| To determine whether the excessive application of fertilizer in the soil harden and compare and improve GPS satellites and BDS single frequency satellite reflected signal inversion accuracy and reliability of the soil moisture,using traditional regression method,the random forest law and based on the feature selection of random forest law inversion of soil moisture,compare different inversion method on the same satellite and the pros and cons of different kinds of satellite inversion method same sex;The variation of multipath reflection components before and after fertilization was analyzed,and the accuracy and reliability of the three inversion models were verified.For the satellites collected and screened in this paper and the specific fertilization environment,the main conclusions include:1)The correlation between interference characteristic parameters of satellite signals and measured soil moisture: the frequency of each satellite signal is negatively correlated with soil moisture,and the correlation between the frequency,amplitude and phase of multi-path reflection signal of C28 satellite in BDS satellite and measured soil moisture is higher than that of other interference characteristic parameters of BDS satellite signals;The correlation between the frequency and amplitude of G24 satellite signals and soil moisture is higher than that of other GPS satellite signals.After fertilization also has the above law.2)Model accuracy before fertilization: both stochastic forest algorithms can effectively improve the reliability of soil moisture inversion model established by traditional regression method for BDS satellite,compared with the optimal model of traditional regression method,the accuracy of traditional random forest model has the largest improvement for C20 satellite,and the determination coefficient R~2 has been increased by 29.95%.For GPS satellite,the accuracy of G05 satellite has the largest improvement,and the determination coefficient R~2 has been improved by 31.10% overall.The model accuracy is highest for C28 satellite(R~2=0.9033)and G24 satellite(R~2=0.9109).Only C28 and G05 are superior to the traditional random forest regression model after feature selection,in which the determination coefficient R~2 of C28 increases by 1.42% and the root mean square error RMSE decreases by 6.82%.The determination coefficient R~2 of G05 satellite increases by 1.39%,and the root mean square error RMSE decreases by 3.10%.3)Model accuracy after fertilization: For the traditional random forest regression method,THE model accuracy of C28 in BDS satellite is the highest,and its determination coefficient R~2 is 0.8880,which is 1.73% higher than the traditional regression method R~2.Among GPS satellites,G24 model has the highest accuracy and its coefficient of determination is 0.6693,which is 12.96% higher than the traditional regression method.For the random forest regression method based on feature selection,THE model accuracy of C28 in BDS satellite is the highest,with R2 of 0.7908,which is 9.41% lower than that of the traditional regression method.The accuracy of G24 model is the highest among GPS satellites,which is 11.43% higher than that of traditional regression method.4)Comparison of satellite signal frequency changes and model accuracy before and after fertilization:the frequency increased significantly after fertilization,and the higher the nitrogen content,the higher the frequency;At the same time,the detection depth of electromagnetic wave increased significantly,and the soil water content of 2-5cm surface decreased.The modeling accuracy of C28 and G24 satellites was the highest,and the accuracy of the above three models established by the selected satellites after fertilization was lower than that before fertilization.The traditional random forest regression method decreased by 0.0153 and 0.1235,respectively,and the traditional random forest regression method based on feature selection decreased by 0.2416 and 0.1927,respectively.Figure [57] Table [14] Reference [82]... |
PDF Full Text Request