A Research On Classification And Quantitative Inversion Of Physiological Structure Parameters For Marsh Vegetation Based On Aerospace Remote Sensing Data

Marsh vegetation classification and leaf chlorophyll content(CCC)remote sensing inversion is of great significance for the assessment and management of marsh wetlands and the formulation of reasonable protection and restoration policies.However,the current research still has the following problems:(1)The spatial distribution of marsh vegetation is complex and the spatial heterogeneity is poor,which brings great challenges to the high-precision classification of marsh vegetation;(2)Due to the complexity habitat of marsh vegetation,it is difficult to collect the CCC measured data for marsh vegetation that matches pixel scales of the remote sensing images,which brings great uncertainty to the results of the CCC remote sensing inversion for marsh vegetation.In this study,the Honghe National Nature Reserve is selected as the research area.Based on multi-source remote sensing data,high-precision marsh vegetation classification and near real-time CCC remote sensing inversion are achieved.The research content and main conclusions of this article are as follows:(1)An object-based multi-dimensional dataset based on GF-1 PMS and ZY-3 MS satellite remote sensing data is constructed.The object-based Random Forest(RF)algorithm is used for marsh vegetation classification.The results show that the use of multi-dimensional datasets improved the classification accuracy of marsh vegetation(1%?5%),but there are many redundant variables in the texture information.(2)The parameters(mtry and ntree)of the object-based RF algorithm is optimized to construct an optimal model suitable for marsh vegetation classification.Variable selection on the multidimensional datasets is performed to eliminate redundant variables.The results show that parameter optimization improved the stability and classification accuracy of RF algorithm.The elimination of redundant variables from the multidimensional datasets based on the three variable selection algorithms could effectively improve the classification efficiency and accuracy(2%?7%).Among them,the Recursive Feature Elimination(RFE)algorithm has the best performance.(3)A new method for UAV-assisted collection of marsh vegetation CCC sample data that matches pixel scales of remote sensing images is proposed.By extracting pure vegetation pixels,the uncertainty of obtaining CCC sample data at remote sensing images pixel-scale from UAV-scale CCC inversion results is reduced.The results show that the method of UAV-assisted collection of CCC sample data for marsh vegetation quantified the collection accuracy(R~2?0.86,RMSE?6.98)while expanding the number of samples.Extracting pure vegetation pixels reduced the uncertainty of CCC sample data collection at pixel scales of remote sensing images.There are significant differences between the two groups of CCC sample data before and after extracting pure vegetation pixels(P-value<0.05).(4)The application performance of GF-1 WFV,Landsat-8 OLI and Sentinel-2 MSI satellite remote sensing data in marsh vegetation CCC inversion are evaluated by the RF regression algorithm.The accuracy of CCC inversion is improved through parameter optimization of the RF regression model.The results show that parameter optimization further improves the accuracy of CCC inversion.Sentinel-2 MSI has the highest CCC inversion accuracy(R~2=0.79,RMSE=10.96)for marsh vegetation,because it contains the red-edge band that is more sensitive to vegetation characteristics.