Model performance and output variables for phenological events across land cover types in the Northwestern Plains, 1989-2014

Many aspects of recurring plant developmental events – vegetation phenology – are measured by remote sensing. By consistently measuring the timing and magnitude of the growing season, it is possible to study the complex relationships among drivers of the seasonal cycle of vegetation, including legacy conditions. We studied the role of current and legacy climate, and contextual factors on the land surface phenology of the U.S. Northern Great Plains. Specifically, we used annual and seasonal climate variables (e.g., temperature, precipitation, growing degree days, and vapor pressure deficit) covering the current year and the past four years derived from the PRISM climate dataset. We also included soils, disturbance, and within pixel land cover heterogeneity as additional independent variables that may also influence phenology. We assessed three phenological measures from the AVHRR satellite sensor (utilizing the years from 1989 to 2014), start of season, peak productivity, and season-long productivity, using variable selection techniques utilizing nested random forest models (VSURF). We identified the top variables for the phenological measures for pixels with predominantly grassland, shrubland, and evergreen forest land cover classes. Datasets in this release include tables from the associated manuscript and supplementary information found in the Larger Work section. Specifically, included tables cover the modeling parameters, final factors derived from the random forest variable selection process, and model performance measures.