Jianyong Wu

Water-quality monitoring is increasingly becoming data-driven

Microplastics are pervasive contaminants in marine and coastal ecosystems. Although experimental and human biomonitoring studies suggest biologic plausibility for chronic health effects, population-level evidence linking environmental microplastic exposure to cancer incidence remains limited. We conducted a county-level ecological study of 255 U.S. coastal counties linking NOAA/NCEI marine microplastic exposure data with NCI/CDC State Cancer Profiles cancer incidence estimates. Counties were categorized by mean marine microplastic concentration as low, medium, or high exposure. Outcomes included incidence rates per 100,000 population for colorectal, lung, pancreatic, liver and intrahepatic bile duct, esophageal, kidney, thyroid, breast, prostate, and cervical cancers. Poisson regression was used to estimate incidence rate ratios (IRRs) comparing medium- and high-exposure counties with low-exposure counties, adjusting for county-level demographic, socioeconomic, healthcare access, and geographic covariates. Among 255 coastal counties, 53 (20.8%) had low exposure, 36 (14.1%) medium exposure, and 166 (65.1%) high exposure. High-exposure counties were more frequently located along the Atlantic and Gulf coasts and had a lower median household income, higher poverty, and higher uninsured rates than low-exposure counties. In adjusted models, higher microplastic exposure was not associated with a broad or consistent increase in cancer incidence across sites. For pancreatic cancer, the IRR comparing high- with low-exposure counties was 1.18 (95% CI 0.99–1.26), with no statistically significant overall association across exposure categories (p = 0.09). Marine microplastic exposure was geographically and socioeconomically patterned across U.S. coastal counties, but current county-level exposure metrics did not reveal a consistent population-level cancer signal. Improved exposure assessment and longitudinal individual-level studies are needed.

Artificial intelligence (AI) is increasingly used in water research. However, many AI models, particularly complex machine learning models, often generate outcomes that are difficult for humans to interpret. Explainable artificial intelligence (XAI) has been developed to address these challenges by providing transparent and human-interpretable explanations of model behavior and predictions. We conducted a structured narrative review using predefined searches of Web of Science Core Collection and Scopus to synthesize empirical XAI applications across six water-research domains: hydrological processes, water quality and pollution, groundwater systems, urban water systems, climate–water interactions, and water and wastewater treatment. The review covers feature-importance methods, SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), partial dependence plots (PDPs), individual conditional expectation (ICE) plots, accumulated local effects (ALE) plots, counterfactual explanations, and deep-learning attribution methods. Building on previous reviews and perspectives focused on particular water domains or methodological priorities, we provide a cross-domain synthesis of XAI spanning natural and engineered water systems, with emphasis on method selection, model and data compatibility, explanation reliability, and operational implementation. These capabilities, however, must be interpreted with appropriate caution because XAI explanations remain conditional on the data, fitted model, and explanation method, and therefore should not be treated as evidence of causal mechanisms or environmental controls. Recognizing these limitations, we provide practical guidance for selecting and evaluating XAI methods and outline priorities for developing reliable, scalable, and operationally useful AI systems for water research and management.

Microbial contamination in urban watersheds poses significant challenges for public health, yet traditional monitoring offers limited insight into identifying contamination sources or environmental drivers. This study integrates chip-based digital polymerase chain reaction (cdPCR), a sensitive microbial source tracking (MST) assay, with explainable artificial intelligence (AI) to identify dominant fecal sources and key environmental drivers in the Rock Creek watershed, Washington, DC. Biweekly water samples were collected over 13 months (2021–2022) at three sites and analyzed for four host-associated MST markers (HF183, Rum2Bac, DG3, and GFD). After fitting Random Forest models to E. coli concentration data, SHapley Additive exPlanations (SHAP), an explainable AI tool, was used to quantify variable importance and local effects. Although HF183 exhibited the highest concentration by cdPCR, SHAP identified the canine-associated DG3 marker as the key predictor, particularly during storm events, suggesting that canine-associated fecal inputs were a major contributor to elevated E. coli concentrations. Storm events, turbidity, and minimum vapor pressure deficit were the most influential environmental drivers. SHAP further revealed substantial temporal and spatial heterogeneity in fecal contamination, indicating episodic contributions from multiple sources. This novel integration of digital PCR and explainable AI enhances understanding of fecal contamination dynamics and delivers actionable insights for precision water quality management.

