Monitoring the impact of climate extremes and COVID-19 on statewise sentiment alterations in water pollution complaints

The study examines how the COVID-19 pandemic and climatic extremes (droughts and floods) influence public sentiment—emotion and attitude—expressed in water pollution complaint reports in Alabama (2012–2021). COVID-19 prevention policies (social distancing, lockdowns) altered daily life and mental health, increasing anxiety and depression, which may change how people perceive and report environmental incidents. Climatic extremes also degrade water quality and impact mental health via loss, displacement, and stress, potentially shifting emotions/attitudes toward pollution events. Alabama, part of the ACF basin and historically affected by droughts/floods, has limited environmental management resources, making it vulnerable and suitable for participatory surveillance through ADEM’s complaint platform. The research questions are: (1) How did historical climatic extremes alter sentiment before COVID-19? (2) How did pandemic-era sentiment differ from historical extremes? (3) How did socioeconomic factors modulate sentiment during extremes and COVID-19?

Prior work links pandemics and prevention policies to altered risk perception, anxiety, and depression, affecting public responses to environmental hazards. Climatic extremes (droughts, floods) threaten water quality/systems and mental health through socioeconomic stressors and displacement. The southeastern U.S. ACF basin has experienced severe droughts and floods, sparking interstate water conflicts and highlighting governance challenges, particularly in resource-limited Alabama. Participatory surveillance systems have been effective across domains (health, marketing, politics, urban planning) and have shown value in environmental monitoring (e.g., air pollution complaint sentiment), suggesting utility for water quality surveillance. However, empirical studies leveraging public water pollution complaint texts to monitor sentiment are scarce.

Data: 10,690 publicly available water pollution complaints from 67 Alabama counties (2012–2021) were scraped from ADEM using BeautifulSoup. Each record includes county, complaint ID/time/method, and text. Retrieval dates: up to 2020 on July 20, 2021; 2021 data on July 4, 2022. Complaints methods: phone (45.46%), web (39.75%), email (13.27%), mail (0.96%), in-person (0.54%), fax (0.02%). Extreme event data: US Drought Monitor (drought areal extent), NOAA/NCEI (flood frequencies), and COVID Tracking Project (state case counts). Period definitions: Drought D-2017 (May 2016–Apr 2017), Floods F-2016 (Oct 2015–Apr 2016; flash/pluvial dominant) and F-2019 (Aug 2018–Jul 2019; riverine/fluvial dominant), COVID-19 period C-2020 (Jan–Dec 2020) and C-2021 (Jan–Dec 2021). Analytical approach: Text preprocessing removed noise (e.g., prepositions, verbs, names). Sentiment analysis used TextBlob to compute polarity (−1 to 1; negative/positive thresholds ≤ −0.1/≥ 0.1) and subjectivity (0 to 1; objective/subjective thresholds ≤ 0.2/≥ 0.4). State-level and socioeconomic stratification: compared top 15 GDP counties (≈70% of state GDP in 2019) vs other 52 counties. To assess significance, applied bootstrapping: for each target period, randomly sampled with replacement complaints 10,000 times to estimate distributions of percentages of positive/negative and objective/subjective complaints; compared to bootstrapped distributions from the entire study period. Percentage ratio of 1.0 indicates no change; 95% bootstrap interval fully above/below 1.0 indicates significant change. Additionally, seasonal patterns and correlations with precipitation/temperature were examined. To isolate impacts of extremes/COVID from specific local incidents, 262 incident-driven complaints in 2019 and 2021 peaks were excluded from certain assessments.

