Large fisheries declines linked to compound and extreme climate events

Abstract

Climate extremes are increasingly disrupting marine ecosystems and fisheries. However, evidence on extreme ocean temperature effects on fish stocks and fisheries is mixed, and the combined impacts of heat and productivity extremes remain unclear. Using three decades of global data encompassing 6659 time-series for 1246 species across 254 regions, we conduct a risk-based analysis to quantify how local extreme high temperatures and low ocean productivity that these species are exposed to, alone and together, affect local fisheries catches. These events, particularly when compounded, sharply increase the likelihood of local extreme low catch events, especially in tropical and subtropical regions. Species critical to food security and conservation are disproportionately affected, with widespread risks in socio-economically vulnerable countries. Without adaptation, extreme events’ risks are projected to strongly intensify by the mid-21st century. Strengthening monitoring and climate-responsive fisheries management is urgently needed to build resilience.

Climate change is increasing the frequency of extreme ocean conditions that disrupt marine ecosystems and fisheries. This study finds that compound marine heat and low-productivity extremes are linked to low fish catches worldwide, with impacts varying across regions and fish groups.

Introduction

The increasing frequency and intensity of extreme ocean climate events^1,2 are causing localized declines in marine life, altering ecosystem structure, and disrupting fisheries^3–5. These changes result in substantial socio-economic impacts such as fishery closures, revenue losses, and diminished livelihoods^6–8. Marine heatwaves, or extreme high ocean temperatures (TempX), have emerged as prominent drivers of fisheries disruption, though empirical evidence remains geographically and taxonomically limited^7,9. In contrast, the effects of biogeochemical extremes¹⁰, such as extreme low net primary production (NPPX), and their interactions with high TempX are far less understood, especially at spatial and temporal scales relevant for fisheries management^11–14. Despite advances in physical oceanography and ecosystem modelling, empirical, global-scale tests of how such local compound extremes translate into realized fisheries outcomes have been lacking. Advancing our understanding of how individual and compounded marine extremes affect fisheries is essential for assessing emerging risks to seafood security and for guiding strategies to enhance the resilience of ocean management under a rapidly changing climate¹⁵.

Temperature and net primary production (NPP) are key drivers of the distribution, biomass and production of species important to fisheries^16–18. However, a recent meta-analysis of biomass and temperature time-series found that marine heatwaves have muted and variable impacts on demersal fish populations⁹. In contrast, observed high TempX has driven rapid biomass declines and fisheries closures of commercially important species (e.g.,^19,20). Theoretical models further project that extreme temperature and its compounding effects with biogeochemical extremes reduce fish biomass^6,13. These contrasting empirical and modelling results may reflect limitations in available datasets and the complex, region- and species- specific interactions among environmental variables, ecosystems and other human drivers^9,21. This underscores a critical need to reconcile the evidence base and robustly assess the risks that climate-driven extremes experienced by marine species pose to fisheries at the local scale.

This study quantifies the effects of annual temperature extremes (TempX), productivity extremes (NPPX), and their compounded co-occurrence on local fish stocks and fisheries worldwide. Here, environmental and fisheries extremes arise from variability superimposed on long-term mean trends (see Supplementary Fig. 1, Methods). We focus on annual extremes as these events often have the greatest impacts on fisheries⁶. Annual-scale anomalies capture persistent exposure and cumulative ecological and socio-economic responses and are not intended to replicate short-duration marine heatwave definitions often used in marine heatwave hazard studies^22,23. By integrating species biogeography with fisheries catch, temperature and net primary production time-series data from 1993 to 2019 with statistical modelling (see Methods), we evaluate two main hypotheses: (1) high TempX and low NPPX are associated with extreme low fisheries catches across 254 EEZs and 1,246 exploited fishes and invertebrates, (2) the strength and direction of these associations vary across regions and functional groups, and (3) the likelihood of extreme low fisheries catch events will increase under future climate change.

Our analysis was structured in four tiers (see Supplementary Fig. 2 and Method). First, we pooled all observations across species–EEZ groups and treated each species–EEZ–year combination as an independent sample. Second, we model the probability of an extreme low-catch event as a function of binary indicators for extreme temperature and NPP anomalies, their lagged values, and combinations of these drivers, including additive and two-way interaction.

Third, we conducted EEZ-specific analyses by fitting the same model set separately within each EEZ (i.e., using only species–year observations available for that EEZ) to quantify geographic variation in effect sizes. In addition, we grouped species into functional categories and repeated the analysis to evaluate responses at the functional-group level.

In this work, we adopt a risk-based framework applied to 6,659 fisheries and environmental time-series data to quantify how environmental extremes alter the likelihood of extreme low fisheries catches across species at the functional groups and Exclusive Economic Zone levels, as well as globally aggregated. Using a relative-probability approach analogous to epidemiological risk analysis (see Method), we estimate how much the odds of extreme low catch events, defined as the lowest decile of catch anomalies, increase during years of TempX and/or NPPX. This framework captures the probabilistic nature of extreme and compound climate risks, revealing not only whether extremes coincide with drops in fisheries production, but also how strongly they amplify the underlying risk compared to non-extreme conditions. It therefore bridges empirical observation and theory by reconciling the frequency of observed co-occurrences with the modelled amplification of low catch likelihood under extreme or compound stressors. We show that both TempX and NPPX significantly increase the likelihood of extreme low catches, with compound events producing the strongest effects. These risks are spatially heterogeneous, disproportionately affect several ecologically and socio-economically important fisheries, and are projected to intensify under future climate change. Our findings identify compound ocean extremes as an emerging global risk to fisheries and highlight the need for climate-responsive fisheries management and adaptation.

