Meteorological Data Fusion Approach for Modeling Crop Water Productivity Based on Ensemble Machine Learning

Abstract

Crop water productivity modeling is an increasingly popular rapid decision making tool to optimize water resource management in agriculture for the decision makers. This work aimed to model, predict, and simulate the crop water productivity (CWP) for grain yields of both wheat and maize. Climate datasets were collected over the period from 1969 to 2019, including: mean temperature (Tmean), maximum temperature (Tmax), minimum temperature (Tmin), relative humidity (H), solar radiation (SR), sunshine hours (Ssh), wind speed (WS), and day length (DL). Five machine learning (ML) methods were applied, including random forest (RF), support vector regression (SVM), bagged trees (BT), boosted trees (BoT), and matern 5/2 Gaussian process (MG). Models implemented by MG, including Tmean, SR, WS, and DL (Model 3); Tmax, Tmin, Tmean, SR, Ssh, WS, H, and DL (Model 8); Tmean, and SR (Model 9), were found optimal (r2 = 0.85) for forecasting CWP for wheat. Moreover, results of CWP for maize showed that the BT model, a combination of SR, WS, H, and Tmin data, achieved a high correlation coefficient of 0.82 compared to others. The outcomes demonstrated several high performance ML-based alternative CWP estimation methods in case of limited climatic data supporting decision making for designers, developers, and managers of water resources.