Thermal sharpening of Landsat-8 TIRS surface temperatures for inland water bodies based on different VNIR land cover classifications

Water temperatures are crucial for biological and physiological processes and are an integral part of water quality monitoring. Thermal infrared remote sensing can help to acquire extensive temperature information at low cost, but for inland water bodies, the available satellite data often has an insufficient spatial resolution to monitor water surface temperatures (WST). Hence, thermal sharpening to increase temperature data availability is evaluated in this study. Landsat-8 TIRS data was sharpened based on land cover information from Landsat-8 OLI data and a gyrocopter survey over the research area in Koblenz, Germany. This area presents a challenging composition of two rivers, infrastructure, urban and vegetation areas. The water and non-water pixels were identified with a pixel based, two-step modified parallelpiped classification algorithm. The sharpening water thermal imagery (SWTI) algorithm is available for water bodies to unmix temperatures of water-land boundary pixels based on the relation between their radiance and land cover composition. It was modified in order to test its applicability to the research area. Two auxiliary datasets were used to investigate the effect of large resolution differences between temperature and land cover data. In most cases the SWTI improves the RMSE and in all cases reduces the average difference of the Landsat-8 compared to reference WST. The quality of the results depends mainly on the congruency of the input data. The SWTI also proved to be superior to the cubic convolution algorithm when increasing the spatial resolution over water areas and reduced the required minimum river width.