Main

The evolving vision of sustainability and food systems transformation requires the need to minimize the ecological impact of every fish caught and maximize its societal benefit1. In low- and middle-income countries (LMICs) and among populations vulnerable to malnutrition and diet-related diseases, that societal value may best be realized by maximizing the contribution of fisheries to nutrition and health. This can be achieved by minimizing waste and loss2 and by managing the fishery to supply the micronutrients most needed to support nutrition and healthy diets in the populations accessing that aquatic food3. The specific management of nutrient flows in agriculture has long been highlighted4,5, and recent studies emphasize the role of blue foods specifically in providing diets that improve social, environmental and human health outcomes6,7. Fish and other harvested aquatic animals comprise thousands of species, and while they all have similarly high protein content, they have very different micronutrient compositions8,9,10. This offers the possibility of managing fisheries to optimize the supply of the nutrients most lacking and most needed to improve the health outcomes of human diets. The nutrient yield from most of the world’s marine and inland fisheries could be better targeted11, and nutrient-sensitive fisheries management (NSFM) that also incentivizes maintenance of biodiverse ecosystems could be a pivotal strategy within this transformation12 but remains largely theoretical13.

While there is advocacy for biodiversity conservation (and corollary protection of nutrient diversity) in small-scale fisheries (SSFs), NSFM proposes the integration of nutritional goals to sustain fish populations and ecosystems whilst also contributing to human health and wellbeing14. This approach recognizes the crucial role that aquatic foods play in providing essential nutrients9,11, addressing micronutrient deficiencies8,15,16,17 and improving diets—particularly among vulnerable populations13,18,19,20—and seeks to optimize the nutritional benefits from fisheries. Fish and other aquatic foods are rich in bioavailable protein and iron as well as a key source of essential nutrients, such as omega-3 fatty acids, vitamin A, calcium and zinc, critical for early childhood development and overall health21. Such contributions cannot be overstated in the context of LMICs, which suffer most from micronutrients deficiencies (including 1.5 billion women and children globally22) and from stunting (affecting 148 million children in 202223). Recent efforts that complement NSFM approaches to improve nutrition in the Global South include a framework on nutrition-sensitive fisheries governance12, trade and foreign fishing assessments on nutrients’ relocation24, spatial analyses of aquatic food flows to inform nutrition-sensitive policies25 and analyses examining the integration of fisheries, food security and nutrition policies26,27,28,29,30. Despite the potential, there has been limited effort to translate this into practical fisheries management tools for LMICs to enhance nutrition within an Ecosystem Approach to Fisheries (EAF)12. Managing fisheries for multi-species maximum nutrient yield has been suggested, theoretically working through a framework for nutrient-based reference points in specific contexts31, but in data- and capacity-limited contexts, an approach to implement this and inform policies is still lacking.

SSFs account for 40% of the global reported catch from capture fisheries32 and play a crucial role in food and nutrition security in LMICs8,9,11,33. SSFs are often subsistence34, targeting a wide range of species with multiple gears across ecologically diverse habitats, beyond commercially lucrative species35. These mixed-species fisheries are too complex and data-poor for conventional fisheries management of single-species stock assessments that estimate the maximum sustainable yield (of generic fish protein) in tons. Conventional fisheries management tends to overlook the intricate species interactions and competition, leading to highly selective strategies that can disrupt ecosystem balance and reduce overall productivity36. EAF aims to balance ecological, economic and human wellbeing goals, providing a more sustainable and resilient fisheries management framework37,38,39 that tacitly advocates for nutrient diversity. However, this multi-dimensionality of EAF policies makes it hard to implement and evaluate, especially in data-constrained scenarios such as SSFs of LMICs.

Data remain a crucial part of EAF and is critical in managing fisheries for recovery and sustainability40, yet the quality and resolution of data from SSFs hamper evidence-based fisheries management. However, emerging digital advancements support robust data systems for SSF monitoring41,42 and present new opportunities for synthesized management tools, such as incorporating nutritional or climatic considerations. Peskas is one such open-source software for managing and visualizing SSF data and has become the official fisheries national monitoring system of Timor-Leste43,44. Moreover, data on nutrient composition of aquatic foods have only recently become available8,9. LMICs such as Timor-Leste, where stunting and micronutrient deficiencies such as anaemia are rife45 and child and maternal dietary quality remain poor46, could benefit from data-driven NSFM approaches to optimize the nutritional contributions of SSFs.

