Research On Fractional Vegetation Cover(FVC) Inversion And FVC Products Validation In The Three-River Source Region

The Three-River Source Region(TRSR)is one of the most significant water conservation bases on the Qinghai-Tibet Plateau(QTP),and it performs critical tasks such as regulating the northern hemisphere’s carbon cycle,regulating climate change,maintaining the plateau’s ecological security barrier,and supplying freshwater resources to the plateau and surrounding areas.Alpine grassland,as the main vegetation in the TRSR,has suffered serious degradation as a result of continuous global warming and increasing human activity,endangering the stability of the QTP’s ecological security barrier and the water security of the people living on the QTP and its surrounding areas.Fractional vegetation cover(FVC)is an essential ecological parameter index for measuring surface vegetation change,as well as an ideal index for dynamic monitoring of the eco-environment.As a result,tracking dynamic changes in FVC is critical for guiding the restoration of the TRSR’s alpine eco-environment as well as the planning and construction of the Sanjiangyuan National ParkThe dynamic changes of FVC in the alpine grassland in the TRSR and the driving or limiting factors affecting the dynamic changes of FVC were first explored based on the multi-source FVC remote sensing products.And the discrepancies in the results of different remote sensing products were quantified and analyzed.Secondly,based on the measured FVC data that can match the scale of satellite remote sensing,multi-source satellite remote sensing data,terrain data,and vegetation index data,machine learning algorithm was used to conducted the evaluation of the accuracy of FVC inversion in the TRSR.A multi-scale FVC product dataset with long-time series and spatial and temporal integrity in the TRSR was produced by using the inversion method with high precision and remote sensing image fusion method.Finally,the accuracy of the current international mainstream FVC products was validated utilizing the measured FVC data.On this basis,the influence of multi-scale validation methods on remote sensing product validation results is evaluated,and the influence of the heterogeneity of the underlying surface on product validation results is discussed.The main conclusions of the study are as follows:(1)The results of eco-environment monitoring and simulation research are greatly affected by the inconsistency of remote sensing products.The dynamic changes of alpine grassland in the TRSR simulated by different remote sensing products have significant discrepancies in statistical values and spatial distribution.The changing trend of FVC of the alpine grassland in about 70%of the study area is controversial.In addition,the constraints or driving factors that different remote sensing products can explain vegetation cover changes are significant discrepancies.(2)Inversion of the 250-meter spatial resolution FVC in the TRSR has high accuracy using a bidirectional long-term short-term memory neural network algorithm constructed from a multi-dimensional feature dataset(R~2=0.886,RMSE=0.093,y=0.868x+0.077),which was higher than random forest algorithm(R~2=0.886,RMSE=0.079,y=0.739x+0.142).In addition,the 30-meter spatial resolution FVC product produced by integrating multi-source remote sensing data,continuous in time and complete in space,in this study is high quality(R~2=0.92,RMSE=0.094).(3)The accuracy of GEOV3 FVC products in the TRSR is the highest(R~2=0.809,RMSE=0.143),better than GEOV1(R~2=0.791,RMSE=0.142),GEOV2(R~2=0.746,RMSE=0.114),GLASS(R~2=0.804,RMSE=0.128),and MuSyQ(R~2=0.750,RMSE=0.121)FVC products.A high-resolution FVC reference map,which was inverted based on the 30m-scale measured FVC data and Landsat-8 data was upscaled to validate the FVC products.The product accuracy evaluation results obtained were higher than the results of direct validation using measured FVC data,resulting in an overestimation of the actual accuracy of FVC products.In addition,the heterogeneity of the underlying surface of the monitoring plots will have a non-negligible impact on the validation of FVC products,but the heterogeneity of the underlying surface of the monitoring plots can be quantified and the monitoring plots with greater heterogeneity can be moved to reduce the uncertainty of the validation results.