P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 287
Abstract
System dynamics provides a powerful support mechanism for resolving problems in highly
complex and dynamic contexts. Effectively building a system dynamics model in a client-group
environment is a challenging task, particularly with a heterogeneous clientele and a variety ofstakeholders influencing the boundaries of the model. In such environments, understanding and
formulating central concerns is as important as it is challenging. This article describes a group
model building initiative designed to study the implications of various policies aimed at revitaliz-ing the fishing industry in Gloucester, Massachusetts. While the iterative and interactive methods
in this group modeling intervention helped fishermen and community members to communicate
effectively with one another, define problems and improve their understanding of the criticalinteractions in the system, the resulting model was used to help the Gloucester Community
Development Cooperation to communicate potential impacts of a prospective fish factory to a
Building system dynamics models with client groups has a long tradition inour field and is well documented (Stenberg 1980; Morecroft and Sterman 1994;Richardson and Andersen 1995; Vennix 1994). In the literature several ap-proaches to group model building are discussed (Richardson and Pugh 1981;Roberts et al. 1983; Vennix 1994) with varying stages on how the process ofconstructing a computer simulation model involves a number of conceptualactivities. In the context of group model interventions, this article discussesthe procedural and conceptual steps, insights and lessons learned from amodel building project for the Gloucester Community Development Coopera-tion (hereafter GCDC).
Gloucester, Massachusetts, is one of the oldest ports in New England. For370 years, its economic prosperity has derived from fishing. During the latetwentieth century, however, Gloucester’s economy has been severely threat-ened by over-fishing. Between 1960 and 1975 ground-fish stocks (the NationalOceanic and Atmospheric Administration [NOAA] classifies ground-fish as amixture of bottom-dwelling species, including Atlantic cod, haddock, redfish,hake and flounder) were severely depleted by both domestic and foreign fishing.Stocks declined rapidly (National Marine Fisheries Service (NMFS) 1999),forcing the federal government to impose controls on the fisheries.
Gloucester Fishery: insights from a groupmodeling intervention
* Correspondence to: Peter Otto, 410 Terrace Road, Schenectady, NY 12306, U.S.A.a School of Business, Dowling College, Oakdale, New York. E-mail: [email protected] Sloan School of Business, Massachusetts Institute of Technology, Cambridge, MA, U.S.A. E-mail: [email protected]
288 System Dynamics Review Volume 20 Number 4 Winter 2004
The resulting constraints on traditional fishing methods posed a challengefor the GCDC. Citizens feared losing their cultural identity along with fishingrevenue. They were apprehensive that empty wharf space would attract realestate developers interested in creating condominiums, motels, and retailoutlets, inalterably changing Gloucester’s landscape. In hopes of revitalizingGloucester’s economy, community leaders began to look for alternate sourcesof revenue.
One of the most promising proposals was establishing a Surimi (pronounced“sir-ree-mee”) factory. Surimi means “minced fish” in Japanese and is tradi-tionally produced from skinless Alaskan Pollack. A new and unique techno-logy, patented by one of the GCDC members, would allow the factory to useherring and mackerel (which are referred to as pelagics) in place of pollack;the fish would then be leached and mixed with sugar and other additives tocreate a frozen product used in the manufacture of imitation crabstick. Herringand mackerel (together with butterfish) are classified by the NOAA as under-exploited stocks and are available in “almost unlimited quantities” off thecoast of New England. The NOAA estimates the current North Atlantic spawn-ing biomass at roughly 1.5 million metric tonnes (MT) for herring and 2million MT for mackerel, with a total allowable catch (TAC) of 317,000 MT ofherring and 326,000 MT of mackerel, implying a yearly allowable mortalityfraction of 16–20 per cent (personal communication in meeting with NOAA).A fourth pelagic species, bluefish, is over-utilized. The total average recentyield of herring, mackerel and butterfish is estimated at 158,000 MT. Thisfigure could easily be quadrupled and still fall below the aggregate long-termharvesting potential (NMFS 1999).
The proposal necessitated a feasibility study, the various stages of which aredescribed below. The project culminated in a system dynamics model, whichserved to improve stakeholders’ understanding of the complex interactionsbetween the community, government regulations and the resources needed tobuild and sustain such a factory.
Project description
Rather than waiting for stocks of traditional white fish species (e.g., cod andhaddock) to return to sustainable harvesting levels, the GCDC decided to focuson new ways of sustaining the economic and social well being of Gloucester.Given the abundance of pelagics, the new factory could easily process 10,000MT of Surimi per year without adversely affecting the North Atlantic biomass.A 10,000 MT Surimi throughput would require harvesting about 4 per centof the “total allowable catch”, or approximately 30,000 MT of herring ormackerel. This estimate yielded the first starting assumption for the GCDC: thesize of the Surimi factory would be small compared to other fish factories inGloucester (at least during the first years of operation). Second, the Surimi
of socio-technical
systems; in particular,
he focuses on thetransition challenges
related to hydrogen
fuel cell vehicles.Jeroen received his
MSc in Physics from
the Delft Universityof Technology and
has worked as a
managementconsultant in the
Netherlands.
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 289
factory would not only generate revenue while stocks of traditional specieswere returning to sustainable harvesting levels but would also help foster abalanced multi-species management of the natural resources. Third, it wasassumed that the Surimi factory would facilitate research and developmentin similar innovative processes, for example extracting oil or fertilizer fromleftover skin and bones.
Current issues for the GCDC project
Shifting the burden from over-fished species, like cod and flounder, to morerobust and sustainable stocks of dark fish may be a laudable initiative. How-ever, the problem with mixed-stock or mixed-species fisheries is that somespecies and stocks are more productive than others. As a result, when quotasare low enough to protect the least productive or most “catchable” population,others are under-harvested and there is great pressure to increase catch rates.Predictably, when the catch rate increases, the less productive populations beginto show signs of exhaustion (National Marine Fisheries Service (NMFS) 1999).
