Impacts, Environmental

Reconstructing time series of fisheries catch for all countries of the world from 1950 (the first year that FAO published its ‘Yearbook' of global fisheries statistics) to 2010 was undertaken by fisheries ‘sectors'. However, because a standardized global definition of fishing sectors based on vessel size does not exist (for example, a vessel considered large-scale (industrial) in a developing country may be considered small-scale (artisanal) in developed countries), reconstructions utilize each country's individual definitions for sectors, or a regional equivalent. These are described in each country reconstruction publication underlying this work. We consider four sectors:

Industrial: large-scale fisheries (using trawlers, purse-seiners, longliners) with high capital input into vessel construction, maintenance and operation, and which may move fishing gear across the seafloor or through the water column using engine power (for example, demersal and pelagic trawlers), irrespective of vessel size. This corresponds to the ‘commercial' sectors of countries such as the USA;

Artisanal: small-scale fisheries whose catch is predominantly sold (hence they are also ‘commercial fisheries'), and which often use a large variety of generally static or stationary (passive) gears. Our definition of artisanal fisheries relies also on adjacency: they are assumed to operate only in domestic waters (that is, in their country's EEZ). Within their EEZ, they are further limited to a coastal area to a maximum of 50 km from the coast or to 200 m depth, whichever comes first. This area is defined as the Inshore Fishing Area (IFA)44 . Note that the definition of an IFA assumes the existence of a small-scale fishery, and thus unpopulated islands, although they may have fisheries in their EEZ (which by our definition are industrial, whatever the gear used), have no IFA;

Subsistence: small-scale non-commercial fisheries whose catch is predominantly consumed by the persons fishing it, and their families (this may also include the ‘take-home' fraction of the catch of commercial fishers, which usually by-passes reporting systems); and

Recreational: small-scale non-commercial fisheries whose major purpose is enjoyment.

In addition to the reconstructions by sector, we also assign catches to either ‘landings' (that is, retained and landed catch) or ‘discards' (that is, discarded catch), and label all catches as either ‘reported' or ‘unreported' with regards to national and FAO data. Thus, reconstructions present ‘catch' as the sum of ‘landings' plus ‘discards'.
Discarded fish and invertebrates are generally assumed to be dead, except for the US fisheries where the fraction of fish and invertebrates reported to survive is generally available on a per species basis45 . Due to a distinct lack of global coverage of information, we do not account for so-called under-water discards, or net-mortality of fishing gears46 . We also do not address mortality caused by ghost-fishing of abandoned or lost fishing gear47 .
For commercially caught jellyfishes (particularly Rhizostomeae, but also other taxa), it has been shown that over 2.5 time more are caught than reported to FAO (mostly as ‘Rhizostoma spp.')48 . This factor is used to estimate missing catches of unidentified jellyfish. However, this additional catch is, pending further study, not allocated to any specific country or FAO area, and is thus counted only in the world's total catch.
We exclude from consideration all catches of marine mammals, reptiles, corals, sponges and marine plants (the bulk of the plant material is not primarily used for human consumption, but for cosmetic or pharmaceutical use). In addition, we do not estimate catches made for the aquarium trade, which can be substantial in some areas in terms of number of individuals, but relatively small in overall tonnage, as most aquarium fish are small or juvenile specimens49 (link).
Most catch reconstructions consist of six steps15 :
(1) Identification, sourcing and comparison of baseline catch times series, that is, (a) FAO reported landings data by FAO statistical areas, taxon and year; and (b) national or regional data series by area, taxon and year. Implicit in this first step is that the spatial entity be identified and named that is to be reported on (for example, EEZ of Germany in the Baltic Sea), something that is not always obvious, and which poses problems to some of our external collaborators, notably those in countries with a claimed EEZ overlapping with that of their neighbour.
