Streamflow temperature estimation based on Landsat thermal infrared and optical bands data

River temperature is an essential water variable for the health of river ecosystems and the survival of many aquatic organisms. Ground observations of water temperature are very limited to describe their spatial and temporal distribution. The thermal infrared (TIR) remote sensing techniques have been useful to partly fill such gaps. In this study, we develop a new method of thermal sharpening for river-temperature estimation (TSRE) by deploying the optical instruments onboard the same satellite with high spatial resolutions. The TSRE method delineates the TIR radiance of a surface area as a function of surface reflectance, remote sensing index (.i.e NDVI), and landcover fractions. River-temperature observations from six stations on Willamette River and Columbia River in the Pacific Northwest of USA are used to calibrate and validate the TSRE method based on 64 Landsat images including ETM+/Band6 (120m resolution) and TIRS/Band10 (100m resolution). Three stations, located on rivers wider than 300 m, are used to calibrate the ETM+ and TIRS based estimation. The other three stations with river width of 30~60 m, 100-200 m, and 200~300 m respectively are used for the validation. The calibration results show that the stations using ETM+/Band6 images have the RMSE of 0.74 °C, 0.58 °C, 1.18 °C, respectively, while the RMSEs are higher as 1.52 °C, 1.32 °C, 2.19°C when using TIRS/Band10 images. It shows that for the rivers wider than three thermal water pixels (>300m), the pure pixels around the stream centerline can be used to retrieve stream temperature directly with very good accuracy. For the three validation stations using ETM+/Band6 images, the RMSE values are 7.41 °C, 2.01°C, 1.67°C without thermal sharpening, while the RMSE values significantly decrease to 3.25 °C, 0.87°C, 0.84°C respectively after applying the thermal sharpening technique. With the TIRS/Band10 images, the validation at the same stations shows similar results, with the RMSE values from 8.64 °C, 3.34°C, 2.98°C to 4.66 °C, 0.74°C, 1.02°C before and after using the TSRE method. This study indicates that the proposed TSRE method could provide river-temperature mapping for more upstream rivers with width less than ~30m, while also significantly improving the accuracy of temperature estimation for wide downstream rivers ( >300m width).