Estimation of Rainfed Wheat Yield Functions Using Climatic Parameters and Multivariate Regression Methods

More than half of the agricultural lands in arid and semi-arid climates are rainfed. Many climatic variables affect the yield of rainfed crops, among them rainfall is the most important variable. The aim of the present study is to determine the yield functions of rainfed wheat in Tabriz, Sarab and Maragheh stations located in the east of Lake Urmia basin, considering the changes of climatic variables during different stages of rainfed wheat growth. In order to model the yield using multivariate regression method, some precipitation variables such as, number of effective precipitation events, vegetation precipitation deficit, reference precipitation and evapotranspiration deficit in rainfed conditions during six stages of rainfed wheat growth including germination; End of germination until the beginning of flowering; Flowering stage; Finishing flowering until the seeds start to fill; Seed filling stage and whole growing season were used. In general, based on the obtained results, precipitation fluctuations have the greatest effect on wheat yield. Therefore, identifying the precipitation regime and analyzing its characteristics is important for assessing yield fluctuations of rainfed crops. Among the growth stages, the fluctuation of the proposed traits in the whole growth season has a greater role in determining yield functions. Yield functions were determined using variables that had a significant correlation with yield. For this purpose, 22-year and 3-year data were used for calibration and validation, respectively. The results of the model efficiency coefficient and normalized root mean square error indicated better effeciency of Enter method in the Sarab (EF=0.55 and NRMSE=0.19) and the Moderate accuracy of Stepwise method in estimating the rainfed wheat yield in Maragheh and Tabriz. In Maragheh and Tabriz, the Stepwise method with average relative error values of 21% and 15.6%, respectively, and in Sarab, the Enter method with an average relative error of 16.5% had better results in yield fitting.