Comparing the Benguela and Humboldt marine upwelling ecosystems with indicators derived from inter-calibrated models

Abstract

Large-scale, mass-balance trophic models have been developed for northern and southern regions of both the Benguela and Humboldt upwelling ecosystems. Four of these Ecopath models were compared and calibrated against one another. A common model structure was established, and a common basis was used to derive poorly known parameter values. The four resulting models represent ecosystems in which the main commercial fish species have been moderately to heavily fished: central-southern Chile (1992), northern-central Peru (1973–1981), South Africa (1980–1989), and Namibia (1995–2000). Quantitative ecosystem indicators derived from these models were compared. Indicators based on large flows (involving low trophic levels) or top predators were not well estimated, because of aggregation problems. Many of the indicators could be contrasted on the basis of differences between the Benguela and Humboldt systems, rather than on the basis of fishing impact. These include integrated values relating to total catches, and trophic levels of key species groups. Indicators based on integrated biomass, total production, and total consumption tended to capture differences between the model for Namibia (where fish populations were severely reduced) and the other models. We conclude that a suite of indicators is required to represent ecosystem state, and that interpretation requires relatively detailed understanding of the different ecosystems.

Introduction

Coastal upwelling ecosystems are characterized by large stocks and productive fisheries (FAO, 1995) for small pelagic species that fluctuate on annual and decadal time scales. The underlying basis for their large productivity is wind-driven coastal upwelling (Strub et al., 1998), but estimates of primary production differ considerably among ecosystems (Carr, 2002), and environmental variability is large. Intensive exploitation affects the total standing stocks of some species, either directly or indirectly. Quantitative ecosystem indicators are needed to help establish whether differences among systems can be attributed to fishing. Previous comparative work used Ecopath models to compare four marine upwelling systems (northern-central Peru, Namibia, northwest Africa, California). The main results of those analyses indicated that the models clustered together by system rather than by species dominance (Jarre-Teichmann, 1998; Jarre-Teichmann and Christensen, 1998; Jarre-Teichmann et al., 1998). This implies that quantitative indicators for fisheries management that aim to track fishing impacts might be confounded by inter-system differences.

The two main coastal upwelling systems in the southern hemisphere (Figure 1) are located off the west coasts of South America (the Humboldt) and Africa (the Benguela). Ecopath models are available for central-southern Chile (Neira and Arancibia, 2004), northern-central Peru (Jarre et al., 1991; Jarre-Teichmann et al., 1998), the southern Benguela (Shannon et al., 2003), and the northern Benguela (Heymans et al., 2004; Roux and Shannon, 2004). However, a rigorous comparison of these models has not been possible, because of extant but questionable differences in input parameters and trophic structures. First, we standardize the existing models to reflect a common structure and consistent parameterization, then use a comparative approach to determine which indicators provide consistent information about changes in ecosystem state as a consequence of fishing.

Methods

Existing models

Neira and Arancibia (2004) modelled the central-southern Chilean ecosystem (33–39°S; extending up to 30 nautical miles offshore and covering 50 000 km²) for a single year (1992). That year represents a period when the ecosystem was dominated by medium-sized pelagic fish (horse mackerel and hoki; for Latin names of species see Table 1), with medium to high abundance of small-sized pelagic fish (common sardine and anchovy; Neira et al., 2004). The ecosystem was affected by a moderate ENSO event, but the main fish stocks were regarded as healthy (i.e. not yet fully exploited or overexploited). The area covers the main fishing ground of both the purse-seine and the industrial trawling fleets.

The model for the northern-central Peruvian ecosystem (4–14°S; extending 60 nautical miles offshore, approximately 82 000 km²) represents the period 1973–1981 (Jarre-Teichmann et al., 1998). That period follows the collapse of the huge anchoveta fishery of the late 1960s, while sardine biomass was increasing. The period did not include major El Niño events. The area modelled covers the main distribution area of small pelagic fish over the Peruvian shelf only, and excludes part of the biomass of horse mackerel, mackerel, and mesopelagic fish, all of which are distributed farther offshore.

