北京市黑臭水体治理的动态遥感监测及影响因素分析

为及时、准确掌握黑臭水体治理进展，基于“北京二号”影像数据和同期的野外综合水体实测数据，采用深度学习算法对黑臭水体进行识别，并引入地理探测器对黑臭水体影响因素进行定量分析。结果表明：基于Faster R-CNN算法的黑臭水体遥感识别，总准确率达到90%左右，短时间内(5~33 h)即可完成北京市建成区黑臭水体的筛查工作；在空间维度上，黑臭水体主要分布在中心城区以外，并在通州区、朝阳区和大兴区较为集中；在时间维度上，专项治理期间(2015—2018年)内，黑臭水体的数量和长度总体趋势都是递减的，但偶尔也有反黑现象；2018年底，在全市建成区范围内，已全面消除黑臭现象；在一年内，第1季度水体环境最好，第2季度次之，第3季度最差，从第4季度开始好转；在北京市大兴区，土壤全氮量(贡献率为32.07%)和周边养殖场排污(贡献率为27.04%)是黑臭水体形成的主要影响因素，高程(贡献率为8%)、土壤类型(贡献率为7.6%)和土地利用类型(贡献率为6.1%)的贡献率较弱。由此可以看出，基于Faster R-CNN算法识别影像中的黑臭水体识别准确率高，可及时、准确地监测城市黑臭水体治理情况，使用地理探测器可定量分析并确定各影响因素的贡献率。本研究成果可为城市黑臭水体的动态监测和治理提供有力的技术支撑。

Dynamic remote sensing monitoring and its influence factors analysis for urban black and odorous water body management and treatment in Beijing,China

To timely and accurately grasp the progress of black and odorous water body management and treatment, as well as its spatial-temporal distribution, based on the "Beijing No.2" image data and the field monitoring data of surface water quality detected at the same time, the deep learning algorithm was used to identify the black and odorous water body with different water quality, and the geographic detector was introduced to quantitatively analyze the causes of black and odorous water bodies. The results show that: the total accuracy of the remote sensing identification of the black and odorous water body based on fast r-cnn algorithm was about 90%, and the screening of the black and odorous water body in the built-up areas in Beijing could be fulfilled within a few hours. In spatial dimension, the black and odorous water body mainly distributed outside the central city, and concentrated in the North Canal and Daqinghe River Basin. In temporal dimension within special treatment, the number and length of black and odorous water generally decreased, while the back to black and odorous water body occurred occasionally. At the end of 2018, the black and odorous water bodied were eliminated in the built-up areas in Beijing. Among the four quarters of one year, the best water environment occurred in the first quater, and then was the second quater, the worst was the third quarter and it changed better in the four quarter. Daxing District was selected as the representative to analyze its causes, and eight indicators of internal and external sources were selected as risk factors. The soil nitrogen content(contribution rate of 32.07%) and sewage discharge from surrounding livestock farm (contribution rate of 27.04%)were the dominant factors, while altitude(contribution rate of 8%), soil type(contribution rate of 7.6%) and land use(contribution rate of 6.1%) contributed less to the black and odorous water body. Therefore, it could accurately identify black and odorous water body in high-resolution remote sensing images based on fast r-cnn algorithm, the technical framework is simple, which will help to timely and comprehensively grasp the distribution and treatment progress of black and odorous water body. At the same time, the quantitative analysis of the causes of black and odorous water body also provides a strong technical support for urban black and odorous water remediation.