Ecological memory modulates land surface phenology and productivity measures

Link To Share This Poster: https://cdmcd.co/dEyWJJ

David J. A. Wood, Northern Rocky Mountain Science Center, U.S. Geological Survey, Billings, MT, David J. A. Wood and Scott L. Powell, Land Resources and Environmental Sciences, Montana State University, Bozeman, MT, Paul C. Stoy, Biological Systems Engineering, University of Wisconsin, Madison, WI, Erik A. Beever, Ecology, Montana State University, Bozeman, MT, Erik A. Beever, Northern Rocky Mountain Science Center, U.S. Geological Survey, Bozeman, MT

Background/Question/Methods
Many aspects of phenology – recurring seasonal events – can be measured by satellite remote sensing. These measurements make it possible to study the climate drivers of the seasonal cycle of vegetation, but our ability to simulate these dynamics using ecosystem models remains constrained due to complex relationships that often involves lagged responses. We studied the role of contemporaneous climate, legacy climate, and contextual factors on the phenology of the U.S. Northern Great Plains, a region with diverse ecosystems including large expanses of extant grasslands. Specifically, we used current year and seasonal climate variables in conjunction with antecedent conditions from PRISM, with soils, disturbance, and land cover variables as independent variables that impact key phenological measures. We studied start and end of season dates, and peak and season-long productivity derived from AVHRR NDVI as our dependent variables. Statistical approaches were complicated by nonlinear responses and correlated inputs, which limited our ability to define an underlying model structure. Therefore, we selected random forests due to its decreased sensitivity to corelated predictor variables and ability to handle complex interactions, and used variable selection based on importance values to narrow down inputs to a set for inference using the R package VSURF.

Results/Conclusions
We selected the top variables for the four phenological measures, which varied between grassland, shrubland, barren, deciduous forest, and evergreen forest classes. In general, models were stronger for vegetation productivity variables (typically R2 > 0.8) than for date-based measures like start of season (typical R2 between 0.4 and 0.6). Across cover types, with variation in the length of the lagged influence, we found antecedent conditions modulated the impact of current-year climate on phenology. In grasslands, mean precipitation over the prior four water years was a key input for peak productivity. Likewise, for shrublands, start of season date was influenced by vapor pressure deficit over the prior four years and summer growing degree days from the prior year, while key antecedent factors for season-long productivity included the mean precipitation over the prior four water years. Vegetation phenology forms a feedback loop with climate and has cascading impacts to other species in the ecosystem and ecosystem processes. These results demonstrate that antecedent conditions over multi-year timeframes influence vegetation phenology. Accounting for ecological memory along with the large variability of drivers among biomes is important for predicting phenological mismatches, impacts to biodiversity, and risks to ecological services.