For most countries, the baseline data are the statistics reported by member countries to FAO. We treat all countries recognized in 2010 (or acting like independent countries with regards to fisheries) by the international community as having existed from 1950 to 2010. This is necessary, given our emphasis on ‘places', that is, on time-series of catches taken from specific ecosystems. This also applies to islands and other territories, many of which were colonies, and which have changed status and borders since 1950.
For several countries, the baseline data are provided by international bodies. In the case of EU countries, the baseline data originate from the International Council for the Exploration of the Sea (ICES), which maintains fisheries statistics by smaller statistical areas, as required given the Common Fisheries Policy of the EU. A similar area is the Antarctic waters and surrounding islands, whose fisheries are managed by the Commission for the Conservation of Antarctic Marine Living Resources (CCAMLR), where catch data are available by relatively small statistical areas50 .
When FAO data are used, care is taken to maintain their assignment to different FAO statistical areas for each country (Supplementary Fig. 1), as they often distinguish between strongly different ecosystems. For example, the Caribbean Sea versus the coast of the Eastern Central Pacific in the case of Panama, Costa Rica, Nicaragua, Honduras and Guatemala. For each maritime country, the area covered extends from the coastline to the edge of the EEZ, including any major coastal lagoons connected to the sea, and the mouths of rivers, that is, estuaries. However, freshwaters are excluded.
(2) Identification of sectors (for example, subsistence, recreational), time periods, species, gears and so on, not covered by (1), that is, missing data components. This is conducted via literature searches and consultations with local experts. This step is one where the contribution of local co-authors and experts is crucial. Potentially, all four sectors defined by us can occur in the marine fisheries of a given coastal country, with the distinction between large-scale and small-scale being the most important25 (link). For any entity, we check whether catches originating from the four sectors were included in the reported baseline of catch data, notably by examining their taxonomic composition, and any metadata, which were particularly detailed in the early decades of the FAO ‘Yearbooks'51 .
The absence of a taxon known to be caught in a country or territory from the baseline data (for example, cockles gleaned by women on the shore of an estuary)26 can also be used to identify a fishery that has been overlooked in the official data collection scheme, as can the absence of reef fishes in the coastal data of a Pacific Island state10 . To avoid double counting, tuna and other large pelagic fishes, unless known to be caught by a local small-scale fishery (and thus in the past not likely reported to a Regional Fisheries Management Organization or RFMO), are not included in this reconstruction step (see below under ‘High Seas and other catches of large pelagic fishes').
Finally, if gears are identified in national data, but a gear known to exist in a given country is not included, then it can be assumed that its catch has been missed, as documented for weirs (hadrah) in the Persian Gulf52 .
(3) Sourcing of alternative information sources on missing sectors identified in (2), via literature searches (peer-reviewed and grey) and consultations with local experts. Information sources include social science studies (anthropology, economics and so on), reports, data sets and expert knowledge. The major initial source of information for catch reconstructions is governments' websites and publications (specifically their Department of Fisheries or equivalent agency), both online and in hard copies. Contrary to what could be expected, it is often not the agency responsible for fisheries research and initial data collection that supplies the catch statistics to FAO, but other agencies, for example, statistical office or agency. As a result, much of the granularity of the original data (that is, catch by sector, by species or by gear) may be lost even before data are prepared for submission to FAO. Furthermore, the data request form sent by FAO each year to each country does not encourage improvements or changes in taxonomic composition, as the form that requests the most recent year's data contains the country's previous years' data in the same composition as submitted in earlier years. This encourages the pooling of detailed data at the national level into the taxonomic categories inherited through earlier (often decades old) FAO reporting schemes, as was discovered, for example, for Bermuda in the early 2000s (ref. 53 ). Thus, by getting back to the original data, much of the original granularity can be regained during reconstructions.
Additional sources of information on national catches are international organizations such as FAO, ICES or SPC (Secretariat of the Pacific Community), or a Regional Fisheries Management Organization (RFMO) such as NAFO (Northwest Atlantic Fisheries Organization), or CCAMLR54 , or current or past regional fisheries development and/or management projects (many of them launched and supported by FAO), such as the Bay of Bengal Large Marine Ecosystem project (BOBLME). All these organizations and projects issue reports and publications describing—sometimes in considerable details—the fisheries of their member countries. Another source of information is the academic literature, now widely accessible through Google Scholar.