To further complicate matters, an increase in pelagic harvests could leadto unexpected interactions in the biomass. At present, little is known aboutthe function of pelagics in the food chain; yet, if pelagic stocks decrease,predators (primarily ground-fish like cod) will find less prey, thus affectingtheir own sustainability. In the past, there were always enough pelagic stocksto harvest. However, the successful launch of a Surimi factory in Gloucestercould inspire other fishing communities to tap into this lucrative busi-ness. According to the GCDC, the total worldwide market for Surimi isapproximately 760,000 MT, growing at 10–20 per cent per year, with Japanconsuming 60 per cent of total production. If European and North Americanconsumers increase demand, Surimi consumption could easily reach stagger-ing levels.
History could repeat itself! Safina (1994) describes the following precedent:encouraged by the North Pacific Fishery Management Council’s open-to-all-at-no-fee policy, the number of factory trawlers, i.e., those capable of catching350,000 pounds in a single haul, jumped from 12 in 1986 to 65 in 1992. Miningthe waters off Alaska, they targeted walleye pollack; landings of 3 billionpounds (valued at $324 million in 1992) made this the largest single-speciesfishery in the world. As a result of this fishing frenzy, pollack stocks havedeclined sharply. Consequently, animals that feed on pollack have also beenbadly affected; the population of sea lions and several seabirds has declined50 to 90 per cent in the region in the last 20 years. In response to increasedpressure on fish stocks, the government imposed tougher regulations, trying tomaintain a healthy balance between harvesting and conservation. However,because of the complexity of these resource systems, establishing effectiveharvest quotas has proven very challenging. Even with sophisticated statisticalanalysis and assessment tools, there is still a high degree of uncertainty.
290 System Dynamics Review Volume 20 Number 4 Winter 2004
Expectations and challenges of the project
Being unable to adequately determine pelagic sustainability and, therefore, thesustainability of a Surimi factory, the GCDC found itself in a highly unstruc-tured decision-making environment and realized it needed assistance. GCDCboard members, including Dr Carmine Gorga (director), Dr Steve Kelleher(patent holder of the Surimi processing technology), Dr Damon Cummings(board member) and Joe Sinagra (fisherman), approached the MassachusettsInstitute of Technology for help in studying the dynamic structure of pelagicsand the impact a Surimi factory would have on the biomass. MIT offered ateam of PhD and Masters students from the Sloan School of Management andthe State University of New York at Albany the opportunity to join the projectas part of their enrollment in a system dynamics course at MIT (“AdvancedApplications in System Dynamics”, Spring 2002).
Over a 15-week (one semester) period, the MIT/UAlbany team met amongthemselves and with the Gloucester representatives to formulate questions,which the MIT/UAlbany team sought to answer through a system dynamicsmodel building process. Questions included:
• What happens to the biomass if consumption of Surimi reaches unprec-edented levels?
• Is there a critical point for the sustainability of pelagic stocks?• Is it possible to achieve sustainability through multi-species management of
the natural resources or can we do little more than shift the burden?
There are no easy answers to these questions because the highly complexstructure of the ecosystem begets so many uncertainties and unknown param-eters. Ideally, a model must capture the evolution of stocks as well as theinteractions between harvesting and biological renewal to determine dynamics.Yet, project participants could agree neither on the relevance of these dynam-ics (e.g., resource recovery rates) nor on their potential impact. The GCDCsought to understand the dynamic, non-linear feedback structure of the system,including the complex socio-economic interactions between the communityand the factory. It is difficult to identify the critical thresholds and sensitivitiesof the system in question, but, if we extrapolate from what occurs in otherecosystems that exhibit increasing scarcity (e.g., forestry and agriculture),we might forecast that pelagic stocks could “overshoot and collapse” undercertain conditions.
Group model building intervention
Our endeavor to guide the client through the project and to provide insightsand recommendations required an iterative and structured process. Meadows
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 291
et al. (1974) and Richardson and Pugh (1981) identified a number of stages inthe model-building process that are organized linearly so as to integrate thelearning that takes place along the way. For the GCDC project, we followedHines’s (2001) standard method, which was the indicated method in the course.The method outlines structural steps and processes for applying system dy-namics in a consulting environment, which is different from building simula-tion models without client interaction. In the following section, we describethe steps involved in, as well as the lessons learned from, our clientgroupmodeling intervention.
Diagnosing the problem
In the first phase of the endeavor, we challenged our clients’ assumptions as towhat the boundary and focus of the project should be. Our clients’ initialconcerns were related to sensitivity issues, for example water and electricityusage; constraints influencing the desired Surimi throughput were core con-cerns. Following the “standard method”, we elicited approximately 60 differ-ent variables influencing the project one way or another. The list was thennarrowed to a number of key variables, clustered in the three high-level sectorsshown in Figure 1. The purpose of clustering the key variables was to focusattention on the interactions between these variables and to help clarify sys-tem boundaries. As soon as we drew the causal loop diagrams, the sectorboundaries as indicated in Figure 1 disappeared.
The next step in the project was to formalize the following problem defini-tion, which reflected our understanding of the situation after intense discus-sion with the client:
The decline of traditional fish species and the curtailing of fishing efforts by theGovernment require the fishing industry of Gloucester to identify alternative re-sources to sustain their industry . . .
. . . A Surimi factory—harvesting fast renewable fish stock—might compensate forthe missing revenues from traditional white fish until their stock returns to a sustain-able level . . .
The problem definition, in essence, reflected a consensus as to what the projectshould address and how the boundaries of the simulation model should bedrawn. In addition, we visualized the problem dynamically; the referencemode for the expected behavior of the system is shown in Figure 2.
The reference mode shown in Figure 2 captures both the decline in revenuefrom ground-fish prior to 2003, due to dwindling fish stocks and governmentregulation, and expectations concerning future revenue from pelagics. By 2004,the new Surimi factory should begin to realize profits. Based on our interpreta-tion of the structural elements in the system, we hypothesize that traditionalfish stocks will rebound as multi-species management and conservation ease
292 System Dynamics Review Volume 20 Number 4 Winter 2004
Fig. 1. Key variables
and sectors
influencing the Surimiproject
pressures. An underlying assumption is that, once the factory is near completion,a number of Gloucester fishermen will decide to harvest pelagics. Once trawlersare retrofitted to make them suitable for harvesting pelagics, pressure on ground-fish stocks should decrease. This assumption depends, however, on a constantnumber of trawlers; if the retrofitted trawlers are merely replaced by otherground-fish trawlers, stocks will continue to decline.
