MAR 580: Models for Marine Ecosystem Based Management

Lecture 4

Qualitative Modeling

Model Structure Uncertainty

Acknowledgements: Robert Wildermuth, ICES WGNARS members

​

20 September 2022

This week’s outline

• Qualitative modeling for EBM
• Social-Ecological Systems Modeling
• Loop analysis
• Evaluate model structure uncertainty

Challenges of ecosystem (or any) models:�finding the sweet spot

(Collie et al. 2014. Fish & Fisheries)

Bias-variance tradeoff

Challenges of ecosystem (or any) models:�finding the sweet spot

(Collie et al. 2014. Fish & Fisheries)

Bias-variance tradeoff

Can simple be useful?

​

​

​

​

​

​

​

Simple models are easy to interpret AND explain.

Despite not being realistic, are they robust?

Conceptual ecosystem models

Highlight links between species and key ecosystem drivers, components, and goals.

​

Understand how human well-being is affected by changing conditions.

​

NOAA NEFSC

What is most important for your system?

What can you do with a conceptual model?

• Formalize dialogue & discussion about objectives, system linkages.
• Include relationships across disciplines
• Give everyone a common understanding of the system (better graphics)
• Do basic what-if analyses
• Address uncertainty

Why Qualitative models?

• Development of models for complex systems is constrained by
• data availability,
• mechanistic understanding,
• capacity and capability.
• Desire tools for exploration of system structure and function without need for fine detail in functional representations or parameterization.
• Advantages:
• rapidly formulate ideas about system function,
• capture feedback effects in qualitative predictions,
• precursor to inform the development of more detailed quantitative models.

​

ICES Working Group on the� Northwest Atlantic Regional Sea (WGNARS)

Modules built with discipline-specific groups

Human dimensions

Food web

Physical environment

Georges Bank submodels

NOAA NEFSC

Then merged into the working�conceptual model: Georges Bank

All links in the network are documented

Georges Bank Conceptual Model

U.S. Department of Commerce | National Oceanic and Atmospheric Administration | NOAA Fisheries | Page 14

What-if analysis: different model structures

Wildermuth et al 2018. CJFAS.

​

​

​

Council/stakeholder process

Specifies MSE objectives,

Performance measures,

Range of strategies

​

Scientists

develop tools

​

​

​

Council Decision Support:

• Tradeoffs between objectives
• Potential management strategy performance considering
• key interactions
• risks
• uncertainties

S.Gaichas

Why include stakeholders at all?

• Help stakeholders identify objectives
• are the ‘right questions’ are being addressed?
• Facilitate interactions, increased awareness of what tools can(‘t) do.
• Provide stakeholders outside standard management process chance to participate.
• Break down siloing.
• Clarify roles.

Thebaud et al. 2017. ICES JMS.

partial slide credit: André Punt

History of qualitative modeling

• Originally developed by Levins (1974) and Puccia and Levins (1985)
• formalized analysis of feedbacks in network models.
• Broadly applied in ecology since, e.g.
• develop hypotheses for species interactions (Dambacher et al. 1999),
• explore life expectancy changes in perturbed communities (Dambacher et al. 2005),
• determine indicators for exploited systems (Dambacher et al. 2009, Metcalf et al. 2011),
• eradication scenarios for invasive species (Raymond et al. 2011),
• alternative stable states in ecological systems (Marzloff et al. 2011),
• Ecosystem impacts of aquaculture development (Reum et al. 2015a, 2015b)
• Effect of eutrophication & crab management in coastal systems (Carey et al. 2014)
• System effects of fisheries management strategies and climate change (DePiper et al. 2016, Tam et al. 2016, Wildermuth et al. 2016)
• Marine spatial planning for EBM (Wildermuth et al. 2021)
• Socio-Environmental tradeoffs with Offshore Wind (Haraldsson & Niquil 2021)
• Methods for formal assessment of uncertainty in qualitative network models
• Hosack et al. 2008
• Melbourne-Thomas et al. 2012, 2014

​

(Main) Types of Qualitative methods

• Loop analysis
• Extension using simulation and Bayesian Belief Networks
• Fuzzy cognitive mapping

​

• Numerous software packages

e.g.