A good source of information for the earlier decades (especially the 1950s and 1960s) for countries that were part of former colonial empires (especially British or French) are the colonial archives in London (British Colonial Office) and the ‘Archives Nationales d'Outre-Mer', in Aix-en-Provence, and the publications of ORSTOM (Office de la recherche scientifique et technique d'outre-mer), for former French colonies. A further source of information and data are non-fisheries sources, including household and/or nutritional surveys, which are occasionally used for estimating unreported subsistence catches. Our global network of local collaborators is also crucial in this respect, as they have access to key data sets, publications and local knowledge not available elsewhere, often in languages other than English.
Supplementary Figure 2 shows a plot of the publications used for slightly over 110 reconstructions against their date of publication. Although, recent publications predominate, older publications firmly anchor the 1950s catch estimates of many reconstructions. On average, around 35 unique publications were used per reconstruction (not counting online sources and personal communications).
Potential language bias is taken seriously in the Sea Around Us, to ensure that data are collated in languages other than English. Besides team members who read Chinese, others speak Arabic, Danish, Filipino/Tagalog, French, German, Hindi, Japanese, Portuguese, Russian, Spanish, Swedish and Turkish. To deal with other languages, research assistants are hired who speak, for example, Korean or Malay/Indonesian. We also rely on our multilingual network of colleagues and friends throughout the world, for example, for Greek or Thai. While it is true that English has now become the undisputed language of science55 , other languages are used by billions of people, and assembling knowledge about the fisheries of the world is not possible without the capacity to explore the literature in languages other than English.
(4) Development of data ‘anchor points' in time for each missing data item, and expansion of anchor point data to country-wide catch estimates. ‘Anchor' points are catch estimates usually pertaining to a single year and sector, and often to an area not exactly matching the limits of the EEZ or IFA in question. Thus, an anchor point pertaining to a fraction of the coastline of a given country may need to be expanded to the country as a whole. For expansion, we use fisher or population density, or relative IFA or shelf area as raising factor, as appropriate given the local condition. In all cases, we consider that case studies underlying or providing the anchor point data may had a case-selection bias (for example, representing an exceptionally good area or community for study, compared with other areas in the same country), and thus use raising factors very conservatively.
(5) Interpolation for time periods between data anchor points, either linearly or assumption based for commercial fisheries, and generally via per capita (or per fisher) catch rates for non-commercial sectors. Fisheries are often difficult to govern, as they are social activities involving multiple actors. In particular, fishing effort is often difficult to reduce, at least in the short term. Thus, if anchor points are available for years separated by multi-year intervals, it usually will be more reasonable to assume that the underlying fishing activity continues in the intervening years with no data. We tread this ‘continuity' assumption as a default proposition. Exceptions to such continuity assumptions are major environmental impacts such a hurricanes or tsunamis56 , or major socio-political disturbances, such as military conflicts or civil wars57 , which we explicitly consider with regards to the use of raising factors and the structure of time series estimates. In such cases, our reconstructions mark the event through a temporary change (for example, decline) in the catch time series, which is documented in the text of each catch reconstruction. At the very least, this provides pointers for future research on the relationship between fishery catches and natural catastrophes or conflicts. We note that the absence of such signals (such as a reduction in catch for a year or two) in the officially reported catch statistics for countries having experienced a major natural or socio-political disturbance can be a sign that their official catch data may not accurately reflect what occurs on the ground. This contributes to the emergence of ‘poor numbers'40 . Overall, our reconstructions assume—when no information to the contrary is available—that commercial catches (that is, industrial and artisanal) can be linearly interpolated between anchor points, while non-commercial catches (that is, subsistence and recreational) can generally be interpolated between anchor points using non-linear trends in human population numbers or number of fishers over time (via per capita rates).