Indeed, with open access to a common resource—in this case pelagics—over-exploitation remains a threat. Such depredations would inevitably leadto the “tragedy of the commons” famously described by Gordon (1954) andHardin (1968). Moxnes (1998) relates: “. . . the commons problem is widelyheld to be the cause of mismanagement of common renewable resources”.Over the last few years, the NOAA has devoted a great deal of effort to effect-ively and efficiently managing the North Atlantic biomass. The resurgence inground-fish in the waters off New England is tangible proof of the NOAA’ssuccess in preserving the biomass.
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 293
Fig. 2. A reference
mode to capture the
expected dynamics inthe system
System boundary and dynamic hypotheses
The project involved a number of stakeholders with different agendas andinterests. While the Gloucester fishing industry is looking for ways to increaserevenue, the community wants to keep and create jobs. Communication withand between these groups was not always easy, yet their involvement wascrucial. The figures given below (Figures 3 and 4) capture the expected behavior
Fig. 3. Dynamic hypothesis “control and utilization”
294 System Dynamics Review Volume 20 Number 4 Winter 2004
of the system; we used these in our discussions with the client in order toobtain consensus on structural consistency and system boundaries. Employ-ing a reference mode along with causal loop diagrams helped us captureimportant system dynamics. The ability to visualize the problem further helpedthe various groups draw important system boundaries.
The reference mode detailed in Figure 2 assumed two conditions: first, thatthe fishing industry understands the feedback structure that leads to over-investment and over-utilization; and, second, that there are policies in place toensure sustainability. It is subject to debate just how realistic these assump-tions truly are. To express alternate scenarios in the same system, we formu-lated a number of dynamic hypotheses in line with the “standard method”.The primary purpose of a dynamic hypothesis is to explain a reference modewhile making assumptions explicit. In that sense, dynamic hypotheses aretheories that a certain structure or process will contribute to certain behaviorpatterns (Hines 2000).
The dynamic hypothesis illustrated in Figure 3 captures the influence of thecontrol and over-utilization loops on the “total allowable catch” (TAC) figuresprovided by the government. In line with the standard method, we did notidentify stocks in our dynamic hypotheses; as we believed that one can onlyintroduce so many new concepts at a time—in this first stage of the projectthe additional value would not weigh against the potential risk of confusion.The reinforcing loop (“overshoot and collapse”) contributes to the “decay”, or
Fig. 4. Dynamic Hypothesis “Community Quality of Living”
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 295
“fear”, scenario; the two balancing (control) loops stabilize the system (indi-cated by the “hope” scenario in our graph). Thus, our hypothesis states thatincreasing attractiveness of pelagics and decreasing availability of white orground-fish could lead to resource depletion. The underlying assumption ofthis hypothesis is that the sustainability of the system can only be maintainedby government action and not by self-regulation in the fishing industry. Whilegovernment control for ground fish is based on daily landings, pelagic fishstocks are assessed only once a year. Thus “total allowable catch” ( TAC) forwhite fish is a floating number, which is based on actual landings and thencompared with the most current fish stock estimation from NOAA. The differ-ence in stock assessment leads to different delays from government control.
The dynamic hypothesis illustrated in Figure 4 shows the dynamics of thevariable “community quality of living”, a term used by the client to describethe impact of the Surimi factory on life in Gloucester. This variable is multi-faceted and has to do with job creation and retention and revenue generation(from white fish and the Surimi revenues)—factors enabling Gloucester toretain its essential character as a thriving port.
The hope scenario (in Figures 3 and 4) implies that there are enough renew-able resources in the ecosystem (both white and dark fish) to justify reinvest-ment in the Surimi plant and to create improved stability (“community qualityof living”). As noted above, the factory would also act as an “incubator”,facilitating research in new fish processing technologies. Reinvestment in thefactory would provide funding for these research activities, resulting in risingstability. The behavior of the slope “fear1” suggests that the factory could betoo successful, i.e., that increased revenue could foster increased competitionfrom other fishing communities, resulting in depletion of stocks and/or un-equal wealth distribution within the Gloucester fishing community. With the“fear2” scenario, we hypothesize a delay in takeoff due to a lack of Food andDrug Administration (FDA) approval, poor sales or increased competition.
Scope of the project
While the dynamic hypotheses helped us capture the structure and expectedbehavior pattern of the system, we soon realized that, by using these diagrams,we pushed the system boundaries too far and the project became so complexthat it exceeded our initial purpose. In response, we redefined the scope of theproject and focused on capturing the most relevant structural details of thesystem. In consultation with the client, we used the diagram in Figure 5 todefine the scope of the project more accurately. The diagram not only provideda functional description of the modules, but also helped the team stay withinthe boundaries during conceptualization of the simulation model. We did notuse a system dynamics stock and flow structure but tried to simplify a high-level view of the system to communicate with the client more easily. However,the boxes in the diagram suggest “accumulations” while the arrows indicate a
296 System Dynamics Review Volume 20 Number 4 Winter 2004
Fig. 5. Scope and
model structure to
capture the systemboundaries
causal link. For example, the arrows from the box “Pelagic Trawlers” to “HerringStock” and “Mackerel Stock” suggest that a change in the number of trawlerschanges the stock levels.
Figure 5 reveals the suggested scope of the project. The model structure iscomposed of nine modules, which the group identified as relevant for ourproject:
1. Processing Quality of Surimi, which could differ depending on the composi-tion of the biomass;
2. Factory Capacity—initially adjusted to 10,000 MT of output per year;3. Total Output—the actual output of the factory per year;4. Pelagic Trawlers—the number of boats employed by the factory;5. Relative Attactiveness of Pelagics;6. Ground-fish Stock—primarily white fish;7. Herring Stock;8. Mackerel Stock;9. Ground-fish Trawlers—the number of boats harvesting white fish.
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 297
This script1 turned out to be very useful for the group to determine the systemboundaries as well as to lay out the ground for the model structure.
