Modelling Actual Evapotranspiration Using Cloud-based Computation and Artificial Intelligence

Actual evapotranspiration serves as a fundamental for water resource management. Although commonly used, methods for measuring ETa often have drawbacks, including exorbitant costs, complex calibration, and limitations in geographic or temporal resolution. The latest development of Artificial Intelligence with Machine Learning has introduced a promising method for estimating ETa. AI models optimize ETa predictions by combining satellite imagery and meteorological data. Historical observations can be used to train the model. In this paper, different ML algorithms were implemented using preprocessed data generated by Google Earth Engine, has been used to process and analyze geospatial data. To improve ETa forecast accuracy, the system included a number of crucial procedures. To find the required input data, existing datasets were first examined. The next step was to create a cloud-based Google Earth Engine application for preparing data. To guarantee accuracy and consistency, a variety of datasets were gathered in various forms and then aligned with measurement data. Results showed that AI-driven models, especially the Random Forest method, greatly increase the accuracy of ETa predictions. An AI model was created using the processed data. Four regression models are compared based on Mean Squared Error and R2 score as a proportion of the variance. The RF regressor outperforms the others with the lowest MSE 0.358 and highest the R2 score 0.752, making it the most reliable model. In contrast, linear regression performs the worst, with the highest MSE 0.670 and a relatively low R2 score 0.530. The decision tree regressor and SVR show moderate performance, with MSE values of 0.614 and 0.528 and R2 scores of 0.574 and 0.634, respectively. The notable improvement in accuracy and power of the WaPOR dataset demonstrates how data properties affect model performance.