Linear mixed modeling and artificial neural network techniques for predicting wind drift and evaporation losses under moving sprinkler irrigation systems

Abstract

Pressurized irrigation systems, center pivots, and linear moves are used worldwide on a large scale. Accurate predictions of wind drift and evaporation losses (WDEL) could help in improving the system’s uniformity and efficiency. The current study evaluates data analysis techniques for accurately estimating WDEL under moving sprinkler irrigation systems. A total of 72 experiments (2015–2017) were conducted at the research and extension center in Prosser, WA, under a wide variety of climate conditions. Two data analysis techniques, namely linear mixed modeling (LMM) and artificial neural networks (ANN), were used to identify the significant drivers of WDEL from the given weather-related inputs. Four published datasets were also used to check the generalization capabilities of the developed models. The results revealed an average of ~ 20% WDEL under Prosser, WA, conditions. Vapor pressure deficit and wind speed were the only significant weather variables at a 0.05 level of significance. Both in training and in testing, the ANN models (root mean squared error (RMSE = 2%)) worked better than the LMM (RMSE = 5%). Testing results revealed the high generalization and predictive power of ANN models with a RMSE of 1% for the (Yazar 1984) datasets. The best LMM model was with the Sanchez et al. (2011) dataset with a RMSE of 14%. The above results showed that ANN models can be used to accurately predict WDEL. This should help in further research for efficiency improvements in sprinkler irrigation systems.