Landsat Thermal Calibration for River Temperature Monitoring: A 25-Year Case Study of the Vistula River

Gauging stations on large rivers are too sparse to resolve spatial patterns of water temperature; satellite thermal imagery can fill this gap. We calibrated 25 years (2000–2024, May–November) of Landsat 5, 7, 8 and 9 surface temperature against in situ water temperature at three gauging stations on the lower Vistula River, Poland, using 942 matched pairs (876 retained after quality control). The linear calibration achieved a coefficient of determination (R2) of 0.91, a root mean square error (RMSE) of 1.37 °C and a mean absolute error (MAE) of 1.10 °C, and it transferred between stations (leave-one-station-out R2 = 0.90, RMSE = 1.46 °C) and was consistent across all four sensors (R2 = 0.87–0.92). A two-term seasonal correction captured the physically interpretable seasonal error structure and reduced error by 14%, whereas 20 regression algorithms offered no gain. Applied to the full archive, the calibration revealed six recurrent warm reaches (+1.0 to +1.4 °C) and a cooler outlet reach along the lower Vistula, August surface temperature warming of about 1.1 °C per decade, and placed August 2020 among the six warmest Augusts. Landsat thermal data thus extend point-based monitoring to the river scale, and the framework is applicable to gauged rivers wider than about 200 m during the ice-free season.