GEE-based Analysis Of Water Body Extraction And Its Geographical Influence In The Selenga River Basin In Mongolia

The Mongolian Plateau is an arid and semi-arid region,so the river systems have a great impact on the resource environment and the ecological environment of this region.The Selenga River basin is at the core of China-Mongolia-Russia Economic Corridor.Having accurate spatial and temporal information of waterbody distribution of this river basin is of great significance for studying the effects of resource and environment in this region.This paper studies the portion of the Selenga River basin in Mongolia.Based on GEE(Google Earth Engine)platform and using Sentinel-2 images as the data source,different machine learning methods were used to extract waterbody information.Different waterbody extraction methods were then evaluated and a suitable one for extracting waterbody for the Mongolia Plateau region was identified.This provided a methodological support for studying the evolution mechanism of curved rivers.Combined with the river network data in this region,the 2018 river distribution information of the Selenga River basin was obtained and used for correlation analysis with the statistical information(population,animal husbandry and size of cropland)in the corresponding administrative regions,and with the size of river areas in the Selenga River basin,to conduct a geographical impact analysis.The main research findings of this paper are as follows:(1)Different machine learning models were constructed using feature datasets of Normalized Difference Water Index(NDWI),Modified Normalized Difference Water Index(MNDWI)and Automated Water Extraction Index(AWEI)to see which model can be used in the information extraction for a curved river in a plateau region.The results showed that given the computing support of a new cloud platform(such as GEE)providing image data source and automated processing,all machine learning methods can extract the waterbody information for a plateau region automatically,quickly and effectively.(2)After comparing the suitability of Support Vector Machine(SVM)model and Deep Neural Networks(DNN)model used in waterbody extraction for the Mongolian Plateau,it was found that all machine learning methods can effectively complete the extraction of hydrological information for a large region,especially for large-sized waterbodies,and they all can extract curved rivers and branched rivers accurately.But in the extraction for mountain rivers,DNN model had significantly less flow break and void than SVM model.The results of these 2 models were both better than the waterbody dataset from Joint Research Centre(JRC).Waterbody extraction by DNN model had an overall accuracy of 97.65%,with Kappa coefficient being0.8755.(3)Combining the DEM river network extraction with the waterbody extraction results from DNN model,the 2019 river distribution information for the Selenga River basin in Mongolia was obtained.Using spatial analysis function from Arc GIS to analyze the secondary basin in each administrative region,the spatial distribution of the rivers was obtained.Combining with population,livestock and geographical factors including cropland information,an overlay analysis was conducted to explore the relationship between the river system features and the geographical locations.This analysis found that the provinces with higher population density also had larger size of river areas.The livestock density was consistent with the population density.The two provinces with larger cropland percentage,Darhan province and Selenga province,are both located in the main stream basin of the Selenga River and are greatly affected by the water source.