values over 0.83. The integration of ancillary data improved the performance of the boosted artificial neural networks by approximately 5% and the random forests by 9%. The latter showed overall higher accuracy; however, boosted artificial neural networks showed better generalization ability and lower overfitting tendencies. The results reveal the merit of applying ensemble methods to integrated spectral and ancillary data of similar desert landscapes for achieving high classification accuracies.