Prediction of wind drift and evaporation losses from sprinkler irrigation using neural network and multiple regression techniques

Wind drift and evaporation losses (WDEL) play a significant role in the development of water conservation strategies in sprinkler irrigation. In this study, artificial neural network (ANN) and multiple linear regression (MLR) models were developed by taking data collected from published studies on predicted WDEL for several design, operational, and meteorological conditions of variables in sprinkler irrigation. Five combinations of input variables, including riser height, operating pressure, main nozzle diameter, auxiliary nozzle diameter (da), water discharge by main nozzle, water discharge by auxiliary nozzles, wind speed (WS), air temperature, and relative humidity were used to create prediction models for WDEL. The ANN and MLR models were trained and tested on 70% and 30% of the data points, respectively. The accuracy of the models was assessed by the coefficients of correlation (r), overall indices of model performance (OI), root mean square errors (RMSE), and mean absolute errors (MAE). Statistical results showed that the ANN and MLR models with all input variables had the best predicting capabilities. When comparing the results of different ANN and MLR models, it was seen that the ANN models had more success in predicting WDEL. The ANN models gave higher r (0.843–0.956) and OI (0.794–0.909) values, and lower RMSE (2.662%–4.886%) and MAE (2.197%–3.729%) values compared to the MLR models in the training stage. The MLR models’ r values ranged from 0.794 to 0.864, OI values ranged from 0.747 to 0.816, RMSE values ranged from 4.562% to 5.514%, and MAE values ranged from 3.513% to 4.414%. Furthermore, a contribution analysis found that the design parameter da and the climatic parameter WS were considered to obtain the most robust estimation model. It can be stated that the ANN model is a more suitable tool than the MLR model for the prediction of WDEL from sprinkler-irrigation.