PaperlayerView exact revision on arXiv

Source-linked AI summary

Assessing flood-related wellbeing from public discourse in Ireland using large language model as judge, 2012--2025

Róisín Luo, Karyn Morrissey

arXiv:2610.06330v1cs.CY

TL;DR

Flood-wellbeing research has limited evidence on how multiple domains and resilience mechanisms operate in recurrent exposure, while common elicited methods constrain population-scale study. This paper applies an LLM reasoning pipeline and bespoke three-domain instrument to Irish public discourse from 2012–2025, finding stressor activation and appraisal–efficacy and social-buffering pathways.

  • Problem

    Flood-wellbeing research has given comparatively limited attention to integrated activation and mediation mechanisms under recurrent exposure, while surveys and interviews constrain large-scale population-level study.

    Source: §1 · pp. 3–4

  • Method

    The study applies a bespoke three-domain instrument and an LLM reasoning pipeline to large-scale, non-elicited Irish public discourse from 2012–2025.

    Source: §1 · pp. 3–4, §5 · p. 10

  • Results

    Flood stressors directly activate multiple wellbeing domains, with appraisal–efficacy and social-buffering mediation pathways identified in Irish flood discourse.

    Source: §8 · pp. 14–15, §9 · p. 15

  • Takeaways & Limitations

    Unobtrusive public discourse provides a population-level and longitudinal perspective on flood-related wellbeing in Ireland.

    Source: §9 · p. 15

    The analysis is framed around the proposed direct-activation and mediation pathways linking flood stressors with wellbeing domains.

    Source: §4 · p. 9

Abstract

from arXiv · show

Flooding is an environmental hazard that produces social and psychological consequences for exposed individuals, shaping threat appraisal, perceived coping capacity, access to social support, and confidence in institutional response. This paper studies flood-related wellbeing in Ireland from 2012 to 2025 using large-scale, unobtrusive public discourse from Meta Content Library. We develop a bespoke flood-wellbeing instrument organised with three domains: \emph{affective--cognitive distress appraisal}, \emph{diminished perceived efficacy and agency}, and \emph{perceived relational--institutional connectedness}. We then design a flood-wellbeing reasoning pipeline with large language models to extract multi-indicator, multi-domain flood-wellbeing measures from public discourse at scale. The analysis reveals both stressor activation and social--ecological resilience mediation pathways in Irish flood discourse. Flood stressors directly activate affective--cognitive distress appraisal, diminished perceived efficacy and agency, and perceived relational--institutional connectedness. The analysis further identifies two major mediation pathways: an appraisal--efficacy pathway linking flood exposure to threat appraisal, affective burden, uncertainty, and perceived coping capacity; and a social-buffering pathway in which social and institutional protective resources are associated with perceived efficacy and appraisal of flood-related adversity. The paper contributes an unobtrusive discourse-based approach to flood-wellbeing research by introducing a bespoke instrument, identifying activation and mediation pathways in flood discourse, and producing a temporal dataset for studying wellbeing, adaptation, and resilience under recurrent flood exposure.

1 PUBLIC DISCOURSE

The paper studies flood-related wellbeing in Irish public discourse, developing a three-domain instrument and identifying activation and resilience-mediation pathways.

1 PUBLIC DISCOURSE

  • The study develops a bespoke instrument for analysing flood-related wellbeing across three domains.It also creates a multi-indicator, multi-domain temporal dataset for Ireland covering 2012–2025.Source: §1 · p. 1
  • The analysis identifies flood-stressor activation and social–ecological resilience mediation pathways in Irish flood discourse.These pathways are examined through the paper’s flood-wellbeing reasoning pipeline.Source: §1 · p. 1
  • The resulting temporal dataset supports study of flood-related wellbeing over 2012–2025.The dataset is described as multi-indicator and multi-domain.Source: §1 · p. 1

Highlights

The paper contributes an instrument, an account of activation and mediation pathways, and a temporal dataset for Irish flood-wellbeing research.

Highlights

  • A bespoke instrument measures flood-related wellbeing across three theory-informed domains for unobtrusive discourse analysis.The domains are affective–cognitive distress appraisal, diminished perceived efficacy and agency, and perceived relational–institutional connectedness.Source: Highlights · p. 2
  • Flood stressors activate all three wellbeing domains, while appraisal–efficacy and social-buffering pathways structure flood-related wellbeing.These activation and mediation pathways surface in Irish flood-related discourse.Source: Highlights · p. 2
  • A multi-indicator, multi-domain temporal dataset of flood-related wellbeing in Ireland spans 2012–2025.It is intended to support longitudinal research on wellbeing, adaptation, resilience, and public responses to flooding.Source: Highlights · p. 2

1. Introduction

Flood-wellbeing research has relied on costly, researcher-elicited evidence and has given limited attention to integrated activation and mediation mechanisms. This paper uses unobtrusive public discourse, a bespoke instrument, and an LLM reasoning pipeline to examine those pathways in Ireland.

