Abstract

A recurring pattern of declining mean trophic level of fisheries landings, termed “fishing down the food web,” is thought to be indicative of the serial replacement of high-trophic-level fisheries with less valuable, low-trophic-level fisheries as the former become depleted to economic extinction. An alternative to this view, that declining mean trophic levels indicate the serial addition of low-trophic-level fisheries (“fishing through the food web”), may be equally severe because it ultimately leads to conflicting demands for ecosystem services. By analyzing trends in fishery landings in 48 large marine ecosystems worldwide, we find that fishing down the food web was pervasive (present in 30 ecosystems) but that the sequential addition mechanism was by far the most common one underlying declines in the mean trophic level of landings. Specifically, only 9 ecosystems showed declining catches of upper-trophic-level species, compared with 21 ecosystems that exhibited either no significant change (n = 6) or significant increases (n = 15) in upper-trophic-level catches when fishing down the food web was occurring. Only in the North Atlantic were ecosystems regularly subjected to sequential collapse and replacement of fisheries. We suggest that efforts to promote sustainable use of marine resources will benefit from a fuller consideration of all processes giving rise to fishing down the food web.

Results

We deemed declines in mean trophic level >0.15 to be evidence of ecologically significant fishing down the food web. From an energetic perspective, this decline in trophic level represents an ≈50% decrease in the primary production required to sustain a given amount of catch (15). Of the 48 large marine ecosystems (LMEs) worldwide that had suitable data, 30 showed evidence of substantial fishing down the food web, with an average decline of 0.42 trophic level. These declines were substantially larger than those originally documented in the analysis of more spatially aggregated data (6), consistent with the notion that original estimates from poorer-quality data generally underestimated the magnitude and frequency of fishing down the food web (16).

Visual exploration of catch data revealed evidence for both mechanisms underlying fishing down the food web (Fig. 1). The Scotian Shelf provides a typical example of the sequential collapse/replacement mode (Fig. 1A). The mean trophic level in fisheries landings declined markedly beginning in 1987, corresponding with the initial collapse of groundfish stocks. This collapse was succeeded by a decline in herring (Clupea harengus) landings, ultimately leading to increased exploitation of northern prawn (Pandalus borealis). In contrast, the Patagonian Shelf exhibited a similar decline in mean trophic level between 1980 and 2001, but landings of high-trophic-level species (namely, Argentine hake,Merluccius hubbsi) generally increased over this period (Fig. 1B). Most of the decline in trophic level in this ecosystem was due to the addition of a new fishery targeting the shortfin squid (Ilex argentinus).

By fitting statistical models to time series of catches of high-trophic-level species we derived estimates of the mean annual rate of change of high-trophic-level catches (β; yr⁻¹) for each of the 30 ecosystems where fishing down the food web occurred. High-trophic-level catches were calculated in two ways. Apex predators were defined as species with trophic level >4. Upper-trophic-level species were defined within each LME as those species having trophic levels greater than the mean trophic level at the onset of fishing down the food web. We report here only those estimates calculated from apex predator catches, but the results derived from analysis of all upper trophic levels were similar (Table 2, which is published as supporting information on the PNAS web site). Maximum likelihood estimates of β ranged from −12.8% yr⁻¹ to 7.0% yr⁻¹, with the two lowest estimates associated with the collapse of Atlantic cod (Gadus morhua) stocks in the Scotian Shelf and the Newfoundland–Labrador Shelf. Only nine instances of fishing down the food web were associated with a statistically significant decline in apex predator catches (P < 0.05) (Fig. 2), and only four of these estimates were less than −6.7% yr⁻¹(a rate equivalent to a 50% decline in catches over a 10-year period). In contrast, 6 ecosystems showed no statistically significant change (P > 0.05), and 15 showed significant increases in apex predator catches during the period when fishing down the food web was occurring (Fig. 2). Considering catches in all ecosystems worldwide together, the mean trophic level was stable from 1950 to 1956, declined from 3.44 to 3.16 between 1956 and 1986, and has remained stable since 1986. During the time period that the mean trophic level was declining, apex predator catches increased by 1.8% yr⁻¹ (P < 0.001;Fig. 2). Analyses of lower-trophic-level catches in these ecosystems confirm that declining mean trophic level was associated with rapid increases in lower-trophic-level catches, except for instances in which high-trophic-level catches declined (Table 2). Thus, the data identify the sequential addition mode as the most common process underlying fishing down the food web, representing more than two-thirds of all cases.

Because a close examination of the estimates presented inFig. 2 revealed similarities among ecosystems within similar biogeographic regions, we explored whether there were differences in the mode of fishing down the food web among different ocean regions. Six oceanographic regions had multiple ecosystems exhibiting fishing down the food web: North Pacific, Tropical Pacific, Indian Ocean, North Atlantic, Tropical Atlantic, and South Atlantic. The metaanalysis of β across ecosystems within ocean regions supported the notion that there were region-scale patterns of variation in β. For the apex predator catches, only the North Atlantic region had a mean β that was negative (P < 0.05; mean = −4.5% yr⁻¹;Fig. 3). For all other regions, the mean β was positive and was statistically greater than 0 for the Tropical Pacific, Indian Ocean, and the South Atlantic regions (P < 0.05;Fig. 3). Contrasts among regions were virtually identical when trends in all upper-trophic-level fisheries were examined: again, only the North Atlantic region had a mean β that was <0; all others had a mean β that was >0, and these estimates were statistically significant for the Indian Ocean and the South Atlantic region. These results suggest that the North Atlantic is a hotspot for the sequential collapse/replacement mode of fishing down the food web, whereas the sequential addition mode generally describes the process underlying changes in mean trophic level throughout the rest of the world.