As previously noted, the purpose of building the model was to clarify therelationship between the fishery and fishing-induced changes in the variousstocks. For example, if ground-fish stocks increase as a result of a reboundin the cod population, the relative attractiveness of pelagics (to fishermen)decreases; ground-fish generate a much higher profit per boat. As a result,fewer trawlers will harvest pelagics, thereby reducing the total output of thefactory. If ground-fish stocks decay or if the government issues tighter quotas,attractiveness will increase and more trawlers will harvest pelagics.
If pelagics are perceived to be under-exploited, fishermen will attemptto harvest as much as they can, for as long as they can. By the time thegovernment gets around to issuing quotas, it could be too late to stop the decayof the biomass. With our model, we seek to locate optimal harvest rates, inorder to determine just how much can be harvested without jeopardizingstability. Ideally, we will be able to identify thresholds for the factory aswell, predicting how it can respond to changing economic and ecologicalconditions.
Even though the team understood the process of harvesting and subsequentinteractions between the different modules, getting reliable data was not easy.The NOAA provided access to certain data (landings, total allowable catch,sustainable catch per species, etc.), yet the quality of these data is subject todebate. Also, as previously noted, the precise function of pelagics in the foodchain is unknown. The NOAA is currently investigating how it can analyze“digest data” to provide a better understanding of stock dynamics betweenpredators and prey.
Model building process
In the foregoing sections, we have provided details of problem identification,system boundaries issues, and the formulation of dynamic hypotheses. Thescripts we used in the first phase of our intervention, e.g., reference modesand dynamic hypotheses, helped us to communicate with our client andstakeholders and framed the basis for the simulation model. The componentsof the model we present in the following sections were used in discussionswith our client and experts from NOAA to confirm structural consistency andto capture the important dynamics for this project (the complete model isavailable at http://www.albany.edu/~potto/research/index.htm).
Following the standard method (Hines 2001), we initiated the model build-ing process by subjectively selecting a dynamic hypothesis that scored the“highest” in such characteristics as tangibility of parameters and perceivedease of design. The client was involved neither in model building nor inthe determination of mathematical relations (except for some of the table
298 System Dynamics Review Volume 20 Number 4 Winter 2004
functions). The client did, however, provide useful data, particularly variousscaling factors and operational/process details for the factory.
The first hypothesis constructed was “control and utilization” (see Figure 3).Bearing in mind our initial project goal of forming “structural insights”, weiteratively adjusted the structure and parameter values based on “rough cal-ibration” and responses to pulse and step inputs, and then tested the validity ofthe hypothesis. After we incorporated two additional hypotheses, structuralcomplexity unintentionally rose. So we went one step back and used the same“sectors” that were used to elicit and structure the key variables. We did so inaccordance with the belief that utility is significantly improved at the detailedlevel for functionally related concepts such as factory behavior, communica-tion and trust building with the client. Dynamic complexity was also incorpor-ated at this time. The types of insights that resulted from this process differedfrom the hypotheses that emerged from discussions with the client. In thefollowing section, we discuss vital aspects of the operationalization of themodel, including some of the critical issues and insights that arose duringmodel building, testing and analysis.
Resource sector—resource dynamics
For the simulation model, we incorporated three types of fish stocks: ground-fish, mackerel and herring. We were obliged to model the two pelagic stocksseparately because differing regeneration dynamics influence the growth anddecay of the individual stocks.
Since our initial objective was to gain general insights, we felt compelled tokeep structures as simple as possible. The structure for representing regenera-tion and harvesting is nearly identical for all three stocks; Figure 6 shows theimplementation for mackerel. As a basis for the resource dynamics, we used astandard system dynamics approach to population dynamics (Swart 1990;Ruth 1995; Sterman 2000), a procedure widely employed by theoretical biolo-gists. We assumed one stock: mackerel stock (MK) in units of metric tonnes(MT). We concluded that adding more complexity was unnecessary and weconcur with Moxnes’ (2000) argument for simplicity in modeling this type ofstock (even though his purpose differed from ours). For calibration, however,we took more detail into account (see discussion of model validation).
Figure 6 demonstrates the level of detail for the mackerel sub-sector used todiscuss the model at the operational level. Structures for herring and groundfish are identical. Within the group we defined the “carrying capacity” torepresent the natural equilibrium population biomass in metric tons [MTO],being the population that is attained in the absence of harvesting. For mackerelthe carrying capacity is estimated by the NOAA to be 2.4 million MT, for theNorth Atlantic coast (NOAA 2001). We assumed the natural fractional changerates (“MK Fractional Death Rate” and “MK fractional birth rate”, [dmnl/month]) to depend non-linearly on the relative density of the population
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 299
(Mackerel Stock/MK Carrying Capacity [dmnl]). Sketches for these and othertable functions were made within the client group (conforming with Ford andSterman, 1998) and resulted in full agreement on their qualitative shapes. Theresulting table function “effect of relative density on the fractional death rate”(inset 1, Figure 6), represents the result of the group’s assumption of a deathfraction that is constant as long as the population is low, but increases atincreasing rate when crowding passes the carrying capacity; the assumptionbehind this being increased harvesting by other fish and reduced availabilityof food. In contrast, the shape of the “effect of relative density on fractionalbirth rates” (inset 2, Figure 6) represents an assumption that birth fractions areconstant at their “reproductive potential” for medium to high densities, butdecrease for low populations. As depicted here, the fractional birth rate falls to75 per cent of its original value as the population equals 25 per cent of the totalcarrying capacity. This decline in population performance at low densities is
Fig. 6. Structural components of resource dynamics for mackerel fish stock
300 System Dynamics Review Volume 20 Number 4 Winter 2004
known to occur for a variety reasons for different species, including fish, and istermed “Allee’s principle” in the population ecology literature (Odum 1971).More importantly, while we did not discuss the cause of decline in birthfraction in detail amongst the group, it matched the intuition of several clientpeople, based on experiences with sustained decline after extreme depletion.
With data provided by the NOAA (2001) the parameters “Normal FractionalBirth” and “Normal Fractional Death”, which are equal,2 were determined andthe reference points for the table functions were set. This effort did not rejectany of the qualitative assumptions for the table functions or other parts of themodel structure. The Normal Fractional Birth equals the reproductive poten-tial and was found to be 0.041 [dmnl/month] for mackerel and 0.03 for herring.