Using a 6-year catch data series44 (77,438 fishing trips) linked to NutrientFishbase and the Global Food Composition Database for Fish and Shellfish nutrient composition databases47, we developed an NSFM framework. We selected women of reproductive age (WRA) for nutrient analysis, a standard reference group in nutrition-sensitive fisheries research11,17, representing a large, nutritionally vulnerable group targeted by public health interventions. Our approach modelled the relationship between catches from different fishing methods and corresponding nutritional outcomes, offering a tool to visualize and optimize catch nutrient supply. The framework can be replicated using multi-species fisheries catch data from other contexts, while the model enables applied policy recommendations.

Results

Nutrient supply from SSFs

Our analysis confirmed that small pelagic fishes, particularly mackerels, flying fish and sardines/pilchards, were the most important contributors to recommended human dietary intake, providing relatively high concentrations of essential nutrients such as protein, zinc, calcium, omega-3 fatty acids, iron and vitamin A, compared with other fish groups (Fig. 1a). Marine invertebrates typically caught by gleaners, such as crabs, cockles and octopus, ranked particularly high in nutrient density, illustrating the potential of a diverse range of aquatic foods in enhancing dietary quality. Small pelagic fishes contributed most to the overall nutrient yield, owing to their higher nutrient density and volumes caught (Fig. 1b). By linking nutrient density to catch landing prices, we found that small pelagic fishes were also the most affordable, priced at under US$2 per kilogram. In contrast, larger pelagic fishes such as tuna/bonito were more expensive despite their modest nutritional density (Fig. 1c).

Fig. 1: Nutrient density, nutritional yield and affordability of marine foods from Timor-Leste fisheries.
Fig. 1: Nutrient density, nutritional yield and affordability of marine foods from Timor-Leste fisheries.
Full size image

a,d, The bars show the percentage contribution of a 100-g serving of raw edible flesh from marine foods to daily RNI for six nutrients for WRA (19–50 years old), where each bar is the mean RNI contribution across all species that form each functional group (a) or the combination of species typically caught relative to habitat and gear type (d), weighted by their contribution to total catch. The cumulative NDS (x axis) is the sum of the percentage contributions for the six nutrients. The plot in a ranks functional fish groups by cumulative NDS: small pelagic teleost fish (red), large pelagic (green), small demersal (yellow), large demersal (blue), sharks and rays (black) and marine invertebrates (grey). Bars are colour-segmented by nutrient, with white labels indicating the percentage contribution of each nutrient. Mean annual edible catch (metric tons, 2018–2023) is shown at the end of each bar, with transparency scaled to relative catch contribution. b, The cumulative nutritional yield (edible metric tons) from 2018 to 2023 for small pelagic (SP), large pelagic (LP), small demersal (SD), large demersal (LD), sharks and rays (SR) and marine invertebrates (MI), defined as the supply volume of calcium, omega-3, iron, protein, vitamin A and zinc; values for the lower four functional groups are very small relative to the other groups on this scale. c, A comparative analysis of NDS versus economic accessibility for key fish groups. The scatter plot relates cumulative NDS (x axis) to median landing price per kilogram (y axis) for the most important fish group within each functional category. The dot size reflects the relative catch percentage of each group, serving as an index of affordability and availability. d, The NDS for habitats (left) and gear types (right), based on a 100-g portion. Bars are colour-segmented by nutrient, with white labels showing each nutrient’s percentage contribution.

Different marine habitats and fishing gears influenced nutrient yields considerably (Fig. 1d and Extended Data Fig. 1). Results show that deploying fish aggregating devices (FADs) enhances the capture of species high in calcium, iron and omega-3, key nutrients for addressing micronutrient deficiencies in the region. Catches from reef habitats also showed high average nutrient density, with notable contributions from vitamin A and calcium.

