Potential impacts of climate change on agriculture and fisheries production in 72 tropical coastal communities

The study addresses how climate change may simultaneously affect two key food production sectors—fisheries and agriculture—in tropical coastal communities, and how social vulnerability dimensions shape potential impacts. Prior assessments often consider sectors in isolation and at national scales, missing sub-national variability and the reality that many households engage in both sectors. The authors frame vulnerability as a combination of exposure and sensitivity (with adaptive capacity not included here), and focus on potential impacts (exposure plus sensitivity). They pose three research questions: (1) What are the potential impacts of projected changes to fisheries catch potential and agriculture on coastal communities? (2) To what extent can climate mitigation reduce these potential impacts? (3) Are lower socioeconomic status communities more likely to face higher potential impacts than wealthier ones? The study’s purpose is to integrate biophysical projections with socioeconomic data at community scales to inform climate adaptation and policy that reflect local realities and cross-sector dependencies.

The paper situates its work within literature documenting projected tropical losses in plant growing days and marine biomass, and extensive research on social vulnerability frameworks emphasizing exposure, sensitivity, and adaptive capacity. Previous studies have evaluated climate impacts on fisheries and agriculture separately, and only recently have joint assessments emerged, mainly at national scales, which obscure local engagement and substitutability between sectors. Literature also shows that as socioeconomic status increases, dependence on natural resource-based livelihoods tends to decrease, affecting sensitivity. The authors also reference work on alternative livelihoods, noting frequent failures due to socio-cultural and psychological factors, and discuss how these insights bear on sensitivity and adaptation options in coastal communities.

Design and sites: The study integrates socioeconomic survey data from 72 tropical coastal communities across Indonesia (n=25), Madagascar (n=6), Papua New Guinea (n=10), the Philippines (n=25), and Tanzania/Zanzibar (n=6), collected between 2009 and 2015 across five projects evaluating coastal resource management initiatives. Sites were purposively selected to span socioeconomic conditions in rural and peri-urban villages; strictly urban sites were excluded. Sampling and surveys: Between 13 and 150 households were surveyed per site (3,008 households total), using systematic, random, or full census approaches. Respondents (often household heads) provided informed verbal consent. Occupational data were collected by listing all livelihood activities contributing food or income and ranking them by importance. Occupations were grouped into fisheries (fishing, mariculture, gleaning, fish trading), agriculture (farming, cash crops), and off-sector (salaried, informal, tourism, other). Material wealth: A material style of life index (household asset-based wealth) was constructed from presence/absence of 16 material items (e.g., electricity; housing materials). A principal component analysis was applied; PC1 (34.2% variance) was taken as the wealth score, averaged to the community level and scaled 0–1. Exploratory time-series for two PNG communities (Muluk and Ahus; 2001/2002–2018) assessed temporal stability. Sensitivity metrics: Sensitivity was quantified separately for agriculture (S_A), fisheries (S_F), and joint agriculture–fisheries (S_AF). Each integrates: proportion of households engaged in the focal sector; cross-sector linkages (households also engaged outside the focal sector); and the directionality of importance rankings between sectors and off-sector activities. Formal equations (Eqs. 1–3) weight sectoral participation by whether the focal sector is ranked higher than alternatives. Exploratory analysis of S_AF over time in the two PNG communities indicated stability over decadal scales. Exposure metrics: Exposure represents projected mid-century changes relative to a historical baseline. Fisheries exposure used relative change in simulated total consumer biomass (trophic level >1) from the FishMIP ensemble under CMIP6 forcings (GFDL-ESM4, IPSL-CM6A-LR). For each community, values were extracted from the 20 nearest 1° ocean grid cells (sensitivity analyses tested alternative cell counts). Nine ecosystem models contributed (APECOSM, BOATS, DBEM, DBPM, EcoOcean, EcoTroph, FEISTY, Macroecological, ZooMSS), yielding up to 16 runs. Agriculture exposure used Global Gridded Crop Model Intercomparison (GGCMI) Phase 3 outputs (0.5°), with an 11×11 cell land window centered on each site. Four crop models (EPIC-IIASA, LPJmL, PDSSAT, PEPIC) driven by five CMIP6 ESMs projected relative yield changes for rain-fed rice, maize, and cassava (cassava from LPJmL only). Agriculture exposure is the mean projected change across the three crops. Both sectors used ISIMIP Fast Track Phase 3 protocols, comparing 2046–2056 vs 1983–2013 under SSP1-2.6 and SSP5-8.5. Potential impact: Calculated as the Euclidean distance from the origin in exposure–sensitivity space, computed separately by sector and jointly where relevant. A composite exposure metric E_AF was defined as the average of fisheries and agriculture exposure. Validation and sensitivity tests: The authors report ensemble model evaluation from prior studies (e.g., FishMIP models vs SAUP catches; GGCMI vs FAOSTAT yields) and present model run agreement (share of runs agreeing on change direction) per site as an uncertainty indicator. Agricultural exposure weighted by observed production patterns (SPAM2005) was compared to unweighted estimates, showing no significant differences. A large set of 4,746 randomly selected coastal cells (10% sample; population density >25/km²) across the study countries was used to test whether study sites were biased in exposure. Statistical analysis: Linear mixed-effects models (country as random effect) tested differences between fisheries and agriculture exposure and sensitivity, evaluated mitigation scenario effects (SSP5-8.5 vs SSP1-2.6), and related potential impacts to material style of life. Paired differences maintained within-site comparison when testing sector differences. Model uncertainty was communicated via standard errors and bootstrap (1,000 reps) for relationships with wealth; marginal and conditional R² were reported.

