Evapotranspiration Modeling Using Different Tree Based Ensembled Machine Learning Algorithm

The present study investigates and evaluate the scope and potential of modern computing tools and techniques such as ensembled machine learning methods in estimating ETo. Five different type of machine learning model namely (i) decision tree, (ii) Random Forest (RF), (iii) Adaptive Boosting (AdaBoost), (iv) Gradient Boosting Machine (GBM) and (v) Extreme Gradient Boosting (XGBoost) were compared for performance in estimating daily P-M ETo values. The RF, GBM and XGBoost model performed extremely well on the criteria of weighted standard error of estimate (WSEE) which is less than 0.25 mm/d. Furthermore, the ensembled machine learning model substantiated by boosting algorithm (XGBoost) significantly enhance the performance in estimating P-M ETo (WSEE is less than 0.17 mm/d). Moreover, the sensitivity analysis suggested that the data requirement for XGBoost is commonly available at most of the places unlike P-M ETo model. Given the generalization capability of the model, it can be successfully implemented for other similar location where comprehensive data are not available.

Automated summary

The present study investigates the potential of modern computing tools, particularly ensembled machine learning methods, in estimating ETo. The results show that the RF, GBM, and XGBoost models perform well in estimating daily P-M ETo values. The ensembled machine learning model, particularly the XGBoost model, significantly improves the estimation performance and has the capability to be used in locations with limited comprehensive data.