1. Introduction

  • Survey and interview studies are costly and time-intensive, often producing relatively small samples that may be selection-biased and miss wider non-elicited discourse.These constraints limit their feasibility for population-level research.Source: §1 · p. 3
  • The study analyses large-scale, non-elicited Irish public posts from Meta Content Library with an LLM pipeline that extracts multi-indicator, multi-domain wellbeing measures.The corpus comprises approximately 24,000+ flood-related public posts spanning 2012–2025.Source: §1 · p. 4
  • Flood stressors directly activate affective appraisal, coping efficacy, and relational–institutional connectedness in Irish discourse.The paper further describes appraisal–efficacy and social-buffering mediation pathways; the latter links protective resources to perceived efficacy and appraisal of adversity.Source: §1 · p. 4
  • Flood-wellbeing research has given limited attention to how stressors activate multiple domains and how appraisal, efficacy, and social protection mediate wellbeing in public discourse over time.The paper positions its findings as evidence that these pathways surface in Irish flood discourse and can be studied longitudinally.Source: §1 · p. 4
  • The bespoke instrument organises flood-related wellbeing into affective–cognitive distress appraisal, diminished perceived efficacy and agency, and relational–institutional connectedness.It is designed for unobtrusive discourse-based study.Source: §1 · p. 4
  • The paper constructs a secondary multi-indicator, multi-domain temporal dataset of Irish flood-related wellbeing for 2012–2025.The dataset is intended to support longitudinal research on wellbeing, adaptation, resilience, and recurrent flood exposure.Source: §1 · p. 4

2. Background

The background frames flood-wellbeing as multidimensional and identifies limited study of integrated activation and mediation under recurrent exposure. The instrument translates that framing into three domains and an evidence-based ordinal assessment.

2. Background

  • Prior disaster research links psychological adjustment to interacting appraisal and coping efficacy, while social-buffering accounts describe protective support and resources.The background notes comparatively limited research on how these mechanisms co-occur under recurrent flood exposure.Source: §2 · p. 5
  • Flood-wellbeing research has examined distress and disruption across cases, but has paid less attention to integrated activation and mediation mechanisms.Survey and interview designs can be costly and limited in population-scale coverage; sentiment-based social-media approaches typically reduce response to one polarity dimension.Source: §2 · p. 5

3. Development of bespoke instrument

  • The instrument operationalises flood-related wellbeing for large-scale unstructured, non-elicited public discourse using LLM-based reasoning.Its theoretical framing aligns stress appraisal, self-efficacy, social buffering, and post-disaster resilience.Source: §3 · pp. 6–7
  • BE− captures affective arousal, threat appraisal, and emotional dysregulation, with indicators including stress, anxiety, and hopelessness.The instrument presents it as the affective–cognitive distress appraisal domain.Source: §3 · p. 7
  • DO− captures diminished coping efficacy, personal agency, and everyday functioning, with indicators of efficacy deficits, disruption, and resource constraints.The cited passage connects these indicators to reductions in adaptive capacity.Source: §3 · pp. 7–8
  • CONNECT+ captures social embeddedness, support mobilisation, and institutional confidence through indicators of connectedness, protective support, and institutional trust.The instrument uses an evidence-based ordinal scale to rate how strongly discourse substantiates constructs, rather than estimating clinical severity.Source: §3 · p. 8

4. Activation and mediation pathways

The section frames flood stressors as activating three wellbeing domains and examines appraisal–efficacy and social-buffering pathways. Reported associations place the direct CONNECT+ pathway to distress below pathways involving perceived efficacy.