- Complaint volumes: Stable 2012–2018; peaks in 2019 (local fish kill incidents) and pandemic-era surges in mid-2020 and late 2021 (up to 215 in September 2021). Top 15 GDP counties contributed ~50% of complaints. Seasonal cycle: higher Jan–Aug (~100/month), decreasing to ~60 in December, aligned with precipitation/temperature patterns. Higher complaint counts in higher population density and GDP counties. - Baseline sentiment (2012–2021): Approximately 12% positive and 34% negative complaints statewide; about 26% subjective and 32% objective (residents in top 15 counties posted more objective and fewer negative complaints than the other 52 counties). - Flood 2016 (F-2016; flash/pluvial dominant): No significant changes in emotion or attitude at state level or by county groups. - Drought 2017 (D-2017; flash drought): Significant statewide increase in negative emotion by +35% (95% CI: +15% to +60%); attitude shifted toward neutrality with ~−10% in both subjective and objective shares (i.e., fewer polarized attitudes). Other 52 counties exhibited higher negative percentages than top 15 counties. - Flood 2019 (F-2019; riverine/fluvial): More negative (≈+25%, 95% CI: +3% to +50%) and less subjective (≈−25%, 95% CI: −1% to −40%) complaints; top 15 counties had fewer objective complaints during flood periods than the full period. - COVID-19 (2020): Statewide negative complaints decreased ~−30% (95% CI: −40% to −20%); subjective complaints increased significantly: top 15 counties +20% (95% CI: +10% to +38%), other 52 counties +15% (95% CI: +1% to +34%). Suggested drivers include reduced pollution and outdoor exposure due to lockdowns and a shift of public attention to health. - COVID-19 (2021): 2020 sentiment shifts persisted, indicating longer-term effects. A few of the other 52 counties showed increased negative and subjective complaints, suggesting socioeconomic heterogeneity. - Additional check on incident severity (2019, 2021 subsets): Environmental incidents increased positive complaint percentages, indicating incident severity can influence sentiment.

Findings show that both the type/severity of climatic extremes and the COVID-19 pandemic significantly altered sentiment in water pollution complaints, with differential effects by socioeconomic context. Flash drought (2017) and riverine flooding (2019) increased negativity, with drought also reducing polarized attitudes, while flash-flood-dominated 2016 had negligible sentiment effects. During COVID-19 (2020–2021), negative emotions decreased and subjectivity increased, likely reflecting reduced exposure/pollution and reallocation of public attention to health. Socioeconomic development moderated these effects: top GDP counties tended to be more objective and less negative overall, and pandemic-era increases in subjectivity were stronger in wealthier counties. Emotion appears more volatile (“shorter shelf life”) than attitude, which may be shaped by longer-term factors like income and education. These insights can guide authorities in adaptive resource allocation, risk communication, and public engagement, leveraging participatory surveillance to detect shifts in public emotion/attitude during crises.

This study provides observational evidence that climate extremes and the COVID-19 pandemic altered both emotions and attitudes expressed in public water pollution complaints in Alabama. Droughts and riverine floods heightened negativity pre-pandemic, whereas COVID-19 reduced negative sentiment and increased subjectivity, with persistent effects into 2021 and notable socioeconomic differences. The work demonstrates the utility of participatory surveillance data and text-based sentiment analysis for environmental governance, offering a cost-effective complement to traditional monitoring. Future research should incorporate measures of incident severity, conduct targeted surveys/interviews to elucidate mechanisms behind sentiment shifts, and integrate interdisciplinary approaches linking environmental conditions, socioeconomic structures, and mental health to improve risk communication and water quality management.

- Complaint data lacked detailed measures of incident severity, limiting the ability to disentangle sentiment changes driven by event severity from broader contextual factors. - Observational design with bootstrapping cannot establish causality; unobserved confounders (e.g., media coverage, policy changes, local enforcement) may influence sentiment and reporting behavior. - Generalizability may be limited to Alabama’s context (socioeconomic structure, environmental governance capacity). - Sentiment analysis using TextBlob and thresholding introduces measurement uncertainty and may not capture nuanced emotions or sarcasm. - Socioeconomic stratification by top 15 GDP counties is a coarse proxy; finer-grained demographic and socioeconomic data were not incorporated. - Exclusion of incident-driven spikes (2019, 2021) in some analyses, while intended to isolate effects, may omit relevant real-world influences on sentiment.