Implicit Stochastic Optimization Approach For Irrigation Scheduling Of Winter Wheat

Optimization of irrigation scheduling is significant to increase water use efficiency. Based on soil water balance model for the dynamic simulations of soil water content and field evapotranspiration during crop growing period, an optimization model for crop irrigation scheduling was developed, and was optimized using dynamic programming (DP) and genetic algorithm (GA). Furthermore, referring the application of reservoir regulation function, the irrigation optimization function was gained by using the implicit stochastic optimal regulation method for the decision of irrigation.The meteorology data and the soil and crop parameters of Xiaohe Irrigation Station in Shanxi were used in this research. The stochastic characteristics and correlation of evapotranspiration series and precipitation series in Xiaohe Irrigation Station were analyzed. Based on soil water balance model for the dynamic simulations of soil water content and field evapotranspiration and Jensen model of crop water-production function, an optimization model for crop irrigation scheduling was developed with the maximal relative yield as the decision objective, and was optimized using DP and GA. The results of DP and GA are similar with that of GA being better. The optimal irrigation scheduling under different situations appear the similar characteristic, i.e., irrigation tends to be applied during the heading stage of crops, which shows that heading stage is the key period for winter wheat growing. Field evapotranspiration and crop yield increase with the irrigation volume, but marginal benefit and the use efficiency of irrigation water decrease.Using the method of implicit stochastic optimization and comparing the concept of reservoir regulation function, the irrigation optimization function was proposed. Based on the optimal results under historical meteorology data, the function was established by statistical analysis, and was to describe therelationship between irrigation decision and different relative influencing factors. Then, qualitative and quantitative prediction can be acquired by using the irrigation optimization function.Qualitative prediction for irrigation was decided through logistic model. Logistic model calculates the probability of current irrigation by precipitation history and initial soil water content, and then compares this probability with a certain appointed value to determine whether to irrigate or not. The accuracy of logistic model is over 80% and higher. Quantitative prediction of the irrigation water amount was estimated respectively by partial least-squares regression(PLSR) and artificial neural networks(ANN). The results show that, ANN is more accurate than PLSR. But the irrigation water amount estimated by both PLSR and ANN is more than the optimal results near the minimal water amount of irrigation while being close to the optimal results under other situations.