4. Activation and mediation pathways

  • The conceptual framework treats flood exposure as an environmental stressor activating interconnected psychological, social, and institutional resilience processes.It proposes simultaneous activation of multiple wellbeing domains, with distress further shaped by efficacy-related and social-protective pathways.Source: §4 · p. 9
  • The proposed activation model links flood stressors to affective–cognitive distress, diminished efficacy and agency, and relational–institutional connectedness.The three domains are BE−, DO−, and CONNECT+; the hypothesis tests their association with climate stressors.Source: §4 · p. 9
  • The appraisal–efficacy hypothesis proposes that DO− mediates the association between flood stressors and BE− through deficits in coping efficacy, daily functioning, and resource access.Source: §4 · p. 9
  • The social-buffering hypothesis proposes a flood stressors → CONNECT+ → DO− → BE− pathway in which protective resources co-emerge with efficacy disruption.Source: §4 · p. 9
  • The direct CONNECT+→BE− association appears weaker than the appraisal–efficacy and extended social-buffering pathways, with CONNECT+ interpreted as consolidating efficacy and agency.The interpretation places social support, protective mobilisation, and institutional confidence primarily upstream of DO− rather than as direct mediators of distress.Source: §4 · p. 9

5. Data

The data combine Irish public posts from 2012–2025 with precipitation, named-storm, and dated-flood reference series. LLM scoring retains a minority of processed records after relevance and evidence screening.

5. Data

  • The MCL–Irish–Flood–Discourse corpus contains approximately 24,000+ Facebook and Instagram posts about Irish flood events from 2012–2025.Retrieval combined flood, rain, or storm terms with at least one Irish geographic marker.Source: §5 · p. 10
  • Monthly CSO rainfall from a 15-station network was scaled by 1.139 to align with the 1,255 mm gridded long-term average.For 2023 onward, annual gridded totals were apportioned monthly using the 1991–2020 seasonal cycle.Source: §5 · p. 10
  • 13.4% of processed posts passed the relevance and evidence gate for scoring, while 85.8% were non-relevant and 0.8% lacked sufficient evidence.Source: Figure 4 · p. 11
  • The named-storm reference includes 78 storms from 2015–2016 through 2025–2026 and is structurally zero before the naming scheme began in November 2015.Source: §5 · p. 11
  • The dated-flood reference covers 31 major events from 2011–2026, using best-effort monthly multiplicity counts for distinct sub-events.The authors identify month-of-occurrence patterns, rather than multiplicity counts, as the primary downstream signal.Source: §5 · p. 11

6. Wellbeing reasoning

The reasoning pipeline uses an instrument-defined rubric to assess curated discourse with an LLM and aggregate evidence scores into yearly domain summaries. The analyses report score distributions and co-movement between wellbeing domains and climate reference series.

6. Wellbeing reasoning

  • Score 1 is the dominant category across most indicators, while I2 Anxiety, I4 Low efficacy, and I9 Institutional trust have no score-3 observations.Source: Figure 5 · p. 12
  • The pipeline applies a theory-informed instrument and evidence-based rating criteria to curated MCL discourse using an LLM for multidimensional indicator extraction.Preprocessing includes removing posts in the upper 5% of post length; structured prompting motivates the use of LLMs as constraint-guided raters.Source: §6 · pp. 11–12
  • Each domain score sums its three indicator scores, and yearly means and population standard deviations use records with positive domain evidence.Zero denotes no admissible evidence, not low intensity, and is excluded from subsequent analyses; summing within records weights records with jointly expressed facets more heavily.Source: §6 · p. 12
  • Wellbeing domains broadly track elevated precipitation, major flood events, and named-storm activity, and are substantially correlated with one another.Figure 6 presents aggregated domain scores alongside climate reference series for 2012–2025.Source: Figure 6 · p. 13
  • A zero indicator score denotes absent admissible discourse evidence rather than absence of the underlying construct, so it is treated as missing data.The lowest substantive score is 1, and zero-domain records are excluded from domain-level temporal summaries.Source: §6 · pp. 12–13

7. Results

Flood-related discourse scores co-moved with climate activity and with one another across 2012–2025. The correlations suggest concurrent activation of distress, efficacy-related adversity, and social-protective processes, while low-intensity evidence predominated among records retained for scoring.

Scoring statistics

  • Of approximately 222,566 processed records, 13.4% passed the relevance and evidence gate, 85.8% were excluded as non-relevant, and 0.8% lacked sufficient wellbeing evidence.Among retained records, score 1 was the dominant category across indicators; I2: Anxiety, I4: Low efficacy, and I9: Institutional trust had no maximum-score observations.Source: §7 · p. 13

Activation of flood–wellbeing domain

  • Across flood-active periods from 2015–2016 through 2024–2025, BE−, DO−, and CONNECT+ rose concurrently rather than moving in opposing directions.The pattern included periods associated with major storms and flooding; aggregate connectedness and support did not coincide with lower distress or efficacy-related impacts.Source: §7 · pp. 13–14