Fisheries’ contribution to healthy diets

Approximately 62.9% of Timorese WRA (337,144, approximately one-quarter of the population) could potentially meet their protein balanced recommended nutrient intake (RNI) from edible annual catches, assuming seafood was distributed equitably among WRA (29.4 g day−1). Catches from SSFs have the potential to contribute the RNI of zinc to 31.3% of WRA and to 18.3% and 16.4% for calcium and omega-3, respectively (Extended Data Table 1). In recognition that single-nutrient analyses are of limited usefulness for quality diets, we present findings for multiple nutrient combinations. In Timor-Leste, where animal protein and micronutrient intakes are estimated to be below requirements46,48, marine catches could supply the RNI for four nutrients to ~55,000 WRA, and to ~22,000 for the six examined nutrients. In terms of satisfying the national food-based dietary guidelines’ recommendation of 250 g of cooked fish per adult per week49, current fisheries production has the potential to contribute to two-thirds, 65.4%, of all WRA (Extended Data Table 2). Municipal-level totals show that five municipalities catch enough fish to exceed the recommended quantities for WRA in the local population (Extended Data Fig. 2), suggesting trade and distribution opportunities with deficit municipalities and highlighting the need to develop nutrition-oriented value chains. Whilst current catches can support WRA’s nutrition, a substantial supply gap remains for meeting the broader population’s needs from aquatic foods.

FNP prediction

Using clustering techniques (elbow and silhouette methods) on nutrient density data from the Dili and Atauro municipalities of Timor-Leste, we identified three distinct fishery nutrient profiles (FNPs) applicable to both gill nets and other gears datasets (Extended Data Fig. 3). These profiles cluster fishing trips on the basis of the nutrient density of their catches (Box 1) and were statistically significant, as confirmed by permutational multivariate analysis of variance (PERMANOVA) analysis (Extended Data Table 3). Figure 2 illustrates the distribution of nutrient density scores (NDS) among these profiles for each gear type, highlighting variations in nutrient contributions across different fishing methods. In the gill nets dataset (Fig. 2a), FNP-3 stood out with the highest levels of calcium, iron and omega-3 fatty acids, surpassing FNP-1 and FNP-2. FNP-2 was notably higher in vitamin A compared with the other profiles. While protein levels were consistent across all profiles, FNP-1 exhibited higher zinc levels than FNP-2 and FNP-3. In the dataset for other gears (Fig. 2b), FNP-2 demonstrated markedly higher levels of protein, omega-3 and iron. In contrast, FNP-1 and FNP-3 showed elevated levels of zinc and vitamin A. Calcium distribution was even among all profiles.

Fig. 2: Distribution of cumulative NDS among the FNPs from k-means clustering in the Dili and Atauro municipalities.
Fig. 2: Distribution of cumulative NDS among the FNPs from k-means clustering in the Dili and Atauro municipalities.
Full size image

a,b, Plots delineating the NDS per 1 kg of catch using exclusively gill nets (a) and other gears (b) for various nutrients across the identified k-means clusters (N = 3), indicating contributions to the RNI. Each large point represents the mean NDS for a specific nutrient in each profile, while the smaller points illustrate the total distribution of the NDS within each profile.

The extreme gradient boosting (XGBoost) models accurately predicted FNPs on the basis of fishing gear (mesh size in the gill nets model), habitat, season and vessel type. The gill nets model achieved an area under the curve (AUC) score of 0.92 and the other gears model achieved an AUC of 0.91 (Extended Data Fig. 4). The other gears model demonstrated slightly better reliability in classification, particularly in its accuracy and specificity (Extended Data Table 4), effectively distinguishing FNPs across different fishing methods.

Feature importance (Extended Data Fig. 5) showed that in the gill nets model, the interaction between habitat and mesh size was the most influential, with mesh size alone also contributing considerably. In the other gears model, gear type was the most influential predictor, closely followed by its interaction with habitat. Shapley additive explanations (SHAP) values then revealed how those top predictors influenced the outcome of each FNP. Gill nets mesh sizes <40 mm were associated with FNP-2 and FNP-3 across various habitats (Fig. 3a). FNP-2 showed higher predictive influence in reefs, beaches and pelagic areas, indicating a strong likelihood of predicting this FNP when fishing in these environments. FNP-3 was also associated with smaller mesh sizes, particularly within pelagic zones, followed by mangroves and FADs. At mesh sizes ranging from 40 to 60 mm, FNP-3 was predominant, especially in reefs and FADs. In addition, FNP-1 showed some association within the 40–60 mm mesh size range, particularly in pelagic habitats and, to a lesser extent, in mangroves. Between 60 and 80 mm, FNP-1 became more characteristic across all habitats, especially in littoral environments such as seagrass, reefs and beaches. Larger mesh sizes were predominantly associated with FNP-2, especially in mangrove and reef habitats.

