Hammond, J.C. Diaz, J. Goodling, P.J. 20260130 tabular digital data and application/service https://doi.org/10.5066/P13HFSYE Machine learning (ML) models were used to forecast streamflow drought from 1 to 13 weeks into the future at more than 3,000 streamgage locations across the conterminous United States. Long short-term memory [LSTM] neural network models were used to predict weekly streamflow percentiles with independent models for each forecast horizon. In this data release we specifically include an LSTM model where only percentiles below 50% (LSTM<50) are prioritized. Model performance was evaluated regionally and nationally for drought occurrence (the classification performance for a future date) and for drought onset and termination (performance identifying drought starts and ends). This data release contains model outputs (model_outputs.zip) and model code (model_code.zip). The model_outputs.zip folder contains files with model performance data (performance_data.feather) and observed and predicted streamflow percentile time series data (timeseries_data.feather). The model_code.zip folder contains Jupyter notebooks with Python scripts for running the LSTM models. Given the need for information on current and forecast hydrological drought conditions, we developed an experimental ML tool to forecast streamflow drought occurrence, onset, and duration for more than 3,000 gaged locations across the Conterminous United States. This work highlights the challenges and opportunities to further advance hydrological drought forecasting and supports an experimental operational streamflow drought assessment and forecast tool that provides previously unavailable streamflow drought forecasts to enable the public to better anticipate and prepare for hydrological drought impacts. This product is the second version of a tool that may be substantially improved in future iterations. 20001002 20200330 ground condition Complete None planned -124.7565 -66.9491 49.3845 24.5232 ISO 19115 Topic Category inlandWaters USGS Thesaurus water resources None autoregressive integrated moving average deep learning drought gradient boosting machine hydrology long short-term memory machine learning modeling neural network streamflow time series USGS Metadata Identifier USGS:695bf56fd4be02126e7fb5a6 None Conterminous United States No access constraints. Please see "Distribution Information" for details. These data are marked with a Creative Common CC0 1.0 Universal License. These data are in the public domain and do not have any use constraints. Users are advised to read the dataset's metadata thoroughly to understand appropriate use and data limitations. John C Hammond USGS - Northeast Region Research Hydrologist mailing and physical