Results

Significant global effects of univariate and compounded extremes on catches

Globally, across all EEZs, our analysis indicates that univariate extreme environmental conditions significantly increase the likelihood of extreme low fisheries catches (Fig. 1, Supplementary Table 1). Here, extreme events are defined as annual mean anomalies falling beyond the 10th (for NPP) or 90th (for temperature) percentile of the historical distribution (see Method). Odds ratio (OR) quantifies the strength of association between environmental extremes and the likelihood of an extreme low fisheries catch event, with OR > 1 indicating increased odds relative to baseline (non-extreme) conditions in the time-series. We estimated odds ratios using logistic regression models, where the occurrence of an extreme low catch (binary response) was modelled as a function of extreme environmental conditions. For example, OR of 1.5 indicates that the odds of occurrence of low catches are 50% higher during extreme conditions compared to non-extreme years. In contrast, OR below 1 indicates a reduced odds of low catches during extreme conditions. Two consecutive years with high TempX (i.e., concurrent and one-year lagged TempX) showed the largest estimated increase in the odds of an extreme low catch, by 40% (95% CI: odds ratio = 20–65%, Supplementary Table 1) compared to non-extreme conditions. Using the most extreme conditions in the historical record did not further elevate the odds of extreme low catches beyond those associated with TempX that were identified using the 90th percentile temperature threshold (Fig. 1). In contrast, univariate NPPX events significantly affected catch anomalies only when the lowest NPP in a time-series were classified as extreme.

Fig. 1. Associations between environmental extremes and extremely low fish catches.

Estimated Odds Ratios (OR) for the occurrence of extreme low fish catch events in relation to extreme high temperature (T), extreme low net primary production (N), and their 1-year lagged effects (T_lag and N_lag) across all exclusive economic zones (EEZs) and exploited fishes and invertebrates over the period 2002–2019. Variables marked with a cross (†) and in darker red and green colour represent the most extreme high temperature or low NPP events while the lighter red and green colours represent extreme events (10 percentile outliers) occurring within the time series. Grey dots are statistically not significant (p > 0.05) while the red and blue dot indicates the estimated OR is significantly above 1 and below 1 (p < 0.05), respectively. (+) refers to additive effects while (X) is interaction terms capturing the compound effects of T, N and their lags in the generalized linear modelling (see “Method”). Points represent estimated odds ratios, and error bars represent 95% confidence intervals of the estimated odds ratios. Estimates were derived from 6659 fisheries–environment time series representing 1246 exploited species across 254 EEZs; see Methods for details of the statistical analyses and sample selection. Odds ratios were estimated using model-averaged generalized linear models with a binomial error distribution and logit link. Statistical significance was evaluated using two-sided tests. Exact p-values and associated statistics are provided in Supplementary Table 1. Source data are provided as a Source Data file.

Compound TempX with NPPX events amplified the risk of extreme low catches (Fig. 1). Several combinations of co-occurring or sequential TempX and NPPX events significantly increased the odds of extreme low catch events beyond those of univariate extremes. For instance, the occurrence of lagged TempX with concurrent NPPX that were identified as the lowest NPP of the time-series yielded odds ratios up to 1.8 (CI = 1.5–2.1), indicating nearly an 80% higher likelihood of extreme low catch relative to non-extreme conditions (Supplementary Table 1).

The impacts of high TempX on catches can be attributed to the pervasive effects of temperature on marine species and ecosystems^24,25 (see Supplementary Table 2 and Supplementary Fig. 3). Extreme high temperature impairs physiological performance, growth, and reproduction in fish and invertebrates, particularly during early life stages, thereby reducing cohort strength and subsequent year-class abundance^19,24–26. These impacts may appear with a time lag as affected cohorts enter the fishery^4,27, and recovery from prolonged extremes can take multiple years^28,29. Low NPPX can compound these effects by limiting food availability, disrupting trophic interactions, and reducing prey resilience^17,30. Delayed catch responses may also reflect lags in fisheries and their management, such as the timing of fishery closures or effort reductions³¹. The synergistic stress from high TempX and low NPPX thus create a cascade of social-ecological disturbances that undermines the abundance and availability of fishery catches, with potential localized exceptions due to shifts in species composition or fishing behaviour. The additive effect between temperature and lagged productivity extremes exhibited slightly lower odds ratios relative to non-extreme conditions, suggesting a modest buffering effect in some contexts. However, the magnitude of this effect was small relative to the reinforcing interactions, indicating that antagonistic responses are secondary to the dominant pattern of increased risk under compound extremes.

While our main analysis focuses on surface-layer extremes, additional analyses examining extreme conditions at different depths show that surface-layer extremes are strongly associated with demersal catch declines, likely due to tight oceanographic and ecological coupling between surface and bottom layers that facilitates the propagation of thermal and productivity stress through the water column (see Supplementary Fig. 3). Using bottom-layer temperature anomalies yields consistent patterns and does not alter the main conclusions of this analysis. These results indicate that the association between annual-scale environmental extremes and fisheries declines is robust across depth layers. We tested the effects of “reverse” extremes (i.e., low TempX and high NPPX), the results of which are presented in a subsequent section.

Sensitivity to alternative NPP products and fishing effort

We tested the robustness of our results to alternative estimates of NPP by repeating the global analysis using five additional commonly used NPP products derived from different biogeochemical formulations and parameterizations (see “Methods”). Because several alternative products are available only from 1998 onward, sensitivity analyses were conducted over the restricted period, whereas the primary analysis spans from 1993. We found that while the interaction between NPPX and NPP product was statistically significant (ΔDeviance = 50.3, df = 5, p < 0.001), the magnitude of the estimated ORs for both the main NPP effect and its interaction with temperature extremes remained consistently close to unity (Supplementary Fig. 4). Between-product variability in estimated odds ratios was minimal (coefficient of variation = 0.024), indicating that alternative NPP formulations altered effect sizes by approximately 2–3%. The slight attenuation of the compound interaction in the restricted-period analysis is consistent with the exclusion of the strong 1997–98 El Niño event and with percentile-based extreme definitions being sensitive to shorter time-series distributions. Confidence intervals broadly overlapped among NPP products, and no formulation altered the direction or qualitative interpretation of either the independent or compound effects on catch anomalies. These results show that at the global-scale, our conclusions regarding the roles of high NPPX and their interaction with TempX are robust to structural differences among alternative productivity estimates. Future studies could examine how alternative NPP estimations influence climate–fisheries relationships at regional or ecosystem scales, where local oceanographic processes and community structure may amplify differences among NPP estimates.