One can readily derive much of the dynamics for the population from tablefunctions 1 and 2. First, when the relative density of the population equals 1(population equals carrying capacity), the birth rate equals the death rate,which was by construction. Second, at higher densities, as the death fractionincreases dramatically, the population will always bounce back to the carryingcapacity. The net reproduction rate is at a maximum when the population is atabout 75 per cent of the carrying capacity; for mackerel the reproduction rate atthis point is about 40,000 MTO/month, for herring 15,000. Furthermore, whenthe population falls below 15 per cent of the carrying capacity, the dynamicsresult in a downward spiral leading to permanent decline. For herring thetipping point is at 5 per cent of the population. Note that, while one can raisequestions about “full extinction” beyond a tipping point, the existence of anysort of Allee-effect (the phenomenon of density dependent growth rates) willhave serious implications for the speed of recovery, even in the absence of atipping-point.
The maximum amount trawlers can harvest is their total fleet capacity (MT/month). The actual catch per trip (which lasts from several days to a week)depends on the relative abundance of the stock, which is a function of bothtechnology and population (this will be discussed in detail in the “Resourceregulation” section below). The fractional yield represents the average frac-tional boat filling (assuming the technology to be constant over time). The“effect of relative density on yield” (inset 3, Figure 6) represents the notion ofa declining catch per effort (day on the ocean) at lower population levels.Initially, this effect is very limited as people can increase their efforts per day,for instance by doing more rounds per day, but, ultimately, at very low re-source abundance, yield and value are zero. Fractional yield is also a functionof fish species: some species are caught more easily than others; harvestinglocations and technology also differ, but the shape of the curves is assumedto be identical (see inset 3 in Figure 6). Under conditions of abundance, thecurve flattens (saturation). Improved technology would shift the curve to theleft. The final variables we considered, i.e., “other effects”, include extremeecological fluctuations and normal harvesting rates in the North Atlantic Sea.For the time being, the fractional impact of “other effects” on the population
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 301
[dmn/mt] is considered an external parameter, i.e., independent of suchfactors as “attractiveness of pelagics”.
In a meeting that included our client and the authors, this model sector was“validated” by marine biologists from NOAA in Woods Hole, Massachusetts,who confirmed the assumptions on the “simple model”, the characteristicshapes of the table functions discussed above (including the decline in birthfraction for low densities), the assumed parameter values and the dynamics ofthe calibrated model.
Currently, the NOAA supports the GCDC proposal. However, as pressureon pelagic stocks increases, regulation will increase as well (though variousdelays will, undoubtedly, occur). This process is captured in the regulationsegment (see Figure 7); again, mackerel is taken as the sample. The government
Fig. 7. Model structure for government regulation
302 System Dynamics Review Volume 20 Number 4 Winter 2004
currently regulates fishing through two mechanisms. First, total allowablecatch (TAC) is a yearly allocation of biomass in MT per region, regulatedthrough a maximum total mortality rate of the older population. Second,specific numbers of “days at sea” are assigned to individual boats in order tocope with the well-discussed social pressures arising from limitations oncommon resources (Hardin 1968). We assume that both policies will be used toregulate pelagic harvesting. These and many other details were hammered outin our meetings with the NOAA.
The NOAA estimates stocks by combining historic data with current yield;the latter is based on “harvesting samples” taken by boats with constant fishingtechnology over time. The actual process is complex and numerous estimatorsexist. It is assumed that the yield curve is a useful proxy for this uncertaintyelement (Table Perceived Effect of Relative Density on MK Yield, with thesame slope as for the table function in inset 3 in Figure 6). For now, we assumethat the NOAA can accurately estimate the RD/Fractional Yield-Curve. Notethat miscalculation of this curve can lead to severe over-fishing.
Samples and measurements are taken quarterly (to understand seasonalfluctuations), while regulation decisions (“Time to Regulate Boat days” and“Time to Regulate TAC”) are made on a yearly basis. These factors, togetherwith existing fleet capacity and factory requirements, will determine the totalnumber of days at sea (MK Fractional Fleet days at Sea). This determinationalso involves a decision delay, yet we assume it to be small relative to the otherdelays and do not incorporate it. If perceived stocks are sufficient, “TotalAllowable Catch” will increase and part of it will be “released” to the fishingfleet outside the protected zone of the United States. This process has beenincorporated into the actual model, but omitted here.
An interesting insight resulted from testing this segment: even when boat-days at sea and total allowable catch are correctly estimated (i.e., decisionrules accord perfectly with stock availability), strong oscillations occur infishing activity. Based on discussions with fishermen, we assumed an aggres-sive rule for determining the current number of days at sea; fishermen willmake sure they harvest their proper share of ground-fish, especially when trustin regulation data is low. Because of decision delays and ongoing correction ofthe TAC, readjustments are pervasive at the end of the fishing season andseasonal fluctuations occur (according to observed data). Yet, however inter-esting or relevant they may be in general, we decided not to take seasonaladjustments (in contrast to the long-term oscillations) into account.
Key dynamics: balancing resources
A key insight for the group as a whole was qualitative understanding of thedynamics of the various resources in combination with harvesting. This under-standing provided a focal point for discussing further modifications to themodel and also led to the structural conceptualization as shown in Figure 8.
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 303
Qualitatively, the structural relations can be captured in the “butterfly stock-flow diagram” of the figure (table functions, normalizing parameters and otherdetails have been omitted).
In this “butterfly” structure we capture the dynamics for ground-fish on theleft and pelagics on the right. The upper sub-structures (R1, R2, B3) on bothright and left represent the resource dynamics discussed above. “Attractive-ness” results from several parameters that depend either directly or indirectlyon the stock: marginal cost (a function of yield per unit of effort plus techno-logy and a normalizing parameter); marginal revenue (a function of yield anda normalizing parameter); Total Allowable Catch; and Boat Days at Sea (asmeasured by the yield). Several delays are involved. Attractiveness deter-mines the swap between ground-fish and pelagic trawlers, which results in aseparate delay; this couples the left and the right parts.
