基于GEE的中国西南地区水体变化分析
Analysis of Water Body Change in Southwest China Based on Google Earth Engine

湖泊河流等水体是区域气候和环境变化的敏感指示器，其面积变化能客观反映内陆流域内的水量平衡过程，基于遥感大数据平台，快速提取水体变化信息，分析影响因素分析，具有较重要的研究意义。研究以Landsat数据与MODIS数据为数据源，通过水体指数模型与随机森林分类法进行西南地区的水体提取与分析，结果表明：(1) Landsat数据水体提取的效果好于MODIS数据，误提率与过提率均低于MODIS数据，Landsat数据水体指数模型方法提取水体的总体精度为86.1%，误提率为16.8%，过提率为11.0%，而MODIS数据水体提取模型的总体精度为81.9%，误提率为21.8%，过提率为14.4%；(2) 水体指数模型法所提取水体效果较随机森林分类法精度更高，随机森林方法在Landsat数据下总体精度为85.7%，误提率与19.2%，高于水体提取模型法的16.8%，但过提率为9.4%，较水体指数模型法低；(3) 基于Landsat数据通过M-K趋势检验法得出了西南地区2000~2020年水体面积呈现显著增加的趋势，总面积从2000年的28963.14平方公里增长到2020年的33859.17平方公里，水体主要分布于青藏高原地区与云贵高原地区湖泊与其发育的众多河流的中上游流域范围内。
Lakes, rivers, and other water bodies are sensitive indicators of regional climate and environmental change. Their area variations can objectively reflect the water balance processes within inland basins. Based on a big data platform for remote sensing, rapidly extracting information on water body changes and analyzing the influencing factors holds significant research importance. Taking Landsat data and MODIS data as data sources, the water body in Southwest China is extracted and analyzed using the water body index model and random forest classification. The results show that: (1) the effect of water body extraction from Landsat data is better than MODIS data, and the false extraction rate and over extraction rate are lower than MODIS data. The overall accuracy of water body extraction from Landsat data is 86.1%, the false extraction rate is 16.8%, and the over extraction rate is 11.0%. The overall accuracy of MODIS data water extraction model is 81.9%, the false extraction rate is 21.8%, and the over extraction rate is 14.4%; (2) The effect of water body extracted by water body index model method is higher than that of random forest classification method. Under Landsat data, the overall accuracy of the random forest method is 85.7%, and the false extraction rate is 19.2%, which is higher than 16.8% of water body extraction model method, but the over extraction rate is 9.4%, which is lower than that of water body index model method; (3) The M-K trend test method shows that the water area in Southwest China shows a significant increase trend based on Landsat data from 2000 to 2020, with the total area increasing from 28963.14 km² in 2000 to 33859.17 km² in 2020. The water body is mainly distributed in the lakes in the Qinghai-Xizang Plateau and the Yunnan-Kweichow Plateau and lots of middle and upper reaches of the river basin.