To evaluate whether climate-related associations were confounded by variation in fishing intensity, we conducted additional analyses incorporating fishing effort as a covariate (see “Method”). Annual fishing effort was quantified as effective active fishing hours aggregated at the species–EEZ level. We defined extreme high effort events using percentile-based thresholds the same as the approach used for TempX and NPPX. Because effort time series were not available for all species–EEZ units, this sensitivity analysis was conducted on the subset with available effort data. Changes in estimated odds ratios for temperature and NPP extremes were evaluated to assess whether inclusion of effort altered climate effect sizes. Including extreme fishing effort as an additional predictor improved model fit in the data subset. Notably, extreme high fishing effort was associated with reduced odds ratios of extreme low catches, consistent with the expectation that intensified fishing activity can elevate catch levels in the short term. However, estimated odds ratios for temperature and productivity extremes changed by <3% after accounting for effort (Supplementary Fig. 5). The direction and magnitude of climate effects remained stable, indicating that associations between climate extremes and catch anomalies are not explained by simultaneous variation in fishing intensity.

Spatial patterns in the effects of extreme environmental conditions

Our spatial analysis revealed that high TempX significantly increased the odds of extreme low catches across some, but not all, EEZs (Fig. 2A, Supplementary Table 2). Substantive elevation of odds ratios (OR ≥ 2) was most apparent along the eastern boundary upwelling systems, particularly the Humboldt and California Currents, as well as in the tropical eastern Pacific, including Peru, Ecuador, and Central America. Significant hotspots are also prevalent across small island nations in the Pacific, such as the Solomon Islands and Northwestern Hawaiian Islands, and in parts of the Indian Ocean, including Seychelles and Maldives (Fig. 2A, Supplementary Table 2). Additionally, semi-enclosed seas such as the Mediterranean Sea, Red Sea, and Persian Gulf exhibited consistently high odds of extreme low catches during TempX events, highlighting regional sensitivity to thermal extremes. These regions are also vulnerable to long-term ocean warming due to the narrow thermal tolerance of local species and limited adaptive capacity^32,33. Our results suggest that these vulnerabilities may extend to acute temperature extremes, not just long-term warming trends, reinforcing concerns that climate-driven thermal variability poses a disproportionate threat to these fisheries.

Fig. 2. Global maps of the estimated effects of environmental extremes on the odds ratio of extreme low-catch events relative to normal conditions between 2002 and 2019.

A Extreme high temperature effects; B extreme low NPP effects; C combined effect of co-occurring high temperature and low NPP extremes. Analysis included events that occurred as 90 percentile and 10 percentile of the high temperature and low NPP anomalies, respectively. Dotted areas represent non-significant effects (p > 0.05). EEZ area without colour indicates that estimated odds ratio is not available, primarily due to insufficient compound TempX and NPPX co-occurrences in the time period for robust statistical analysis. Odds ratios were estimated using model-averaged generalized linear models with a binomial error distribution and logit link. Statistical significance was evaluated using two-sided tests. Complete statistical summaries are provided in Supplementary Tables 2–4. Source data are provided as a Source Data file. Base maps were obtained using the rnaturalearthdata R package from the public-domain Natural Earth dataset (https://www.naturalearthdata.com/downloads/50m-cultural-vectors/).

Low NPPX showed distinct spatial patterns from temperature extremes (Fig. 2B). Elevated odds of extreme low catches were significantly (p < 0.05) associated with low NPPX in part of the temperate and subarctic regions, including the Barents Sea, Arctic Canada, North Sea, and northwest African coast, as well as parts of the Caribbean (Supplementary Table 3). Low NPPX also significantly (p < 0.05) increased the likelihood of low catches in the western Arabian Sea and around Mauritius, and several small island states in the South Pacific (Supplementary Table 3). These patterns reflect the ecological and biogeochemical role of primary production in supporting fish biomass and trophic transfer, particularly in regions where productivity is highly variable annually due to fluctuations in monsoons, nutrient upwelling, or ice dynamics^17,18. Overall, low NPPX may affect fisheries primarily through bottom-up pathways, whereby reduced primary production constrains food availability and energy transfer to higher trophic levels. Ecosystem modelling studies further indicate that such reductions can be amplified through trophic interactions, leading to disproportionate impacts on consumer biomass relative to changes at the base of the food web³⁴. Thus, low NPPX affects fisheries through mechanisms that may be distinct from thermal stress.

Compounded high TempX and low NPPX amplified OR of extreme low catches beyond those associated with univariate extremes across many EEZs (Fig. 2C). These EEZs are found in regions including the eastern tropical Pacific, South China Sea, and eastern Indian Ocean. In several areas, including the EEZ of Australia, significant increases in low catch odds occurred only when both temperature and productivity extremes co-occurred, highlighting the nonlinear and compounding nature of multi-stressor impacts on fisheries. In some regions, TempX and NPPX are each significantly associated with elevated odds of low catches, but their co-occurrence is less frequent (Supplementary Table 3). Because the compound analysis relies only on years in which both extremes occur together, this reduces statistical power and widens confidence intervals, so some EEZs may lose statistical significance in the compound case even when univariate effects are significant.

Many non-significant areas (p > 0.05) were concentrated in shelf seas at intermediate and high latitudes, including the North Pacific, the northwest and north Atlantic and the south Atlantic (shaded area, Fig. 2, Supplementary Tables 2–4). Several mechanisms may attenuate detectable associations at the EEZ catch scale. Species turnover, including both gains and losses, within diverse shelf assemblages driven by biogeographic shifts under TempX and NPPX may have dampened the signal at the EEZ level³⁵. Also, climate-induced shifts in species distribution may have already altered catch composition, with warm-adapted species potentially more resilient to acute extremes³⁶. While our analysis focuses on the probability of extreme low-catch events, climate extremes can also generate “winners,” including cases where species expand into newly favourable habitats and create novel fishing opportunities, as observed for longfin squid in the Gulf of Maine during the 2012 marine heatwave²⁰. In some EEZs, lower quality or resolution of catch statistics may also reduce the ability to detect robust relationships. These factors may help explain the weaker apparent influence of high TempX, low NPPX, or their co-occurrence in these shelf sea regions, without implying an absence of ecological impact.

In addition, a limited number of EEZs exhibited effects in the opposite direction (odds ratio <1), indicating potential antagonistic or buffering interactions between climate drivers. However, the spatial extent of these effects was limited compared to the dominant trend of increased risk under annual-scale extreme climate stressors, and in most cases, it is because of the association with significant negative effects of TempX (e.g., the Azores) or NPPX (e.g., Central Indonesia). Consistent with the spatial patterns, the frequency distribution of grid-cell odds ratios shows that the majority of cells have OR > 1, with the highest proportion in the OR ≥ 2 category, particularly for compound extremes (Supplementary Fig. S6).

