Modeling Of Water Production Function Of Winter Wheat

Winter wheat is one of the most important crops in North China. The studyof crop water-fertilizer production function is significant for water-savingagriculture. Based on field experiments, Jensen's models of water productionfunction were established with several linear or nonlinear methods, as well asartificial neural networks (ANN) models of water- fertilizer production functiontrained with the combination of genetic algorithms (GA) and back-propagationalgorithms (BPA). The field experiment results of winter wheat at both Yongledian Station inBeijing and Xiaohe Station in Shanxi were used in this research. Two-wayanalysis of variance and polynomial regression of irrigation and fertilization toyield based on the field water balance analysis showed that the fertilization hadno significant impact on yield for Yongledian Station. The methods of ordinary least square regression (OLSR), partial least squareregression (PLSR), genetic algorithms (GA), and EXCEL programming solver(EPS) were adopted to establish Jensen's models for Xiaohe station andYongledian station. PLSR is most favorable when the independent variables arehighly correlated, otherwise OLSR, GA and EPS are advisable. According to theconstant induced by PLSR, yield modification coefficient was introduced toJensen's models, which yielded better fitting for the experiment results.Considering the feature of Jensen's model, variable reference treatment methodto generate the relative data was proposed to deal with field data with fewreplication and poor precision. From the results of Jensen's models with differentfertilization at Yongledian station, variable reference treatment method wasproved to be more effective than fixed reference treatment method. ANN models, with the weights and bias trained with hybrid algorithms ofGA and BPA, were used to simulate the yield response to water and fertilization.In general, the results of ANN models were consistent with those of Jensen'smodels.