Prediction of the Reference Evapotranspiration Data from Raipur Weather Station in Chhattisgarh using Decision Tree-Based Machine Learning Techniques

Abstract

Agricultural lands need a lot of water for yield production. Water scarcity is becoming a significant problem on Earth. Reference evapotranspiration ( $$\text {ET}_0$$ ) estimation has a vital role in water management for irrigation purposes. This paper implements four variants of decision tree (DT)-based machine learning (ML) algorithms, including DT, random forest (RF), ExtraTrees regression (ET), and gradient boosting regression (GBR) to predict $$\text {ET}_0$$ using two feature sets of the dataset from the Raipur weather station in Chhattisgarh, India, collected from the India Meteorological Department (IMD). Results showed that GBR performs better than other variants in featureset-1 considering temperatures as input (MAE = 1.6660, MSE = 1.9684, and RMSE = 1.332), and in featureset-2 with all input variables, ExtraTrees regression is performing better than others (MAE, MSE, and RMSE values are 1.0972, 1.7754, and 1.3441, respectively). RMSE values compared with the two neural network-based prediction algorithms. In featureset-1, GBR is more efficient than others, whereas in featureset-2, differential evolution-based radial basis function neural network (RBFDE) is the winner. Comparisons conclude that decision tree-based ML methods could be a better choice with fewer input variables.