Abstract: Remote sensing water indices are important basis for surface eco-environment analysis,and are critical to large-scale waterbody interpretation and hydrological monitoring.A variety of water indices focus on large-scale,short-term,non-urbanized regions,while the extraction of urban water bodies has been plagued by the applicability of these indices.The study collected Landsat images from 1986 to 2020 in Shenzhen city,and adopted three commonly used water indices(TCW,NDWI,MNDWI) and five improved water indices(WI2015,AWEInsh,ANDWI,MBWI,WI2020).The interpretation ability of these water indices in the urban noise area were analyzed by taking five regions of building shadow,high albedo building,low albedo building,tidal-flat and mountain shadow as examples.The results show that:(1)Producer accuracy of water indices are negatively correlated with user accuracy(Otsu threshold),and TCW,NDWI,ANDWI,AWEInsh can effectively filter architectural noise and mountain shadow,but unrecognize small river water bodies,MNDWI,WI2015,MBWI,WI2020 are the opposite;(2)The spectral transformation type and difference type water indices consider the absolute value of multiple bands with low separability,and they are suitable for high-quality image mapping.The ratio type and logical operation type water indices calculate the relative value of multiple bands,and suitable for long-term monitoring;(3)The method of multi-temporal sampling verification in typical cases highlights the interpretation ability and threshold applicability of water indices,and the NDWI is most applicable in urbanized areas,while the WI2020 achieves the highest interpretation accuracy in non-urbanized areas.The above research provides a certain theoretical reference and practical basis for the selection of water index and accuracy verification for urban hydrological monitoring.