Urban runoff chemical composition is highly variable and results from both localized and regional effects such as physical structure, landscape, and climate. A wide range of physical and chemical conditions are present across the gradient of human-developed landscapes, even within a single city or neighborhood. These conditions greatly affect urban biogeochemical cycling and introduce various anthropogenic contaminants, which together create detectable patterns of chemical mixtures across a city. Our work focuses on detecting these signature chemical patterns and understanding their relationship to catchment-scale conditions within the urban mosaic. The patterns are then used to predict changes in water quality associated with climate and development. We used a three-step machine learning workflow to examine how site and watershed attributes drive the composition of urban runoff. Using hierarchical clustering, we first identified 6 stormwater signatures — distinct combinations of 9 common water quality indicators (TSS, TDS, Pb, Zn, Cu, TP, TKN, NO3+NO2, and BOD) — from 1,302 storm events in 26 US cities measured between 1992 to 2003 and reported in the National Stormwater Quality Database (NSQD). Next, we combined the NSQD data with environmental information for each storm event (land cover, land use, topography, climate, and weather). Using a decision tree algorithm, we examined how location, storm event, and watershed attributes are associated with the identified chemical signatures.
Results/Conclusions
We found strong evidence for correlations between the 6 stormwater signatures and watershed attributes, including spatial and temporal dimensions. We classified these attributes into watershed typologies — typical urban mini-biomes that produce one of the 6 signatures, or alternately, produce a seasonal combination of signatures. Chemical signatures in colder and wetter climates indicate generally cleaner water quality across all indicators, whereas warmer, drier climates indicate increased biogeochemical hotspots and hot “moments” that reflect a direct influence from urban landscapes. This also holds true for localized areas within the urban mosaic; individual sites show seasonal alternation between signatures, with cleaner signatures associated with cooler temperatures, greater canopy cover, and lower imperviousness. Results from this research expose fundamental interrelationships between place and pollution that are relevant to understanding how stormwater chemistry changes across a gradient of urban development.