Machine learning and regression-based techniques for predicting sprinkler irrigation's wind drift and evaporation losses

Abstract

Wind drift and evaporation losses (WDEL), which can occur as a result of operational and meteorological factors, are two of the most significant sprinkler-irrigation losses that can occur even in a well-managed irrigation system. A proper understanding of factors that influence WDEL in sprinkle irrigation is critical for developing water conservation strategies that significantly impact the quality and return on investment of irrigation projects. The specific objective of this research was to determine the predictive ability of five soft computing approaches (artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), multivariate adaptive regression spline (MARS), probabilistic linear regression (PLR), and support vector regression (SVR)) for predicting WDEL on a sprinkler irrigation system under design, operational, and meteorological conditions. Datasets were collected from previously published studies conducted under a variety of conditions. The results showed that the five approaches yielded statistically different WDEL predictions. The ANN model produced the most accurate WDEL predictions compared to the other models with the training and testing dataset. The ANFIS, MARS, PLR, and SVR models' performance ranks were found to be inconsistent across a variety of statistical performance criteria. Hence, Shannon's entropy-based decision theory was used to rank these models. The MARS model was ranked second (0.896), followed by the ANFIS model (0.865), the PLR model (0.833), and the SVR model (0.794). The design variable “auxiliary nozzle diameter” and climate variable “wind speed” both had high contribution ratios (17.5% and 12.19%, respectively) in WDEL modeling to produce a robust predictive model. In general, the developed models, particularly the ANN model, demonstrated a high degree of accuracy in estimating the WDEL of sprinkler irrigation systems.