Radical and rapid effort reductions as a result of an intentional policy decision and implementation do not occur widely. One example we are aware of is the trawl ban of 1980 in Western Indonesia58 . The ban had little or no impact on official Indonesian fisheries statistics for Western Indonesia, another indication that these statistics may have little to do with the realities on the ground. FAO hints at this being widespread in the Western Central Pacific and the Eastern Indian Ocean (the only FAO areas where reported catches appear to be increasing) when they note that ‘while some countries (i.e., the Russian Federation, India and Malaysia) have reported decreases in some years, marine catches submitted to FAO by Myanmar, Vietnam, Indonesia and China show continuous growth, i.e., in some cases resulting in an astonishing decadal increase (e.g., Myanmar up 121 percent, and Vietnam up 47 percent)'.42
(6) Estimation of total catch times series. A reconstruction is completed when the estimated catch time series derived through steps 2–5 are combined and harmonized with the reported catch of step 1. Generally, this results in an increase of the overall catch, but several cases exist where the reconstructed total catch is lower than the reported catch. The best documented case of this is that of mainland China14 (link), whose over-reported catches for local waters in the Northwest Pacific are compensated for by under-reported catches taken by Chinese distant water fleets fishing elsewhere. In the 2000s, Chinese distant water fleets operated in the EEZs of over 90 countries, that is, in most parts of the world's oceans5 . Harmonizing reconstructed catches with the reported baselines goes hand-in-hand with documenting the entire reconstruction procedure. Thus, every reconstruction is documented and published, either in the peer-reviewed scientific literature, or as detailed technical reports in the publicly accessible and indexed Fisheries Centre Research Reports series or the Fisheries Centre Working Paper series, or other regional organization reports (Supplementary Table 5).
Several reconstructions were conducted in the mid- to late 2000s, when official reported data (that is, FAO statistics or national data) were not available to 2010 (refs 15 , 59 ). All these cases are updated to 2010, in line with each country's individual reconstruction approach to estimating missing catch data. Thus, all reconstructions are brought to 2010 to ensure identical time coverage (Supplementary Table 5).
Since these six points were originally proposed, a seventh point has come to the fore that cannot be ignored10 :
(7) Quantifying the uncertainty associated with each reconstruction. In fisheries research, catch data are rarely associated with a measure of uncertainty, at least not in the form resembling confidence intervals. This may reflect the fact that the issue with catch data is not a lack of precision (that is, whether we could expect to produce similar results upon re-estimation), but about accuracy, that is, attempting to eliminate a systematic bias, a type of error which statistical theory does not really address.
We deal with this issue through a procedure related to ‘pedigrees'60 and the approach used by the Intergovernmental Panel on Climate Change to quantify the uncertainty in its assessments61 . The authors of the reconstructions are asked to attribute a ‘score' expressing their evaluation of the quality of the time series data to each fisheries sector (industrial, artisanal and so on) for each of the three time periods (1950–1969, 1970–1989 and 1990–2010). These ‘scores' are (1) ‘very low', (2) ‘low', (3) ‘high' and (4) ‘very high' (Table 1). There is a deliberate absence of an uninformative ‘medium' score, to avoid the effective ‘non-choice' that this option would represent. Each of these scores is assigned a percentage uncertainty range (Table 1). Thereafter, the overall mean weighted percentage uncertainty (over all countries and sectors) was computed (Fig. 1).

To link environmental impacts to the 332 commodity foods in FCID, we followed a four-step
process (see table 1). First, we used data from
original research on specific foods inventoried in the literature review, as described above.
The mean, standard deviation, minimum and maximum values for CED and GHGE at farm gate and at
processor gate were calculated for each specific food, and then matched to the FCID. Studies of
heated greenhouse vegetable production or those of beef from dairy herds were not included in
our averages because information on market share of these production methods is unavailable or
unreliable. Second, if we did not have an original research report on an FCID food, we turned
to reports with previously-compiled food LCA data to supply environmental impacts [23 –29 (link)]. These resources contained data not captured in the literature review, perhaps due
to non-English language reports or proprietary sources. Overall, for stage 1 and 2 of the
linkage process, CED matches were made for 35% of the food commodities, and GHGE matches for
47%. Third, remaining FCID foods were populated with values from similar foods as proxies.
