Extracted Features and Preprocessed Remote Sensing and In Situ Data for Long-Term Streamflow Forecasting

This dataset supports the development and evaluation of multi-step-ahead probabilistic streamflow forecasting models by integrating in situ and remote sensing data across two hydrologically diverse basins in the United States: the John River basin in Maine and the Clearwater River basin in Idaho. These basins represent distinct hydroclimatic regimes and drainage areas, offering a robust testbed for generalizable hydrological modeling. Daily historical streamflow records were retrieved from the United States Geological Survey (USGS) database, covering extensive periods (1946\u20132025 for Maine and 1910\u20132025 for Idaho). Each streamflow time series was preprocessed using a 30-day lag based on cross-correlation analysis to preserve temporal dependencies.In addition to ground observations, the dataset incorporates two auxiliary climate datasets to enhance predictive performance: the NOAA Optimum Interpolation Sea Surface Temperature (OISST) dataset and the NOAA nClimGrid-Daily Version 1 dataset. The OISST dataset captures large-scale ocean-atmosphere interactions, such as El Ni\u00f1o-Southern Oscillation (ENSO) dynamics, while the nClimGrid-Daily dataset provides high-resolution daily temperature and precipitation fields across the contiguous United States. Together, these data sources support probabilistic machine learning frameworks such as Transformer and Informer models by enriching input features with both localized meteorological drivers and broader teleconnection signals. This comprehensive, multi-source dataset is intended for benchmarking long-range streamflow forecasting under uncertainty and advancing flood risk assessment tools.