The model of the southern Benguela ecosystem (28°E–29°S; extending to the 500 m isobath and covering 220 000 km²; Shannon et al., 2003) covered the 1980s, a period when anchovy was more abundant than sardine. During that decade, many directed ecological studies of the pelagic ecosystem were conducted and data quality was good. The stocks of round herring, horse mackerel, and Cape hake were believed to be healthy, whereas the pelagic resources were probably fully exploited (Cochrane and Payne, 1998). The area spans the upwelling region along the South African west coast, as well as the shallow Agulhas Bank. The reason for including the latter in the model is that most stocks migrate through the entire area.

The northern Benguela ecosystem has experienced different phases since the beginning of industrial fishing in the mid-20th century. Originally, sardine dominated the pelagic component, but after its depletion in the late 1960s and early 1970s, the system became dominated by pelagic gobies, jellyfish, and horse mackerel. The model (15–29°S; extending between 20 m and 600 m deep; approximately 186 500 km²; Roux and Shannon, 2004) describes the state of the ecosystem during the period 1995–2000, which followed environmental anomalies that further depleted the commercial pelagic species. The limits cover the distribution areas of the main pelagic and demersal species in the region, but exclude littoral communities.

Common model structure

The original Ecopath models represented 21–31 groups, mainly because of varying degrees of aggregation of planktonic and demersal components. A common model structure was derived (Table 1), and the parameter values for each model under this common structure were compared, with the objective of separating biological information from modelling artefacts. Details of how the original parameters and input variables were estimated are given in the original papers describing the models. Here, consistent input parameter values were assigned by consensus among the authors if no biological reason appeared to exist for choosing different values (Table 1). One important change to model structure concerned microzooplankton. This group was only present in the Benguela models, so was added to the Humboldt models, copying the input parameters and diet compositions. The changes agreed affected all four models, but there were no major difficulties in rebalancing the models with the new set of parameter values. The inter-calibration exercise resulted in four adjusted models having consistent structures based on 27 model groups, though differing in terms of representation of some single-species groups, in the relative sizes of the groups, and in the flows among groups (diet matrices; Moloney and Jarre, 2003).

Indicators

The balanced Ecopath models were used to calculate ecosystem indicators. Some of these are directly available within the Ecopath software, whereas others involve comparisons of integrated flows and ratios, and have been used previously for ecosystem comparisons (Jarre et al., 1991; Jarre-Teichmann et al., 1998; Shannon and Jarre-Teichmann, 1999; Shannon et al., 2003). The indicators presented here include integrated biomass values and flows, catches, and ratios among these. Trophic levels were also examined, and trophic spectra (Bozec et al., 2005; Gascuel et al., 2005) for biomass and catch were calculated for each model by smoothing plots of consumer biomass and catches along fractional values (i.e. increments of 0.1) of trophic levels calculated by Ecopath.

To quantify the impacts of fishing, 12 indicators (Table 2) were selected on the basis of recommendations by Rochet and Trenkel (2003) and Cury et al. (2005). The first indicator was the total biomass of consumers, excluding plankton and macrobenthos, which was expected to decrease with an increase in fishing. Two indicators were based on biomass ratios; the ratio between planktivores and piscivores was expected to increase with fishing, whereas that between piscivores and total consumers was expected to decrease. The flows from trophic level I that are required to sustain the fishery are expected to decrease with fishing, because of the fishing-down-the-foodweb syndrome once high-value, piscivorous species have been removed (Pauly et al., 1998). This effect should also be reflected in increased catch ratios of small pelagics to (large hake + large pelagics), and a reduced trophic level of the total catch, and of the community. With fishing, the proportion of total mortality (Z) that is caused by fishing (F) compared with predation would be expected to increase, and this was examined through F/Z ratios for large hake 1 (Merluccius spp. >24 cm total length TL) and anchovy. Fishing is also expected to remove old, slow-growing fish from the population, so P/B ratios by species should increase (also tested for large hake 1 and anchovy).

