Random forest and donor-based simulated streamflow drought for 3,198 gages across the CONUS, 1982-2020

This metadata record describes random forest model simulated daily drought status at 3,198 stream gage locations distributed across the CONUS from 1982-04-01 to 2020-03-31. The models were trained on data from 1984-10-01 to 2015-09-30 and tested on data from 1982 to 1984 and 2016 to 2020. The models used explanatory variable inputs described in Wieczorek (2023) (doi.org/10.5066/P98IG8LO) to model streamflow drought conditions, which was calculated by converting daily streamflow values to percentiles using the Weibull plotting position following methods described in Simeone (2022). Additionally, droughts were conceptualized and modeled as binary events where streamflow below the 20th percentile is considered drought. In addition to the random forest model outputs, additional outputs are provided in each file from selected donor models. These donor models were selected based on similarity in basin physiography using selected GAGES-II characteristics following Konapala & Mishra (2020). The top 1% of models in terms of basin similarity with a Cohen's Kappa of at least 0.4 were selected, resulting in 19 donor output time series per file.