Fisheries sub-goal

We changed the methods for calculating the status of the fisheries sub-goal of the food provision goal in response to improved models and available data, and both informal and formal suggestions [14–15]. The indicator aims to assess the amount of wild-caught seafood that can be sustainably harvested, with sustainability defined by multi-species yield, and with penalties assigned for both over- and under-harvesting. This approach requires establishing a reference point at which harvest is maximized within sustainable bounds. Previously this reference point was derived from an estimate of multi-species maximum sustainable yield (mMSY; [7, 11]), modified from Srinivasan et al. [16] to provide stock status assessments using catch data only (i.e., a ‘data-limited’ approach where variables generally required by formal stock assessment methods are unknown, as is the case for most commercially exploited species across the majority of areas reported to the Food and Agriculture Organization; FAO). In the current assessment, we modified our approach in multiple ways.

First, we changed the model used to assess the status of each individual stock. Since the initial assessment [7], several new data-limited approaches have been developed to assess fisheries that leverage globally-available catch data [17–21]. Building on these methodological advances, we developed a new approach to assessing food provision from wild caught fisheries that is based on estimating population biomass relative to the biomass that can deliver maximum sustainable yield (B/B_MSY) for each landed stock. Estimates of B/B_MSY were obtained by applying a model developed by Martell & Frœse [19], modified according to methods described in Rosenberg et al. [21], and hereafter referred to as the “catch-MSY” method. This method was chosen over other data-limited methods because simulation-tests demonstrated that it performed well in predicting stock status for simulated stocks having a broad range of life history traits and different known sources of uncertainty, i.e., environmental stochasticity, time-series length, initial depletion, and temporal autocorrelation [21]. The original catch-MSY method [19] is derived from stock reduction analysis [22], whereby a time series of catch is combined with an estimate of the final biomass relative to an unfished or initial biomass state (i.e., depletion level) in order to estimate historical biomass trends. A Schaefer surplus production model is used to produce ‘viable’ combinations of the intrinsic rate of growth, r, and the carrying capacity, K. ‘Viable’ was defined as any pair of r and K that did not allow the stock biomass to collapse or to exceed carrying capacity. In the original formulation of Martell & Froese [19], the geometric mean values of r and K were used to derive an estimate of MSY. Rosenberg et al. [21] modified this method by producing a biomass time series for each of the viable r-K pairs using the surplus production model. The arithmetic mean biomass time series was selected and the current year stock abundance (B) relative to the abundance that achieves MSY (B_MSY) produced a measure, B/B_MSY.

The model applies a constraint on the prior distribution used by the Bayesian model to predict final biomass (i.e., 0.01–0.4) based on whether the ratio of catch in the final year relative to historical peak catch is less than 0.5. When the ratio is greater than 0.5, the ratio is set 0.3–0.7. In applying this approach to the global catch data, we found this prior distribution caused the model to frequently estimate a decline in B/B_MSY for stocks with declining catch in the final years of the time-series. Explorations suggested that these included cases of managed fisheries where reduced catch was likely due to declining effort rather than declining population biomass. As an alternative, we applied a uniform prior, thereby removing this constraint. This resulted in estimates of biomass that were increasing when catch in the final years was declining (i.e., biomass increases due to a reduction in fishing pressure). This is an unlikely outcome in poorly regulated fisheries. Therefore, for analyses here we applied these two formulations of the prior distribution discriminately based on the level of management of a given stock in a given region; we assumed that the original constrained prior on final biomass is more appropriate in poorly regulated fisheries, while places with stronger fisheries management regulations were best modeled using a uniform prior on final biomass. In order to discriminate between these two cases, we assigned a governance-based resilience score to each stock, S_r. The resilience score was a mean of the fisheries resilience score (see S1 File) across all regions where the stock was caught, weighted by the relative mean catch in each of the regions: (Eq. 3) where n is the number of regions z (EEZs or high seas) in which the stock is caught, r_z is the fisheries resilience score assigned to that region, c_z is the mean catch of that stock in that region through time, and c_j is the mean catch of each of that stock in each of the regions. We estimated B/B_MSY with a uniform prior for all stocks with a resilience score of 0.6 or above, and used the model with the original constrained prior for all stocks with resilience scores less than 0.6. We recognize there is no precedent for using the model this way, and further testing would be valuable to establish more rigorous rules for how the priors are defined. Nevertheless, based on current knowledge and understanding, this approach was the best option. The estimated B/B_MSY was then used to produce stock status scores, SS, for each individual stock.