Discussion

Our results indicated that fishing down the food web is prevalent among marine ecosystems worldwide and that fishing down the food web was most commonly associated with the sequential addition of new fisheries. In contrast, the sequential collapse/replacement mode of fishing down the food web was common in North Atlantic ecosystems but rare elsewhere. The observed frequencies of these alternative mechanisms may not be accurately reflected in the scientific community’s interpretation of fishing down the food web. We analyzed >200 peer-reviewed publications and did not find a single article claiming that fishing down the food web was associated with the sequential addition of new fisheries (Table 1). Instead, our review revealed a scientific community that has embraced the view that fishing down the food web is evidence of unsustainable fishing and human alteration of food web structure and “… is clear evidence of ineffective management” (17). This disconnect between perception and reality, which places undue emphasis on the less common sequential collapse/replacement mechanism, is dangerous because it leads us to ignore the policy implications of the more common sequential addition mechanism.

That high-trophic-level fisheries were maintained (or even grew) during most instances of fishing down the food web does not imply that fish stocks are healthy. We emphasize that our results should not be used to make inferences about stock status for at least two reasons. One, we looked at aggregate catches of species that fell within particular trophic levels. Consequently, our analysis cannot address individual stock status or shifts in community structure that might maintain total apex predator productivity despite depletion of individual species (18). Second, increased catches are likely indicative of increased exploitation rates, which can only act to further reduce stock sizes. Moreover, there is little doubt that many predator stocks are overfished and that fisheries preferentially target large, high-trophic-level species (refs.19 and20; but see refs.21–26 for an in-depth discussion on the status of shark and tuna species). Our analyses simply indicate that fishing down the food web was generally not accompanied by declining catches of high-trophic-level species, suggesting that fishing down the food web was not associated with a depletion of these species sufficient to make them economically extinct, i.e., so depleted that the effort required to capture the remaining fish is more costly than the expected profit. We therefore conclude that, in most instances, the standing stock of upper-trophic-level species was sufficiently abundant to support fisheries during the time periods when fishing down the food web occurred. This observation, coupled with our observations that fishing down the food web was accompanied by increased catches of low-trophic-level species, indicates that fishing down the food web is symptomatic of increased direct (harvest) and indirect (harvest of prey) impacts of fisheries on high-trophic-level species. These multiple impacts may be sustainable during the initial phases of fisheries development but can ultimately lead to collapse of high-trophic-level stocks if fisheries develop unchecked and without consideration of these interactions.

The sequential collapse/replacement mode of fishing down the food web was most common and most extreme among ecosystems within the North Atlantic region. The poor status of North Atlantic fisheries and ecosystems has been well documented (27) as groundfish stocks throughout the region have suffered from the combined effects of overcapitalization and climate-induced declines in stock productivity (28). The consistent and large declines in upper-trophic-level landings suggest that there has indeed been a region-wide depletion of upper-trophic-level species so severe as to make directed fisheries for them unprofitable. Yet our results also suggest that the North Atlantic region is an anomaly in this respect, because upper-trophic-level catches generally increased throughout the rest of the world ocean regions.

As with any attempt to explore global patterns and to derive generalizations, several caveats are required. The first issue is the precision and information content in catch data. Fisheries catch data are generally a poor indicator of stock status because catches are dictated by the abundance of targeted species, their availability to the fishing fleet, the capacity of the fleet, and the efficiency of the fleet at capturing them. These all change as fisheries develop, producing ambiguous and sometimes misleading trends in total landings. For example, high catches of Atlantic cod were maintained for several years in the Northwest Atlantic despite the pending collapse in the early 1990s (29). Furthermore, the catch data forming the basis of our analysis originally derived from information reported to the Food and Agriculture Organization (FAO) by individual countries, and these reports may not always be accurate (30). Notably, many of the catch estimates in the FAO database reported by China may be incorrect, although the catch estimates we used attempted to remove this bias through an explicit adjustment of Chinese catches (31). Finally, the partitioning of FAO region-wide catch data into individual LMEs has been conducted with great care (31), but there are limits to the precision of any method that seeks to disaggregate data of this sort. Despite these limitations, which make catch data relatively imprecise, our main conclusion still stands: the best available evidence suggests that fishing down the food web is most commonly caused by the addition of new fisheries, not by the serial replacement of fisheries.

