Analysis of Energy Flow in US GLOBEC Ecosystems Using End-to-End Models

. End-to-end models were constructed to examine and compare the trophic structure and energy flow in coastal shelf ecosystems of four US Global Ocean Ecosystem Dynamics (GLOBEC) study regions: the Northern California Current, the Central Gulf of Alaska, Georges Bank, and the Southwestern Antarctic Peninsula. High-quality data collected on system components and processes over the life of the program were used as input to the models. Although the US GLOBEC program was species-centric, focused on the study of a selected set of target species of ecological or economic importance, we took a broader community-level approach to describe end-to-end energy flow, from

(e.g., Wiebe et al., 2003;Lough et al., 2005). However, understanding trophodynamic interactions among species has long been recognized as critical to understanding the dynamics of the ecosystem as a whole (e.g., Frank et al., 2005). Multispecies ecosystem models of increasing sophistication are being developed to meet the need for a community-level approach to management of marine resources and ecosystem services subject to fishing pressures and climatic change (Travers et al., 2007;Fogarty et al., 2013, in  ecosystems. An end-to-end model describes the flow of energy (as biomass) through the ecosystem from the input of nutrients, through the production of plankton, fish, seabirds, mammals, and fisheries, to detritus and recycled nutrients. Our primary goal is to identify the main attributes that regulate each system's response to perturbations at multiple trophic levels. We use the models to estimate the relative importance of the different functional groups as energytransfer nodes and to estimate the impact of changes at these nodes. In addition to understanding and comparing ecosystem structure and dynamics, a major goal of this study is to develop an end-toend model platform that can be applied broadly across diverse ecosystems.

Four US GLOBEC Ecosystems
There are striking differences among the GLOBEC ecosystems in bottom depth and topography, circulation and stratification, seasonal cycles, and community composition across all trophic levels. These differences have prompted collection of different data sets and application of different food web models for each ecosystem, making direct end-to-end comparisons of energy flow patterns challenging.

Northern California Current
The Northern California Current (NCC; Figure 1a) is a highly productive seasonal upwelling ecosystem (Huyer, 1983;Checkley and Barth, 2009). On short time scales, lower trophic level dynamics are strongly coupled to the timing, strength, and duration of upwelling (Thomas and Strub, 2001;Thomas and Brickley, 2006). On interannual to interdecadal time scales, basin-scale climate processes (e.g., El Niño-Southern

INTRODUCTION
The broad objective of the Global Ocean Ecosystem Dynamics (GLOBEC) program was to understand the processes that control population variability. The GLOBEC approach was to study linkages between the recruitment variability of target species (e.g., calanoid copepods, euphausiids, cod, haddock, salmon) and environmental processes operating across broad temporal and spatial scales. The inability to conduct controlled experiments is a major impediment to the scientific study of the mechanics of ocean ecosystem dynamics. Ecosystem models provide the best proxy for controlled experiments (deYoung et al., 2010) and offer a way to study the integrated effects of the critical processes that occur on different scales (Fogarty and Powell, 2002).
Species-centric models have proved to be valuable tools for studying the effects of fishery management policies on individual fish stocks (Rothschild, 1986) and the effects of ocean physics on the dynamics of individual species ABSTR ACT. End-to-end models were constructed to examine and compare the trophic structure and energy flow in coastal shelf ecosystems of four US Global Ocean Ecosystem Dynamics (GLOBEC) study regions: the Northern California Current, the Central Gulf of Alaska, Georges Bank, and the Southwestern Antarctic Peninsula.
High-quality data collected on system components and processes over the life of the program were used as input to the models. Although the US GLOBEC program was species-centric, focused on the study of a selected set of target species of ecological or economic importance, we took a broader community-level approach to describe end-to-end energy flow, from nutrient input to fishery production. We built four endto-end models that were structured similarly in terms of functional group composition and time scale. The models were used to identify the mid-trophic level groups that place the greatest demand on lower trophic level production while providing the greatest support to higher trophic level production. In general, euphausiids and planktivorous forage fishes were the critical energy-transfer nodes; however, some differences between ecosystems are apparent. For example, squid provide an important alternative energy pathway to forage fish, moderating the effects of changes to forage fish abundance in scenario analyses in the Central Gulf of Alaska. In the Northern California Current, large scyphozoan jellyfish are important consumers of plankton production, but can divert energy from the rest of the food web when abundant.
Oscillation, Pacific Decadal Oscillation [PDO]) and interregional transport of large water masses strongly influence local ecosystem dynamics (Di Lorenzo et al., 2013, in this issue), control the composition of upwelling source waters (Huyer et al., 2002), and affect the composition of the local mesozooplankton grazer community (Batchelder et al., 2002;Keister and Peterson, 2003). These physical and lower trophic level processes directly affect the production of pelagic fishes (Brodeur and Pearcy, 1992;Ruzicka et al., 2011;Burke et al., 2013), benthic invertebrates (Barth et al., 2007), and local seabird and marine mammal populations (Ainley and Boekelheide, 1990;Keiper et al., 2005). An end-to-end model of the NCC must incorporate both local physical processes (upwellingdriven primary production) and important nonlocal factors that affect community composition across all trophic levels.

