Streams in urban areas are prone to degradation. While urbanization-induced poor water quality is a widely observed and well documented phenomenon, the mechanism to pinpoint local drivers of urban stream degradation, and their relative influence on water quality, is still lacking. Utilizing data from the citizen science project FreshWater Watch, we use a machine learning approach to identify key indicators, potential drivers, and potential controls to water quality across the metropolitan areas of Shanghai, Guangzhou and Hong Kong. Partial dependencies were examined to establish the direction of relationships between predictors and water quality. A random forest classification model indicated that predictors of stream water colour (drivers related to artificial land coverage and agricultural land use coverage) and potential controls related to the presence of bankside vegetation were found to be important in identifying basins with degraded water quality conditions, based on individual measurements of turbidity and nutrient (N-NO3 and P-PO4) concentrations.
Keywords: Citizen science; Eutrophication; Machine learning; Turbidity; Urbanization; Water quality.
Copyright © 2017 Elsevier B.V. All rights reserved.