Forecasting animal population responses to climate change is a common objective, but there are few methods for evaluating confidence in such predictions. The typical approach to predicting population abundance and/or species distributions involves estimation of current species-climate relationships, projecting the future climate state, and then using the current species-climate relationships to project future abundance patterns. Such approaches implicitly assume that species relationships with abiotic variables will hold in the future, which may or may not be true. Thus, forecasting future population abundance based on long-term observational studies is a useful practice but it has limitations, which are not typically acknowledged. This is especially true when attempting to predict events across large spatial scales or at extreme covariate values. We develop a quantitative method to evaluate the accuracy of climate-based population predictions and use this approach to assess the extent of spatio-temporal synchrony among distinct regions within the breeding range of monarch butterflies. Using Bayesian regression models, we estimate the effects of climatic conditions along the spring and summer migratory route on monarch abundances. By running models on subsets of the data, we evaluate how well our model can predict abundances across years assuming the best case scenario: when true climate conditions are known.
Results/Conclusions
Our results reveal that at least 10 years of data are needed for adequate model predictability of average future counts of monarchs. Because annual weather conditions experienced during spring migration primarily drove yearly abundances (as opposed to localized effects experienced during summer breeding) year-specific counts were often difficult to predict reliably, even when using more than a decade of data across hundreds of sampling locations. Our study addresses an important gap in scientific knowledge regarding the assessment of confidence in climate-based ecological predictions. Current climate-species relationships are regularly used to make predictions about future conditions. Yet, despite the long time series and large spatial extent of our data, we demonstrate that there is still substantial uncertainty in predicting year-specific abundances, specifically when predictive climate conditions are outside the range of typical regional conditions. Our assessment method can be used in similar analyses to more confidently interpret ecological responses to climate change. Our results demonstrate the relative importance of climatic drivers in predicting population abundance and the difficulties in producing reliable predictions of animal populations in the face of climate and environmental change.