Specifically, we took an average of either CED or GHGE values from existing entries within a
specific food grouping (e.g. berries, brassicas, brassica greens, citrus, fresh herbs, grains,
other greens, nuts, roots, dried spices, other tree fruit, tropical fruit) to proxy for a
specific food item in that same grouping that was lacking data. Failing this approach, other
proxies of foods with similar form were then assigned. These assignments were based on
similarities of specific crops in their botany and, most importantly, production methods, as
determined by the expertise of our research team. Values that were assigned from other foods in
the database in this third stage accounted for 50% of CED values and 39% of GHGE values.
Fourth, the FCID dataset includes minimally processed forms of fruits and vegetables (e.g.
strawberry juice, dried apples). Where direct LCA matches were not available for these forms,
we applied a mass conversion factor, gathered from nutritional databases [30 , 31 ], to the
base fruit or vegetable in order to approximate the agricultural production burdens of these
processed forms. This stage accounted for the remaining 15% of CED and 15% of GHGE values for
FCID foods. For juices, vinegar and maple syrup, additional sources were used to develop valid
estimates. These additions are detailed in supporting information.
Because of the inconsistency in full life cycle boundary conditions across the literature
review entries, cradle-to-farm gate impact factors were chosen for the vast majority of foods.
This choice is further supported by the fact that these commodity foods, in many cases, become
ingredients in processed, as-consumed foods, and inclusion of life cycle stages downstream from
the farm gate would not necessarily reflect impacts of the actual foods consumed. The
exceptions to this farm gate boundary condition are foods within the FCID listing that require
processing: flours, refined sugars, vegetable oils, etc supporting information contains
additional details on these boundary condition choices, as well as an environmentally extended
input-output based estimate of the cumulative food processing impacts excluded in this
analysis.

ASV tables were classified according to climatic zones (tropical, subtropical, and subalpine). ASV tables were used to determine alpha diversity and community composition of host plants, fungi (including total fungi, ectomycorrhiza, arbuscular mycorrhiza and pathogen) and bacteria (including total bacteria, nitrifier, nitrogen-fixing bacteria and pathogen). Species richness (SR), Shannon–Wiener diversity index (Shannon), and phylogenetic diversity (PD) were calculated to measure the alpha diversity of plant and soil microorganisms. SR was calculated as the total number of species in a plot (Zhang et al., 2021 (link)). Shannon-wiener index is based on relative abundance data, which is affected by both richness and evenness (Shannon, 1948 (link)). PD estimates phylogenetic alpha diversity (Faith, 1992 (link)), defined as the sum of branch lengths (from the terminal to the base of the phylogeny) of all species in a plot (Zhang et al., 2021 (link)). These diversity indexes were calculated using the picante package (Kembel et al., 2010 (link)) and vegan package (Oksanen et al., 2013 ) in R software.
The relationship between alpha diversity and elevation in the three climatic zones was evaluated by linear regression. We compared the Akaike information criterion (AIC) values of simple and multinomial linear regressions and selected models with smaller AIC values for visualization (Supplementary Table S3 in Supplementary material). The models were visualized using the ggplot2 package in R (Wickham, 2016 ).
The association between alpha diversity and environmental variables (elevation, OM, TC, TN, TK, AK, TP, HN, humidity, temperature, and soil water content) was assessed by Pearson correlation analysis. The importance of environmental variables to diversity was evaluated by random forest. The random forest method accommodates collinear predictors by distributing the relevance of a variable across all variables (Yang et al., 2021 (link)). The explanatory power of environmental variables for three alpha diversity metrics was estimated using the linkET package (Huang, 2021 ) and randomForest package (Liaw and Wiener, 2007 ) in R software.