The catch-MSY approach improves upon the method used in Halpern et al. [7] by using a less simplistic relationship between catch and stock status. It is based on a mechanistic understanding of the connection between harvest dynamics and population dynamics and uses this functional link to infer stock depletion levels (see also [20]). Simply put, it takes into account life history traits and observed relationships between catch trajectories and depletion level, rather than just using an empirically-observed relationship between peak catch and MSY. Because the model uses more information from the catch time-series, it may more accurately describe stock status. In the case of developing fisheries that have not yet exploited stocks to their full productive potential (i.e., with catch levels that steadily increase over time), both approaches are flawed. The previous approach assigned a perfect score to these stocks. The new method does not resolve this issue fully, but it can be more informative at least in some cases where the time series is not monotonically increasing.

Second, we modified the approach to modeling fisheries to capture the portfolio effect afforded by preserving catch diversity [23], the status of wild caught fisheries (x_FIS) was calculated as the geometric mean of the stock status scores. The geometric mean allows stocks that are doing poorly, in particular smaller ones, to have a stronger influence on the overall score than they would using an arithmetic mean, even though their catch, C, may contribute relatively little to the overall tonnage of harvested seafood. The behavior of the geometric mean is such that improving a well-performing stock is not rewarded as much as improving one that is doing poorly. We believe this response is desirable because the recovery of stocks in poor condition requires more management effort and can have more important effects on the system than increasing the abundance of a species that is already abundant. Use of the geometric mean operationalizes our view that a healthy ocean sustainably provides a range of benefits to people now and in the future because it values both absolute tons of fish produced, the distribution of that catch among species, and the condition (B/B_MSY) of all harvested species in the system. Thus, it gives credit for preserving the health of the full range of species. The stock status scores for each taxon landed within each FAO major fishing area (A, noted below) were then combined weighting each taxon by its relative contribution to overall catch (C) within each reporting region and year, such that: (Eq. 4) where i is an individual taxon and n is the total number of taxa in the reported catch for that country or region throughout the time-series, and C was calculated as the taxon catch averaged across the time series from the first non-null record within each of our reporting regions.

Because many fish populations straddle the boundaries of EEZs, we calculated B/B_MSY by applying the catch-MSY model to catch aggregated within each major FAO fishing area, A. These values of stock status were then assigned to our reporting regions. Goal status was calculated at the spatial scale of our reporting regions, with each taxon’s stock status weighted by its mean catch relative to each reporting region’s total catch (note that for a geometric mean, weights appear in the exponent). This differs from the previous iteration, where the catch stream from each EEZ (based on catches that were spatially allocated to EEZs by the Sea Around Us project) was analyzed separately and aggregated up to the previous Ocean Health Index reporting regions. Any aggregation method will be biased in some way, but populations with the largest catches are most often straddling stocks, so they are not likely to be subject to aggregation biases. Instead, erroneous aggregation of catch could occur more often with high patchy species that primarily include small, sedentary populations that contribute little to a country or region’s fisheries and thus have little influence on overall catch.

