Research On Spatiotemporal Variation Of Vegetation In The Dongting Lake Wetland

Wetland is one of the largest "carbon pools" on the earth,which play a critical role in the regional ecological environment and climate.Due to the combined influence of natural and human factors,large areas of wetland are degraded and disappeared,facing major threats.It is of great importance for wetland protection and management,restoration,and reconstruction to achieve a high-precision wetland map,analyze the spatiotemporal changes of wetland vegetation,and to prob its driving mechanism by using remote sensing technology.In this study,an object-based Stacking algorithm was proposed for Dongting Lake wetland vegetation mapping based on the multi-temporal spectral reflectance,vegetation indices,phenology,and backscattering coefficient dataset.Additionally,the Spatial and Temporal Non-Local Filter-Based Data Fusion Method(STNLFFM)model was used to reconstruct high spatiotemporal resolution Normalized Difference Vegetation Index(NDVI)data.The Breaks For Additive Seasonal and Trend(BFAST)algorithm was used to analyze the spatiotemporal wetland vegetation changes.Finally,the driving mechanism of spatiotemporal changes of wetland vegetation was quantitatively explored by using a partial correlation algorithm.The main conclusions are as follows:(1)Using multi-source,multi-temporal and multi-feature remote sensing data is conducive to significantly improve the accuracy of wetland vegetation classification.The accuracy(OA=80.73%,Kappa=0.79)of wetland vegetation classification using multi-temporal Sentinel-2 multispectral data was achieved.The classification accuracy was improved(OA=90.67%,Kappa=0.89)by combining the time series Sentinel-1 backscattering coefficient and fusing the NDVI time series.When using multi-temporal Sentinel-2 spectral data,vegetation indices,time-series Sentinel-1 backscattering coefficients,fusion NDVIs,and vegetation phenology parameters,the highest classification accuracy can be obtained(OA=91.37%,Kappa=0.90).(2)The overall accuracy and Kappa coefficient of the proposed object-based Stacking algorithm are 91.37%and 0.90,respectively,which are 4.13%and 0.04 higher than that using the pixel-based method.Moreover,the object-based stacked generalization algorithm is superior to traditional classification algorithms(such as Random Forest,k-Nearest Neighbor,and Support Vector Machine)in mapping wetland vegetation over high heterogeneity areas.Results indicate that the proposed algorithm has the application potential in large-scale wetland vegetation classification.(3)Employing STNLFFM and BFAST algorithms is enable to effectively analyze the spatiotemporal change patterns and differentiation characteristics of a wetland.The STNLFFM spatiotemporal fusion model was used to generate 240 monthly synthetic NDVI images from 2000?2019 and then the BFAST algorithm was applied to capture wetland vegetation changes.The accuracies of wetland vegetation change detection on the spatial and temporal scales are 82.6%and 84.6%,respectively.From 2000 to 2019,more than 50%of the vegetation pixels exhibited at least two breakpoints,with?5%of the vegetation pixels exhibiting eight breakpoints.(4)The driving forces of Dongting Lake wetland vegetation are analyzed using the partial correlation-based algorithm.The Dongting Lake wetland is becoming greener during 2000?2019,with NDVI increases of 0.006 per year.,Climate change is the key driving factor of Dongting Lake wetland vegetation change,which threatens wetland vegetation(59.19%).Regarding climate factors,the influence of solar radiation on vegetation was found to be stronger than that of temperature and precipitation.Human activities have favored wetland vegetation recovery(58.85%).