Fig. 3: Differential influence of mesh size and gear type from XGBoost models.
Fig. 3: Differential influence of mesh size and gear type from XGBoost models.
Full size image

a, The impact of mesh size on the probability of observing various FNPs on the basis of the habitat. Each sub-panel corresponds to a specific FNP (FNP-1, FNP-2 and FNP-3), showing how the contribution of mesh size to the prediction changes across a range of sizes. Data points are colour-coded by habitat (beach, pelagic, FAD, mangrove, reef and seagrass), illustrating how the effect of mesh size varies between environments. The y axis represents mesh size ranges, while the x axis shows the strength of the contribution of mesh size to the prediction of each FNP, represented by SHAP values. b, The role of gear type in predicting FNPs, with habitats influencing the effectiveness of each gear. Each sub-panel corresponds to a specific FNP, with gear types listed on the y axis. The x axis represents the strength of each gear type’s contribution to the prediction of a specific FNP. Larger and more opaque points denote gear types (or mesh sizes for gill net data) with higher SHAP values and stronger contribution to the predictions for each FNP in different habitats.

When excluding gill nets, SHAP analysis revealed strong associations between specific gear types and FNPs (Fig. 3b). FNP-1 was associated with spearfishing in reef and seagrass habitats, which exhibited the highest predictive influence and, to a lesser extent, gleaning. FNP-2 was strongly linked to long lines, especially in pelagic, FAD and mangrove areas, indicating a robust association between this gear type and the FNP-2 profile. For FNP-3, seine nets had the strongest predictive contribution across multiple habitats, including beach, reef and FAD zones. Hand lines also contributed to FNP-3 but with a weaker influence.

Discussion

SSFs are crucial for the livelihoods, culture and wellbeing of half a billion people globally11 and contribute substantially to the RNI of populations34,50. Our study presents an analytical framework and predictive model designed to guide nutrition-sensitive SSF management strategies, fostering interagency cooperation and policy coherence across fisheries, health, environmental and social inclusion domains.

Nutrition profiles associated with fish catches are highly relevant in countries where malnutrition is a critical concern, such as LMICs. Timor-Leste has one of highest child stunting rates globally, 47%45, and widespread anaemia48—typically caused by dietary iron deficiency and particularly affecting women51. Aquatic foods offer rich sources of micronutrients and omega-3 fatty acids8,9,10, essential for optimal brain development52; their consumption is associated with improved child health53 and reductions in micronutrient deficiencies9. Our framework enables countries with marine or inland SSFs to integrate nutrition goals into fisheries policies, helping address these health burdens. Importantly, it provides a practical tool to visualize and quantify the nutritional value of fisheries outputs, offering actionable data for policy and programmatic decisions.

Small pelagic fishes are productive, affordable and nutritious and should be promoted among nutritionally vulnerable groups, underscoring their role in improving healthy diets, as also found in other LMICs17,54,55,56. Nearshore FADs are well documented to increase access to tunas and small pelagic species for Pacific coastal fishers57,58,59,60, which can improve fish consumption when combined with social behaviour change interventions61 and be a productive and cost-effective investment62. We found that catches of small pelagic fishes from FADs increased the yield of calcium, omega-3 and iron, as shown in FNP-3 using gill nets. If deployed close to shore, FADs can be sustainably fished using existing methods and redistribute fishing pressure away from reef-associated fisheries in response to improved catch rates63. This can contribute to reef recovery and the retention of nutrient-rich and diverse reef species for small-scale and women-dominated fisheries such as gleaning64. Redirecting fishing efforts towards these systems indicates an opportunity to align livelihoods, nutrition and conservation goals.

The framework can inform multi-sectoral strategies in LMICs with SSFs, such as investing in FADs and linking trade networks with government-endorsed nutrition programmes, such as mother support groups65 or school meals66. Our analysis of nutrient yield relative to women’s recommended intake highlights how fisheries production can meet populations’ nutritional needs. In Timor-Leste, this analysis revealed supply gaps that can be addressed through sustainable fisheries and aquaculture development61,62,67. These findings support the integration of nutrition objectives into fisheries policy, advancing the agenda for NSFM.