The taxonomic level of each reported taxon was assigned to 1 of 6 categories derived from its ISSCAAP code (FAO International Standard Statistical Classification of Aquatic Animals and Plants, http://www.fao.org/fishery/collection/asfis/en, see ‘S5.68. Spatially-allocated catch data’ in S1 File for more details). B/B_MSY values could be directly calculated only when catch was reported at the species level, i.e., taxon group level 6, as the time-series of catch across miscellaneous taxa is unlikely to fit required model assumptions (the method of deriving stock status scores for higher level taxa is described below). Overall, we were able to evaluate a total of 1874 stocks. The estimated species level values of B/B_MSY were used to derive a stock status score, SS, such that the best score is achieved for stocks at B/B_MSY = 1, with a 5% error buffer, and the score decreases as the distance of B from B_MSY increases, due to under- or over-exploitation. For each species reported, within each major fishing area A, SS was calculated as: (Eq. 5) where, for B/B_MSY < 1 (-5% buffer), SS declines with direct proportionality to the decline of B with respect to B_MSY, while for B/B_MSY > 1 (+5% buffer), SS declines at a rate α, where α = 0.5, so that as the distance of B from B_MSY increases, SS is penalized by half of that distance. For B/B_MSY > 1 (+5% buffer), β is the minimum score a stock can get, and was set at β = 0.25. The α value ensures that the penalty for under-harvested stocks is half of that for over-harvested stocks (α = 1.0 would assign equal penalty). The β value ensures stocks with B/B_MSY > 1.4 due to, for example, an exceptionally productive year, are not unduly penalized, and also recognizes that goal scores are more easily improved when stocks are under-harvested (i.e., by increasing fishing pressure) than when they are over-harvested and need to be rebuilt. Both parameters α and β were chosen arbitrarily because there is no established convention for this particular approach. Thus, consistent with previous work [7], countries or regions are rewarded for having wild stocks at the biomass that can sustainably deliver the maximum sustainable yield, +/-5% to allow for measurement error, and are penalized for both over- or under-harvesting.

Third, the method of penalization for poorly reported catch statistics was modified. In the previous version [7], we used a taxonomic reporting index, T_C, to incorporate the consequence of under-reporting catch. This indicator was based on the assumption that if a country (or region) reported catching a species, and this species’ area of distribution overlapped with the EEZ of another country, then the other country must also be fishing it even if they do not report it as part of their catch. This T_C variable was intended to capture the proportion of a country’s catch that was not being officially reported and hence not monitored or managed, but it was subject to underestimation in countries with very large coastlines or remote locations, and over-penalizing places with many neighboring countries. In the calculations performed in 2012, this parameter strongly penalized scores in most countries [11]. Here, for taxa reported at a lower resolution than species, we developed a method to estimate stock status that accounts for coarser resolution data. The species-based estimates within the same fishing area and year were used to generate missing scores. An increasing penalty was applied for increasingly coarser taxonomic reporting, as this is considered a sign of minimal monitoring and management, so that, for a given taxonomic aggregation g (i.e., increasing coarseness of taxonomic reporting, when g<6), a proxy value for B/B_MSY was estimated as follows: (Eq. 6) The resulting value was then used to obtain the stock status score as shown in Equation 4. Thus, this new model adopts a more accurate estimate of individual stock status, accounts for multi-species effects by aggregating stock status scores through a geometric mean, and applies penalties for coarse taxonomic reporting as a reflection of weak monitoring effort. However, the philosophical approach first introduced by Halpern et al. [7] is maintained in its essence through two key properties: individual stock status scores are penalized for both over- and under-exploitation, and all species reported are assessed, albeit through a data-limited approach. Further details on this sub-goal model are provided in the Supplementary Methods in S1 File.

Mariculture Sub-goal

We also improved the method for calculating the status and reference point for the mariculture sub-goal of the food provision goal, based on sensitivity analyses exploring how scores were affected by choice of reference point [11]. The new approach calculates a country or region’s mariculture score based on total yield from mariculture (Y_M), the species-specific sustainability of that harvest (S_M,k), and coastal population density (P_C). Yield (and its sustainability) is adjusted by coastal population instead of coastal area (as was done previously in [7]) under the assumption that locally available workforce, coastal access and infrastructure needed for mariculture were proportional to that density. The current status of mariculture (x_MAR) is thus: (Eq. 7) where: (Eq. 8) and the reference point (Y_ref) is: (Eq. 9) with the 95^th percentile value (P₉₅) used due to the high skew in data.