While the relationship between race disparity and green space access is well-established in the United States (U.S.), its spatial-temporal variations and local drivers are still unclear. This study examines how racial disparities in access to green space changed between 2010 and 2020 across the contiguous U.S. and identifies both global (the whole study area) and local (individual spatial units) drivers of these disparities. By integrating high-resolution satellite-derived land cover metrics with census tract-level sociodemographic data, we evaluated these disparities using geospatial analysis, linear regression, and explainable machine learning. We find that the racial disparity in green cover neither narrowed nor widened over the decade, despite substantial national demographic diversification, indicating that demographic change alone was insufficient to reduce inequity. Our results reveal a consistent pattern: census tracts with higher proportions of White residents have significantly greater green cover, while areas with greater non-White populations have considerably less. Feature importance analysis identified socioeconomic vulnerabilities, particularly housing burden, transportation access, and economic disadvantage, as primary drivers of these disparities, highlighting the role of structural barriers in perpetuating environmental injustice. Local explainability analyses further revealed substantial spatial heterogeneity in these relationships: single-parent families remained the dominant local predictor nationally, PM2.5 declined sharply in influence across the western U.S., and forest and shrub cover emerged as a new leading predictor in parts of the Mountain West, Pacific region, and Texas by 2020. Our findings highlight the need for equity-centered policies, targeted interventions, and community-informed urban planning to dismantle these disparities and ensure equitable access to health-promoting natural environments for all communities.

Background: Alzheimer's disease (AD) prevalence varies substantially across United States (US) counties, yet the combined influence of environmental and socioeconomic factors remains incompletely characterized. Objective: This study bridges this gap by employing machine learning combined with explainable artificial intelligence to identify and interpret key environmental drivers of AD prevalence across US counties. Methods: Utilizing county-level AD prevalence data linked with comprehensive environmental and socioeconomic indicators, we applied an eXtreme Gradient Boosting (XGBoost) regression framework. SHapley Additive exPlanations (SHAP) were used to evaluate global and local feature contributions. Results: The final model demonstrated stable performance (RMSE: 0.89 ± 0.04; R ² : 0.62 ± 0.03). Among 15 selected factors, chloride precipitation and median household value emerged as the most influential predictors. Additional significant drivers included land-use metrics (open land, walkability), atmospheric deposition, and localized industrial exposures. SHAP analysis revealed nonlinear, threshold-like increases in predicted AD prevalence at higher chloride levels, while geographic mapping identified regionally distinct exposure profiles. Conclusions: Geospatial analysis and explainable AI demonstrate that AD prevalence is driven by interacting environmental toxicants and socioeconomic disadvantages. This “exposome” perspective highlights that mitigating environmental hazards and systemic inequities is critical for reducing the neurodegenerative disease burden.

Extreme climate events are increasingly impacting human health and vulnerability, particularly in urban areas. This study examines whether perceptions of climate change in Columbus, Ohio, align with observed environmental factors and social determinants of health. Using Natural Language Processing (NLP) on open-ended survey responses from 200 participants, we identified five dominant concern themes: flooding and water health, urban heat and air quality, heatwaves and environmental vulnerability, seasonal shifts and ecosystem loss, and climate impacts on lake health. The analysis of climate records (1985–2023) revealed an increase in the frequency of heatwaves, extreme precipitation, and dry days, coupled with a decline in cold events. Land cover analysis over the same period revealed significant expansions in developed (+30.7%) and barren land (+68.1%), alongside reductions in forest (−14.0%) and shrubland cover (−82.8%). Social vulnerability metrics (2009–2020) indicated growing proportions of elderly residents, immigrants, non-English speakers, households in multi-unit housing or with limited resources, and Hispanic populations. Cross-domain synthesis found perceptions generally aligned with observed trends, particularly for heat, flooding, and environmental vulnerability. Seasonal shifts were partially supported by more heatwaves, fewer cold events, and vegetation decline, although additional indicators such as frost dates would strengthen the evidence. Similarly, concerns over lake health require further ecological data for validation, and air quality would need support from particulate matter concentrations. These findings highlight the value of integrating perceptions with objective climate, land cover, and vulnerability data. Comparing perceptions with measurable trends clarifies which concerns reflect observed change and which may stem from information gaps, supporting targeted adaptation and communication strategies.

Fine particulate matter (PM_2.5) from wildfire smoke has emerged as a significant environmental health threat in the United States (U.S.), yet the combined roles of climate extremes and land cover in shaping wildfire smoke PM_2.5 exposure remain poorly understood. Drawing on census tract-level data from 2010 to 2018, we developed an integrative framework that combined time series analysis, multilevel regression, and interpretable machine learning (XGBoost with SHapley Additive exPlanations, SHAP) to examine the drivers of wildfire smoke PM_2.5. Across methods, elevated temperatures, drought conditions, and limited precipitation consistently preceded increases in smoke PM_2.5, particularly in the western and northern U.S. Cold waves and heavy rainfall, in contrast, were associated with modest reductions. The influence of land cover on smoke exposure differed across regions, with forests, wetlands, and grasslands affecting exposure according to local fuel availability and landscape heterogeneity. The XGBoost-SHAP model achieved high predictive performance (R² = 0.77-0.87) and further revealed that compound drivers such as drought amplification and cold wave suppression became increasingly important in later years. Our findings underscore the need for adaptive, geographically targeted smoke mitigation strategies that account for evolving influence of climatic and land-based factors across space and time.