Our study has three key results. First, we find that overall possible impacts on fisheries catch potential is higher than possible impacts on agriculture, but there can be substantial within-country variability in both exposure and sensitivity (Fig. 1). Specifically, exposure under the high-emissions Shared Socioeconomic Pathway 8.5 scenario (which has tracked historic cumulative CO2 emissions17, but has been recently critiqued for over-projecting CO2 emissions and economic growth18) indicates substantive losses by mid-century to fisheries catch potential [Fig. 1; 14.7% +/- 4.3% (SE) mean fisheries catch potential loss]. To put these projected losses in perspective, Sala et al.19 found that strategically protecting 28% of the ocean could increase food provisioning by 5.9 million tonnes, which is just 6.9% of the 84.4 million tons of marine capture globally in 201820. Thus, the mean expected fisheries catch potential losses are approximately double that which could be buffered by strategic conservation. Model run agreement about the directionality of change for projected impacts to fisheries catch potential was high (SSP5-8.5: 84.7 +/- 4.5% (SE); SSP1-2.6: 89.2 +/- 4.06% (SE)). Interestingly, crop models projected that agricultural productivity (based on rice, maize, and cassava- see methods) is expected to experience small average gains across the 72 sites (1.2% +/- 1.5% (SE) mean agricultural gain), with a large response range between sites and crops (Fig S1). However, the average gains are not significantly different from zero (t = -0.80, df = 5.0, p = 0.46), and model run agreement about directionality of change was lower for agriculture (SSP5-8.5: 69.1 +/- 4.82% (SE); SSP1-2.6: 70.4 +/- 3.27% (SE)). These projected agricultural gains are driven exclusively by rice (Supplemental Fig. 1), which has particularly large model disagreement14,21. Excluding rice shows an average decline in agricultural production by mid-century, since maize and cassava show consistent median losses under both SSP1-2.6 and SSP5-8.5 climate scenarios (Supplemental Fig. 1). Significantly greater losses in fisheries catch potential compared to agriculture productivity are apparent not only for our study sites (i.e. 15.9 +/- 5.6% (SE) greater; t = 2.81, df = 4.97, p = 0.0379), but also for a random selection of 4,746 (10% of) coastal locations in our study countries with populations > 25 people per km2 (Fig. 2). Among those random sites, fisheries catch potential losses are an average of 15.6 +/- 5.1% (SE) greater than agriculture productivity changes (t = 3.06, df = 5.00, p = 0.0282). Differences between expected losses at our sites and the randomly selected sites are small for agriculture (Cohen’s D for SSP5 − 8.5 = − 0.31, SSP1 − 2.6 = − 0.35) and negligible for fisheries catch potential (Cohen’s D for SSP5 − 8.5 = − 0.02, SSP1 − 2.6 = − 0.03), indicating that our sites are not particularly biased towards high or low exposure for the study region. Not only is the level of exposure generally higher in fisheries compared to agriculture, but the sensitivity is on average nearly twice as high (Fig. 1A,B; 0.077 +/- 0.007 mean fisheries sensitivity; 0.04+/-0.01 mean agricultural sensitivity; t = 3.0, df = 2.26, p-value = 0.0815).
Our analysis also reveals high within-country variability in potential impacts (i.e. both exposure and sensitivity), particularly for fisheries (Fig. 1) - a finding that may be masked in studies looking at national-level averages15,16. Looking only at the mean expected losses can obscure the more extreme fisheries catch potential losses projected for many communities (Figs. 1,2). For example, under SSP5-8.5, our Indonesian sites are projected to experience very close to the average fisheries catch potential losses among our study sites (15.9 +/- 2.1%SE), but individual sites range from 6.5–32% losses (Fig. 1B). There is also substantial within-country variation in how communities are likely to experience climate change impacts, based on their sensitivity (Fig. 1A,B). For example, in the Philippines, exposure to fisheries is consistently moderate (range 8.9–12.6% loss), but sensitivity ranges from our lowest (0.001) to our highest recorded scores (0.32). There is also within-country variability in model agreement, particularly for the agricultural models in Indonesia, where agricultural model agreement ranges from 50–85% and fisheries model agreement ranges from 56–100% for SSP5-8.5, and 50–80% and 50–94%, respectively, for SSP1-2.6.
The second key result from our integrated assessment reveals that some locations will bear a double burden of losses to fisheries and agriculture simultaneously, but mitigation efforts that reduce greenhouse gas emissions could curb these losses. Specifically, under SSP5-8.5, 64% of our study sites are expected to lose productivity in fisheries and agriculture simultaneously (Fig. 3A), but this would reduce to 37% of sites under the low emissions scenario SSP1-2.6 (Fig. 3B). Again, the effect of mitigation is consistent in the random selection of 4,746 sites (Supplemental Fig. 2), with 70% of randomly selected sites expected to experience a double burden under SSP5 8.5, and 47% under SSP1 2.6. Many of the sites expected to experience the highest losses to both fisheries catch potential and agriculture have moderate to high sensitivity (Fig. 3A, Supplemental Fig. 3), which means the impacts of these changes could be profoundly felt by coastal communities.
