Research On Remote Sensing Estimation Of Forest Quality And Stand Quality Based On Sentinel Data

As the largest ecosystem on earth,forests play a vital role in maintaining ecological balance,promoting biological evolution and community succession.Forest quality reflects the real productivity of forests and site quality reflects the potential productivity of forests,and they are indispensable tools for forest management.Evaluation of forest quality and site quality requires stand parameters as data sources and is a further application of stand parameters.At present,the acquisition of stand parameters is inseparable from forest resource survey work,and the traditional forest resource survey is not only time-consuming and laborious,but also difficult to achieve a large area coverage.At present,remote sensing technology has the advantages of simultaneous observation over a large area,high timeliness,and low cost,which makes it possible to obtain long time and large scale forest parameters efficiently,so it is of great significance to explore remote sensing estimation of forest quality and site quality.In addition,there is a certain spatial distribution pattern between forest quality and site quality,and studying the spatial correlation between them is of great value for regional forest management strategies.In this study,Sentinel-1/2 data were used as the data source,and the ecological"green-core" area of Changzhutan urban cluster was selected as the study area to explore the potential of remote sensing estimation of forest quality and site quality in combination with the ground-based forest resources survey data in the same period.The sensitivity of remote sensing features of different data sources to stand parameters,forest quality and site quality was analyzed,and the optimal set of features related to response variables was screened using random forest importance and stepwise linear regression models,and three estimation models such as vector regression(SVR),random forest regression(RF)and K-nearest neighbor regression(KNN)were combined to estimate forest quality and site quality by direct and indirect methods.The spatial distribution patterns of forest quality and site quality and their influencing factors were discussed from the perspective of remote sensing data,and the main findings are as follows.(1)The sensitivity of remote sensing features extracted from Sentinel-2 to forest quality and site quality is significantly higher than that of remote sensing features extracted from Sentinel-1.According to the importance ranking of remote sensing features,among the top 8 remote sensing factors with high sensitivity to forest quality(PSSRa(Pigment specific simple ratio chlorophyll index),VH,B12,B2,RVI(Ratio vegetation index),B8A_cor,GNDVI(NDVI of green band)and MTCI(Meris terrestrial chlorophyll index)),7 of them were from the remote sensing feature set extracted from Sentinel-2,1 remote sensing factor(VH)was from the remote sensing feature set extracted from Sentinel-1 remote sensing feature set,and PSSRa(Sentinel-2)had the highest importance for forest quality.Among the top 6 remote sensing factors with high sensitivity to site quality(B2,MTCI,RVI,B8A_me,VV_me,and VV),the 4 most important features were from the remote sensing feature set extracted by Sentinel-2,2 features were from the remote sensing feature set extracted by Sentinel-1(VV_me and VV),and B2(Sentinel-2)band has the highest importance for site quality.(2)The estimation accuracy of different stand parameters is closely related to the remote sensing data sources,and the estimation accuracy of combined Sentinel-1 and Sentinel-2 is significantly better than that of a single data source,and the quantitative estimation model has less influence on the accuracy of stand parameters.The coefficient of determination(R2)was distributed between 0.16 and 0.60,and the rRMSE was distributed between 0.18 and 0.38 in the estimation of stand parameters based on single Sentinel-1 data.The coefficient of determination(R2)was distributed between 0.01 and 0.44,and the rRMSE was distributed between 0.22 and 0.45 for the estimated stand parameters based on single Sentinel-2 data.For the estimated stand parameters based on the combination of Sentinel-1 and Sentinel-2,the coefficient of determination(R2)was distributed between 0.12 and 0.61,and the rRMSE was distributed between 0.20 and 0.40.Among the estimation results of different stand parameters,the highest estimation accuracy was the average tree height and average diameter at breast height,with the accuracy distribution between 75%and 80%,followed by their forest crown closure,density and stand age,with the estimation accuracy distribution between 68%and 75%,and the lowest accuracy was the forest stand volume,with the estimation accuracy distribution between 55%and 65%.(3)In the remote sensing estimation results of forest quality and site quality,the estimation accuracy of the indirect method with stand parameters as the remote sensing response variable was significantly higher than that of the direct estimation of forest quality and site quality.The best feature set for direct estimation of forest quality and site quality was selected from the Sentinel-1/2 combined feature set by combining stepwise regression models,in which the highest accuracy of forest quality estimation was supported vector regression(SVR)model(R2=0.28,rRMSE=0.24),and the highest accuracy of site quality estimation was supported vector regression(SVR)model(R2=0.08,rRMSE=0.21).The forest quality and site quality were obtained indirectly through the estimated stand parameters,where the estimation accuracy of forest quality was R2=0.36 and rRMSE=0.19,and that of site quality was R2=0.40 and rRMSE=0.15.Especially,the indirect method can limit the estimated outliers to a greater extent and improve the accuracy and reliability of forest quality and site quality estimation.(4)Forest quality and site quality have significant spatial correlation.The spatial distribution of forest quality and site quality in the study area estimated by combining Sentinel image data mainly shows that:in terms of slope direction,the highest average forest quality grades are on the southeast and west slopes,and the lowest is on the north slope;the highest average site quality grades are also higher on the east and south slopes;in terms of grade,the degree of matching between the spatial distribution grades of forest quality and site quality In terms of rank,the matching degree of spatial distribution of forest quality and site quality in the study area is 9.46%,85%of the area is basically matched,and only 5.53%is not matched.