Mesh size and gear types, in conjunction with habitat, were reliable predictors of FNPs, illustrating how different combinations of fishing strategies can achieve similar nutritional yields. For instance, gill nets with <40-mm mesh in pelagic and mangrove habitats—catching sardines and flying fish—or 50.8 mm in reef and FAD zones—catching short mackerels and scads—returned the highest concentrations of calcium, iron and omega-3 (FNP-3). Combining these configurations in management interventions, such as pairing FAD deployments with distribution of 50.8-mm nets, enhances the likelihood of obtaining FNP-3, as indicated by our predictive models. Long lines in pelagic habitats—catching jacks/trevally and tuna/bonito—returned the highest levels of protein, omega-3 and iron (FNP-2). Promoting government-led capacity building focused on the most nutrient-dense gear–habitat combinations and their regulation could enhance the nutrient yields of catches in different environments. Fisheries managers require a range of tools to meet the challenges posed by climate change and overfishing, while understanding the balance of gear selectivity and fishing pressure can potentially increase the resilience of marine ecosystems36,68.

Traditional management paradigms, such as maximum sustainable yield, are predominantly single-species oriented and fall short in multi-species data-poor SSF settings69. SSFs require a different approach owing to high species diversity, diverse gear usage and daily fisher decision-making37,70. In this setting, fishing trips represent a practical management unit, reflecting individual fishers’ decisions about these parameters, and associated livelihood considerations, such as optimizing catch per unit effort. By focusing on fishing patterns at the trip level, our FNP framework links nutritional outcomes to operational realities of fishers, providing actionable guidance on gear regulation, habitat management and capacity building.

Our results support distributing fishing effort across diverse habitats and gear types to optimize nutrient gains. This aligns with the balanced harvest concept, which explicitly addresses EAF as a measurable indicator, and where moderate fishing mortality on each usable (and societally acceptable) ecological group in proportion to its production can increase total long-term sustainable yields while minimizing impacts on ecosystem structure71. While implementation challenges exist72,73, balanced harvest principles may emerge naturally in SSFs, where demand and prices are evenly distributed across species74,75,76. Yet, to achieve a balance between optimizing nutrient harvests and sustainable yields while conserving ecosystem structure—and with it, biodiversity—we suggest FNPs would be best used as a heuristic tool, rather than a prescriptive management system.

For instance, FNPs can inform incentive-based and participatory governance strategies. In contexts where regulatory capacity (for example, mesh size enforcement or spatial closures) is limited, as in Timor-Leste, FNPs can guide the design of social and economic incentives—such as nearshore FAD deployment, value chain development for nutrient-dense species or integration with nutrition programmes—aligning local fishing practices with broader ecological and nutritional goals. Grounded in the lived realities of SSF communities, this tool offers a scalable pathway to integrate fisheries, health and social inclusion policies.

The framework provides a modelling approach for data-driven nutrition-sensitive SSF planning and management, complementing previous efforts16,31. Our primary contribution to NSFM is methodological; we propose an approach that links nutrient outcomes to specific fishing patterns—defined by gear, habitat and species combinations—at the scale of individual fishing trips. This reframes the core management question in multi-species SSF from ‘how much to fish?’ towards ‘how to fish?’, providing a more ecologically and socially relevant basis for decision-making. While applying the model in other countries may be constrained by data availability, open-source tools such as Peskas reduce technical and financial barriers to digital SSF monitoring44. As more LMICs invest in data systems, including vessel tracking, digital logbooks and citizen science initiatives, the feasibility of implementing this framework at scale increases77.

Future research should investigate ecological (for example, habitat and stock degradation caused by certain gear types and fishing strategies), economic (for example, gear costs and fuel efficiency as catch per unit effort factors) and social (for example, employment, gender equity and governance) considerations to refine FNP-based management strategies. Analyses addressing these integrated objectives would substantially advance the applicability of NSFM policies.

Limitations

Our approach depends on the availability of time series of species- and gear-specific catch data, so modelling with data from other contexts may not be equally robust. For data-scarce and capacity-constrained contexts, options such as the Peskas software44 are available. Nutrient composition values used include Bayesian estimates; future analyses would benefit from using primary data on aquatic foods’ nutrient content from a country or region, including ways to prepare and consume them. Limited gleaning data availability underestimates its nutritional contributions, as ‘marine invertebrates’ in Fig. 1 are particularly nutrient-dense; this women-led non-commercial fishery contributes to coastal nutrition security, but its inclusion in fisheries monitoring is limited by normative barriers78,79. Purchasing power, distribution and consumption patterns shape who benefits from fisheries’ nutrient yields; inclusive policies are critical to overcome structural and social barriers and ensure equitable access to aquatic foods.