Central Gulf of Alaska
The Central Gulf of Alaska (CGOA) system ( Figure 1b) is a highly productive downwelling system (Stabeno et al., 2004). Offshore surface waters that are advected onto the shelf during downwelling events originate from the highnutrient, low-chlorophyll (HNLC), ironlimited region of the North Pacific gyre.
Mixing of HNLC waters with iron-replete shelf waters drives the production cycle (Fiechter et al., 2009). Fish and marine mammal populations have changed dramatically over the past 40 years, with some species shifts correlating well with the 1976-1977 PDO shift (Francis et al., 1998;Anderson and Piatt, 1999).
Connecting these physical and lower trophic level processes with what appear to be strong shifts among mid and upper trophic level interactions in this ecosystem (Gaichas et al., 2011) is an important challenge for end-to-end modeling.

Georges Bank
Georges Bank (GB) is a shallow bank offshore of Cape Cod (Figure 1c). It has long been the site of economically important fisheries, including cod (Gadus morhua), haddock   (Flagg, 1987). A pronounced diatom bloom usually occurs in early spring, supporting production of the large calanoid copepod Calanus finmarchicus.
Both the phytoplankton and zooplankton communities shift to smaller forms during the remainder of the annual cycle (Davis, 1984). Strong interactions between benthic and pelagic components at several trophic levels complicate end-to-end analysis (Steele et al., 2007).

Southwestern Antarctic Peninsula
The southwestern Antarctic Peninsula (sWAP) ecosystem supports roughly half of the total Antarctic krill (Euphausia superba) population (Atkinson et al., 2004) and some of the largest populations of vertebrate predators in the Southern Ocean region (Everson, 1977(Everson, , 1984. Although nitrogen is not considered to be limiting, micronutrients (including iron) and sunlight are. Interannually variable seasonal sea ice cover reduces solar irradiance into the upper water column, limiting overall system production and impacting the ecology of the entire ecosystem (Longhurst, 1998;Ducklow et al., 2007). The sWAP (Figure 1d) is connected to the larger Antarctic ecosystem at several trophic levels. It is thought to be an upstream source for recruits to the krill population around South Georgia (Fach et al., 2006). Satellite tracking studies show that seabird and marine mammal predators move and forage throughout the greater Antarctic Peninsula region (Catry et al., 2004;Croxall et al., 2005;Phillips et al., 2005;Biuw et al., 2007). An end-to-end model of the sWAP ecosystem must incorporate important local physical processes and must take into account intra-regional connectivity within the greater Antarctic Peninsula-Scotia Sea ecosystem.

MODELS AND METHODS Building the Food Web Models
The basic information needed to build a food web model consists of: (1) diet information for each functional group, which defines the topology of the food web network, and (2) terms for biomasses and physiological rates, which define the rate of energy flow through each trophic linkage. Except for GB, the models were initially constructed as Ecopath food web models (Christensen and Walters, 2004; http://www.ecopath. org). Ecopath models infer the strength of individual trophic linkages from the energy demand of consumers upon their prey. The logic behind this "top-down" approach is that data availability and quality are typically better for upper trophic level consumers and fisheries than for low and mid-trophic level groups. It is then mathematically simple to transform a top-down linear expression of predation pressure (Ecopath) into a bottom-up map of energy flow from lower trophic level producers to upper trophic level consumers (Steele, 2009).
With the inclusion of external nutrient fluxes as input for uptake by phytoplankton, nutrient recycling via bacterial metabolism of detritus and consumer metabolism, and an accounting for production losses from the system via physical export, an end-to-end ecosystem model may be constructed (Steele and Ruzicka, 2011). From diverse model origins, all four ecosystems were described within similar end-to-end model frameworks.
For a comparative ecosystem study, care must be taken to (1) define functional groups similarly across models, (2) define model domains on similar temporal and spatial scales as appropriate to the data, (3) be aware of connectivity to neighboring systems, and (4) account for uncertainty and variability among parameters. Figure 2 shows the food webs of each US GLOBEC region; Table 1 provides the details about the underlying data sets used to build each model. The full parameter sets defining each model are available in the supplementary material for Ruzicka et al. (2013).