MD/DE/DC WSC UMBC Campus,UMBC Research Park

Catonsville MD 21228 443-498-5568 jhammond@usgs.gov The project producing this data release was funded by the USGS Water Resources Mission Area Drought Program, with computing resources provided by USGS Cloud Hosting Solutions. Amazon SageMaker AI, R, Python, PyTorch, Jupyter, and Git were used to develop and run the models. Hammond, J.C. 2025 tabular digital data https://www.sciencebase.gov U.S. Geological Survey data release https://doi.org/10.5066/P1X5VH96 No formal attribute accuracy tests were conducted. No formal logical accuracy tests were conducted. Dataset is considered complete for the information presented, as described in the abstract. Users are advised to read the rest of the metadata record carefully for additional details. A formal accuracy assessment of the horizontal positional information in the dataset was not conducted. A formal accuracy assessment of the vertical positional information in the dataset was not conducted. Falcone, J. 2011 tabular digital data https://doi.org/10.5066/P96CPHOT Digital 2011 publication date Falcone, 2011 Additional data used to interpret the model output in the related primary publication Hammond, J.C. 2024 tabular digital data https://doi.org/10.5066/P9JWJPNC Digital 2024 publication date Hammond, 2024 Additional data used to interpret the model output in the related primary publication Hammond, J.C. 2025 tabular digital data https://doi.org/10.5066/P1X5VH96 Digital 2025 publication date Hammond, 2025 Input used by the models and additional data used to interpret the model output in the related primary publication Price, A.N. Jones, C.N. Hammond, J.C. Zimmer, M.A. Zipper, S.C. 2021 tabular digital data https://doi.org/10.1029/2021GL093298 Digital 2021 publication date Price and others., 2021 Additional data used to interpret the model output in the related primary publication Simeone, C.E. 2022 tabular digital data https://doi.org/10.5066/P92FAASD Digital 2022 publication date Simeone, 2022 Additional data used to interpret the model output in the related primary publication U.S. Geological Survey 2025 tabular digital data https://doi.org/10.5066/F7P55KJN Digital 2025 publication date USGS, 2025 Additional data used to interpret the model output in the related primary publication Wieczorek, M.E. Jackson, S.E. Schwarz, G.E. Sekellick, A.J. Staub, L.E. 2018 tabular digital data https://doi.org/10.5066/F7765D7V Digital 2018 publication date Wieczorek and others., 2018 Input used by the models and additional data used to interpret the model output in the related primary publication Wieczorek, M.E. Wolock, D.M. McCarthy, P.M. 2021 tabular digital data https://doi.org/10.5066/P92S9ZX6 Digital 2021 publication date Wieczorek and others., 2021 Input used by the models and additional data used to interpret the model output in the related primary publication John C Hammond Aaron J Heldmyer Jeremy A Diaz Phillip J Goodling Ryan R McShane 2025 dataset https://www.sciencebase.gov U.S. Geological Survey https://doi.org/10.5066/p132nswy Digital and/or Hardcopy 20250930 publication date Old model code Previous version code We obtained daily streamflow data for 2000-2020 from the USGS National Water Information System (USGS, 2025). We then converted 7-day average daily observed streamflow to percentile values to identify drought via consistent thresholds across streamgages. We computed de-seasonalized streamflow percentiles with the unbiased Weibull plotting position, applying a variable threshold for each day of the year that used only the values for a 30-day window surrounding that day from all years of record (for example, percentiles for August 15 used values from August 1 to August 30 for all years of record). We prepared watershed average time series of several gridded datasets (Hammond, 2025) as well as watershed average values of static watershed properties (conus_static_inputs_gages.csv; Hammond, 2025; Wieczorek and others., 2018; and Wieczorek and others., 2021). Neural network (LSTM) models were trained and evaluated using streamflow and explanatory variable data from the period 2000-2020 for CONUS. Separate regression models were built to predict weekly streamflow percentiles for each lead time (1, 2, 4, 9, and 13 weeks). We trained our models on a central period from October 1, 2002, to September 30, 2018, leaving the first part of the record (October 1, 2000-September 30, 2002) and the last part of the record (October 1, 2018-March 30, 2020) for model testing. Models were evaluated for their performance in matching the correct drought intensity category (moderate drought less than 20th percentile, severe drought less than 10th percentile, extreme drought less than 5th percentile) as well as for their ability to correctly forecast the onset and termination of drought events. See Hammond and others. (2025) for more detail. The notebooks and README contained in the compressed file (model_code.zip) contain code, comments, and instructions for using AWS (Amazon Web Services) to load data from S3 (Simple Storage Service) buckets and train models in SageMaker AI on EC2 (Elastic Compute Cloud) instances. 20250801 performance_data.feather Feather file containing model performance data. U.S. Geological Survey site_id Site identifier. USGS streamgage station identification number. U.S. Geological Survey NA No data U.S. Geological Survey USGS National Water Information System https://doi.org/10.5066/F7P55KJN model_id Model identifier. U.S. Geological Survey NA No data U.S. Geological Survey LSTM<50 Model identifier for long short-term memory neural network models trained with only streamflow observations below the 50th percentile. U.S. Geological Survey nday_forecast Number of days ahead of model forecast. U.S. Geological Survey NA No data U.S. Geological Survey 7 91 days kap_10 Cohen's kappa of streamflow predictions classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey -0.251 1 unitless kap_05 Cohen's kappa of streamflow predictions classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey -0.118 1 unitless kap_20 Cohen's kappa of streamflow predictions classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey -0.236 1 unitless bal_acc_10 Balanced accuracy of streamflow predictions classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0.388 1 unitless bal_acc_05 Balanced accuracy of streamflow predictions classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0.443 1 unitless bal_acc_20 Balanced accuracy of streamflow predictions classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0.369 1 unitless n_true_positive_10 Number of true positives of streamflow predictions classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 131 number n_true_positive_05 Number of true positives of streamflow predictions classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 91 number n_true_positive_20 Number of true positives of streamflow predictions classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 131 number n_true_negative_10 Number of true negatives of streamflow predictions not classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 24 181 number n_true_negative_05 Number of true negatives of streamflow predictions not classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 56 181 number n_true_negative_20 Number of true negatives of streamflow predictions not classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 20 181 