Abstract:Wheat is one of the major food crops in China. To further estimate the yield of winter wheat accurately, Guanzhong Plain in Shaanxi Province was used as the study area, vegetation temperature condition index (VTCI), leaf area index (LAI) and fraction of photosynthetically active radiation (FPAR), which were closely related to water stress and photosynthesis at the main growth stage were selected as remotely sensed characteristic parameters, and the improved particle swarm optimized wavelet neural network (IPSO-WNN) was used to improve the shortcomings of gradient descent method which tended to fall into local optimum and construct winter wheat yield estimation model. The results showed that the IPSO-WNN model had a coefficient of determination (R2) of 0.66 and a mean absolute percentage error (MAPE) of 7.59%. Compared with the BPNN (R2=0.46, MAPE was 11.80%) and WNN (R2=0.52, MAPE was 9.80%), the IPSO-WNN can further improve the accuracy of the yield estimation and enhance the robustness of the model. It was explored by sensitivity analysis that the input parameters had a strong influence on winter wheat yield, and it was found that FPAR at the heading-filling stage had the greatest effect on winter wheat yield, followed by VTCI at the jointing stage, LAI at the heading-filling and milk maturity stages and FPAR at the green-up and jointing stages. The I index of winter wheat was obtained from IPSO-WNN output, and a yield estimation model between I and statistical yield was constructed to estimate the yield of winter wheat in the Guanzhong Plain. The results showed that the R2 between estimated yield and statistical yield was 0.63 and root mean square error (RMSE) was 505.50kg/hm2, and the problem of “l(fā)ow yield and high estimation” of the yield estimation model was solved. Therefore, the yield estimation model constructed based on IPSO-WNN can estimate the yield of winter wheat in the Guanzhong Plain more accurately.
