Integration of remote sensing and artificial intelligencetechniques for estimation of evapotranspiration

In this research a Machine Learning (ML) algorithm was implemented to estimate Evapotranspiration (ET) aiming to eliminate time consuming steps of classical ET estimation methods. The Artificial Neural Network (ANN) model was used and trained using Remotely Sensed data as input variables and a calibrated Mapping EvapoTranspiration at high Resolution with Internalized Calibration (METRIC) model as response ET data. The source of multispectral data was Landsat 8 platform. In the first step METRIC model was calibrated for a selected study area which was Eskişehir/Alpu. Secondly, the ANN model was trained and tested using calibrated ET data and input RS data. The trained model was then applied to the whole study area and the results where validated using confusion matrixes. At the next steps of the study, the performance of the ANN model was compared with the available products including Eddy Covariance Flux data and a widely referenced MODIS product. The results of the study showed that ANN can be implemented for ET estimations with high performance even better than the best RS available product like METRIC. The ANN model showed significant performance in farm based (smaller scale) ET estimation which is a highly important application in water resource management systems. Further study on the comparison with EC and MODIS data showed that those techniques fail to predict small scale ET but still applicable in larger area cumulative ET estimations. It was also concluded that precise validation of any RS or ML algorithm demands new technologies for direct rasterized field ET measurements, something that is not available yet.