Accurate ecological forecasts provide useful insights to inform policy and management, but building models to produce these forecasts is challenging. Modelling approaches can vary from mechanistic models that attempt to capture the underlying ecological processes to purely phenomenological or statistical models that rely on inferences from data. These different approaches are likely to have different strengths depending on the metric being predicted, the amount of data for training, and the time horizon of the prediction. In particular, too strong reliance on past data may lead to incorrect inferences about the future of ecosystems under novel conditions, such as those induced by climate change and anthropogenic disturbances. We here study several models of different paradigms, including neutral models, to predict Mountain pine beetle (MPB) infestations in Alberta, Canada. MPB life history makes mathematical modeling challenging, as they use complicated chemical signalling processes and occasionally disperse over very long distances above the tree canopy. During a recent hyperepidemic in neighbouring British Columbia, MPB were able to overcome the natural border of the Rocky Mountains and spread into Alberta. Alberta dedicated extensive resources to monitor and control this spread, including helicopter surveys of infested trees. We use this data to study the predictive accuracy of several models that range in complexity and mechanistic basis. We discuss general trends in model performance with the aim of providing practice advice about the types of models that may achieve the greatest predictive accuracy given different data availability, target year, and forecast horizons.
