Remote Sensing Quantitative Inversion Of Vegetation Cover And Management Measure Factor Based On Multi-angle Data And PROSAIL Model

Vegetation cover and management measures factor(C-factor)is the most sensitive and variable factor in soil erosion model.Using normalized different vegetation index(NDVI)to extract vegetation coverage to estimate C-factor is the most commonly used method.However,the traditional vegetation coverage ignores the vertical structure information of vegetation,and the inversion accuracy is low.Leaf area index(LAI),which can reflect the information of horizontal vegetation cover and vertical structure,is more suitable for quantitative evaluation of soil erosion.In this study,Mount Zijin and Mufu Mountain in Nanjing city were selected as the study areas.Through the methods of field experiments,remote sensing image,radiation transfer model and mathematical models,this study established the LAI inversion model of random forest model based on radiation transfer model and multi-angle PROBA/CHRIS remote sensing data,and established a quantitative coupling model between LAI and C-factor,which provided a new method for C-factor remote sensing quantitative inversion.The main research results are as follows:(1)Selection of optimum vegetation index for vegetation erosion prevention.The average value of C-factor for different vegetation types measured in the field was0.0262.The mean value of C-factor was coniferous forest(0.0391)>broadleaf forest(0.0266)>coniferous-broadleaf forest(0.0194).The soil and water conservation ability of coniferous-broadleaf forest was the strongest.The forest land with complete vegetation structure had strong water storage and soil conservation capacity,and the risk of soil erosion was small.The regression results of six vegetation indices at forest scale and ten remote sensing vegetation indices at regional scale with C-factor showed that the regression effect between LAI and C factor was the best,and the regression equation with the highest accuracy was C=0.4337exp(-0.84LAI).(2)Reflectance simulation and accuracy evaluation based on multi-angle PROBA/CHRIS remote sensing data and PROSAIL model.The root mean square error(RMSE)between the leaf reflectance and the measured leaf reflectance simulated by the PROSPECT model was 0.0347,which was highly accurate and can be used to simulate the canopy reflectance of the forest.The parameter sensitivity analysis of the PROSAIL model was carried out by qualitative analysis and sensitivity calculation,and the sensitivity intensity of each parameter was determined as LAI>C_ab>C_m>S_L>N>C_w.The sensitivity of different observation angles of multi-angle PROBA/CHRIS remote sensing data to LAI was 55°>36°>0°>-55°>-36°.The sensitivity of the forward observation angle was greater than the backward direction,and the larger the observation angle,the greater the sensitivity.The accuracy of the canopy reflectance simulated by the PROSAIL model was 0°>36°>-36°>55°>-55°.(3)Construction of LAI inversion model based on the random forest model.In this study,we selected eight vegetation indices and analyzed the correlation between LAI and vegetation indices calculated by the simulated canopy spectral reflectance at different observation angles.The results showed that PVI had the highest correlation with LAI,and the correlation coefficient reached above 0.89.In the LAI inversion models based on the single angle data,the accuracy of the forward observation angle of 55°was the highest with the R²,RMSE and MAPE were 0.9157,0.2357 and 0.0426,respectively.In the LAI inversion models based on the multi-angle data,the three angles combination of 0°,36°and 55°had the highest accuracy with the R²,RMSE and MAPE were 0.9184,0.2319 and 0.0415,respectively.The accuracy of LAI inversion models increased with the observation angle.However,after more than three angles,the accuracy decreased.The reason may be that too much angle will brought about some problems such as the spectral information redundancy,blade and soil shadow and so on.Compared with no-linear regression model and Landsat 8 OLI data,the random forest model based on multi-angle remote sensing data and radiative transfer model can significantly improve the accuracy of LAI inversion.The estimated LAI value of forest in the western part of Mount Zijin ranged from approximately 0.554 to 5.709 by using the model of three angles combination,with a mean of 3.04.And the forest in the western part of Mount Zijin showed that the LAI in the northern and southern parts was higher than those in the middle part.(4)Remote sensing quantitative inversion and application of the C-factor.In this study,twelve kinds of C-factor inversion models based on traditional vegetation coverage,LAI,vegetation indices and bands were selected,and the accuracies of the models were evaluated and analyzed by sampling points in the multi-angle images.The results showed that the model based on multi-angle remote sensing data and LAI had the highest accuracy.The estimated C value of forest in the western of Mount Zijin was between 0.006 and 0.174,and the mean value was 0.034.By comparing the inversion results of the models based on traditional vegetation coverage and LAI,it can be seen that the estimated C value based on the traditional vegetation coverage inversion is small overall,and the RMSE was 0.0238.However,the estimated C value based on multi-angle LAI had the high accuracy,and the RMSE were 0.0074.