Remote Sensing Estimation Of Winter Wheat Yield Based On Suitable Spatial Scale Selection And Growth Model

Winter wheat is one of the main food crops in China.Timely and accurate monitoring of winter wheat planting information by remote sensing technology is conducive to understanding and mastering the spatial growth of winter wheat and taking reasonable cultivation and management measures to promote the increase of winter wheat production.In this study,the winter wheat in Shuyang County,Suqian City,Jiangsu Province was selected.The domestic GF-1/PMS panchromatic image and HJ-1A/CCD multispectral image were fused to generate multiple spatial scale fusion images.After image quality evaluation and spectral characteristics comparison,the image scale suitable for Jiangsu winter wheat field pattern was selected.Different training samples and image combinations were designed for object-oriented classification,and suitable spatial scale images were used for winter wheat identification and growth monitoring.Winter wheat leaf area index（LAI）and biomass were used as assimilative variables of remote sensing information and winter wheat yield estimation model,and PSO optimization algorithm was introduced to optimize the model parameters to obtain the single-point scale winter wheat yield estimation.A point-plane conversion model of winter wheat production was established based on BP neural network and spectral characteristics of remote sensing images,and the winter wheat production of Shuyang County was estimated by remote sensing.The image scale method suitable for the planting distribution pattern of winter wheat in Jiangsu field,the method of winter wheat identification and planting area extraction based on appropriate spatial scale and the remote sensing estimation method of winter wheat yield based on suitable spatial scale image and growth model were formed.The results were as follows:（1）The Brovey method was applied to the fusion of GF-1/PMS panchromatic image and HJ-1A/CCD multispectral image in Shuyang County,and five scale fusion images（2 m×2 m,8m×8 m,12 m×12 m,16 m×16 m and 20 m×20 m）were generated,and the fusion image effect was evaluated.There is little difference between the mean values of the five scales fusion images.The average gradient and correlation coefficient of the 16 m×16 m fusion image are the largest,and the standard deviation is second only to that of the 12 m×12 m fusion image.Combined with the vegetation index analysis,the RVI,NDVI and DVI of the 16 m×16 m fusion image are the closest to the measured values,and the performance capacity of the vegetation spectrum is 92.87%,82.67%and 92.31%,respectively.The spectral information is rich,the spectral fidelity is high,and the fusion effect is the best.The image scale of 16 m×16 m is suitable for the distribution characteristics of winter wheat plots in Jiangsu Province,which is conducive to the identification and growth monitoring of winter wheat.（2）Based on the GF-1/WFV image of Shuyang County and the re-sampled HJ-1A/CCD image,a training sample 1 and a training sample 2 were established,respectively,and three classification combinations were formed with the two Shuyang image images to classify the terrain objects of Shuyang County and extract the winter wheat planting area.The results showed that the overall accuracy of classification combinationⅠ（training sample 1 and GF-1/WFV image）was93%,Kappa coefficient was 0.91,and the extraction of winter wheat planting area was 88473.63ha,with an accuracy of 94.37%.The overall accuracy of classification combinationⅡ（training sample 1 and HJ-1A/CCD image）was 87%,the Kappa coefficient was 0.83,and the extracted planting area of winter wheat was 103998.96 ha,with an accuracy of 89.07%.The overall accuracy of classification groupⅢ（training sample 2 and HJ-1A/CCD image）was 82%,Kappa coefficient was 0.76,and the extracted planting area of winter wheat was 109808.19 ha,with an accuracy of82.88%.Classification combination I had the highest classification accuracy and the highest extraction accuracy of winter wheat planting area.In conclusion,the method of winter wheat planting area extraction based on the combination of appropriate spatial scale and object-oriented classification developed in this study has good extraction effect,and can meet the technical requirements of county agricultural departments for obtaining information of winter wheat planting area at county level.（3）Leaf Area Index（LAI）and biomass were both used as model assimilation variables.PSO algorithm was introduced to optimize the parameters of the model.After running the optimized parameters,the model obtained the single-point scale winter wheat yield estimation.The R²,RMSE and RAE of the optimized parameters were 0.91,436.79 kg/ha and 5.98%respectively.BP neural network and remote sensing spectral characteristics were used to establish the winter wheat yield point surface conversion model(WWYTM_BP).The 3-6-1 BP neural network model has the highest conversion accuracy,with R² of 0.95,RMSE of 580.01 kg/ha and RAE of5.35%.The winter wheat yield of Shuyang County was estimated by using appropriate spatial scale images.The 30 estimated values were selected and their accuracy was verified with the measured values at the same location points.The R²,RMSE and RAE were 0.89,484.53 kg/ha and 6.71%respectively.The paper made a classification map of winter wheat production in Shuyang County and calculated the field area data of each yield grade.According to the data,the area of GradeⅠyield field was 8.12%of the total planting area of winter wheat,that of GradeⅡyield field was 52.46%,that of gradeⅢyield field was 25.07%,and that of gradeⅣyield field was 14.35%.The results showed that the winter wheat yield estimation method based on the combination of suitable spatial scale image and growth model had a good effect.