Functional group-specific sensitivity to extremes

The sensitivity of fisheries to climate extremes varied by functional groups and ecosystem types, reflecting variation in responses due to differences in ecological traits and biogeography.

From an ecological perspective, differential sensitivity among ecosystem types and functional groups would be expected a priori. Coral reef–associated species often exhibit narrow thermal tolerances and strong dependence on structurally complex reef habitats that are vulnerable to bleaching-related degradation under extreme warming events³⁷. Similarly, small pelagic species in upwelling and eastern boundary current systems, as well as large pelagic fishes around oceanic islands, are strongly influenced by variability in temperature and primary production and associated prey fields. Their recruitment success and biomass dynamics are tightly linked to plankton productivity and trophic transfer efficiency, suggesting heightened sensitivity to productivity extremes that disrupt food-web structure³⁸. In addition, large impacts of extreme temperature events on benthic invertebrates have been documented, including the recent collapse of Bering Sea snow crab following a marine heatwave, where elevated bottom temperatures increased metabolic demand and likely contributed to starvation-driven mortality¹⁹. These examples illustrate that vulnerability to climate extremes arises through multiple ecological pathways, including thermal tolerance limits, habitat dependence, metabolic stress, and trophic coupling. Therefore, we selected case studies to explore functional group–specific patterns of linkages between TempX and NPPX on fisheries declines identified in our global analysis.

Across all extreme event types, sharks and rays, small pelagics, and commercial invertebrates were among the most sensitive groups, with the highest median odds ratios of extreme low catches (Fig. 3A, Supplementary Table 5). These groups also experienced the most frequent and severe observed declines, with median catch drops exceeding 40% during extreme events (Fig. 3B). Functional groups with broader distributions or higher mobility, such as benthopelagic and large pelagic fishes, exhibited more variable responses. In addition, reef-associated fishes showed heightened and widespread risk from NPPX and compounded TempX with NPPX, potentially due to their strong dependence on biogenic habitats, with the indirect effects stemming from coral bleaching, habitat degradation, and disrupted trophic interactions in addition to the direct temperature impacts from environmental extremes.

Fig. 3. Functional group-specific sensitivities to extreme high temperature and low net primary production events.

A Odds ratios (ORs) for extreme low catches associated with temperature extremes (TempX), NPP extremes (NPPX), and compounded TempX–NPPX events across seven functional groups. Bars show median ORs across species–Exclusive Economic Zone (EEZ) combinations. Colours indicate the proportion of EEZs where significant effects were detected (A) and the total number of EEZs where the species groups are exposed to extreme conditions (B). B Corresponding observed median percentage declines in catch during extreme low catch years, highlighting the severity of catch reductions. The vertical dashed line in panel A indicates OR = 1 (no effect), and in panel B indicates 0% change in catch. For both panels, the centre line represents the median across species–EEZ combinations, the box bounds represent the 25th and 75th percentiles (interquartile range, IQR), the whiskers extend to the most extreme values within 1.5× IQR of the box, and red points denote observations beyond the whiskers (outliers). Sample sizes underlying each functional group, including the numbers of species, Exclusive Economic Zones (EEZs), and species–EEZ combinations contributing to the analyses, are provided in Supplementary Table 5. Source data are provided as a Source Data file.

These patterns of sensitivity to extremes and their potential implications for countries vulnerable to climate change were exemplified in distinct case studies (Fig. 4, see Supplementary Fig. 7 and Supplementary Table 6 for illustrative examples for each case studies). Small pelagic fisheries in Peru, dominated by anchoveta (Engraulis ringens), showed extreme sensitivity to prolonged high temperature anomalies, with odds of extreme low catches increasing more than ninefold and median declines around 60% (Fig. 4A), which is consistent with the close association of anchoveta biomass and its fisheries to El Niño Southern Oscillation³⁶. Our analysis indicates that in the Solomon Islands, medium and large pelagic fisheries experienced up to a 41-fold increase in the odds of low catches under NPP extremes (Fig. 4B), consistent with observed declines in NPP and the eastward displacement of tuna, such as skipjack tuna (Katsuwonus pelamis), during El Niño events³⁹. Reef-associated fisheries in the Maldives were highly sensitive to lagged temperature extremes, with odds ratios exceeding 4 and typical catch losses of ~25%; in line with the very high vulnerability of tropical coral reef ecosystems in Indian Ocean, and globally, to heatwaves and other climatic extremes^40,41 (Fig. 4C). Moreover, in the Galápagos Islands, shark and ray fisheries, such as those for blue shark (Prionace glauca), were particularly sensitive to singular high-temperature extremes, which disrupted thermocline structure and shifted vertical species distributions, affecting their catchability by fisheries (Fig. 4D). Marine heatwaves have led to sharp declines in shark abundance in the waters around the Galápagos Islands, particularly in shallow habitats, due to altered thermal structure and prey availability⁴². In temperate systems, commercial invertebrate fisheries in the United Kingdom were particularly vulnerable to compounded TempX and NPPX events (Fig. 4E). Although compound extremes are rarely isolated as the sole drivers of documented invertebrate declines in UK waters (Supplementary Table 6), evidence indicates that ocean warming, acidification, and productivity changes impair growth, recruitment, and survival of shellfish species⁴³. Recent marine heatwaves in UK and adjacent waters have disrupted regional marine ecosystems and raised concerns about fisheries vulnerability, with impacts mediated by mechanisms such as enhanced stratification, oxygen limitation, altered plankton dynamics, and shifts in species distributions^44,45.

Fig. 4. Case study examples of significant fisheries responses to temperature and productivity extremes.

Case study examples show ORs and 95% confidence intervals for specific fisheries with significant responses: small pelagic fishes in Peru (A), medium and large pelagic fishes in the Solomon Islands (B), reef-associated fishes in the Maldives (C), sharks and rays in the Galápagos Islands (D), and commercial invertebrates in the United Kingdom (E). Points represent estimated odds ratios and horizontal error bars represent 95% confidence intervals. Symbols denote the type of extreme events that had significant effects: T (90th percentile of the temperature time-series), T† (highest temperature in the time-series), N (10th percentile of the NPP time-series), and N† (lowest NPP in the time-series), with lags and compound (+ for additive and x for interactive) events indicated where relevant. Source data are provided as a Source Data file.