Metrics and Scenarios
Basic metrics are extracted from observations of each ecosystem and from food web models to describe the overall size  (Steele, 2009;Steele and Ruzicka, 2011).
A structural scenario is constructed by changing the relative consumption rate of one or more consumer group(s) upon any specified prey group. In the scenarios presented here, the imposed change comes at the direct expense of (or benefit to) any consumer group competing for the same prey. The total consumer pressure on a given prey group was not changed and transfer efficiencies were held constant, implying no changes to group physiologies (assimilation efficiencies, growth efficiencies, and weight-specific production rates) nor to   Figure 2. Food webs for each US GLOBEC region. Color shows the footprint and reach of the planktivorous fishes (forage fishes). Footprint (green) is the fraction of each group's production consumed by the planktivorous fishes. Reach (red) is the fraction of each consumer's production that has originated with the planktivorous forage fishes via all direct and indirect pathways.   (2007) GLOBEC: Ashjian et al. (2004Ashjian et al. ( , 2008 Marrari et al. (2011) Daly (2004) (2007) NOAA: Keller et al. (2008) Stock assessments NOAA: Britt and Martin (2001) NOAA: Azarovitz (1981) NEFC (1988) Smith (

RE SULTS Food Web Metrics
Ecosystem Size and Production (Table 2 FOOTPRINT: The relative importance of a group as a consumer expressed as the energy demand of the consumer upon one or more producers. A consumer may have a footprint upon a producer even if it does not directly prey upon that producer. A commonly encountered footprint in the literature is the "primary production required" (PPR) to support a fishery or consumer group of particular interest.
REACH: The relative importance of a group as a producer expressed as the fraction of the group's production that reaches one or more consumer groups via all direct and indirect food web pathways.
FOOD WEB EFFICIENCY: How efficiently energy is transferred through a food web, considering all alternate energy pathways and physiological losses at each link in the web. Food web efficiency, expressed as the realized production rate of each functional group per unit of primary production, is insensitive to differences in overall ecosystem size. In the NCC, gelatinous zooplankton have a particularly large footprint on system production-much larger than in the other systems. Large scyphozoan jellyfish such as the sea nettle (Chrysaora fuscecens) can attain very high densities during late summer months (Suchman et al., 2012). They also have an apparently large reach, though much of it can be attributed to predation among the different classes of gelatinous zooplankton (e.g., larger jellyfish preying upon salps and larvaceans). If large jellyfish are considered separately, their footprint is almost 4% of total system production while their contribution back to the system represents only 0.05% of total consumer production in the system. In this system, jellyfish might be considered a trophic dead end: they consume much more in comparison to what they return to the ecosystem. Table 2. Model-derived mean annual production rates (t C km -2 yr -1 ).  Table 3. Ecosystem-scale Footprint and Reach metrics of mid-trophic level groups. Footprint = percentage of total system production supporting each consumer group. Reach = percentage of total system consumer production that passes through each mid-trophic level group.

NCC
(Flows to and from detritus groups excluded.) (See Figure 3). *Reach in excess of footprint represents detritus feeding and recycling of "lost production" back into the food web NCC = Northern California Current. CGOA = Central Gulf of Alaska. GB = Georges Bank. sWAP = Southwestern Antarctic Peninsula Food Web Efficiency (Figure 4) The NCC and the CGOA are significant producers of forage fishes, producing almost twice the biomass of small planktivorous fish per unit of phytoplankton production than the GB and sWAP ecosystems. The NCC is also a large producer of "piscivorous" fishes, such as Pacific hake (Merluccius productus), that have mixed diets of fish and euphausiids (Miller et al., 2010). Omnivory across trophic levels may contribute to the higher efficiency of fishery production in the NCC. On Georges Bank, more of the energy in the system supports the production of demersal fishes, for example, cod (G. morhua) and haddock (M. aeglefinus), than production of pelagic fishes.

