OOS 15-10 - An experimental test of population predictions based on historical climate-demography correlations

Tuesday, August 8, 2017: 4:40 PM
Portland Blrm 256, Oregon Convention Center
Andrew R. Kleinhesselink, Department of Wildland Resources, Utah State University, Logan, UT, John B. Bradford, Southwest Biological Science Center, U.S. Geological Survey, Flagstaff, AZ, Caitlin M. Andrews, Southwest Biological Science Center, USGS, Flagstaff, AZ and Peter B. Adler, Department of Wildland Resources and the Ecology Center, Utah State University, Logan, UT

Annual variation in the demographic rates of species is often correlated with annual climate variation. However, ecologists rarely quantify whether these correlations can accurately predict how populations respond to climate change experiments. We conducted a field experiment to test our ability to predict the effects of variation in soil moisture on three dominant perennial grasses and one shrub in a sagebrush steppe community in Idaho, USA. We fit Bayesian models for the growth, survival and recruitment of each species to 26 years of observational data. Our models included the effects of plant size, competition, and three seasonal soil moisture variables. We then generated one-step-ahead predictions from these models for the growth, survival and recruitment of each species for each year of a five year irrigation and rain-out shelter experiment. We compared the predictive skill of models that only included the effects of competition and plant size to that of models that also included the effects of the soil moisture variables. If including annual variation in soil moisture improves predictions, as measured by mean square error (MSE) and log-posterior predictive density (lppd), it would demonstrate the predictive value of historical climate and demography correlations.


Rain-out shelters decreased and irrigation increased spring soil moisture by roughly 40%. Rain-out shelters decreased growth and survival of all three grasses, but increased growth and survival of the shrub and these effects were positively correlated with the predicted effects (r = 0.77). In contrast, the observed and predicted effects of irrigation were not as strongly correlated (r = 0.46). This was largely due to overpredicting the positive effect of irrigation on the growth and survival of the grasses. Adding soil moisture covariates improved predictive skill (increased lppd) in six out of twelve vital rate models, but the improvements were slight (<3% of lppd). We also used individual based population models to generate one-step-ahead predictions of total cover per plot for each species in each year. For two of the four species, including soil moisture improved the accuracy of these population-level predictions (lower MSE), but these improvements were small (~2% of MSE). While we had some success using observational data to predict the direction of demographic responses to experimental climate manipulation, our study demonstrates that translating this qualitative understanding into improved quantitative predictions is more challenging and may require even more data