Random Forest and Multilayer Perceptron hybrid models integrated with the genetic algorithm for predicting pan evaporation of target site using a limited set of neighboring reference station data

This study explores the application of machine learning algorithms for the prediction of pan evaporation (Ep), which is a critical factor in water resource management for the assessment of water demand and usage. Specifically, this research evaluates the effectiveness of two base models: Random Forest (RF) and Multi-Layer Perceptron (MLP) and their optimized counterparts using a Genetic Algorithm (GA), designated as GA-RF and GA-MLP, for modeling Ep at a target station using data from adjacent stations. The datasets were split into a training set (70%) and a testing set (30%). The models’ performances were judged using three statistical measures: Correlation Coefficient (CC), Scattered Index (SI), and Willmott’s Index of agreement (WI). The enhanced models, particularly GA-MLP-5, showed superior performance with a CC of 0.8704, SI of 0.2539, and WI of 0.9212, indicating the potent ability of GA to refine RF and MLP models for predictive accuracy. Additionally, sensitivity analysis via the GA-RF indicates the varying influence of Ep from neighboring stations on the target station, shedding light on key predictors for effective water management. Conclusively, this study demonstrates that the hybrid models have significant potential in accurate Ep estimation and can be expanded to predict other meteorological variables, offering valuable tools for water resource management strategies.