Structural Scenarios
Forage Fishes (Small Pelagic Planktivores) In the Northern California Current model, doubling consumption by forage fishes (sardine, anchovy, herring, smelts) directly benefitted groups that prey directly upon forage fish: seabirds, baleen and odontocete whales, and pinnipeds ( Figure 5a). Seabirds in particular benefited, and competitor groups (piscivorous fishes, demersal fishes, squid) were negatively impacted. While piscivorous fishes (dominated by Pacific hake) should be expected to benefit directly from increased forage fish abundance, there is a high degree of omnivory in the NCC where piscivorous fish also prey heavily upon euphausiids (Miller et al., 2010).
This scenario indicates that any benefit to Pacific hake from increased forage fish abundance may be more than offset by . Green bars are the footprints, the fraction of total system production consumed by the group of interest. Red bars are the reach, the fraction of total system consumer production that is produced by (or passes through) the group of interest. (See Table 3  increased competition for euphausiids. In the Central Gulf of Alaska, forage fish abundance (walleye pollock, herring, capelin, eulachon, sandlance, myctophids) could only be increased by about 60% (Figure 5b). Prey resources were insufficient to support more planktivores without restructuring trophic relationships within the food web or increasing food web efficiency. In contrast to the NCC, most top predators suffered in this scenario: only pinnipeds benefited. Why this would be so may be explained by the response of squid, which are a more important energy transfer node in the CGOA model (see Table 3). Increased competition with planktivorous fishes reduces realized squid production and the efficiency of energy transfer to seabird and mammal predators.
On Georges Bank, doubling forage fish abundance (Atlantic herring) had a smaller effect than in the NCC or the CGOA (Figure 5c). Odontocetes benefited directly from increased prey abundance while baleen whales and demersal fishes suffered from increased competition with forage fish for zooplankton.

In the southwestern Antarctic
Peninsula area, the planktivorous fishes (nototheniids, myctophids) could only increase by about 60% without restructuring trophic relationships or increasing food web efficiency (Figure 5d). No group benefited substantially. The sWAP groups most impacted were those that prey heavily upon euphausiids: penguins, crabeater seals (Lobodon carcinophagus), squid, and baleen whales.
Gelatinous Zooplankton (Larvaceans, Salps, Ctenophores, Large Scyphozoans) In the NCC, all groups were impacted negatively by doubling gelatinous zooplankton abundance (Figure 6a). As the footprint and reach metrics show (Table 3), gelatinous zooplankton consume much of the total system production but pass relatively little upward in the NCC food web. The impact of gelatinous zooplankton was much stronger here than in the other US GLOBEC ecosystems. In the CGOA, except for an increase in demersal fish production, increased gelatinous zooplankton abundance had very little effect (Figure 6b

4% reductions in production.
A potential future sWAP scenario may be considered in which warming temperature, decreasing sea ice, and a shift in the phytoplankton community toward smaller cells favors salp production over krill (Loeb et al., 1997;Ducklow et al., 2007). Redirection of phytoplankton production away from krill by 50% and toward salps would lead to reductions in the production of intermediate and top trophic levels of 20-30% ( Figure 6d). Such a salp-dominated system would not support the seabird and mammal populations we observe today .

Baleen Whales
In all four ecosystems, a fivefold increase in baleen whale abundance had much smaller effects than did doubling of forage fish abundance or gelatinous zooplankton abundance (Figure 7a-d).
Piscivorous fishes in the CGOA have diets richer in small pelagic fishes than they do in the NCC or GB, and piscivores in the CGOA were more heavily impacted by direct competition with baleen whales than in the other regions.
Increased baleen whale abundance had a smaller impact in the sWAP ecosystem than in the other ecosystems. Seabirds, penguins, and pinnipeds were the most heavily impacted sWAP groups, with all exhibiting a small decline in production rate.

Dynamic Scenarios
The effects of increased forage fish abundance are evaluated as the relative change in biomasses in the non-altered base runs and the perturbed scenario runs. Figure 8 shows cesses. Biological processes of particular concern include those that define nutrient recycling rates (e.g., detritus and bacterial dynamics), define connectivity with neighboring ecosystems at upper trophic levels (migration), and contribute to population size and structure and community composition (recruitment dynamics). Physical processes that must be considered are the local processes that drive nutrient input and support primary production (e.g., vertical mixing and upwelling) and the regional-scale pro-    comparative modeling activities such as those presented here, more detailed models of specific processes may be developed to improve understanding of ecosystem structure, mechanics, and response to environmental variability and anthropogenic perturbation.

ACKNOWLEDGEMENTS
We especially thank our late colleague and co-author, John Steele, for inspiring this effort from end to end. We also thank Susan Ruzicka