Change Characteristics Of Temporal And Spatial Distribution Of Wetland Plant Communities In The Western Songnen Plain Based On Sentinel Time-series Data

Wetlands are one of the most biodiverse ecosystems,playing an important role in climate regulation,water purification,flood control,drought resistance,and habitat provision.Therefore,the health status of the wetland ecosystem is related to the safety and stability of regional ecology.However,under the influence of climate change and human activities,wetland degradation and area loss have become serious environmental problems.It become one of the important factors restricting the realization of the United Nations Sustainable Development Goals.The wetland plant community is considered the basic unit of wetlands.Obtaining accurate spatial distribution information of wetland plant communities is important for wetland health assessment,sustainable habitat management,and wetland dynamic change monitoring.At present,in order to prevent further loss and degradation of wetlands,it is urgent to strengthen the monitoring of wetland dynamic changes at the community scale.There are abundant wetland resources in the western Songnen Plain.The health of wetlands is not only related to the ecological security of northeast China but also crucial to the survival and breeding of migratory birds along the East Asia-Australia waterfowl migration corridor.Therefore,in this study,with the help of the Google Earth Engine(GEE)platform,Sentinel 1/2 time-series and DEM data were used to extract the distribution data set of wetland plant communities in the western Songnen Plain from 2016 to 2022.Based on classification results,the spatiotemporal evolution characteristics of different wetland plant communities were analyzed.Firstly,the sample data set in 2016-2022 was constructed by combining field investigation and sample migration based on the change detection method.Then,a feature vector set was built including phenological,timeseries,polarization,and terrain features,which can reflect unique spectrum,polarization,and terrain response information of different wetland plant communities.The established random forest classification model was used to classify wetland plant communities in the western Songnen Plain from 2016 to 2022.The classification results were then post-processed and the accuracy was evaluated.Finally,based on the annual distribution data set of wetland plant communities,the temporal and spatial evolution characteristics of different wetland plant communities in the western Songnen Plain,Zhalong,Momoge,and Xianghai reserves were analyzed.The main conclusions of this paper are as follows:(1)The sample migration method based on change detection can achieve accurate and rapid marking of wetland plant community samples,which will help the launch of dynamic monitoring of wetlands.Based on the measured sample data obtained from the field wetland survey in 2022,the sample data of five wetland plant communities were expanded,namely P.australis,T.orientalis,S.triquater,S.glauca,and C.meyeriana.This process was realized by comparing the spectral differences of multispectral remote sensing images and the differences of NDVI time-series curves of different wetland plant communities in different seasons.Then,based on the expanded sample data of2022,the sample migration method based on change detection was used to realize the migration of sample data from 2016 to 2021.Finally,the sample data set from 2016 to2022 was constructed.Except for the small number of C.meyeriana in 2016-2018,the sample migration of other wetland plant communities has achieved better results.(2)The random forest algorithms using feature vector sets including phenological,time-series,polarization,and terrain features can accurately identify different wetland plant communities.Moreover,in various characteristics,phenological and time-series features have important contributions to classification,especially IRS_NDVI,BV_NDWI,BV_NDVI,and SA_NDVI.The results of accuracy verification showed that the classification results have high accuracy.The average overall accuracy of the classification from 2016 to 2022 was 89.5%,and the average Kappa coefficient was0.87.The overall accuracy and Kappa coefficient of the classification in 2022 were the highest,which were 93.9% and 0.92 respectively.The overall accuracy and Kappa coefficient of 2016 were the lowest,81.8% and 0.77 respectively.(3)Studies have shown that the area of P.australis,T.orientalis,and C.meyeriana increased first and then decreased in western Songnen Plain during 2016 and 2022.The area of P.australis increased first and then decreased,while the area changes of S.triquater generally showed a gradually reduced trend in Zhalong reserve.The area of P.australis generally showed a gradually reduced change trend,the area of S.glauca increased first and then decreased,and the area of C.meyeriana had gradually increased in Momoge reserve.The area of P.australis generally showed a gradually increased change trend,T.orientalis decreased first and then increased,and S.triquater gradually decreased in Xianghai reserve.The factors that lead to the spatiotemporal evolution of wetland plant community distribution include not only natural climatic factors but also human activities.In addition,the limitations of remote sensing classification technology can also lead to the “fake-change” of wetland plant communities.