Combining the two resources (allowing for retrofits in the butterfly structure,i.e., converting ground-fish trawlers into pelagic trawlers) makes the situationeven more delicate. In order to make a sensible determination of economicfeasibility and resource allocation policies, several parameters of the variouspopulations must be known. What are the recovery rates for each in times ofdecreased pressure? How long does it take the population to get from R1 backto a stable equilibrium? When we substitute population Y for X, what is thevalue of the substituted population? How large are the several perception andadjustment delays, compared to regeneration time constants? What types of
Fig. 8. Overview of
model structure for
balancing resources
304 System Dynamics Review Volume 20 Number 4 Winter 2004
decision errors are being made? Where are they being made? Can we observeirrational behavior in response to limited or untrustworthy information? Howwill the NOAA respond to changing conditions? What sort of policies will theyissue?
The actual dynamics of the structure, as show in in Figure 8, depend uponvarious time constants and shapes of the net birth rate curves, determining theequilibrium and collapse values at given harvest rates for both species. Thesefactors determined the level of detail chosen in various sectors of the model(pursuant to group discussion). It became clear that if we wanted to think interms of two-source supply chains, policies would need to be adjusted accord-ingly. The effect of increased attractiveness on outside fishermen has not evenbeen broached. Nor has potential resistance from environmentalists seeking tolower the total allowable catch rate nation-wide. As a result of this exercise, werealized that more information was needed pertaining to the basics of theresource dynamics. Questioning the boundaries of the model is one thing; it isinfinitely more difficult to increase the scope of the model.
Results from using the model
In the remainder of the article, we will discuss the results from testing differentpolicies, using the simulation model, as well as the insights we gained fromthe group modeling intervention. Even though client group intervention andmodel building are highly interconnected, we discuss the insights the groupgained separately.
As noted earlier, the purpose of building the model was to ascertainsustainability levels for pelagics under different conditions and to relate fishstock dynamics to the build-up of a Surimi factory. Gloucester could easilyharvest the biomass required to produce the desired output of Surimi; we don’tneed a simulation model to tell us this. But what happens if pelagics become atarget for other Surimi factories that try to imitate Gloucester’s success? Atpresent, the technology for processing Surimi from dark fish is patented. Yet,for how long will this be the case? What if factory trawlers deplete pelagicstocks before the government can intervene? Even if the government andfishermen work together to conserve pelagic stocks, irresponsible fishing byforeign ships could still lead to depletion or extinction.
Conditions for the base run
The relevant time horizon of the model was estimated at 240 months. Thefactory begins operations at 12 months (the time it would require to build thefactory) with sufficient capacity to meet the projected 10,000 MT output. Inother words, the sustainability of the factory, not the start-up process, was ourmain focus. In the base run, we simulated the conditions needed to achieve the
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 305
Fig. 9. Stock levels for
herring and mackerel
under base conditions
Fig. 10. Expectedfinancial performance
of factory under base
conditions
desired factory throughput of 10,000 MT of Surimi per year, which requiresprocessing 30,000 MT of pelagics (about 30 per cent herring and 70 per centmackerel).
The graphs in Figures 9 and 10 demonstrate the results of the base run. Asnoted above, the factory would only need to harvest about 4 per cent of the
306 System Dynamics Review Volume 20 Number 4 Winter 2004
total allowable catch. Thus, it is not surprising that our graphs evince stableequilibrium in both fish stocks. In testing different assumptions to determineinitial conditions, we gained vital information regarding the desired fleetcapacity of pelagic trawlers. In order to achieve the desired throughput, theSurimi factory would need to employ four pelagic trawlers for almost 365 daysa year. Because there are no regulations for fleet days at sea for pelagic trawlers,the four boats could actually be at sea 365 days a year (in theory at least). Thedollar amounts contained in Figure 10 serve only to illustrate the scope of theproject; we do not elaborate upon financial particulars in this article.
Harvest shock
The harvest shock scenarios (see Figures 11 and 12) are based on changingmarket conditions. If, for whatever reason, pelagics become more attractive,other fishing communities could be tempted to tap into this market. We usedhistorical data provided by the NOAA to simulate the outcome of a possibleharvest shock. The data reflect historic landings between 1968 and 1975 whenfactory trawlers put a great deal of pressure on pelagic stocks.
Fig. 11. Stock levels
when the harvest rateis increased
The attempted harvest rates in our simulation are, for herring, an increase to20,000 MT/month from time 36 to 160; and for mackerel an increase to 40,000MT/month from time 36 to 180. At the current population levels these rates areon the order of a total allowable catch of about 300,000 MT/year. Further,while the exogenous pressure is higher for mackerel than for herring, mackerel
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 307
has a higher reproductive potential and, from the discussion of Figure 6, wesee that the indicated rates are below the maximum sustainable yield. This isnot the case for herring. However, these conditions of long suppression do notpush the population beyond the downward spiral, as can be seen from the factthat the herring population remains above the critical point of 5 per cent andthe (slow) recovery after the harvesting shock ceases.
Increased pressure affects herring to a greater extent than mackerel. Mack-erel reproduce more rapidly than herring; thus, mackerel stocks show greaterresistance to increased harvesting. The dynamics in our model suggest that thecontinued pressure from the increased harvest rate would lead to long-termsuppression of the herring population. As a result of a decreasing herringstock, the factory would shift its focus to mackerel, thus keeping the requiredthroughput to meet market demand. However, financial results are suppressed,because of the lower yields.
In testing different conditions, we were able to identify rough limits togrowth for the Surimi factory, in case other fishing communities in Massachu-setts or Maine would follow a successful start-up in Gloucester or harvestingpelagics would gain attractiveness for other reasons. Our client was able to usethe results from the simulation to support the financial assumptions for thebuild-up of the factory and was also able both to understand risks and toencourage trust in the project within the community. The model also encour-aged our client to examine proposed management interventions and helpeddefine acceptable sets of management options carried forward through thepolicy process. Assumptions behind policy changes were made explicit and
Fig. 12. Financial
performance of factory
with increased harvestrates
308 System Dynamics Review Volume 20 Number 4 Winter 2004
were agreed to be necessary subject to additional evaluation and improvement.Modeling also led to systematic identification of information deficienciesthat were addressed by research. Our client also used the simulation models inoutreach and education programs with stakeholders and community members.Together with our client, we facilitated a number of meetings in Gloucester,where we used the model to show cause and effect of different types of policiesfor harvesting fish stocks.
