Professor Northern Arizona University, United States
Abstract: Climate change is altering the timing of seasonal vegetation cycles (phenology), with cascading consequences on larger ecosystem processes, such as the exchange of carbon, water, and energy. Therefore, understanding the drivers of vegetation phenology is critical to predicting the ecological impacts of climate change. While numerous phenology models exist to predict the timing of the start of the growing season (SOS), fewer have been developed to predict the end of the growing season (EOS). The EOS models that do exist can adequately predict EOS in forested ecosystems using air temperature and photoperiod but perform poorly in grassland ecosystems. Thus, the drivers of EOS in grassland ecosystems are not well-understood, creating significant challenges in simulating grassland vegetation dynamics within larger earth system models. The objective of this study was to develop an improved EOS model for grassland ecosystems. We used repeat digital imagery from the PhenoCam Network to extract EOS dates for 44 diverse grassland sites (212 site-years) across North America. We fit this data to 20 new EOS models, as well as 3 existing models. In addition to temperature and/or photoperiod climate drivers, many of the new models included water availability (a limiting factor for grassland plant growth), as well as SOS date as a biological driver. Model parameters were optimized using generalized simulated annealing within the R package “phenor,” and the best-fit model was selected via AIC. All new EOS models (RMSE = 22-33 days between observed and predicted dates) performed substantially better than existing ones (RMSE = 43-46 days) across all sites (~200-day spread in EOS date). The top new model predicted EOS after surpassing an established threshold of either accumulated cold temperatures or consecutive dry days, but only after a certain number of days following SOS. All model fits were substantially better when SOS date was included, indicating a strong correlation between start- and end-of-season timing. Further, since plants are adapted to their regional climates, model performance was improved by independently optimizing parameters for 6 distinct climate regions (RMSE = 4-19 days). While the best model structure varied slightly by region, all included similar drivers as the overall top model for all sites. Thus, across diverse grassland sites, EOS is triggered by both climate (temperature, water) and biological (SOS) drivers. Incorporating these new EOS models into earth system models should substantially improve predictions of grassland dynamics and associated ecosystem processes, both now and under future climate change.