Over a third of our sites (36% under SSP5-8.5) are expected to experience increases in agriculture (due to CO2 fertilization effects that fuel potential increases particularly in rice yields) while experiencing losses in fisheries catch potential. For these sites, a question of critical concern is whether the potential gains in agriculture could help offset the losses in fisheries catch potential. The answer to this lies in part in the degree of substitutability between sectors. Our survey of 3008 households reveals high variation among countries, and even within some countries in the degree of household occupational multiplicity incorporating both agriculture and fisheries sectors (Table 1). 31% of households in our study engaged in both fishing and agriculture, though this ranged from 10% of households in the Philippines to 77% of households in Papua New Guinea. This means that the degree to which agricultural gains might possibly offset some fisheries losses at the household scale is very context dependent. Our survey also revealed that 17% of households were involved in agriculture but not fisheries, ranging from 33% in Madagascar to 3% in our Papua New Guinean study communities. Alternatively, more than a third of households surveyed in Indonesia and Philippines were involved in fisheries but not agriculture (36% and 37% respectively), compared to a low value of 16% in Madagascar. In 12% of the Philippines communities surveyed (n = 3), not a single household was engaged in agriculture. Thus, for 32% of households across our sample, including some entire communities, potential agricultural gains will not offset potential fisheries losses. In these locations building adaptive capacity to buffer change will be critical9.
Table 1
Proportion of surveyed households in each study country engaged in both agriculture and fisheries, agriculture but not fisheries, and fisheries but not agriculture. Note, proportions do not add up to 1 because some households were not engaged in agriculture or fisheries.
country | Number of households | Agriculture and fisheries | Agriculture, no fisheries | Fisheries, no agriculture |
indonesia | 1140 | 0.25 | 0.18 | 0.36 |
madagascar | 339 | 0.42 | 0.33 | 0.16 |
papua new guinea | 318 | 0.77 | 0.03 | 0.18 |
philippines | 973 | 0.11 | 0.18 | 0.37 |
tanzania | 238 | 0.69 | 0.04 | 0.26 |
Our third key result is that coastal communities with lower socioeconomic status are more likely to experience potential impacts than communities of higher socioeconomic status across the climate mitigation scenarios (SSP1-2.6 and SSP5-8.5; Fig. 4). Specifically, we examined the relationship between the average material style of life (a metric of wealth based on material assets; see methods) in a community and the relative potential impacts of simultaneous fisheries catch potential and agriculture losses (measured as the Euclidean distance of sensitivity and exposure from the origin). Importantly, socioeconomic status is related to both sensitivity and exposure (Supplemental Fig. 4). In other words, low socioeconomic status communities tend to have higher sensitivity to fisheries and agriculture than the wealthy, and are significantly more likely to be exposed to climate change impacts. Our findings regarding the relationship between socioeconomic status and sensitivity are consistent with a broad body of literature that shows how people tend to move away from natural resource dependent occupations as they become wealthier10,22−25. One potential interpretation of our findings is that alternative livelihood programs (e.g. jobs outside the fisheries or agricultural sectors, such as the service industry) could reduce sensitivity in lower socioeconomic status communities. However, decades of research on livelihood diversification has highlighted a multitude of reasons why alternative livelihood projects frequently fail26, including that they do not provide high levels of non-economic satisfactions (e.g., social, psychological, and cultural)27–29, as well as cultural barriers to switching occupations (e.g. caste systems)30, and attachment to identity and place31. Alternative occupations need to provide some of the same satisfactions, including basic needs (safety, income), social and psychological needs (time away from home, community in which you live, etc.), and self-actualization (adventure, challenge, opportunity to be own boss, etc.). For example, fishing attracts individuals manifesting a personality configuration referred to as an externalizing disposition, which is characterized by a need for challenges, adventure, and risk. Fishing can be extremely satisfying for people with this personality complex, while many alternative occupations can lead to job dissatisfaction, which has negative social and psychological consequences32,33. Research has shown that recreational fishing captain or guide jobs produce some of the same satisfactions as fishing and have been successfully introduced as alternative occupations33. Despite these limited successes, alternative livelihood programs frequently fail and are not a viable substitute for mitigating climate change for the ~ 6 million coral reef fishers globally34.