- Fisheries losses exceed agriculture changes on average: Under SSP5-8.5, communities experienced mean fisheries catch potential losses of 14.7% ± 4.3% SE by mid-century. Agricultural productivity showed small average gains of 1.2% ± 1.5% SE, not significantly different from zero (t = -0.80, df = 5.0, p = 0.46), driven exclusively by rice; excluding rice, maize and cassava exhibited consistent median losses under both SSP1-2.6 and SSP5-8.5. - The difference between sectors is statistically significant: Fisheries losses were greater than agriculture changes by 15.9% ± 5.6% SE at study sites (t = 2.81, df = 4.97, p = 0.0379). In a random sample of 4,746 coastal locations within the study countries, fisheries losses exceeded agriculture changes by 15.6% ± 5.1% SE (t = 3.06, df = 5.00, p = 0.0282). Exposure distributions at study vs random sites were similar (small to negligible Cohen’s D). - Sensitivity higher for fisheries than agriculture: Mean fisheries sensitivity 0.077 ± 0.007 vs agricultural sensitivity 0.04 ± 0.01 (t ≈ 3.0, df = 2.26, p = 0.0815; trend toward higher fisheries sensitivity). - High within-country variability: Despite Indonesian sites averaging ~15.9% ± 2.1% fisheries losses (SSP5-8.5), site-specific losses ranged 6.5–32%. In the Philippines, fisheries exposure was moderate (8.9–12.6% loss), but sensitivity spanned 0.001–0.32. Model run agreement was generally high for fisheries (SSP5-8.5: 84.7% ± 4.5%; SSP1-2.6: 89.2% ± 4.06%) and lower for agriculture (SSP5-8.5: 69.1% ± 4.82%; SSP1-2.6: 70.4% ± 3.27%). - Double burden common, mitigable: Under SSP5-8.5, 64% of study sites are projected to face simultaneous losses in fisheries and agriculture; this declines to 37% under SSP1-2.6. In the random coastal sample, the double burden drops from 70% to 47% under mitigation. Sites with the highest combined losses often had moderate to high sensitivity, amplifying potential impacts. - Cross-sector engagement shapes substitutability: Overall, 31% of households engaged in both sectors (range: 10% Philippines to 77% Papua New Guinea); 17% engaged in agriculture only (33% Madagascar to 3% PNG); and many engaged in fisheries only (36% Indonesia; 37% Philippines). In 12% of Philippine communities, no households engaged in agriculture, implying agricultural gains cannot offset fisheries losses for those communities. - Socioeconomic disparities: Communities with lower material wealth face greater potential impacts, reflecting both higher sensitivity (greater dependence on natural-resource livelihoods) and higher exposure. Relationships between wealth and potential impacts were significant across scenarios (details in Fig. 4).