Tourism & Recreation Goal

The tourism & recreation goal aims to capture the number of people, and the quality of their experience, visiting coastal and marine areas and attractions. The economic benefits of coastal tourism industries (i.e., jobs, wages, revenue), which can be important to coastal economies, are assessed separately as part of the coastal livelihoods & economies goal. Few non-economic indicators of tourism and recreation exist at the global scale, and thus the original approach in the 2012 assessment approximated this goal by measuring the number of international tourists arriving by airplane to coastal countries or regions, adjusting these values to the region’s population density to allow comparability across regions, and accounting for their average length of stay. This approach was sub-optimal, in part, because it did not account for domestic tourism, which is a large part of tourism in many countries, especially large countries such as Brazil, Canada, Russia, Australia and the USA. In the 2013 assessment we develop a different model to capture tourism and recreation, one that better accounts for both international and domestic tourism.

We used employment in the tourism sector as a reasonable proxy measure for the total number of people engaged in coastal tourism and recreation activities. Employment within this sector should respond dynamically to the number of people participating in tourist activities, based on the assumption that the number of hotel employees, travel agents and employees of other affiliated professions will increase or decrease with changing tourism demand within different regions.

Ideally there would be data available specifically for employment in coastal tourism industries, however the best data available at a global scale report total number of jobs, not just coastal jobs, within the travel and tourism industries (World Travel and Tourism Council (WTTC)). These data include jobs for both leisure and business that are directly connected to the tourism industry, including accommodation services, food and beverage services, retail trade, transportation services, and cultural, sports and recreational services, but exclude investment industries and suppliers [24]. Unfortunately it was not possible to determine the proportion of jobs affiliated with strictly leisure tourism. However, some (unknown) proportion of business travelers also enjoy the coast for leisure during their visit to coastal areas, such that we assumed all travel and tourism employment was related to tourism and recreation values. Regional assessments of the Index can make use of better-resolved data and more direct measures of how people enjoy and recreate in coastal areas, as has been done within the US West Coast [13], where data for participation in 19 different coastal recreational activities were available.

To approximate coastal travel and tourism employment using WTTC data, we calculated the proportion of direct employment in the tourism industry (E_d) relative to total labor force (E_t). As in 2012, we used the travel and tourism competitiveness index (TTCI) from the World Economic Forum [25] to capture the sustainability (S_t) of the tourism industry.

Therefore, the status of tourism & recreation (x_TR) is: (Eq. 10) where E_d is defined as the proportion of employees directly involved in the travel and tourism industry (E_t) relative to the total employees in that country (or region), calculated as the country’s total labor force (L_t) corrected by the percent of the population that is unemployed (U_t), such that: (Eq. 11) Because we do not know how employment patterns vary geographically within sectors for each country, we assume that the proportion employed in the tourism industry is the same in coastal areas as it is away from the coast, and thus E_i is the same whether applied solely to coastal areas or to the entire country. As such, the status of this goal could be increased by increasing a) the number people employed in the tourist industry relative to unemployment-corrected changes in the labor force within the whole country or b) the sustainability of tourism and recreation (measured with the TTCI).

Data for E_t existed for 148 regions (i.e., data were missing for 63 reporting regions; see Fig. C in S1 File). To fill the gaps for missing regions we used final goal scores rather than E_t values for the 148 regions and then followed the gap-filling guidelines described in section 6 of S1 File. We avoided gap-filling the E_t data layer because doing so created cases where the number of tourism jobs exceeded the reported labor force (data from the World Bank).

Given perfect data, one would use the best score across all regions and all years as the reference point (Saba in 2008). However, the highest-scoring regions were outliers in the distribution of scores (the second best performing region was 53% of the maximum and the third was 18%) and most scores were clustered around zero; we therefore rescaled all scores to the 90^th percentile score (the 21^st ranked region, Belize, with a score of 3). All regions above this score received a current status score of 100.