There is a need for fisheries policies that maximize micronutrient mass80. Nations with substantial fisheries exports should prioritize domestic production for national consumption, especially where nutritionally vulnerable populations exist. But just as with broader ecosystem approaches and multi-species management, the major challenge to NSFM is implementation. We have demonstrated that a nutrition modelling approach can provide practical and applicable intervention points for NSFM. Improved data on food-critical fisheries focused on catch volumes, fishing methods and composition, as well as local nutrient data, will be important for practical implementation of NSFM. Finally, there is a need for national policymakers to shift attention away from production and revenue as indicators of fisheries development success and, instead, emphasize the role of ‘blue foods’ for social, environmental and human health outcomes27.

Methods

Study site and fisheries characterization

The SSF sector of Timor-Leste plays a modest role in the country’s formal economy yet is crucial for food security. Comprising approximately 4,554 vessels, the fleet is predominantly motorized (about 60%) and consists of canoes (40%)81. Fishing activities are primarily nearshore, targeting fringing coral reefs and the pelagic forereef zone within 5 km of the coast. The most common fishing gear employed is gill nets, which catch a diverse composition of fish species. More selected gear types used in reef areas include gleaning, speargun, hook and line, long line and seine nets. In the pelagic, FADs are utilized in some areas to concentrate small pelagic fish and make them easier to catch.

SSFs make crucial contributions to the livelihoods of coastal communities, particularly in rural areas where poverty rates are higher. Fishing not only provides income but also serves as a vital source of nutrition in a country with one of the highest malnutrition rates globally45. The average annual catch is estimated at 6,781 tons, with municipalities such as Atauro, Lautem, Bobonaro and Manatuto being the most productive. The catch composition is diverse, with larger fish species fetching higher market prices, particularly in Dili, where proximity to the capital reduces transportation costs.

The fisheries sector provides crucial resilience to system shocks such as the COVID-19 pandemic, where resilience was attributed to limited export activities and a modest tourism sector, which insulated local fisheries from broader economic shocks81. The Peskas monitoring system, developed in partnership with the Timor-Leste Ministry of Agriculture and Fisheries, has enhanced data collection and management, allowing for better decision-making and sustainable practices within the sector.

As Timor-Leste continues to invest in its fisheries sector, it is essential to balance modernization efforts with the needs of local communities, ensuring that initiatives address food security, dignified livelihoods and environmental sustainability. The ongoing development of the SSF sector can serve as a model for other Small Island Developing States facing similar challenges.

Analysis overview

First, we analysed the nutritional characteristics of Timor-Leste SSFs through descriptive analyses of (1) the nutritional density of the main species caught (by functional group and volume), (2) the cumulative yield (by nutrient from overall catches), (3) the affordability of key species vis-à-vis their nutrient density and (4) the nutrient density of catches (by marine habitat and gear type). This is the proposed NSFM analytical framework, which evaluates the nutritional properties of the overall fisheries composition (species/groups), their relative importance (production), accessibility (price) and key fisheries parameters (habitat and gears).

For analyses involving cumulative nutrient yields, we applied group-specific edible portion coefficients (Extended Data Table 5) to estimate the share of total catch weight that is typically consumed. These coefficients, derived from FAO Fisheries technical paper no. 309 (T0219E)82 and the Aquatic Foods Composition Database83, reflect raw edible fractions (mainly muscle tissue, with skin and soft bones where commonly consumed).

Second, we estimated the contributions of marine catches to human nutrition by assessing the number of people meeting the RNI and the recommended quantity of fish from the national food-based dietary guidelines, which are crucial for integrating public health nutrition policies. We focused on a population subgroup that requires more nutrient-rich diets, WRA, and conducted the analyses nationally and sub-nationally. This theoretical exercise illustrates how catches by geography can contribute to dietary quality of subpopulations and highlights opportunities for fish distribution between regions to maximize nutritional impact and equity gains. We present the percentage of WRA achieving 20% of the daily RNI for the six modelled nutrients from SSF catches11. This acknowledges that an adequate diet should include diverse food groups, and that fish alone is neither a realistic nor desirable sole source of 100% of a person’s RNI. Dietary diversity guidelines recommend consuming foods from at least five food groups84, and aquatic foods should be promoted as part of a diverse and balanced diet. We refer to the 20% RNI measure as ‘balanced RNI’.

Third, we proposed a method to model nutrition scenarios for managing SSFs. This involved identifying and validating FNPs specific to the fishery. We then predicted FNP outcomes on the basis of fishing gear, habitat, season and vessel type. Last, we explored how gear types and habitats interact to shape FNPs.