The integrated assessment demonstrates that climate change can simultaneously undermine fisheries and agriculture at the community scale, with fisheries generally more exposed and communities more sensitive to fisheries than agriculture. This dual-sector perspective reveals a substantial fraction of communities facing a double burden, limiting scope for intra-household livelihood switching to buffer shocks. While mitigation (SSP1-2.6) substantially reduces the fraction of communities facing double losses, many low-wealth communities remain disproportionately affected because they rely more on natural resource-based livelihoods, increasing sensitivity and exposure. Although alternative livelihoods are often proposed to reduce sensitivity, extensive evidence indicates such programs frequently fail due to socio-psychological, cultural, and economic mismatches with fishers’ identities and local contexts. Therefore, mitigation and context-appropriate adaptation that recognize cross-sector dependencies, social constraints, and within-country variability are essential. The findings underscore the need for local-scale, cross-sector climate impact assessments to inform targeted adaptation strategies and investments in adaptive capacity that align with community realities.

This study provides a sub-national, cross-sector assessment of potential climate change impacts on fisheries and agriculture in 72 Indo-Pacific coastal communities by integrating ensemble model projections with socioeconomic surveys. Key contributions include: (1) evidence that fisheries face greater average projected losses than agriculture, with substantial within-country variability; (2) identification of a prevalent double burden of simultaneous sector losses that is substantially reduced under strong mitigation; and (3) demonstration that lower-wealth communities face higher potential impacts due to greater sensitivity and exposure. These insights highlight the importance of assessing climate impacts at community scales and across sectors to inform mitigation priorities and context-specific adaptation planning. Future research should incorporate standardized measures of adaptive capacity, expand crop and fishery habitat representation (e.g., coral reefs, seagrasses), use higher-resolution models and additional scenarios, and evaluate broader climate hazards (e.g., sea-level rise, heat extremes) and their compounding effects on well-being.

- Temporal mismatch: Exposure was projected dynamically to mid-century, while sensitivity and material wealth were measured at a single time point. Limited panel analyses in two PNG communities suggest decadal stability, but future community-level trajectories remain uncertain. - Model scope and resolution: Global fisheries models lack explicit representation of key coastal habitats (e.g., coral reefs, seagrasses) and operate at coarse spatial resolution (~1°). Agriculture modeling covered only three crops (rice, maize, cassava) and assumed no adaptation in management. Fisheries exposure used total consumer biomass as a proxy for potential yield, which may not fully align with harvestable biomass. - Uncertainty and assumptions: Ensemble model run agreement varies (especially for agriculture), with rice driving positive agricultural responses amid high inter-model disagreement. Fisheries models do not explicitly resolve top-down predation effects on phytoplankton; crop models assume fixed fertilizer inputs and growing seasons. - Sensitivity construct: Sensitivity was operationalized primarily via economic dependence on sectors and ranking of occupations, not capturing demographic, psychological, or cultural dimensions comprehensively. - Adaptive capacity not included: The study focused on exposure and sensitivity; standardized indicators of adaptive capacity (assets, flexibility, social organization, learning, socio-cognitive, agency) were unavailable across sites. - Other climate impacts omitted: Potentially dominant impacts such as sea-level rise and deadly heat waves were not integrated. Scenario coverage was limited to SSP1-2.6 and SSP5-8.5 due to data availability at the time.