Ecological forecasting under climatic data uncertainty: a case study in phenological modeling

Forecasting ecological responses to climate change represents a challenge to the ecological community because models are often site-specific and climate data are lacking at appropriate spatial and temporal resolutions. We use a case study approach to demonstrate uncertainties in ecological predictions related to the driving climatic input data. We use observational records, derived observational datasets (e.g. interpolated observations from local weather stations and gridded data products) and output from general circulation models (GCM) in conjunction with site based phenology models to estimate the first flowering date (FFD) for three woody flowering species. Using derived observations over the modern time period, we find that cold biases and temperature trends lead to biased FFD simulations for all three species. Observational datasets resolved at the daily time step result in better FFD predictions compared to simulations using monthly resolution. Simulations using output from an ensemble of GCM and regional climate models over modern and future time periods have large intra-ensemble spreads and tend to underestimate observed FFD trends for the modern period. These results indicate that certain forcing datasets may be missing key features needed to generate accurate hindcasts at the local scale (e.g. trends, temporal resolution), and that standard modeling techniques (e.g. downscaling, ensemble mean, etc) may not necessarily improve the prediction of the ecological response. Studies attempting to simulate local ecological processes under modern and future climate forcing therefore need to quantify and propagate the climate data uncertainties in their simulations.