Completed Daily Streamflow Records for 164 Maryland-Area USGS Stations Using Calibrated Generative AI

This dataset contains completed daily discharge records for 164 United States Geological Survey streamflow stations across Maryland and adjacent hydrologic settings for 2010–2024. The dataset was produced using a calibrated conditional diffusion framework for sparse daily streamflow record completion. Observed discharge values are preserved where available. Missing daily discharge values are completed using the calibrated conditional-diffusion predictive median, with calibrated 5th and 95th percentile uncertainty bounds provided for imputed values. The final dataset contains 898,556 station-day records, including 762,903 observed records and 135,653 imputed records. The upload also includes a reproducible Python code package containing the conditional diffusion model, station-blocked validation workflow, baseline models, uncertainty calibration procedure, final completion workflow, and figure-generation support. Baseline models include climatology, Gaussian residual sampling, and random-forest residual bootstrap. This dataset supports the manuscript titled “Calibrated Generative AI for Completing Sparse Daily Streamflow Records at Under-Monitored Stations.”