SYMP 22-2 - What is the useful "lifetime" of a species distribution model used in conservation decision-making? How to keep pace with rapidly changing methods and standards

Species distribution models (SDMs) are commonly used in conservation decision-making to estimate, for example, current and potential future distributions, areas of high priority habitat, and exposure to change in land cover and climate. These decisions initiate actions that often entrain sunk costs of funds and effort over many decades. While conservation plans often require regular reviews for deciding changes of course, the models on which these decisions are based may “age” in the sense that newer data and methods could improve predictions over the original. Across the past two decades the growth in biodiversity data access and volume and the rate of conceptual and methodological advances has grown immensely, making the “aging” of model predictions all the more likely. Here, I review the general reliable “lifetime” of predictions from SDMs used in conservation assessments and decision-making.

Model predictions can become outdated due to: 1) improved inputs (occurrence and environmental data); 2) improvements in existing methods (e.g., appreciation for the need to “tune” models, manner in which absence data is verified or background sites selected); 3) development of new methods (e.g., coupling SDMs with sophisticated dispersal and/or genetic models, multi-species distribution models); 4) appreciation for different aspects of biodiversity relevant to implementation distribution models (e.g., genetic and phenotypic differentiation within species); and 5) interpretation of model output (i.e., does a model prediction reflect probability of occurrence or general environmental “suitability”?). I first review how existing studies have aged in each of these areas, then outline ways to “future-proof” predictions. From easy-to-hard to implement, these include: 1) forming clear review processes focused on data and modeling methods to assess if they are still reliable; 2) sensitivity analysis to assess what advancements would make current predictions obsolete; 3) development of automated modeling workflows that can provide new predictions as data is acquired, coupled with policy “triggers” that indicate the need for revising course; 4) and “crystal-ball” methods for assessing, from a hypothetical future standpoint, the costs of having designed conservation assessments and actions around models that were wrong in particular ways.