Mediation of flood–wellbeing domain

  • BE− and DO− had the strongest domain–domain correlation (r=0.60), while CONNECT+ correlated positively with DO− (r=0.50) and BE− (r=0.25).Figure 7 summarizes correlations between aggregated domain scores and climate series, and among the domains.Source: §7 · p. 14, Figure 7 · p. 14
  • Among climate references, precipitation correlated with BE− (r=0.60), while named storms correlated most strongly with CONNECT+ (r=0.44).Flood-event frequency was also positively correlated with BE− (r=0.58), DO− (r=0.42), and CONNECT+ (r=0.23).Source: §7 · p. 14

8. Discussion

The discussion presents Irish flood discourse as involving direct activation across three wellbeing domains and interconnected appraisal–efficacy and social-buffering pathways. It further argues that social and institutional resources are more closely linked to efficacy and agency than to an immediate reduction in distress.

Validation of Hypothesis I: direct activation pathway

  • Climate stressors were positively associated with all three domains; BE− showed the strongest activation, correlating with precipitation (r=0.60) and flood frequency (r=0.58).DO− correlations were r=0.24 with precipitation and r=0.42 with flood frequency; CONNECT+ associations were weaker for these measures.Source: §8 · pp. 14–15
  • CONNECT+ correlated most strongly with named-storm activity (r=0.44), a pattern the authors suggest may reflect media attention, emergency communication, and community mobilisation.The discussion interprets the results as activation of appraisal, efficacy, and social-protective processes rather than distress alone.Source: §8 · p. 15

Validation of Hypothesis II & III: social–ecological resilience structure

  • The appraisal–efficacy pathway is represented by the positive DO−–BE− association (r=0.60), linking efficacy-related deficits with heightened stress, anxiety, and hopelessness.The discussion describes a second social-buffering pathway through the positive CONNECT+–DO− association (r=0.50) and an indirect CONNECT+–BE− linkage through DO−.Source: §8 · p. 15

Understanding the role of institutional support

  • The direct CONNECT+–BE− association (r=0.25) was substantially weaker than the DO−–BE− and CONNECT+–DO− relationships.The authors interpret social and institutional resources as relating more directly to coping, functioning, and agency, which in turn shape affective burden.Source: §8 · p. 15

9. Conclusion

The conclusion argues that flood-related wellbeing in Ireland is better understood through interconnected distress, efficacy, and relational–institutional domains than as a single psychological outcome. The study uses public discourse from 2012–2025 to describe these domains and their resilience pathways, and presents discourse analysis as a population-level, longitudinal complement to surveys.

9. Conclusion

  • Flood exposure is associated in public discourse with interconnected distress, adaptive-capacity deficits, and relational–institutional recovery resources, organized through appraisal–efficacy and social-buffering pathways.The conclusion argues these mechanisms form an integrated resilience structure under recurrent flood exposure, rather than isolated processes.Source: §9 · p. 15
  • Unobtrusive public discourse offers a population-level and longitudinal perspective on wellbeing, adaptation, and resilience that complements survey-based approaches.The paper frames this approach as a way to investigate psychosocial responses to climate-related hazards as they unfold over time.Source: §9 · p. 16

A. Climate reference: annual national precipitation (2012–2025)

The climate-reference materials document annual precipitation, named storms, and major Irish flood events, alongside annual wellbeing-score tables. Their captions specify the sources, coverage, and score summaries used in the analysis.

A. Climate reference: annual national precipitation (2012–2025)

B. Climate reference: named storm (2012–2025)

  • The named-storm table covers the 2015–2026 storm seasons; the joint naming scheme began in November 2015, so no named storms are reported for 2012–2015.The table caption also describes maximum gust speeds and Beaufort-class color coding; the supplied text lists wind categories from Strong Gale through Hurricane.Source: Table 5 · p. 19, B. Climate reference: named storm… · p. 19

C. Climate reference: Irish flood event (2012–2025)

  • The major-flood-events table covers Ireland from 2012–2025 and classifies events as fluvial, coastal, or pluvial.Its event count represents distinct sub-events bundled within one calendar month, matching the Figure 6 flood-event reference series.Source: Table 6 · p. 21

D. Annual flood-related wellbeing scores (2012–2025)

  • Annual wellbeing tables report per-indicator mean and standard deviation with non-zero record counts, and aggregated domain-score mean and standard deviation with non-zero domain counts.A score of 0 denotes no evidence and is excluded from the per-indicator table; dashes indicate no non-zero observations.Source: Table 7 · p. 22, Table 8 · p. 22
Loading 2610.06330v1…