On the basis of historical fishing patterns and the assumed states of the model ecosystems for the periods considered, the models were ranked by impact of fishing, from moderate to high: 1. South Africa; 2. Chile; 3. Peru; 4. Namibia. Two hypotheses were tested: (i) that model scores for indicators quantifying fishing impacts match exactly those assumed for the ecosystems; (ii) that differences caused by fishing impacts are overshadowed by differences between the Benguela and Humboldt ecosystems. However, differences within the two systems are such that indicator scores for South Africa should always be smaller than for Namibia, and Chile should have smaller scores than Peru.

Results and discussion

Similar and unique model groups

There are similar species groups within all four ecosystems (Table 1). All four models have an anchovy group, at least one hake group, and at least one group for horse mackerel. Where species were combined into broad functional groups, some tended to be similar among the four ecosystems, e.g. the various plankton groups, mesopelagics, seabirds, pinnipeds, and cetaceans. Some of the highly aggregated functional groups, such as demersal benthic- and pelagic feeders, are found in all four ecosystems, but have very different species compositions. There are also differences in model groups among ecosystems. The Chilean ecosystem does not have a sardine (Sardinops) group, whereas sardine is (or has been) an important component of the other three. Three of the four ecosystems have their own “special” small pelagic species: common sardine in Chile, round herring in South Africa, pelagic goby in Namibia. The Peruvian ecosystem does not appear to have an equivalent species in addition to sardine and anchovy. The Humboldt and South African ecosystems have characteristic large pelagic species: hoki off Chile, mackerel off Peru, chub mackerel off South Africa. When the South African chub mackerel stock was large, the species was believed to migrate to Namibia, but this was not the case for the period being modelled here. Finally, the Benguela ecosystems have two hake species, whereas the Humboldt ecosystems only have one.

Integrated biomass values and flows

The model for Namibia has the largest value of total biomass (Figure 2a). The estimates are strongly influenced by values assigned to the plankton groups, which constitute 52–85% of the total biomass in the models. The plankton biomass estimates are subject to large errors, because of difficulties in integrating across space-time scales that differ greatly from the scales of measurement. They also contribute large amounts to the detritus group. Therefore, we excluded the plankton and detritus groups from total biomass estimates (Figure 2b), but still found that the largest value (Namibia) was more than twice the size of the next largest one (Peru). The smallest value was for the model for South Africa. The large biomass for Namibia results from the gelatinous zooplankton and macrobenthos groups. However, the trophic role of gelatinous zooplankton in that ecosystem is poorly understood at present, whereas the macrobenthos group is only large because the pelagic goby, a major component, is a benthic feeder. The biomasses represent model estimates rather than measured values, so the difference in biomass shown by the model for Namibia might be somewhat artificial, because the benthos is not a central focus of the models. Nevertheless, it does illustrate the importance of pelagic–benthic coupling in this ecosystem, which is an important difference among the four ecosystems, and it also reflects the changed state of the Namibian ecosystem.

Integrated flows indicate that the Chile model has the greatest primary production, followed by Peru, South Africa, and Namibia (Figure 2c), which sequence is not observed in any of the other flows. Peru supports the largest fish catches (Figure 2d), followed closely by Chile, with catches in the two Benguela models being much smaller. The relatively large catches in the Humboldt models appear logical when compared with primary production, because small pelagics in those systems are phytoplanktivorous (Chile; Arrizaga et al., 1993), or have at least a large fraction of phytoplankton in their diet (Peru), whereas small pelagic fish in the Benguela systems are zooplanktivorous (James, 1987; van der Lingen, 2002). However, one might expect the Chile ecosystem to require relatively less primary production than the Peru ecosystem. This raises the question why primary production is so large in the model for Chile, and what would be its fate? Only 30% of primary production in this system is used, whereas 70%, 43%, and 64% are used in the Peru, South Africa, and Namibia models, respectively (Moloney and Jarre, 2003). Because the ecotrophic efficiency of 30% for the Chile model was set to satisfy requirements of flows to detritus, while those for the others were calculated, the difference might well be a modelling artefact.