Another insight obtained from the simulation is related to time delays for stockassessment. Specifically, the fact that pelagic stocks are assessed on an annualbasis, which could lead to misperceptions (actual vs. assumed pelagic stocksize, for example) resulting in harvest quotas being set above sustainable levels.
Model validation
Our client was somewhat skeptical at the beginning of the project as to whetherthis methodology would provide valuable insights, and thus the validationprocess had the effect of instilling confidence in the model. We sought toresolve any remaining uncertainty through further discussion with the client.We explained how we calibrated the model, using historic data from theNOAA; this established confidence in structural and behavioral consistency.Sterman (2000), Richardson and Pugh (1981), and Forrester (1961) have allargued that no model can ever truly be validated because every model repre-sents a simplification of reality, not reality itself. However, the goal of modelvalidation in system dynamics is to determine whether a model is appropriatefor a given purpose and whether model users should have confidence in it.This is accomplished through testing and calibration.
Forrester and Senge (1980) describe 17 tests for building confidence in asystem dynamics model. Sterman (2000) offers 12 tests, examining models onboth structural and behavioral grounds. Other tests focus on collaborativemodel building projects that include both modelers and model users. Richardsonand Pugh divide confidence-building tests into those that test for suitabilityand those that test for consistency. Suitability tests determine whether themodel is appropriate for the problem it addresses, while consistency testsexamine whether the model is consistent with the particular aspect of reality itattempts to capture (Richardson and Pugh 1981).
Group members, especially those with a background in mathematics, hadquestioned our initial assumptions regarding “simplicity of structure”. Yet,here too, their fears were allayed by successful calibration; we calibratedresource stocks on the basis of a two-stock model for each type (“YoungStocks” and “Mature Stocks”). We did so in order to align our data with thoseof the NOAA, which possessed detailed information on both stocks and land-ings along with the NOAA’s own parameter estimations for the last 30 years. Inthe end, the two stock models were converted back into a one stock model andrecalibrated for the sake of efficiency.
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 309
The factory, of course, remained the focal point of our analysis. We weresupplied with detailed information (based on a 10,000 MT throughput) frompilot plants in Iceland and Chile. Our assumptions regarding quality and otherscaling factors (not discussed) yielded important operating data for the basecase (10,000 MT) while expansion results (based on throughput estimatesof 20,000 MT and 30,000 MT) turned out to be consistent. As an additionalprecaution, we ran “integrated” sensitivity analysis on critical parameters,using Vensim software capabilities.
Results from the group modeling intervention
While exercising the simulation model helped the client to identify thresholdsand levers to operate the Surimi factory, the group model intervention led toan improved understanding of the complex interactions in this system. Wewere able to visualize a complex system in an easy understandable way, usingcausal loop diagrams, graphs over time, and dynamic hypotheses to capturethe expected hope and fear behavior of the system. In combining the dynamichypothesis with the hope and fear mode of the system in one script, the clientgroup was able to articulate their interpretation and intuition of the systembehavior and challenged the assumptions from the expected hope and fearmodes. These scripts easily communicated with stakeholders and provedto be a valuable tool throughout the initial stage of the project to gain insightsinto the dynamics of the relationship between fishermen, the community andpelagics fish stocks. The reference mode (as shown in Figure 2) led to animportant insight for stakeholders from the fishing community in Gloucester,on how the Surimi factory would compensate for the lack of revenues fromwhite fish.
The project also raises the issue on when and how to formulate the systemboundary, in the case of a “messy problem”. Strictly following the loop-by-loop “standard method” implies keeping flexibility high to make use of insightsgained and knowledge acquired along the process (Otto and Struben 2003).This results in a late formulation of the boundary. Although this formed thecore of our method and we definitely benefited from this perspective in ourproject, we also felt the need for some structure along the way. This resulted,for example, in the “indicated project scope” as shown in Figure 5.
The experience we have gained from this group modeling intervention led tothe conclusion that combining the dynamic hypothesis with a hope and fearscenario, as described in the standard method, in one script provides not onlya check for “meaningful and focused representation”, but also directly pro-vides an opportunity for the client group and stakeholders to obtain importantinsights into the structure and behavior of the system, even before using thesimulation model. This script can be used throughout the process for differentpurposes and thus provides a very powerful addition to those normally used in
310 System Dynamics Review Volume 20 Number 4 Winter 2004
group model building initiatives. The experience also showed that, while theframework of steps and processes guides the team through the process, amodeler still needs a certain amount of intuition to facilitate a group modelinitiative, as indicated in the literature (Andersen and Richardson 1997).Specifically, this flexible process on what to do, when and how leaves room forintuition and, thus, emphasizes the specific learning experience in a modelbuilding initiative. A critical condition for this is a client who is willing tolearn from these steps, rather than from the conclusion.
Discussion
By expanding the boundaries of the model beyond the constraints of theSurimi factory, we helped our client shift the focus of the project while gaininginsights into the volatility of the ecosystem and the levers determining growthand decay of fish stocks. At the end of the project, we were able to answerquestions pertaining to the deterministic constraints of the factory, i.e., theinitial concerns of the client. The model captured the relevant structure of thesystem, thus enabling us to identify harvesting thresholds for pelagics. Localinitiatives, as discussed in this paper, are highly dependent upon regulatorypolicies and changes therein. Therefore, building trust and co-operation amongall stakeholders is critical for success. This corresponds with standard cri-tiques of “authoritative regulation” (Ostrom et al. 1999).
The fact that we were engaged in this project on a voluntary basis may haveinfluenced confidence-building: a professional consulting team will (inten-tionally or not) garner trust based on the image of their firm. Our positionallowed us to act as facilitators without having to deliver “stuff”, thus ensuringan open and relaxed learning environment. Building confidence in the modelitself was bumpy and gradual yet, ultimately, it had a strong organizing andinternalizing impact. In particular the dialog and the discussions that involveddetailed study of the assumptions behind the model and the parameters, themodel itself and the behavior with marine biologists from NOAA in WoodsHole, Massachusetts, helped the client to gain confidence that the dynamicsthat resulted from the model were right for the right reason. On the other hand,we always addressed the simplicity of the model in our discussions withexperts and emphasized that our objective was to gain qualitative insightsfrom the group modeling intervention.