number n_pred_drought_10 Number of streamflow predictions classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 133 number n_pred_drought_05 Number of streamflow predictions classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 93 number n_pred_drought_20 Number of streamflow predictions classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 136 number n_pred_nondrought_10 Number of streamflow predictions not classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 29 181 number n_pred_nondrought_05 Number of streamflow predictions not classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 70 181 number n_pred_nondrought_20 Number of streamflow predictions not classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 23 181 number n_obs_drought_10 Number of streamflow observations classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 131 number n_obs_drought_05 Number of streamflow observations classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 91 number n_obs_drought_20 Number of streamflow observations classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 131 number pct_obs_drought_10 Percentage of streamflow observations classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 0.724 percent pct_obs_drought_05 Percentage of streamflow observations classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 0.503 percent pct_obs_drought_20 Percentage of streamflow observations classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 0.771 percent n_obs_nondrought_10 Number of streamflow observations not classified as drought below the 10th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 50 181 number n_obs_nondrought_05 Number of streamflow observations not classified as drought below the 5th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 66 181 number n_obs_nondrought_20 Number of streamflow observations not classified as drought below the 20th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 36 181 number rsq Coefficient of determination of streamflow predictions for models trained with all streamflow observations. U.S. Geological Survey NA No data U.S. Geological Survey 0 0.932 unitless rsq_below_30 Coefficient of determination of streamflow predictions for models trained with only streamflow observations below the 30th percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 1 unitless rmse Root mean square error of streamflow predictions for models trained with all streamflow observations. Lower values are better. U.S. Geological Survey NA No data U.S. Geological Survey 3.759 49.411 percentile rmse_below_30 Root mean square error of streamflow predictions for models trained with only streamflow observations below the 30th percentile. Lower values are better. U.S. Geological Survey NA No data U.S. Geological Survey 0.413 31.128 percentile kge Kling-Gupta Efficiency of streamflow predictions for models trained with all streamflow observations. U.S. Geological Survey NA No data U.S. Geological Survey -0.691 0.85 unitless kge_below_30 Kling-Gupta Efficiency of streamflow predictions for models trained with only streamflow observations below the 30th percentile. U.S. Geological Survey NA No data U.S. Geological Survey -12.282 0.977 unitless timeseries_data.feather Feather file containing observed and predicted streamflow percentile time series data. U.S. Geological Survey site_id Site identifier. USGS streamgage station identification number. U.S. Geological Survey NA No data U.S. Geological Survey USGS National Water Information System https://doi.org/10.5066/F7P55KJN reference_datetime Issue date (YYYY-MM-DD) of model forecast. U.S. Geological Survey NA No data U.S. Geological Survey 2000-10-02 2020-03-23 date datetime Valid date (YYYY-MM-DD) of model forecast. U.S. Geological Survey NA No data U.S. Geological Survey 2000-10-09 2020-03-30 date model_id Model identifier. U.S. Geological Survey NA No data U.S. Geological Survey LSTM<50 Model identifier for long short-term memory neural network models trained with only streamflow observations below the 50th percentile. U.S. Geological Survey observation Observed streamflow percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0.0805802 99.9194198 percentile prediction Predicted streamflow percentile. U.S. Geological Survey NA No data U.S. Geological Survey 0 100 percentile variable Variable (target) used in model forecast. U.S. Geological Survey NA No data U.S. Geological Survey var Variable (target) used in LightGBM model forecast for the percentile of the 7-day rolling mean streamflow using the variable threshold method with a 30-day window of Julian dates. U.S. Geological Survey weibull_jd_30d_wndw_7d Variable (target) used in LSTM model forecast for the percentile of the 7-day rolling mean streamflow using the variable threshold method with a 30-day window of Julian dates. U.S. Geological Survey split Split of time series used in model forecast. U.S. Geological Survey NA No data U.S. Geological Survey test Split of time series used in model forecast for testing dates (from 2000-10-02 to 2020-03-30). U.S. Geological Survey duration Duration of model forecast. U.S. Geological Survey NA No data U.S. Geological Survey P1D Duration of model forecast for daily forecast. U.S. Geological Survey P1W Duration of model forecast for weekly forecast. U.S. Geological Survey family Family of probability distribution described by 'parameter'. U.S. Geological Survey NA No data U.S. Geological Survey summary Family of probability distribution described by 'parameter' for summary statistics. U.S. Geological Survey ensemble Family of probability distribution described by 'parameter' for ensemble of models. U.S. Geological Survey parameter Parameter of distribution specified in 'family'. U.S. Geological Survey NA No data U.S. Geological Survey pred_interv_05.0 Parameter of distribution specified in 'family' for 5th percentile of distribution. U.S. Geological Survey median Parameter of distribution specified in 'family' for 50th percentile of distribution. U.S. Geological Survey pred_interv_95.0 Parameter of distribution specified in 'family' for 95th percentile of distribution. U.S. Geological Survey 1 Parameter of distribution specified in 'family' for singular realization of ensemble. U.S. Geological Survey nday_forecast Number of days ahead of model forecast. U.S. Geological Survey NA No data U.S. Geological Survey 7 91 days U.S. Geological Survey - ScienceBase mailing

Denver Federal Center, Building 810, Mail Stop 302

Denver CO 80225 USA 1-888-275-8747 sciencebase@usgs.gov Unless otherwise stated, all data, metadata and related materials are considered to satisfy the quality standards relative to the purpose for which the data were collected. Although these data and associated metadata have been reviewed for accuracy and completeness and approved for release by the U.S. Geological Survey (USGS), no warranty expressed or implied is made regarding the display or utility of the data for other purposes, nor on all computer systems, nor shall the act of distribution constitute any such warranty. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. Digital Data https://doi.org/10.5066/P132NSWY None 20260122 John C Hammond USGS - Northeast Region Research Hydrologist mailing and physical

MD/DE/DC WSC UMBC Campus,UMBC Research Park

Catonsville MD 21228 443-498-5568 jhammond@usgs.gov FGDC Content Standard for Digital Geospatial Metadata FGDC-STD-001-1998