Another possible inaccuracy in our method lies in the fact that we used a single trophic level to describe an entire species. Recent work has indicated that size-selective fishing reduces the average trophic level of fish stocks by removing the largest individuals (32), because trophic level is generally positively correlated with body size (33). However, this process is unlikely to have a large effect on our main conclusions. Most of the cases of fishing down the food web were due to the serial addition of new fisheries with large differences in trophic level (commonly >1.0 trophic level). These contrasts in trophic level greatly exceed the more subtle declines caused by shifts in species’ size structures.

We restricted our analyses to marine ecosystems because there exists a large database of fishes catches that permitted this analysis. For this reason, we cannot generalize our conclusions to freshwater ecosystems. Fishing down the food web has been documented in inland freshwater waters (34), and many freshwater ecosystems are severely overfished because of intense recreational fishing pressure (35). Moreover, reduced body sizes resulting from size-selective fishing is a continuing problem facing freshwater fisheries managers. We therefore speculate that the sequential collapse/replacement model may be more prevalent in freshwater ecosystems, but further data analyses are needed to evaluate this claim.

Achieving sustainable use of marine fisheries and ecosystems will not be easy, but it will be enhanced by a better recognition of the scope of the problems facing us. Our analyses indicate that the sequential addition mode is by far the most common explanation for fishing down the food web. Perhaps the most important policy consideration of the sequential addition mode is that, in most ecosystems of the world, several trophic levels are now exploited simultaneously. These diverse fisheries impose conflicting demands on marine ecosystems that are not generally well represented in single-species management plans that do not consider the effects of these alternative fisheries on each other. As the structure of fisheries and the management environment evolve, the scientific community faces a new challenge of conducting broad-scale ecological research to support the development of more holistic, ecologically based approaches to fisheries management.

Materials and Methods

Analysis.

We identified fishing down the food web as any instance in which mean trophic level exhibited a decline by at least 0.15. The time periods associated with each fishing down the food web event were defined as the year of the highest (start) and lowest (end) observed trophic level. This procedure gave us an estimate of the largest possible decline in mean trophic level of the catch.

To provide a quantitative approach characterizing modes of fishing down the food web, we partitioned the total catches for each ecosystem into upper and lower trophic levels and described the trends in upper-trophic-level catches while fishing down the food web was occurring. This partitioning involved identifying a threshold trophic level and then summing catches for all landings exceeding this threshold. Because any partition of a continuous variable (i.e., trophic level) has the potential to impose artifacts, we used two separate methods to identify the threshold value. The first method used the mean trophic level at the onset of the decline in each ecosystem. We refer to catches calculated in this manner as upper-trophic-level catches. The second method examined the dynamics of apex predators only, which are those species with a trophic level ≥4.0. This threshold value is based on recent interest in the status of large predator species worldwide (3,7).

We fit an exponential model to the time series of upper-trophic-level and apex predator catches during the time periods when fishing down the food web occurred. This model was of the formC(t) =C(0)exp(βt), whereC(t) is the catch rate during yeart,C(0) is the catch rate for the initial year, andt is the number of years since the mean trophic level initiated its decline. Estimates of β were made by using robust linear regression of log(C(t)) vs.t (39).

We estimated region-specific estimates of β by using a random-effects metaanalysis that combined estimates of β for all ecosystems in the Indian Ocean, North Pacific, Tropical Pacific, North Atlantic, South Atlantic, and tropical Atlantic. In all instances, tropical regions refer to ecosystems that reside predominantly within 20° N and 20° S (Caribbean Sea, Guinea Current, Sulu-Celbes Sea, South China Sea, and Pacific Central America Coast), and North and South designate ecosystems that reside outside of this region. The random-effects model is described in detail in ref.40, but we present a synthesis of these methods here. This method presumes that β is distributed among ecosystems within a region according to a normal distribution with a mean β̄ and variance σ_β². The metaanalysis can therefore be viewed as an attempt to estimate the central tendency of β within a region.

The total variance of β for each ecosystem (v_i*) reflects both the variance associated with measurement uncertainty (i.e., the square of the standard error =v_i) and the variance of each ecosystem β around the population mean β̄ (σ_β²).

The population variance of β is calculated from thek independent measurements of β and their associated estimation variances (v_i).

β̄ is calculated as a weighted average of the sample estimates:

where the variance of this estimate equals

Statistical significance is based on the ratio of β̄/var(β̄)^0.5, which, under the central limit theorem and null hypothesis of no change in catches, is normally distributed with a mean of 0 and a standard deviation of 1. Significance is therefore judged from critical values of theZ statistic (1.96 for α = 0.05).

Supplementary Material

Acknowledgments

We thank the members of the Sea Around Us project (particularly Reg Watson, Daniel Pauly, and Villy Christensen) for providing us with their catch estimates and for helpful discussion. We also thank Jim Kitchell, Villy Christensen, Daniel Pauly, Carl Walters, Steve Martell, Marc Mangel, Phil Levin, Daniel Schindler, Bob Francis, Steve Munch, and two anonymous reviewers for helpful comments on earlier drafts of this work. This work was supported by National Science Foundation Biological Oceanography Program Award 0220941.

Abbreviation

LME

large marine ecosystem.

Footnotes

References

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.