Plant and microbial community compositions were ordinated using nonmetric multidimensional scaling (NMDS) with Bray–Curtis dissimilarity matrices using the metaMDS function in the Vegan package (Oksanen et al., 2013 ). The association of species community composition with environmental factors was evaluated by distance-based redundancy analysis (dbRDA). In this analysis, we performed Hellinger transformation on microbial ASV tables. Diversity indices were correlated with environmental variables using the rdacca.hp function in R (Lai et al., 2022 (link)). Canonical analyzes (RDA, canonical correspondence analysis, and dbRDA) are the best multivariate statistical approaches for investigating explanatory factors for the matrix of response variables. The overall explanatory power of environmental variables can be calculated using R packages; however, it is challenging to accurately calculate the explanatory power of individual variables because of covariance across variables. The rdacca.hp function reduces collinearity among environmental factors.
The impact of environmental factors on microbial diversity was assessed using pathway analysis via a piecewise structural equation model approach (Grace et al., 2012 (link)). First, we performed principal component analysis of environmental factors (climatic and soil properties), plant diversity, and microbial diversity to extract data on the first axis (PC1, explained variance >70%). Then, two models (A and B) were fitted to the data to examine whether the effects of environmental factors on microbial diversity were direct (model A, in which environmental factors directly affected microbial and plant diversity) or indirect (model B, in which environmental factors directly affected plant diversity, which in turn impacted microbial diversity). These analyzes were performed using the FactoMineR package (Lê et al., 2008 (link)) and the piecewiseSEM package (Lefcheck, 2015 (link)) in R.
All statistical analyzes were conducted in R version 4.1.2 (R Core Team, 2021 ).

Bacterial OTUs with relative abundances greater than 0.1% were defined as abundant taxa; those with relative abundances below 0.01% were defined as rare taxa, and those with relative abundances between 0.01 and 0.1% were moderate OTUs (Jiao and Lu, 2020 (link); Yang L. et al., 2022 (link)). After all data passed the preliminary Shapiro–Wilk test (p > 0.05), a t-test was used to assess the differences between the parameters (Hou et al., 2019 (link)). Partial least squares discriminant analysis (PLS-DA) with variable importance (VIP) values was used to identify the difference in soil metabolites (Darnaud et al., 2021 (link)). The PLS-DA model was validated with a permutation test (n = 200), R² data are larger than Q² data, and the intercept of Y and Q² was less than 0 (R² intercept = 0.9576, Q² intercept = −0.3811), which means the model was fit and not overfitting (Chen et al., 2021 (link)). The alpha diversity indices including Chao1 and inverse Simpson index (or InvSimpson for short) were calculated by the R program package “vegan” (Oksanen et al., 2007 ). Nonmetric multidimensional scaling (NMDS) based on Bray–Curtis distance was conducted by the “vegan” package.
The network analysis was inferred using the SparCC-based algorithm Fastspar with a bootstrap procedure repeated 100 times, and only strong (r > 0.6) and significant (p < 0.01) correlations between OTUs were retained (Watts et al., 2019 (link)). The network nodes with high closeness (or betweenness) centrality value were identified as keystone hubs in the network: the network hubs were highly connected, both in general and within a module; the module hubs were highly connected within a module, the connectors provided links among multiple modules, and the peripherals had few links to other species (Poudel et al., 2016 (link)). Variation partitioning analysis (VPA) was used to assess the impact of environmental factors on bacterial communities using the “vegan” package (Wang et al., 2019 (link)). Before the performance of VPA, we assessed the collinearity of the variables by calculating the variance inflation factor (VIF). The factors were included in the VPA analyses, only when the collinearity VIF < 10 (Liao et al., 2019 (link)). The predictors of bacterial richness were identified via random forest modeling using the “randomForest” package in R (Liaw and Wiener, 2002 ). The importance of each predictor was determined by the increase in the mean square error (InMSE) and was averaged over 5,000 trees (Xu et al., 2018 (link)).