Data and Additional Analyses

Table I in S1 File lists all data layers, indicating which of the 77 layers were a) new data (8% of data layers used), b) updates of data sources used previously [7] but now containing new year(s) of data (47%), c) used previously but not updated because no new values have been reported (42%), and d) used previously but no longer included (4%). Details on all these data are provided in S1 File. Changes in category b include several data layers that were revised retrospectively for previous years, similar to how countries adjust economic data for past quarters and years. Such revisions were most notable for data from the Food and Agriculture Organization (FAO), used in the fisheries, mariculture and natural products goals. Additionally, Fig. C in S1 File provides a summary of which goals for each region required some level of spatial gap-filling due to non-reporting, with results summarized as the proportion of layers per goal, per region, that required gap-filling.

To assess how Index scores compared to other measures of the health of a region, we applied a regression model to determine whether the variation in OHI scores varied with national gross domestic product from 2012 (GDP data, USD; [9]), population data from 2011 [9], and Human Development Index (HDI) scores from 2012 (HDI, [26]). GDP and population variables were natural log- transformed prior to analysis. We compared models with every combination of these variables and selected the best model based on the Bayesian Information Criterion (BIC). To assess the relationship between goal scores, we determined the Pearson correlation coefficients between each pair of goals for all 221 reporting regions and fit scatterplots of each pair using locally-weighted polynomial regression (the lowess function in R).

Methodological changes

A key change in how Index scores were calculated and reported, compared to the first global assessment in 2012 [7], was the inclusion of 50 new reporting regions. Previously, these regions were aggregated into groups of primarily territorial holdings, but often the individual territories or islands are separated from each other by hundreds to thousands of kilometers and have different cultural and economic interactions with ocean ecosystems. Many are also relatively data-limited regions, which is why they were previously aggregated. To calculate Index scores for these regions, we developed a number of data gap-filling procedures (see S1 File). Consequently, even though scores are now reported for these regions, they generally have higher uncertainty than the scores for other reporting regions, so results should be interpreted with caution (see Fig. C in S1 File).

Methodological improvements in how the fisheries and mariculture sub-goals and the tourism & recreation goal were calculated provide more robust assessments of these goals than was previously possible. For all three cases, global area-weighted average scores increased relative to previous methods [7], although scores for individual countries showed decreases as well as increases. Thus, in these cases, previous data-limited approaches underestimated the overall likely health of these goals; future improvement in methods or data quality for other goals could have the opposite consequence, for example if more extensive or finer-scale data revealed greater levels of pollution or habitat loss. In either case, methodological improvements produce results with higher certainty. Therefore, although changes in assessed results highlight possible sources of error, and future methodological improvements may cause other small changes, imperfect but informed results are more useful than no information. Furthermore, for an Index to remain relevant and based on the best available science, it needs to adapt to improved data and scientific understanding. When such improvements occur, recalculating past scores with the new information, as we did here, allows for direct comparisons among years.

Policy implications

Many countries around the world have enacted policy encouraging or mandating actions that promote sustainable ocean ecosystems in a way largely synonymous with how the Index defines healthy oceans. In particular, the European Union has set an objective of ‘good environmental status’ [5], and the United States and Australia both have stated objectives of ‘healthy oceans’ that focus on environmentally sustainable development [1,6]. As such, the Index provides a unique quantitative tool for assessing progress towards meeting these comprehensive objectives.

Looking forward, we have identified three ways to increase the utility of the Ocean Health Index in informing and guiding management decisions. First, continuing to conduct annual assessments will allow for deeper insights into longer-term trends and stronger attribution of change to specific actions in particular locations. Thus, over time the Index can help assess management effectiveness. Second, translation of our data gap filling assessments (Fig. C in S1 File) into guidance on priorities for additional monitoring to fill those gaps and more quantitative measures of uncertainty will provide managers with key information about what is known with greater and less certainty, allowing for more strategic planning and investment. This is a current focus of our research and will be reported elsewhere. Third, regional applications of the Index at sub-national, national, or multinational scales based on local data and indicators provides more accurate regional assessments than possible through global assessments such as the one reported here, and thus better guidance for local management. As the number of these regional assessments increases, an opportunity emerges to leverage the lessons learned and assessment methodologies from those efforts for new regional applications. Several regional assessments have recently been completed [12–13, 28], and others are underway.