Overall, the functional-group contrasts illustrate the diverse mechanisms linking TempX, NPPX and low catches. The high sensitivity of sharks, rays, and small pelagics to co-occurring TempX–NPPX events may arise from habitat compression and altered vertical structure during stratification events, which constrain foraging and migration corridors^46,47. For demersal and reef-associated species, strong coupling between surface and bottom layers means that surface productivity shocks can propagate to benthic prey and habitat conditions. Beyond these physical drivers, other global and regional environmental, socio-economic, and management factors, such as changes in market demand, fishing effort, policy enforcement, and ecosystem interactions, also contributed to extreme low catches in many cases, often acting in combination with climatic extremes in complex and region-specific ways (Supplementary Table 6). Thus, our study underscores the importance of local monitoring and adaptive management tailored to ecosystem context.

Our analysis of the “reverse” extremes, i.e., cold-year (low temperature) and high productivity (high NPP) events, revealed that these conditions do not produce mirror-image high-catch anomalies to those low catches being driven by high TempX and low NPPX (see Supplementary Fig. 8). Low TempX and high NPPX, whether singular or compound, yielded inconsistent and limited increases in extreme catch odds. Such asymmetric responses have been documented economically for the impacts of El Niño Southern Oscillation, with losses incurred during El Niño are rarely offset by gains in La Niña periods⁴⁸. Biological lags in stock recovery and socio-economic delays further suppress rebounds³¹. In addition, the effects of high TempX and low NPPX extremes on low versus high extreme catch outcomes amongst EEZ-functional groups were only weakly positively correlated (Pearson’s r = 0.047, p < 0.001). While some fish stocks might have benefited from extreme conditions, these gains tend to occur in different species or regions than those that were negatively affected. These findings underscore the importance of diversification of fishing activities as a strategy for adapting to the growing risks posed by climatic extremes⁴⁹.

Projecting the future risks of low catches under TempX and NPPX

We applied temperature and NPP projections from three Earth system models from the Coupled Model Intercomparison Project Phase 6 to our statistical model of extreme catch ORs to predict occurrences of future stock declines under the ‘strong mitigation, low emissions’ (SSP1-RCP2.6) and ‘no mitigation, high emissions’ (SSP5-RCP8.5) scenarios (see Method). Future projections assume that the historical statistical association between environmental extremes and catch anomalies remains structurally stable through the mid-21^st century. We therefore restrict projections to 2060 to minimize extrapolation under potential ecological and management regime shifts. The results show that the share of fish stocks experiencing extreme low catches in vulnerable EEZs is projected to increase from about 11% in the early 1990s to almost 16% by the mid-21st century, with similar trajectories under both low- and high-emission scenarios (Fig. 5). The similarity of trajectories between SSP1-2.6 and SSP5-8.5 reflects climate-ocean-system inertia and the dominance of near-term warming already committed under both scenarios, rather than implying that mitigation is unimportant. This suggests that uncertainties associated with emissions pathways in projecting the mid-century impacts of temperature (TempX) and productivity (NPPX) extremes on catches are relatively low. Climate-driven disruptions to fisheries are already unfolding and are expected to intensify in the next few decades regardless of mitigation, underscoring the urgency of adaptation.

Fig. 5. Projected increases in extreme low catch events under climate change.

Time series of the percentage of stocks (EEZ–species combinations) experiencing extreme low catches defined as catches falling below the lower bounds expected from non-extreme years under two Shared Socioeconomic Pathways (SSP1-2.6 in blue, SSP5-8.5 in orange). Solid lines represent the mean projections across three Earth System Models (ESMs) based on fitted statistical relationships. Shaded ribbons denote the combined uncertainty from the statistical model (95% confidence intervals) and inter-model spread among the ESMs. Grey line and area represent historical periods of the ESMs. Source data are provided as a Source Data file.

Reconciling conflicting evidence on temperature extremes and fish stock responses

Our analysis suggests plausible explanation for the recent inconsistencies in the literature regarding the impact of marine heatwaves on fish stock biomass. Specifically, we examined demersal fish species in EEZs overlapping with those assessed by ref. ⁹ and confirmed that high TempX alone exert limited and variable influence on the odds of extreme low catches of this subset of stocks between the 1990s and 2010s (see Supplementary Fig. 9). In contrast, low NPPX, particularly when lagged or combined across multiple years, significantly elevated the odds of low catches (OR > 1, Supplementary Table 7). Notably, the strongest signal emerged when TempX and NPPX occurred in sequence, with lagged compounding effects nearly doubling the likelihood of low catch anomalies. This suggests that productivity disruptions, not examined by Fredston et al.⁹, rather than heatwaves alone, may be associated with catch declines and potentially underlying stock stress in some systems. Fredston et al. used standardized scientific survey data to estimate biomass changes. In contrast, our analysis relies on reconstructed fisheries catch data that include commercial, artisanal, and subsistence removals and therefore represent realized fisheries extractions rather than direct biomass indices. While catch changes may partly reflect biomass dynamics, they are also affected by fishing effort, and socio-economic and management influences. Also, a key distinction between this study and Fredston et al. lies in the temporal scale of extreme definitions. Whereas Fredston et al. focus on short-duration bottom-temperature heatwaves (≥5 days above the 95th percentile) aggregated into annual metrics, we identify annual-mean extreme anomalies. Annual mean anomalies can reflect cumulative and persistent exposure of shorter events and better match the temporal annual resolution of fisheries production data, which may reflect delayed ecological and socio-economic responses. Short-duration heatwaves may not consistently produce detectable biomass shifts in survey data, whereas persistent or compound annual-scale extremes may be more strongly associated with fisheries production outcomes that integrate ecological and socio-economic responses. Short-term marine extremes can propagate through ecological processes and generate longer-term impacts when events recur or persist²⁸. Both perspectives therefore provide valuable but distinct insights into climate risks to marine systems.

