Reproducibility data for “Where and When Surface Soil Moisture Can Explain Operational Drought”: GM-LSTM station predictions, SPI baselines, and USDM labels

Reproducibility data for the station-scale results of the manuscript “Where and When Surface Soil Moisture Can Explain Operational Drought: A Real-Time Entropy Diagnostic of Information Limits” (Bischof, Zehe & Loritz, Karlsruhe Institute of Technology).

This record contains the three intermediate datasets required to reproduce the station-scale figures and tables (Figure 1, Appendix A, Appendix E) directly from the analysis code at https://github.com/BalazsBis/gmlstm_drought, without retraining the Gaussian-Mixture LSTM. It complements the published ISMN model-input dataset (doi:10.5281/zenodo.18016491), which holds the training inputs.

Contents

• predictions_stations.tar.gz – held-out, per-station predictive distributions from the GM-LSTM under 5-fold spatial cross-validation (fold_1 … fold_5), plus the model checkpoints. These are the diagnostic outputs (mixture parameters / entropy) read by notebooks/03_station_diagnostics.ipynb and scripts/tableE2_event_metrics.py.
• spi.tar.gz – the Standardized Precipitation Index (SPI) baseline series per cross-validation fold (gridMET-derived), used as the reference forecaster in the CRPS-vs-SPI comparison (Appendix E).
• data_usdm.tar.gz – the weekly U.S. Drought Monitor labels rasterised to the analysis grid (usdm_grid.csv) and the grid lookup table (grid_id.csv), used to score predictions against operational drought categories.

Usage. Download and extract each archive into the repository root (spi/, predictions_stations/, data_usdm/), then run notebooks/03_station_diagnostics.ipynb as described in the repository README.

Sources.

U.S. Drought Monitor (USDM). The U.S. Drought Monitor is jointly produced by the National Drought Mitigation Center at the University of Nebraska-Lincoln, the United States Department of Agriculture, and the National Oceanic and Atmospheric Administration. Data and maps courtesy of NDMC. The weekly drought-category data were obtained from the USDM GIS data archive (https://droughtmonitor.unl.edu/DmData/GISData.aspx) and rasterised to the analysis grid. See: Svoboda, M., et al. (2002), The Drought Monitor, Bulletin of the American Meteorological Society, 83(8), 1181–1190, https://doi.org/10.1175/1520-0477-83.8.1181.