Data used

We utilized fishing catch data obtained from Peskas (https://timor.peskas.org), an open-source web portal offering insights into Timor-Leste SSFs44. This platform compiles catch data collected by local enumerators alongside vessel tracking data, enabling temporal and spatial analysis of fishing patterns. To date, the dataset comprises over 85,000 records of fishing trips along the Timor-Leste coastline, encompassing information about the fishing habitat, gear type used, number of fishers involved, catch composition, catch weight and economic value. Launched in 2016 through a collaboration with the Timor-Leste Ministry of Agriculture and Fisheries, Peskas functions as a near-real-time, low-cost, open-access monitoring system primarily targeting SSFs43. Data are continuously processed and validated according to a defined workflow. For this study, we used data collected from January 2018 to December 2023 (n = 77,438).

Nutritional values for catch data were derived using the Fishbase Nutrient Analysis Tool85. This tool employs a Bayesian hierarchical model, incorporating phylogenetic information to represent the interconnectedness of fish species, and trait-based data, reflecting critical aspects such as fish diet, thermal regime and energy demands. It predicts muscle flesh concentrations of seven essential nutrients: calcium, iron, omega-3 fatty acids, protein, selenium, vitamin A and zinc, for global marine and inland fish species. The nutritional yield for each catch was ascertained by merging edible weight estimates for 55 fish groups defined according to the ASFIS List of Species for Fishery Statistics Purposes86 with the model’s nutrient concentration predictions. We used the median nutrient concentrations (calcium, iron, omega-3 fatty acids, protein, vitamin A and zinc) for each species within the fish groups as a representative indicator of each fish group’s nutritional value. For non-fish groups such as octopuses, squids, cockles, shrimps, crabs and lobsters, where NutrientFishbase repository models lacked nutritional data, we used raw edible-portion values from the Global Food Composition Database for Fish and Shellfish47. Values for selenium were excluded from the analysis because its exceptionally high contributions to nutritional requirements across species obscured other results, and it is not considered a nutrient of major public health concern.

Finally, for each fishing trip, we calculated the NDS11 as the sum of RNI-equivalent contributions of six key nutrients per kilogram of catch. For each species in the landed catch, we multiplied its catch weight by the modelled nutrient concentration per 100 g of edible muscle, converted those masses into percentages of the RNI for WRA (19–50 years old)87 and then summed across calcium, iron, omega-3 fatty acids, protein, vitamin A and zinc to produce the trip-level NDS. By evaluating the combined contributions of calcium, iron, omega-3 fatty acids, protein, vitamin A and zinc, we were able to provide a broad assessment of the nutritional value of the catch. The RNI calculations are based on several assumptions. Protein requirements are based on World Health Organization (WHO) recommendations of 0.83 g kg−1 day−1 for adult females and assume an average body weight of 55 kg (ref. 88). This body weight is based on an observed mean body mass index of 20.6 for adult women in Timor-Leste48 and an assumed mean height of 163 cm based on WHO growth standards. Omega-3 polyunsaturated fatty acid requirements are based on Food and Agriculture Organization recommendations of 0.5–2% (taken at the midpoint of 1.25%) of dietary energy from omega-3s for adults89, assuming an average energy requirement of 8,700 kJ day−1 for a 55-kg adult female and physical activity level of 1.6 (ref. 88). Iron, zinc, calcium and vitamin A requirements are based on Food and Agriculture Organization/WHO global recommendations for an adult female aged 19–50 years, assuming a 10% dietary bioavailability of iron and a moderate bioavailability of zinc87.

Modelling and analyses

To investigate the impact of gear type, habitat, season and vessel type on the nutritional quality of catches in SSFs, we applied a combination of k-means clustering and machine learning techniques. We used high-resolution fisheries data from Dili and Atauro, two municipalities with the most complete and reliable trip-level records, and organized the dataset into two subsets: one focusing on gill net usage and the other encompassing all other gear types. The other gear category included hand and long lines, cast and seine nets, beach seines, traps and manual collection (gleaning). Because gill nets were the majority of recorded trips, we separated gill nets and other gear to ensure adequate sample sizes for clustering and subsequent modelling. In this classification, FADs were treated as artificial habitats rather than gear types, since multiple fishing gears (for example, long lines, gill nets and seine nets) can be deployed around them. This segmentation was essential for analysing the differential impacts of various fishing gear types on the nutritional outcomes of the catches.

