Investigating of the climatic parameters effectiveness rate on barley water requirement using the random forest algorithm, Bayesian multiple linear regression and cross-correlation function

Abstract

Due to the pressure on water resources, especially in agricultural sectors, evaluation of the effective strategies on decrease in water consumption has an applicable and important role in field water management. In this research, the effectiveness rate of the 7 climatic parameters on the water requirement of barley (amount of potential evapotranspiration of barley (PETB) during the growth period) using the random forest algorithm (RF), Bayesian multiple linear regression (BR) and cross-correlation function (CCF) was investigated and prioritized. The results of this research can help the managers to control effective climatic variables on PETB. In this paper, the climatic data series of 8 stations with different climate conditions (with two replicates for each climate condition) in Iran during 1968–2017 was used. The results showed the linear regression of simulated PETB using AquaCrop model and predicted PETB using the RF and BR models had no difference with perfect reliable in 0.05 significant levels (T-Statistics varies from 0.010–0.103 and 0.000–0.842 in BR and RF models, respectively, and the R-square (R2) between simulated and predicted PETB were significant in 0.01 levels at all stations (R2 varies from 0.843–0.996 and 0.930–0.999 in BR and RF models, respectively; therefore, the RF and BR models had a good accuracy to predict PETB based on the climatic parameters and the accuracy of RF was more than BR model. The results indicated that based on the all statistical methods the wind speed had the most impact on the PETB (at 62.5% of stations) and based on the RF, BR and CCF methods the average of minimum temperature, the average of sunshine and the average of precipitation had the lowest impact on the PETB, respectively.