Catches

Disregarding for the moment the model for Namibia, and excluding contributions from the plankton groups, the patterns in total biomass (Figure 2b), production (Figure 2e), and consumption (Figure 2f) among the three remaining models are similar to those for total catches (Figure 2d). Note, however, the relatively small catches in the model for South Africa. Why does the southern Benguela sustain smaller catches relative to its primary production than the southern and northern Humboldt ecosystems? First, because small pelagic fish in the Benguela are zooplanktivorous, most fish species and fisheries occupy higher trophic levels than in the Humboldt (Figure 3). This means that each unit of catch would require a greater flow from trophic level I (Figure 4b), despite the fact that the Humboldt models require larger total flows (Figure 4a). Second, the catches of small pelagic fish relative to those of the important demersal species (mainly hake) in the Humboldt are larger than in the Benguela (Figure 5a); the Benguela has more “ecologically expensive” fisheries. Thus, although biomass (Figure 5b) and production (Figure 5c) ratios of small pelagics to large fish scale in a similar way to the catches for the two Humboldt models, catch ratios in the model for South Africa are relatively small. Third, management of the South African pelagic fisheries was conservative in the 1980s (Cochrane et al., 1997), probably more so than for the Humboldt ecosystems, where the fishing industry during the periods modelled was struggling to compensate for the collapse of the huge anchoveta fisheries in the 1960s (Peru), and the pelagic fishery for horse mackerel was still developing (Chile), and stocks were not yet fully exploited (Neira et al., 2004).

Trophic levels and trophic spectra

Despite differences between the Humboldt and Benguela models in the ecological roles of small pelagic fish, when the four models are analysed in terms of discrete trophic levels (Figure 6), the model for Namibia is the one that is most different, with elevated biomass densities in trophic levels II–IV. These result from the large biomass for macrobenthos (level 1.5) supporting the most important small pelagic fish (pelagic goby; level 3.2), which contributes partly to level III and partly to level IV, together with gelatinous zooplankton (level 3.23). Such results indicate that the ecosystem off Namibia is dominated by plankton feeding at a relatively high trophic level.

Gelatinous zooplankton and macrobenthos were excluded from the trophic spectra (Figure 7) because they overshadowed the other biomass values. For the remaining groups, Namibia and South Africa show similar patterns in the biomass of consumers at different positions in the foodwebs (Figure 7b). This consistent trophic signature might be a modelling artefact, because the two models are based on similar studies. However, some groups (e.g. sardine, anchovy, other small pelagics) have different trophic levels in the two models, implying that different species groups occupy different trophic positions. In contrast, the trophic spectra of the two Humboldt models are very different from one another (Figure 7a, c, e). Again, this might be a modelling artefact, or it may represent fundamental differences in ecosystem structure. Anchovy do not have the same trophic level in the two models (2.1 in Chile, 2.7 in Peru), and this discrepancy affects the spectra at higher trophic levels. Catches (Figure 7d) and catch: biomass ratios (Figure 7f) are similar in the two Benguela models.

Consumers

The greatest consumers in all four ecosystems, after excluding most zooplankton and macrobenthos, are the small pelagic fish (Figure 8). In the model for Peru, there are no mesopelagics, because the model did not incorporate the area beyond the shelf. However, it seems likely that small pelagics in this ecosystem would still be the dominant consumers, even if mesopelagic fish had been included. Mesopelagic fish are the second most important consumers off Chile and South Africa, whereas gelatinous zooplankton equals small pelagic fish in the model for Namibia. Large horse mackerel are more important in the Humboldt (Figure 8a, b) than in the Benguela models (Figure 8c, d). Chondrichthyans and demersal fish are the third most important consumer groups in South Africa, but are fifth in Peru and seventh in both Chile and Namibia. This result possibly illustrates the importance of the Agulhas Bank as a shelf habitat in the southern Benguela ecosystem. However, it may also be related to the fact that the biomass of chondrichthyans and many demersal fish species in South Africa was not well known and was therefore estimated. The biomass may have been overestimated.

