Quantifying the impact of climate change on actual and potential evapotranspiration (AET and PET) is essential for water security, agriculture production and environmental management. AET and PET are strongly influenced by local factors such as topography, land cover and soil moisture, which limits the usability of global climate models for their projections. Here, we dynamically downscale CMIP6 models using CCAM to a 10 km resolution over Australia and derive AET and PET at a daily time step using the Morton method and project future changes under SSP126, 245 and 370.</p> <p>The performance of observation- and downscaled climate model-based AET is assessed against measured AET from 26 OzFlux sites in Australia. We show that high resolution downscaled climate models can provide reasonably accurate estimations of AET in Australia, with an ensemble mean error of 17 % for historical period 1981&ndash;2010. This compared favourably to observation- and reanalysis-based products, which reported mean errors ranging from 15.7 % &ndash; 44 %.</p> <p>Annual average end-of-century AET projections for low and intermediate emission scenarios (SSP126 and SSP245) show a decrease of -4.5 % and -3.5 % respectively in Australia, while under high emissions (SSP370) AET was projected to increase by 1.8 %. In contrast, PET was projected to increase by 5.0 % for SSP126, 8.4 % for SSP245 and 11.5 % for SSP370.</p> <p>Using random forest model, we show that the primary controlling factors for changes in AET are precipitation and solar radiation, and solar radiation and maximum temperature for PET. Our results offer new insights into future AET and PET changes estimated using downscaled CMIP6 simulations with implications for agriculture, water supply and natural resources management.