COS 128-10
Do we need detailed demographic data to forecast climate change impacts on plant populations?

Thursday, August 13, 2015: 4:40 PM
342, Baltimore Convention Center
Andrew T. Tredennick, Department of Wildland Resources and the Ecology Center, Utah State University, Logan, UT
Peter B. Adler, Department of Wildland Resources and the Ecology Center, Utah State University, Logan, UT

Population models are essential tools for ecological forecasting, but reconciling the scales at which population models are parameterized and the scales at which environmental changes play out is difficult. Most population models, at least for plant species, are built using data from small, localized plots because parameterizing traditional population models requires tracking the fates of individuals. These models are difficult to scale up from micro to meso-scales because the fitted parameters do not fully represent the spatial and temporal variation present at scales beyond that at which the data are collected. An alternative is to rely on aggregate, population-level data that is easier and less costly to collect. Doing so requires assuming that population-level data accurately represents the aggregate response of the individuals that actually respond to weather. We tested this assumption using population models of four Montana grassland species fit using individual and aggregated forms of the same data. We fit population models with interannual variation in vital rates explained, in part, by climate covariates. We then perturbed the climate covariates to test the sensitivities of species to climate change, and to see if the individual-based and population-based models give similar predictions. 


The individual-level and population-level models were able to reproduce observed dynamics, as assessed using one-step-ahead forecasts. But the performance of the individually based model was less variable and produced consistently better forecasts. For two of our focal species the individual-based and population-based models predicted proportional cover changes of the same sign and similar magnitude. For the other two species the two modeling approaches gave opposing predictions. For example, for Bouteloua gracilis, the individually-based model consistently predicted decreases in cover due to 1% increases in precipitation, temperature, or both, relative to cover predicted by the models with unperturbed climate covariates. Counter to those predictions, the population-based model predicted increases in cover with the climate perturbations we imposed. Discordant model predictions are likely the result of how climate impacts population growth. For example, climate impacts were strongest in the survival vital rate regressions for most species. Unfortunately, the survival process is not well-resolved in aggregated data of percent cover of a given species. While census-type data (e.g., percent cover in plots) allows researchers broader spatio-temporal coverage, the robustness of models based on those data will be strongly dependent on which process (e.g., survival, growth, recruitment) is most sensitive to climate drivers.