When considering the relative impacts of predators of small pelagic fish, the ecosystems are quite different. The fisheries in the Humboldt models (Figure 9a, b) remove relatively large proportions of small pelagics compared with those in the Benguela models (Figure 9c, d), and pinnipeds are bigger consumers in the latter. The most important consumers are different in all models: small hake in Chile, adult horse mackerel in Peru, apex fish predators off Namibia and South Africa. The model for South Africa is noteworthy in having small pelagic fish contributing more evenly to different predator groups than in the other models, even cetaceans playing a relatively large role.

Quantifying impacts of fishing

The scores for the 12 indicators assumed to represent the impacts of fishing did not consistently produce the same trend, and none of them matched the hypothetical scores. Hypothesis 1 was rejected (χ² = 0.52, d.f. = 1, p = 0.47; Table 2). When the Benguela and Humboldt systems were considered separately, but with South Africa < Namibia, and Chile < Peru, the scores were still similar to those that would be expected by chance (χ² = 1.78, d.f. = 1, p = 0.18), suggesting that indicators cannot be used consistently and objectively in an a priori fashion to quantify fishing impacts.

Conclusions

The Ecopath models represent our “best” estimates of the components of the ecosystems and the ways in which they interact, though we recognize that some values used in the models are better estimates than others. No model can ever duplicate reality, and there may be many errors in our model representations of the four ecosystems. Similarly, there will be errors in the indicators derived from the models. Still, progress in our understanding may be achieved by constructing, examining, and comparing such models. The common model structure was useful for intersystem comparisons, because this ensured that species groups were given equivalent status, and that models were not restricted to those groups for which data were available. Such a common model structure could be used for future intersystem comparison of upwelling ecosystems, even if some of the boxes necessarily must remain empty in some of them. For example, small horse mackerel are absent off Chile (spawning occurs in the central south Pacific), and there are no small hake 2 (M. paradoxus ≤24 cm TL) in the northern Benguela (immigration is from the southern Benguela).

The large number of trophic flow indicators assessed (Figures 2–9) appears to be helpful for comparative analyses of trophic structures of ecosystems. Some reflect basic differences among foodwebs. For example, small pelagic fish rely more on zooplankton in the Benguela than in the Humboldt systems. The trophic spectra highlighted structural differences among the models. Similarly, the trophic spectra for catches indicate the trophic levels targeted by fishing at an ecosystem scale. Trophic cascades might be detected by comparing trophic spectra over successive time periods, but this was not possible here with the data available.

The comparison demonstrates that between-system differences in indicators may be more important than differences that could be attributed to heavy vs. medium fishing impacts, confirming the results of Jarre-Teichmann et al. (1998). Many indicators reflect the lower primary production and lower overall productivity in the Benguela ecosystems than in the Humboldt, and the higher abundance of (and stronger flows to and from) anchovy in the latter.

Although the indicators investigated should be able to identify changes in structure or dynamics of the modelled ecosystems, the results highlight the need to have detailed knowledge to allow their interpretation. We therefore support the conclusions of Degnbol and Jarre (2004) that fisheries management in the short and medium term should be based on system-specific knowledge, albeit in an ecosystem context. The trophic indicators at this stage appear most useful for long-term intrasystem comparisons (Cury et al., 2005), and their usefulness should improve when they are available for more than one period per ecosystem. Cross-system comparisons based on global indicators are useful for generalizing ecosystem and fisheries properties, and for modelling possible outcomes of management actions. Although knowledge about any one system may never be perfect, comparative indicators can help highlight uncertainties in structures of particular systems.

This contribution is based on the results of a workshop conducted from 28 October to 1 November 2002 at the University of Cape Town, South Africa. Funding for the workshop was kindly provided by the IRD IDYLE project. We thank Philippe Cury, Christian Mullon, and John Field for useful discussions, Franz Mueter and Yunne Shin for constructive comments, and the University of Cape Town for logistical support.

References