Watershed-scale green infrastructure reduces storm-driven Escherichia coli extremes: Tail-aware modeling and dependence-robust inference for bacterial water-quality management

Meeting recreational water-quality standards for fecal indicator bacteria (FIB) remains challenging in rapidly urbanizing watersheds because storm-driven mobilization generates highly skewed Escherichia coli ( E. coli ) concentrations and sparse monitoring seldom captures the full distribution. We developed a transferable watershed-scale modeling and inference workflow that integrates Soil and Water Assessment Tool (SWAT) hydrology-bacteria simulation, sampling-date model evaluation, tail-focused metrics, threshold screening, and dependence-robust uncertainty estimates. The workflow was demonstrated in the Village Creek-Lake Arlington watershed (Texas, USA) using daily simulations (2011-2019; n = 3,287 days) and Texas Commission on Environmental Quality Station 10786 observations (n = 166). Sampling-date evaluation was acceptable for scenario screening on the log10 scale (R 2 = 0.59; NSE = 0.55; PBIAS = -7.2%; median absolute log 10 error = 0.40; 88% within one order of magnitude), although individual grab-sample predictions remained dispersed. Baseline concentrations were strongly right-skewed [geometric mean (GM) = 132 CFU/100 mL; P 95 = 5,947; P 99 = 14,277]. A dependence-robust log-linear driver model identified precipitation as the dominant control (beta ln[1 + P] = 1.44, p < 0.001), supporting the use of event-pathway green infrastructure (GI) scenarios. Six hypothetical GI scenarios were evaluated across urban land-cover targets. GI effects were modest for long-term geometric means but concentrated in the upper tail: the all-urban 70% scenario reduced GM by 20% and P 95 /P 99 by approximately 70%, whereas the high-adoption all-urban 90% scenario reduced GM by 33%, P95/P99 by approximately 89 – 90%, and exceedance days by 12% (>126 CFU/100 mL), 27% (>235 CFU/100 mL), and 48% (>410 CFU/100 mL). Results indicate that GI can substantially attenuate storm-driven microbial extremes, but full attainment is likely to require complementary chronic-source controls. The workflow is transferable to other mixed land-use watersheds where sparse FIB monitoring and event-driven extremes complicate bacteria pollution control.