Advancing evapotranspiration estimation with remote sensing and artificial intelligence – A review

Reliable estimation of evapotranspiration (ET), is crucial for effective decision-making in sustainable water management. The accuracy of ET estimation using conventional methods is often constrained by the limited availability and poor quality of the meteorological data. Remote sensing (RS) offers a potential solution to address these data gaps by enabling the monitoring and estimation of ET over large areas through satellite images. Several common remote sensing satellites were reviewed based on their relevance and application for ET estimation. The review outlined the essential image pre-processing steps required for such applications, including geometric and radiometric correction. Remote sensing data are also available at multiple processing levels, which have undergone the necessary correction processes. Users can directly access and apply them based on their specific application. Additionally, this review paper discusses substantial parameters, such as land surface temperature, solar radiation and vegetation indices that can be obtained from satellite images, as well as the use of a reanalysis-based method for obtaining relevant variables for ET estimation. Various types of machine learning, including deep learning, hybrid approaches, explainable AI and interpretable machine learning (IML), are discussed. The emerging role of geospatial foundation models (GFM), which are increasingly emphasized, has also been discussed, as they leverage these AI techniques for large-scale geospatial analysis. Finally, challenges and future perspectives of applying remote sensing in ET assessment are elucidated. Ultimately, the aim is to utilize the RS data, artificial intelligence (AI) technologies and relevant measures to derive in-situ ET directly.