Evapotranspiration (ET) is predominant variable for water management in various types of ecosystems, and ET processes in these ecosystems have been assessed through in-situ measuring and modelling methods. However, it is challenging to measure actual ET and upscale it to regional level. In addition, the accuracy of retrieved parameters from models is usually low for karst landscapes, where the underlying surface is more complex than non-karst landscapes. Due to various porosities and conduits, aquifers in karst landscapes typically show remarkable and rapid responses to precipitation events, leading to serious water stress. Therefore, there is an urgent need to quantify water fluxes to provide reliable evidence for the protection and sustainable management of karst water resources. In this study, five plots were built to observe actual ET based on Thermal Dissipation Probes (TDP), re-designed Ventilated-chamber and Micro-lysimeters in a karst catchment in southwest China. Then, three models (Penman-Monteith-Leurning, PML; Remote Sensing-Priestley and Taylor, RS-PT; and Hargreaves) were selected to upscale ET estimation to the regional level based on Landsant-8 and MODIS data. The results showed that: 1) The PML model performed better than other models (p < 0.01) with higher R2 values (0.72 for MODIS images and 0.87 for Landsat-8 images) and smaller RMSE values (1.4 mm·day-1 and 0.8 mm·day-1 for MODIS and Landsat-8 images, respectively); 2) Daily ET exhibited significant seasonal variability and different spatial distribution; 3) ET had a slightly positive correlation with DEM; however, ground temperature had a negative correlation with ET. By combining remote sensing data and upscaling it to the regional level, this study helps improve the accuracy of measured and estimated ET. It suggests that ET is strongly regulated by vegetation coverage and available energy in subtropical humid karst catchments.