In addition, the geographic focus of Fredston et al.⁹ on the Northeast Pacific and North Atlantic regions, where TempX links to catch drops are comparatively weak (Fig. 2), may also limit the generalizability of their findings across other functional groups. In contrast, global modelling studies have identified extreme high temperatures and compound climate events as key drivers of fish biomass and fisheries production declines^13,28. While our studies focus on catches and their declines do not necessarily correspond directly to biomass, our results are consistent with the findings from these modelling studies in suggesting that persistent or compounded temperature and NPP extremes are particularly destabilizing for fisheries. The congruence between modelled probabilities and real-world disruptions reinforces the robustness of our statistical approach and underscores the urgent need to integrate extreme event risk into fisheries management and climate adaptation strategies. Because globally harmonized survey-based biomass datasets at comparable spatial and taxonomic scales are currently limited, reconstructed catch data provide one of the few consistent bases for global empirical risk assessment. Where regional survey data are available, future comparative analyses could further evaluate whether similar risk patterns are detectable in biomass-based indicators.

Considerations of key uncertainties

Definitions of extremes and anomalies are based on historical baselines that may not reflect species-specific thresholds or ecological tipping points⁵⁰. Our statistical models capture associations rather than mechanistic causality, and may not fully account for complex biological responses such as recruitment lags, range shifts, or altered food-web interactions, or linkages with the socio-economic and governance dimensions of fisheries⁵¹ (Supplementary Table 6). Particularly, because reconstructed catch integrates ecological, economic, and governance influences, estimated odds ratios reflect socio-ecological system responses rather than purely biological stock responses. Although fishing effort data were unavailable for all species–EEZ units, sensitivity analyses restricted to the effort-available subset yielded consistent climate effect estimates, suggesting that omission of fleet dynamics in the primary model does not bias inference regarding climate-driven catch extremes. Regional variability in catch reporting, differences in monitoring capacity and management frameworks or actions introduce further uncertainty. Moreover, projections assume that past relationships between environmental extremes and catch anomalies remain stable under future conditions, an assumption challenged by potential species adaptation and evolving fisheries and management responses^52,53. Importantly, co-occurring stressors such as overfishing, habitat degradation, or deoxygenation were not explicitly included, which may underestimate cumulative risks. These stressors can act synergistically with climatic extremes to reduce resilience and amplify mortality or recruitment failure. These uncertainties underscore the importance of integrative, process-based models and improved monitoring to advance our knowledge to support adaptive fisheries governance.

Implications for fisheries management

Our findings underscore the urgent need for climate-adaptive fisheries management that addresses not only long-term environmental trends but also the growing risks of acute disruptions from marine extremes. Several sources of uncertainty warrant consideration (see “Method” for details), but these uncertainties generally suggest that our estimates of climate-driven fishery disruption are conservative. Strategies such as adaptive harvest control rules, early warning systems, diversification of target species, and strengthened monitoring and forecasting capacity will be essential for enhancing resilience. These needs are especially critical for at-risk fisheries in socio-economically vulnerable countries, where resources, scientific infrastructure, and human capacity are often limited. Targeted investments, international support and considerations of diverse global and local drivers affecting marine ecosystems and fisheries will be vital to ensure these regions can respond effectively to the compounding pressures of physical and biogeochemical extremes.

Methods

Fisheries time-series data

We obtained catch data from the Sea Around Us reconstruction database (www.seaaroundus.org), which estimates annual marine fisheries catches (in tonnes) by EEZ and species from 1993 to 2019. We follow the EEZ classification by the Sea Around Us, noting that it subdivides the EEZs of 198 coastal states into 280 regions (e.g., Canada’s EEZ is divided in Arctic, Atlantic and Pacific), including islands territories (updated 1 July 2015, available from https://www.seaaroundus.org). In total, we analyzed 6569 time-series comprising 178,256 catch records. Unlike landing-based datasets that attribute catches to the country of landing, the Sea Around Us data reconstructs where the catch was taken, assigning it to the spatial origin of fishing activity. This distinction is critical for our study, which aims to assess the spatial overlap between environmental extremes (e.g., temperature or NPP anomalies) and the regions where fish were actually harvested (see Supplementary Fig. 2).

The reconstructed database uses a wide variety of sources to estimate total removals, including industrial, artisanal, subsistence, recreational catch, and discards, many of which are absent from official national reports⁵⁴. It incorporates missing components of catch through the use of published and grey literature, fisheries reports, and expert knowledge, thereby producing a more comprehensive time series by species and country.

While catch reconstructions offer critical advantages for spatially explicit analyses, they are subject to caveats^54,55. The methods rely on assumptions and expert input in data-poor situations, which may introduce uncertainty or bias. The reconstruction approach and quality of underlying data also vary across taxa and regions, and current global fisheries statistics tend to exhibit a Global North bias, with better resolution and species identification in those regions than in the Global South. In addition, reconstructed catch data integrate reported landings with estimates of discards and unreported removals based on multiple data sources and modelling assumptions. While such reconstructions provide a more comprehensive representation of total removals, they are subject to uncertainty and methodological debate regarding the estimation of discards and IUU components⁵⁶. In this study, we focus on relative temporal anomalies within species–EEZ time series rather than absolute catch levels, thereby reducing sensitivity to potential systematic biases in level estimates.

Historical sea temperature and net primary production data

We used historical observation-based ocean temperature and model-based net primary production (NPP) datasets from the Copernicus Marine Environment Monitoring Service (CMEMS) to quantify environmental conditions associated with extreme fisheries catch anomalies.

Sea water temperature was derived from the Global Ocean Ensemble Reanalysis product⁵⁷, which provides monthly fields of global ocean temperature from 1993 to 2019 at 0.25° × 0.25° spatial resolution across 50 vertical levels. To match the resolution of other datasets, we bilinearly regridded the temperature data to a 0.5° × 0.5° global grid. Our main analysis used annual mean surface temperature (0–100 m). Moreover, we tested the effects of using depth strata that were more specific to species’ depth distribution: (i) surface layer (0–100 m), (ii) epipelagic layer (0–200 m), (iii) mesopelagic (200 - 1000 m) and (iv) bottom layer (deepest grid cell in the data product). In addition, we tested the effects of using average temperature across the depth strata. Each species was linked to the most relevant thermal layer based on its maximum depth from FishBase and SeaLifeBase: pelagic species with a depth below 200 m were linked to epipelagic temperature, demersal species to bottom-layer temperature, and upper water column species (maximum depth above 100 m) to surface layer (0–100 m).

