Weather–Power Anomaly Atlas for a Remote Hydropower Plant

Remote run-of-river hydropower plants often underperform when debris, sediment, or operational drift alter hydraulic conditions, yet many already archive SCADA and gridded weather data that are rarely exploited for diagnostics. This paper develops a weather–power anomaly atlas for a small hydropower plant using 4368 hourly records of power, OPERATION/STANDBY status, and meteorological variables from 25 November 2021 to 25 May 2022, where the meteorological inputs are drawn from NASA’s Prediction Of Worldwide Energy Resources (POWER) project. A median quantile regression baseline is fitted on OPERATION hours to relate calendar, wet-bulb temperature, and rolling precipitation to expected power, achieving about 63.5 kW median absolute residual, 154.6 kW mean absolute error, and 269.4 kW root mean square error. Residual-based anomaly detection converts large deviations into weather-annotated under- and over-performance events, and OPERATION spell durations are summarized via a Kaplan–Meier survival curve. The atlas shows that the plant operates only 38.5% of the time, with 204 OPERATION spells and a median spell length near 5 h, and that several strong under-performance events follow periods of elevated rainfall. Because it relies on standard statistical tools and routinely logged SCADA variables combined with freely available NASA POWER data, the approach can be readily replicated at similar hydropower sites.