Our study is an important first step in examining the potential simultaneous impacts to fisheries catch potential and agriculture in coastal communities, but has some limitations, some of which could be addressed in future studies. First, our measure of exposure was dynamic (i.e., it was projected into the future), while our measures of sensitivity and material wealth were static (i.e., from a single point in time) and did not consider potential changes over time. Although there are projections of how national-scale measures of wealth (e.g. gross domestic product; GDP) may change in the future, there are no reliable projections for household- or community-scale changes to material wealth or livelihoods. As a supplemental analysis, we examined observed changes in sensitivity and material wealth over 15 and 16 years, respectively, in two Papua New Guinean coastal communities (Fig. 5). We found that, over the observed time frame (2001–2016), which is approximately half that of the predicted time frame of exposure, sensitivity scores were extremely stable, particularly in Ahus (Fig. 5). Similarly, material wealth was also reasonably stable over time, but did reflect a shift in both communities toward more houses being built out of sturdier material (e.g., wood plank walls and floor, metal roofs). Importantly, while there were absolute changes to material wealth in both communities, the relative position stayed very similar. Although these data do not allow us to make inferences about what will happen into the future, they do highlight that, at least in decadal timeframes, these indicators are reasonably stable. One alternative approach may have been to assume that projected national-scale changes to GDP would apply evenly across each coastal community within a country (i.e., adjust the intercept of both material wealth and correlated sensitivity for each country relative to the projected changes in GDP). However, given the wide spread of material wealth and sensitivity scores within countries, we ultimately were less comfortable with the assumptions inherent in the approach (i.e., that national-scale changes would affect all communities in a country equally) than with the caveat that our metrics were static.
Second, there are key limitations and assumptions to the models we used. For example, many tropical small-scale fisheries target seagrass35 and coral reef habitats34, which are not represented in the global ensemble models. Additionally, the ensemble models were developed at relatively low spatial resolution (e.g. 1° cells), and are not designed to capture higher resolution structures and processes. Our approach for dealing with this was to make transparent the degree of ensemble model run agreement about the direction of change, which relies on the assumption that we have greater confidence in projections that have higher model run agreement. Another limitation is that there may be discrepancies between the total consumer biomass (see method) in the absence of fishing that is outputed by the models used here and what would actually be harvested by fishers since total consumer biomass can include both target and non-target species as well as other taxa entirely. Despite these limitations, we assumed that total consumer biomass is directly related to potential fisheries yields11. Likewise, we included just 3 crops in the agricultural models (rice, maize, and cassava), which are key in the study region, with many study countries growing 2 or more of these crops. For example, Indonesia is the 3rd largest producer of rice in the world, and the 6th largest producer of maize and cassava36. However, subsistence agriculture in Papua New Guinea is dominated by banana and yams, for which agricultural yield projections were not available. We used an unweighted average of projected changes in these three crops to represent a portfolio of small-scale agriculture, with a sensitivity test based on agricultural projections weighted by current yields/production area proportions of current yields (Supplemental Fig. 1). Finally, it is important to keep key model assumptions in mind when interpreting these data. For example, the agricultural models assumed no changes in farm management or climate change adaptation over time, while the fisheries models do not explicitly resolve predation impacts from higher trophic levels on phytoplankton.
Third, our sensitivity metric examined a somewhat narrow aspect of what makes people sensitive to climate change. Sensitivity is thought to contain dimensions of economic, demographic, psychological, and cultural dependency37. Our metric was based on people’s engagement in natural resource-based livelihoods, which primarily captures the economic dimensions (although livelihoods do provide cultural and psychological contributions to people26,28,29,31,38).
Fourth, our study explicitly focused on the potential impacts of climate change in 72 Indo-Pacific coastal communities by examining their sensitivity and exposure, but our methodology did not enable us to incorporate adaptive capacity. Adaptive capacity is a latent trait that enables people to adapt to and take advantage of the opportunities created by change39,40, and is critically important in determining the fate of coastal communities under climate change. Adaptive capacity is thought to consist of dimensions of assets, flexibility, social organisation, learning, socio-cognitive, and agency9,41,42. Unfortunately, indicators of these dimensions of adaptive capacity were not collected in a standardised manner across all of the different projects comprising this study.
Fifth, we investigated the potential impacts of climate change on two key food production sectors, but there may be other climate change impacts which have much more profound impacts on people’s wellbeing. For example, sea level rise may destroy homes and other infrastructure43, while heat waves may result in direct mortality44. Lastly, we used shared socioeconomic pathway exploratory scenarios that bracket the full range of scenario variability (SSP5-8.5 and SSP1-2.6). At the time of publication, these were the only scenarios available for both fisheries and agriculture using the ISIMIP Fastrack Phase 3 dataset. Future publications may wish to explore additional scenarios.
Our study quantifies the potential impacts of climate change on key food production sectors in tropical coastal communities across a broad swath of the Indo-Pacific. We find that both exposure and sensitivity to fisheries is generally higher than to agriculture, but some places may experience losses from both sectors simultaneously. These losses may be compounded by other drivers of change, such as overfishing or soil erosion, which is already leading to declining yields45,46. Simultaneous losses to both fisheries catch potential and agriculture will limit people’s opportunity to adapt to changes through switching livelihoods between food production sectors9. This will especially be the case in lower socioeconomic status communities where dependence on natural resources is higher10. Together, our novel integration of model projections and socioeconomic surveys highlight the importance of assessing climate change impacts across sectors, but reveals important mismatches between the scale at which people will experience the impacts of climate change and the scale at which modelled projections about climate change impacts are currently available.