Modeling community hydrogen sulfide exposure in an urban industrial area during routine and extreme events

Hydrogen sulfide (H₂S) is a toxic, odorous gas linked to acute and chronic health effects. In urban-industrial areas like Southern California's South Bay and Harbor communities, H₂S levels can be elevated and highly variable. Following a major warehouse fire in late 2021, H₂S concentrations peaked at levels exceeding California's acute air quality standard (30 ppb) by over 230-fold and remained elevated for weeks, coinciding with spikes in odor and headachecomplaints. This event highlighted how rapidly exposures can shift, motivating methods that can capture exposure patterns under both routine conditions and observed extremes. We developed exposure models for hourly average, daily average, and daily maximum H₂S using community fenceline monitoring data from January 2020 to October 2023, coupled with meteorology, distance to emission sources, land use, and temporal indicators. A key methodological contribution is the application of quantile XGB (QXGB) to estimate the probability of exceeding health-relevant H₂S thresholds conditional on observed predictors, supporting a novel exposure assessment metric. The best-performing models used XGB for daily average and daily maximum concentrations (test R² = 0.89 and 0.70, respectively). Key predictors included wind, humidity, and proximity to major sources. QXGB demonstrated skill in estimating hourly exceedances at both 2 ppb and 30 ppb thresholds, with estimated exceedance counts in the test set closely matching observed counts (e.g. 2 ppb N_obs = 5380 vs N_pred = 5299; 30 ppb N_obs = 120 vs N_pred = 78). Our analysis demonstrates the utility of fenceline monitoring for identifying spatial patterns consistent with likely emission sources and for characterizing both routine and extreme H₂S exposures in overburdened communities. The modeling framework, which combined concentration estimation with threshold exceedance probability estimation and spatial/temporal validation, supports environmental health investigations by improving exposure assessment across heterogeneous conditions, and demonstrates the value of targeted monitoring networks for tracking community-relevant exposures in industrialized urban regions.