The complex transmission ecologies of vector-borne and zoonotic diseases pose challenges to their control, especially in changing landscapes. Human incidence of zoonotic malaria ( Plasmodium knowlesi) is associated with deforestation although mechanisms are unknown. Here, a novel application of a method for predicting disease occurrence that combines machine learning and statistics is used to identify the key spatial scales that define the relationship between zoonotic malaria cases and environmental change. Using data from satellite imagery, a case-control study, and a cross-sectional survey, predictive models of household-level occurrence of P. knowlesi were fitted with 16 variables summarized at 11 spatial scales simultaneously. The method identified a strong and well-defined peak of predictive influence of the proportion of cleared land within 1 km of households on P. knowlesi occurrence. Aspect (1 and 2 km), slope (0.5 km) and canopy regrowth (0.5 km) were important at small scales. By contrast, fragmentation of deforested areas influenced P. knowlesi occurrence probability most strongly at large scales (4 and 5 km). The identification of these spatial scales narrows the field of plausible mechanisms that connect land use change and P. knowlesi, allowing for the refinement of disease occurrence predictions and the design of spatially-targeted interventions.