Despite the limited scope and time frame of the project, imperfect data anduncertainties in parameter validation, we were able to help the GloucesterCommunity Development Cooperation address critical questions pertaining tothe Surimi factory. While the model helped us communicate with the client andother community members, it also demonstrated the limitations facing a fishingindustry looking to expand into new markets. In fact, we realized that addingan economic buffer through diversification (increasing the number of resources
P. Otto and J. Struben: Gloucester Fishery: a Group Modeling Intervention 311
available to the community) could lead to severe ecological deterioration.Since understanding that information is most valuable when it is disseminated,we intend to share our insights with other fishing communities and government.Managing a highly complex ecosystem requires a great deal of coordination,allowing diverse stakeholders to share their knowledge and insights.
Notes
1. Andersen and Richardson (1997) suggest that modelers who engage inmodeling with groups rely on fairly sophisticated pieces of small groupprocess, which they call “scripts”.
2. Normal Fractional Birth and Normal Fractional Death are parameters usedto represent the change fractions at carrying capacity in the absence ofharvesting. Since by definition the population is in equilibrium at thisdensity, they must be equal.
Acknowledgements
The authors wish to thank the managing editor and three anonymous referees for theirdetailed suggestions that improved the quality of this article. We also wish to thankNOAA Woods Hole, the GCDC team, in particular Carmine Gorga, Steven Kelleher, JoeSingara, and Damon Cummings, for their support and cooperation. Furthermore, we areindebted to Jim Hines, Bradley Morrison, and George Richardson for guiding us througha group modeling intervention and/or for detailed discussions on the article.
References
Andersen DF, Richardson GP. 1997. Scripts for group model building. System Dynam-ics Review 13(2): 107–129.
Ford DN, Sterman JD. 1998. Expert knowledge elicitation to improve formal and mentalmodels. System Dynamics Review 14(4): 309–340.
Forrester JW. 1961. Industrial Dynamics. Productivity Press: Portland, OR (Now avail-able from Pegasus Communications, Waltham, MA).
Forrester JW, Senge PM. 1980. Tests for building confidence in system dynamicsmodels. In Legasto AA Jr et al. (eds), System Dynamics Vol. 14. North-Holland: NewYork, pp. 209–228.
Gordon HS. 1954. The economic-theory of a common-property resource—the fishery.Journal of Political Economy 62: 124–142.
Hardin G. 1968. The tragedy of the commons. Science 162: 1243–1248.Hines J. 2000. Course material: “Momentum Policies and Rough Dynamic Hypothesis”:
Sloan School of Management, Massachusetts Institute of Technology, 77 Massachu-setts Avenue, Cambridge, MA 02139-4307.
312 System Dynamics Review Volume 20 Number 4 Winter 2004
——. 2001. Course material: “The Standard Method”: Sloan School of Management,Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA02139-4307.
Meadows DL, Behrens WW III, Meadows DH, Naill RF, Randers J, Erich KO. 1974.Dynamics of Growth in a Finite World. Productivity Press: Cambridge, MA (Nowavailable from Pegasus Communications, Waltham, MA).
Morecroft JDW, Sterman JD (eds). 1994. Modeling for Learning Organizations. Product-ivity Press: Portland, OR (Now available from Pegasus Communications, Waltham,MA).
Moxnes, E. 1998. Not only the tragedy of the commons: misperceptions of bioeconomics.Management Science 44(9): 1234–1248.
——. 2000. Not only the tragedy of the commons: misperceptions of feedback andpolicies for sustainable development. System Dynamics Review 16(4): 325–348.
National Marine Fisheries Service (NMFS). 1999. Our Living Oceans. Report on thestatus of U.S. living marine resources. Available at http://spo.nwr.noaa.gov/unit17.pdf[January 2003].
NOAA. 2001. NOAA Technical Memorandum NMFS-NE-115. Status of FisheryResources off the Northeastern United States for 1998, Clark SH (ed). Available atwww.nefsc.noaa.gov/nefsc/publications/tm/tm115/tmttl.htm [15 November 2002].
Odum EP. 1971. Fundamentals of Ecology 3rd edn. Saunders: Philadelphia.Ostrom E, Burger J, Field CB, Norgaard RB, Policansky D. 1999. Sustainability—revisit-
ing the commons: local lessons, global challenges. Science 284(5412): 278–282.Otto P, Struben J. 2003. The “standard method”: scripts for a group model building
intervention. Paper presented at the 21st International Conference of the SystemDynamics Society, New York. (CD-ROM).
Richardson G, Pugh AL III. 1981. Introduction to System Dynamics Modeling withDYNAMO. MIT Press: Cambridge, MA. (Now available from Pegasus Communications,Waltham, MA).
Richardson GP, Andersen DF. 1995. Teamwork in group model building. SystemDynamics Review 11(2): 113–137.
Roberts NH, Andersen DF, Deal RM, Grant MS, Shaffer WA. 1983. Introduction toComputer Simulation: The System Dynamics Modeling Approach. Addison-Wesley:Reading, MA.
Ruth M. 1995. A system dynamics approach to modeling fisheries management is-sues—implications for spatial dynamics and resolution. System Dynamics Review11(3): 233–243.
Safina C. 1994. Where have all the fishes gone? Issues in Science and Technology(Spring): 37–43.
Stenberg L. 1980. A modeling procedure for the public policy scene. In Randers J (ed.),Elements of the System Dynamics Method. Productivity Press: Cambridge, MA,pp. 257–288. (Now available from Pegasus Communications, Waltham, MA).
Sterman J. 2000. Business Dynamics: Systems Thinking and Modeling for a ComplexWorld. Irwin/McGraw-Hill: Boston.
Swart JH. 1990. A system dynamics approach to predator-prey modeling. SystemDynamics Review 6(1) 94–99.
Vennix JAM. 1994. Building consensus in strategic decision-making: insights from theprocess of group model-building. Paper presented at the 1994 International SystemDynamics Conference, Stirling, Scotland.
