Using PhenoCam data to model drivers of spring green-up in North American grasslands

Background/Question/Methods

Vegetation seasonality heavily influences many ecosystem and climate processes, such as carbon uptake and energy and water cycles. Thus, understanding drivers of vegetation phenology is crucial for predicting the current and future impacts of climate change on ecological systems. Existing models are able to accurately predict the date of spring onset in many forest ecosystems, largely based on photoperiod and accumulated chilling and warming temperatures. However, the climatic drivers of spring onset in grassland ecosystems are poorly understood, and existing models, including large earth system models, are often inaccurate in their predictions of vegetation green-up and brown-down cycles.Our objective was to develop and test new spring onset models for grassland ecosystems. We used long-term datasets of digital imagery from the PhenoCam Network to extract observed spring transition dates for 43 diverse grassland sites across North America (195 site-years). We first fit the data to a suite of existing spring onset models. Then, because grasslands are inherently water-limited, we incorporated a precipitation parameter into those models, as well as developed several new model variations (21 models total). We used the “phenor” R package to optimize parameters using generalized simulated annealing and to determine the best-fit model using AIC.

Results/Conclusions

Overall, models with both a temperature and precipitation parameter fit the data considerably better than models with only one or the other. The best model (RMSE = 16.0 days) was a sequential model, in which a site must first surpass a cumulative precipitation threshold before starting to accumulate temperature; spring onset is the day the temperature threshold is reached. This model fit significantly better than traditional spring models without a precipitation parameter (RMSE = 30.1–37.4 days). Importantly, it performed well with a single set of parameters for all grassland types, from the Great Plains to mediterranean grasslands to arid steppes ( > 250-day spread in spring onset dates). However, model performance was improved when the parameters were independently optimized for four separate climate regions (RMSE = 10.4 days). Thus, both sufficient temperature and precipitation are required for grassland plants to become physiologically active, but optimal thresholds vary by region. Incorporating this new model into larger earth system models should improve predictions of grassland phenology, as well as water and nutrient cycles. By understanding how climate drivers interact to trigger spring green-up, we can make informed predictions about how grasslands might respond to altered precipitation and temperature patterns under climate change.