As a first step, we employed k-means clustering to establish consistent FNPs across the datasets. Fishing trips were grouped on the basis of similarities in their RNI-equivalent contributions of six essential nutrients—calcium, iron, omega-3 fatty acids, protein, vitamin A and zinc. This clustering revealed patterns in nutrient levels, grouping fishing trips with similar nutritional characteristics. The optimal number of clusters, determined to be three distinct FNPs, was identified using the elbow and silhouette methods, which assess the compactness and separation of clusters.

To validate the consistency and significance of the identified FNPs, we conducted a PERMANOVA90. This analysis tested whether nutrient profiles within the same cluster were significantly more similar to each other than to those in different clusters, thereby confirming the robustness of our clustering approach.

After clustering, we developed two XGBoost models to predict FNPs using key fishing-trip attributes—gear type, habitat, season and vessel type—to assess how these factors influenced catch nutritional quality. XGBoost was chosen for its high predictive accuracy and its ability to capture complex nonlinear relationships and feature interactions, particularly when some gear–habitat combinations are rare. For the gill net model, predictors included mesh size (to capture variations in catch composition owing to different net sizes), habitat, their interaction (to capture combined effects), quarter of the year (reflecting seasonal variations) and vessel type (distinguishing between motorized and non-motorized boats). For the other gear model, predictors were gear type, habitat, their interaction, quarter of the year and vessel type.

Both datasets were divided into training (80%) and testing (20%) sets to build and evaluate the models. We applied tenfold cross-validation on the training set to enhance the models’ accuracy and generalizability, reducing the risk of overfitting91. Model tuning involved dynamically adjusting parameters such as the number of trees, tree depth, learning rate, loss reduction, sample size and early stopping criteria to optimize performance. The performance of the XGBoost models was assessed using several metrics, including accuracy, the area under the receiver operating characteristic curve, sensitivity (recall) and specificity. We obtained gain-based feature-importance rankings from each model to identify the most influential predictors.

Next, we calculated SHAP values92 for the predictors ranked highest by gain-based feature importance to interpret their influence on FNP outcomes in each model. SHAP values offer a unified approach to explain the output of machine learning models by quantifying the contribution of each feature to prediction and determining the direction of its impact. In the gill net subset, we focused on the impact of mesh size and habitats on FNPs. This analysis offered insights into how selecting certain mesh sizes could influence the nutritional quality of catches in different environmental contexts. For the other gear subset, we examined how gear type and habitat influenced FNP predictions. SHAP values allowed us to dissect the combined effect of specific gear types in particular habitats on the nutritional outcomes.

All statistical analyses were performed in R (version 4.3.3) using the built-in stats package (version 4.3.3) for k-means clustering, the vegan package (version 2.6-6.1) for PERMANOVA (adonis2) and the tidymodels metapackage (version 1.2.0) for building and tuning XGBoost models. SHAP values were computed with the kernelshap package (version 0.7.0).

Statistics and reproducibility

This study is based on an observational dataset of 77,438 fishing trips recorded between January 2018 and December 2023 through the Peskas monitoring system in Timor-Leste. Sample size was determined by the full availability of validated trip-level records collected during this period; therefore, no statistical method was used to predetermine sample size. All records meeting basic data completeness requirements for trip-level habitat, gear and catch information were included in the analyses. No data were excluded from the analyses, with the exception of fishing trips missing essential metadata required for modelling (for example, missing gear or habitat labels), for which exclusion criteria are reported in the Methods. Because the study uses observational monitoring data, randomization and blinding were not applicable. The analyses were not randomized and investigators were not blinded to allocation during analyses or outcome assessment. To evaluate reproducibility, the complete dataset, preprocessing scripts and analytical workflows used to generate the NDS, identify FNPs and train and validate the machine learning models are openly available (Data availability and Code availability). Reproducibility was further supported through standardized workflows for data cleaning and validation in Peskas, the use of 80/20 training–test splits and tenfold cross-validation for XGBoost models and independent validation of clustering via PERMANOVA. The robustness of clustering was assessed using elbow and silhouette methods to determine optimal cluster number, and cluster significance was evaluated using PERMANOVA. Model reproducibility is supported by transparent hyperparameter tuning and reporting of performance metrics, including area under the receiver operating characteristic curve, accuracy, sensitivity and specificity, for both gill nets and other gear models.

Reporting summary

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