Model outputs and model code for machine learning models forecasting streamflow drought across the Conterminous United States

We applied machine learning (ML) models to forecast streamflow drought from 1 to 13 weeks into the future at more than 3,000 streamgage locations across the Conterminous United States. We applied two machine learning methods (long short-term memory [LSTM] neural network; light gradient boosting [LightGBM] machine) and two benchmark model approaches (persistence; autoregressive integrated moving average [ARIMA]) to predict weekly streamflow percentiles with independent models for each forecast horizon. Both ML models were trained using all percentiles (LSTM-all, LightGBM-all) and only percentiles below 30% (LSTM<30, LightGBM<30). We evaluated model performance regionally and nationally for drought occurrence (the classification performance for a future date) and for drought onset/termination (performance identifying drought starts and ends). This data release contains two zipped archives, one for model outputs (model_outputs.zip), the other for model code (model_code.zip). The model_outputs.zip folder contains files with feature importance data (importance_data.feather); model performance data (performance_data.feather); observed and predicted streamflow percentile time series data (timeseries_data.feather); additional explanatory data (explanatory_data.feather) that were used in the related primary publication to interpret model outputs in this data release; and static watershed attribute data (conus_static_inputs_gages.csv) that were used as model inputs. The model_code.zip folder contains three folders for the different modeling approaches (LSTM, LightGBM, benchmark). The benchmark folder contains two Jupyter notebooks, one for running the persistence models, the other for running the ARIMA models. The LSTM and LightGBM folders contain Jupyter notebooks, with R and Python scripts, for running the LSTM and LightGBM models.