Net primary production (NPP) data were obtained from the biogeochemical hindcast for global ocean produced at Mercator-Ocean⁵⁸, which provides global estimates of biogeochemical variables from 1993 onward using the PISCES model coupled with ocean reanalysis data. NPP is provided monthly at 0.25° × 0.25° spatial resolution. The robustness of the Mercator-Ocean hindcast dataset is supported by quantitative validation metrics detailed in the CMEMS Quality Information Document⁵⁹, which reports seasonal and global agreement for primary productivity and related biogeochemical variables via bias, root-mean-square deviation (RMSD), and Estimated Accuracy Numbers (EANs) against satellite and climatology reference datasets. We selected this dataset because of its consistent temporal coverage, physical-biogeochemical coupling, and relatively high spatial resolution, which together offer robust estimates of NPP anomalies across EEZs and ocean basins. We applied the same bilinear interpolation and regridding approach to convert this dataset to a 0.5° × 0.5° grid, and computed depth integrated annual total NPP by summing daily values for each year and grid cell.

The processed annual temperature and NPP fields were used to characterize environmental exposure for species across EEZs, identify climate extremes, and assess their association with extreme low and high catch anomalies in global fisheries.

Projected future sea water temperature and net primary production under climate change

We used future projections of sea surface temperature (SST) and net primary production (NPP) under climate change to assess future trends in the probability of extreme low catch events. Output from three Earth System Models (ESMs) were used for the analysis: IPSL-CM6A-LR, GFDL-ESM4, and MPI-ESM1-2-HR. Model outputs were taken from emissions scenarios SSP5-8.5 and SSP1-2.6 with simulations performed using the Coupled Model Intercomparison Project 6 (CMIP6) framework.

The raw ESM data were provided on a 1° × 1° regular global grid at a monthly temporal resolution. To harmonize spatial resolution across datasets, we regridded the monthly outputs to a 0.5° × 0.5° regular grid using bilinear interpolation. For near-coastal grid cells where bilinear interpolation introduced missing values due to partial ocean coverage, we applied a distance-weighted extrapolation method. This method employed the haversine formula to calculate distances between grid cell centroids, and applied inverse-distance weighting to estimate values from surrounding valid ocean cells. Following spatial interpolation, the monthly data was aggregated into annual means for SST and sums for NPP (SST and NPP) for each grid cell.

Analyzing the effects of extreme temperature and net primary production on catches

Consistent with Cheung et al.⁶, we define extremes at the annual scale to capture persistent or cumulative exposure to anomalous environmental conditions, including sequences of shorter-duration marine heatwaves. This temporal scale aligns with annual scale fisheries catch reporting and allows detection of lagged ecological and management responses. Prior to identifying extreme events, all environmental time series were statistically detrended to account for long-term climate trends that may confound interannual variability. We fitted generalized additive models (GAMs) with cubic splines separately to each species–EEZ time series to estimate and remove long-term temporal trends prior to defining extreme anomalies. The significance of the temporal trend was evaluated at α = 0.05. If significant, the residuals from the GAM were retained as detrended values; otherwise, mean-centred anomalies were used. This approach allowed us to isolate short-term deviations from long-term trends, thereby improving the robustness of our extreme event definitions. The detrended values were then used to compute anomalies for each variable.

To quantify the effects of extreme temperature and net primary production conditions on fisheries catches, we applied generalized linear modelling (GLM) across individual species–EEZ combinations. Each species–EEZ time series was treated as an independent panel, and inference is therefore conditional on species–EEZ units rather than on individual annual observations pooled across panels. We defined extreme events as annual anomalies in ocean temperature (Temp) and net primary production (NPP) that elevated beyond the 90th (high) or fell below the 10th (low) percentiles of their respective historical anomalies distributions of the detrended time-series. Unlike marine heatwave definitions, which are typically applied to daily temperature fields and require seasonally varying baselines, our analysis focuses on annual anomalies. Therefore, static percentile thresholds relative to the full historical distribution are appropriate for identifying interannual extremes. We also identified extremes using the maximum and minimum values in the time-series for Temp and NPP, respectively. Climate anomalies were matched to each species based on sea surface temperature exposure and the EEZ(s) in which they occurred. Similarly, reversed extremes are defined as the 10th or 90th percentiles of the historical anomalies distributions of the detrended SST and NPP, respectively.

For each species–EEZ time series, we modelled the probability of experiencing an extreme low catch year. This was defined as the bottom decile of the catch anomaly distribution, as a binary response variable using logistic regression. Because catch data do not directly measure biomass, extreme low catch events should be interpreted as production or removal anomalies instead of stock biomass changes. Predictor variables included indicators of extreme high temperature, low NPP, and their 1-year lagged values, along with their addition and interaction terms to capture compound event effects. Additive terms represent independent contributions of concurrent extremes, whereas interaction terms test for non-additive (synergistic or antagonistic) departures from independence. Interpretation of compound effects is therefore based on scenario-level predicted probabilities rather than on the interaction coefficient alone. To evaluate competing hypotheses about the importance of individual and compound stressors, we constructed a predefined candidate model set representing alternative ecological hypotheses about single-driver effects, lagged effects, additive co-occurrence, and interaction-based compound effects. All candidate models were specified a priori and ranked using Akaike Information Criterion (AIC).

Odds ratios (ORs) were derived from the coefficients of generalized linear models (GLMs) with a binomial error distribution and logit link. The GLMs estimated the relationship between the occurrence of extreme low catch events (binary response: 1 =  event, 0 = no event) and predictors representing temperature extremes (TempX), primary production extremes (NPPX), and their combinations. For each predictor, the model coefficient (β) from the logit scale was exponentiated (OR = e^β) to express the change in odds of an extreme low catch event associated with the presence of the extreme condition, relative to baseline (non-extreme) conditions. Because ORs are conditional on model structure, we do not interpret individual regression coefficients as standalone “effects” of single variables across different model formulations. Instead, we defined a scenario matrix of binary combinations of predictors and derived model-averaged contrasts between each extreme scenario and a baseline (no-extreme) condition using AIC weights. We applied this modelling framework to all EEZ-species time-series units globally, and to individual EEZ and functional group separately.

