Study On Marsh Wetland Extraction And Spatiotemporal Change In Heilongjiang Basin Based On Sentinel Image Feature Analysis And Sample Migration

As one of the three major ecosystems in the world,wetlands play an important role in maintaining the regional ecological balance and sustainable development.In recent years,wetlands have been continuously degraded and disappeared due to unreasonable reclamation and transformation,environmental pollution,and excessive use of water resources.The use of scientific and efficient methods for wetland monitoring is a research hotspot and difficulty in the field of wetland ecology.Remote sensing technology is an important method for wetland monitoring,but the traditional image data processing method is inefficient in large-scale wetland monitoring.The application of the new data source and new features of sentinel images is yet to be explored.In this study,the Heilongjiang River Basin in China,which is rich in wetland resources and undergoes dramatic changes,is used as the research area.The role of the Sentinel-2 red-edge bands and the Sentinel-1 radar band for wetland information extraction has yet to be explored.Among many features,wetland information extraction Advantageous features are rarely mentioned.Based on the Google Earth Engine(GEE)cloud platform,using Sentinel-2,Sentinel-1 and Landsat 5 TM satellite images,and using random forest algorithm for the 2018 and 1988 Heilongjiang river swamps in China Wetland information is extracted.A variety of experimental schemes were designed to study the effects of the red-edge bands and radar band on wetland information extraction,and the JM distance was used to find the best features.The spatiotemporal changes of marsh wetlands are analyzed,and the driving forces of marsh wetland changes are analyzed in combination with meteorological data and socio-economic data.The main innovations and conclusions of the study are as follows:(1)The red-edge features of Sentinel-2 and the radar features of Sentinel-1 can improve the classification accuracy of swamp wetlands.The red-edge features and the radar features improved the swamp’s mapping accuracy by 7.56%and 5.04%,respectively,and the user accuracy by 5.01%and 5.39%,respectively.The JM distance is used to optimize the features,and the feature dimension is transformed from 28 to 21 while the classification accuracy is maximum.In terms of the degree of separation of the swamp from other types,red-edge features>other optical features>terrain features>radar features.After random forest tuning,the overall accuracy in 2018 reached 91.54%,the Kappa coefficient was 0.90,and the swamp mapping accuracy and user accuracy were 88.79%and 88.30%,respectively.(2)Using the 2018 training sample point migration to 1988,the overall classification accuracy reached 81.38%when the total number of samples was reduced by 12.5%,and the sample migration achieved good results.After optimization using JM distance,the number of features decreased from 14 to 10.After random forest tuning,the overall accuracy reached 81.38%in 1988,the Kappa coefficient was 0.78,and the swamp mapping accuracy and user accuracy were 80.79%and 78.14%,respectively.(3)From 1988 to 2018,the area of marsh wetlands in the Heilongjiang River Basin decreased by 14.67%.In 2018,the swamp wetland area was 113618.88 km2,a decrease of 19539.23 km2 compared to 1988.In terms of spatial distribution,the areas with the most severe swamp reductions are located in the eastern Heilongjiang River Basin.According to sub-basin statistics,the most severely reduced river basins are below Jiamusi,below Muling estuary,and above Muling estuary,the reduction rates are 55.78%,54.24,%,and 47.76%,respectively.(4)The impact of social and economic factors on the area of marsh wetlands in the Heilongjiang River Basin is greater than that of climate factors.Correlation analysis shows that the correlation between socio-economic factors and marsh wetland area is greater than 0.8,while the correlation between climatic factors is about 0.1.Crop planting,especially the increase in paddy field planting area,the main reason for the decrease in wetland area.The area of paddy field in the whole basin increased by 14170.79km2 from 1988 to 2018.From 1988 to 2018,paddy fields below Jiamusi,above Muleng estuary,and below Muling estuary The area has increased by 6645.91 km2,accounting for 46.89%of the increase in paddy fields in the entire basin.The transformation of swamp wetlands into paddy fields is the direct cause of the shrinking swamp area in the eastern Heilongjiang River Basin.In the mid-1990s,Heilongjiang Province’s decision to reclaim "waste land" accelerated the process.