Ecologists use correlative models to predict how species distributions will respond to environmental change, but these models are unreliable when extrapolating to future environments. To minimize extrapolation, modelers should use mechanistic predictors that reflect range-limiting processes. As generalist herbivores with abundant food, locusts may be limited by thermal effects on digestion. We measured thermal effects on the consumption and defecation of South American locusts (Schistocerca cancellata), and we used these data to project performance in current and future climates. We then integrated the performance projections into species distribution models to predict the distribution of outbreaking locusts based on different predictor sets, modelling methods, and climate scenarios. Contrary to expectations, models with only mechanistic predictors performed worse than those with only macroclimatic predictors; however, the best models were those that used both predictor types. Based on the mechanistic model, these locusts should occur throughout tropical S. America, but species distribution modelling revealed they are restricted to temperate regions. This mismatch between the mechanistic model and the distribution of S. cancellata suggests that the northern range of this species is limited by some factor other than temperature. All models projected that locusts would shift to higher latitudes and altitudes in response to climate warming, with the magnitude of this shift proportional to the amount of warming. Consequently, agriculturists should consider enhanced monitoring and management near the southern range of this species. Partially supported by NSF IOS 1826848.