Model-averaged effect estimates were obtained using AIC weights, from which we derived weighted odds ratios (ORs) and associated 95% confidence intervals for each environmental extreme scenario. We considered effects statistically significant if the confidence interval did not include OR = 1. We presented model results globally and across EEZs and functional groups to assess spatial and taxonomic patterns in climate sensitivity. To identify key combinations of extremes, we developed a scenario matrix of binary combinations of the four predictor variables and computed model-averaged log-odds differences from a “baseline” (non extreme) condition.

As a sensitivity analysis, we additionally fitted generalized estimating equations (GEEs) with an AR(1) correlation structure to account for temporal autocorrelation within species–EEZ panels using the geepack package in R (v. 4.4.0). This analysis examines the effects of accounting for temporal autocorrelation within species–EEZ panels and repeated measures through time on the estimated ORs of extreme low catch in relation to TempX and NPPX. The GEE-based results were consistent with those of the GLMs, showing no significant differences in the estimated odds ratios or their relative rankings across predictor combinations, confirming the robustness of our findings to temporal dependence in the data (see Supplementary Fig. 10).

To evaluate sensitivity to alternative estimates of ocean productivity, we repeated the global and EEZ-level analyses using five alternative NPP products. These products represent commonly used satellite-derived and model-based estimates of ocean net primary production (NPP), each representing distinct biogeochemical formulations and parameterizations: NASA Ocean Biogeochemical Model, the Vertically Generalized Production Model and three other alternative products that are based on the VGPM method including Eppley-VGPM, CbPM-VGPM and CAFÉ-VGPM (see Le Grix et al.¹¹ for description of these NPP products). Because the alternative products are available only from 1998 onward, sensitivity analyses were conducted over the restricted 1998–2019, whereas the primary analysis spans from 1993. For each product, we recalculated NPP anomalies, defined extreme low NPP events using product-specific percentile thresholds derived from the available time series, and re-estimated odds ratios using identical generalized linear modelling procedures. We fitted models including an interaction term between NPP extreme indicators and NPP product to formally test whether estimated effects differed among products. Differences among products were assessed using likelihood ratio tests and comparison of model-averaged odds ratios. This framework allowed us to quantify structural uncertainty associated with alternative productivity formulations while holding the statistical modelling approach constant.

To illustrate the potential confounding effects of fishing pressure, we incorporated spatially explicit fishing effort data from the global database of mapped fishing activity described by Rousseau et al.⁶⁰. This dataset provides annual global fishing effort from 1950–2017 at 0.5° latitude x 0.5° longitude spatial resolution disaggregated by fishing sector, gear type, vessel size class, and targeted functional group. Effort is reconstructed from national fleet statistics and allocated spatially based on gear-specific catch distributions, resulting in gridded estimates of fishing activity comparable with AIS-derived datasets but extending further back in time.

To generate species–EEZ specific annual effective fishing effort time series, we applied the same spatial overlay approach used for SST and NPP. For each species, yearly gridded fishing effort was intersected with the species’ distribution map and EEZ boundaries. As fishing effort in the underlying dataset is reported at the level of targeted functional groups rather than individual species, we assigned effort values to each species according to its corresponding functional group classification. Within each EEZ, effort values were averaged across all grid cells overlapping the species’ range, yielding an annual mean fishing effort exposure for each species–EEZ unit. Effort was then standardized and categorized into binary extreme vs non-extreme states using the same percentile-based threshold procedure applied to environmental drivers. Because the fishing effort dataset does not cover all species–EEZ combinations with sufficient temporal resolution, effort analyses were conducted on the subset of species–EEZ units the environmental, catches and effort data are available.

To assess future trends in the probability of extreme low catch events, we applied the fitted GLMs to Earth system model (ESM) projections of SST and NPP under two Shared Socioeconomic Pathways (SSP1-2.6 and SSP5-8.5). We calculated annual scenario-specific probabilities of low catch events, aggregated these at the EEZ and functional group levels, and derived time-series estimates of the proportion of species with extreme low catch events under each climate scenario. We then calculated the global average proportion amongst the EEZs weighted by the number of species assessed therein. All analyses were implemented in R (version 4.3.2).

We selected SSP1-2.6 and SSP5-8.5 to bracket a wide range of possible climate futures rather than to represent the most likely trajectory. SSP1-2.6 reflects strong mitigation, whereas SSP5-8.5 represents a high-end emissions pathway commonly used in risk-based impact assessments to explore upper-bound climate impacts. These scenarios provide a bounded envelope of potential climate outcomes and allow evaluation of fisheries risk under contrasting mitigation trajectories.

Projections were limited to 2060 because the statistical relationships linking TempX and NPPX to catch anomalies were estimated from the historical period. These relationships may evolve over time due to shifts in species distributions, ecosystem restructuring, adaptive fisheries management, and socio-economic responses. Extending projections further into the late 21st century would require assuming stationarity in climate–fisheries relationships despite likely structural ecological and management changes. To avoid over-extrapolation under such non-stationarity and structural uncertainty, we adopted a mid-century projection horizon.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information ^{(4.2MB, pdf)}

Reporting Summary ^{(2.3MB, pdf)}

Transparent Peer Review file ^{(585.7KB, pdf)}

Source data

Source data ^{(819.2KB, zip)}

Author contributions

Conceptualization: W.L.C.; Methodology: W.L.C., T.L.F., J.P.A., I.M.; Investigation: W.L.C., I.M., T.L.F., J.P.A.; Visualization: W.L.C., J.P.A., T.L.F., I.M.; Funding acquisition and project administration: W.L.C.; W.L.C. wrote the initial draft and all authors contribute to the writing, review and editing of the original draft.

Peer review

Peer review information

Nature Communications thanks Keith Brander and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

We acknowledge funding support from: Natural Sciences and Engineering Research Council of Canada Discovery Grant (RGPIN-2024-05142) (WLC); Social Sciences and Humanity Research Council of Canada Partnership Grant (895-2022-1002) (WLC); EC Horizon 2020 Framework Programme for Research and Innovation (Grant agreement number 101137673) (TLF).

Data availability

The analyses combined multiple publicly available datasets (references provided in the Methods). The processed datasets required to reproduce the statistical analyses and figures are publicly available on Zenodo with this paper. Source data are provided with this paper.

Code availability

All code underlying the results are made available via https://doi.org/10.5281/zenodo.21350761.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77077-z.