• R: LoopAnalyst, Qpress
• Mental Modeler (FCM)

​

Qualitative Modeling to address uncertainty

• Only specify the directions and signs of the interactions between system components.
• Seeks qualitative �predictions of the system’s �response to a perturbation.
• Useful for limited or �inequities in data.
• Can be built in real time �with stakeholders.

​

Dambacher et al. 2003

Build the rest! bit.ly/mebm-qualmod

Seabird

Coastal Habitat

Zooplankton

Fish

Fishery

Phytoplankton

Manager

-

-

-

-

-

-

Qualitative

Modeling

Example

Signed digraphs to community matrix

​

​

​

​

​

​

​

​

​

• Dambacher et al. 2003

Effect of species 2 on species 1

All species here have

negative impact on themselves

e.g. density dependence

Interaction terms among species can be hard to estimate, are often result of several processes.

​

Solution: Replace elements of community matrix with symbolic notation of direction of effect.

Model Structure Uncertainty

• Both nature (the signs [positive, negative, or zero]) and strength (i.e., the magnitudes) of a_ij interactions between populations may be uncertain.
• (1) relates to model structure uncertainty,
• (2) can be thought of as a parameterization problem.

​

• Qualitative modeling focuses on (1) by specifying only the signs of the interaction coefficients of community matrix.
• Seeks only qualitative predictions of the system’s response to a perturbation.

Qualitative ‘loop’ analysis

Complementary feedback cycles

• The computations required in the qualitative assessment of system response to perturbation are based upon feedback ‘loops’ or cycles.

​

Stability of the equilibrium is characterized by the eigenvalues of the community matrix;

if all eigenvalues of A have negative real part, the equilibrium is stable.

​

Importantly, the response of the system to a press perturbation, can be determined from the inverse community matrix A.

​

Net feedbacks on system components associated with positive change in columns.

Total # of loops

Adjoint matrix gives response to perturbations

• Frame management actions as press perturbations.
• Loop analysis can be used to strategically examine a foodweb’s response and provide system-wide understanding.

​

• e.g. Carey et al. (2014) added EBM component as an additional variable with interactions on managed species / components.

​

Prediction weights

Prediction weights

Uncertainty in Response & Model Structure

• Qualitative predictions can be ambiguous due to both positive and negative feedback effects.
• Development of ‘‘prediction weights’’ (derived from ratios of feedback cycles).
• Use of simulation, where linkages are assigned random interaction strengths:
• Outcomes used to verify prediction weightings
• Aggregated from multiple random assignments of interaction strengths (as in Raymond et al. 2011).

​

Exercise: case study scenario, scoping

You’re tasked with developing a conceptual model that can be used as the basis of understanding a fishery ecosystem management plan.

Discuss the scenario with the three below questions in mind. Once you have identified the important components and questions, develop a systems diagram and assign directions (signs) to interactions among system components.

• What aspects of the system suggest an ecosystem approach is needed?
• What are the important system components, issues, and factors to consider in a model of the system?
• What are the important things a management plan that takes an ecosystem approach should (or could) consider? What are possible management objectives?

A profitable industrial pelagic fishery targeting a single species has historically had low bycatch rates, high landings, and apparently coexisted with predators of the target species. Over the past few fairly warm years the fishery has stayed within established catch limits, but bycatch rates of other managed fish have increased, some bird and mammal predator (protected species) populations have decreased, and some fish predator populations have increased. Fishermen are concerned that the targeted pelagic fish has been in poor condition recently (reducing its value), and regional managers are concerned about bycatch and declining protected species populations.

Alternatives to model structure uncertainty

• Use a representative set of models that captures alternative versions of our understanding of the interactions within the system.

​

• Each model within the set is structurally unambiguous

​

• Use a single overarching model structure that includes one or more linkages that are marked as ‘‘uncertain.’’
• Can expand to set of unambiguous models by including or excluding uncertain linkages.

Simulation approach�to qualitative modeling

• Melbourne-Thomas et al. (2012)

​

​

• Enables calculation of response�probabilities without enumerating�T matrix.

• here

• Macquarie island network
• Raymond et al. 2011

Challenges

• All-or-none aspect of predictions.
• Can be accommodated using simulation of weights or via Bayesian networking
• Loop algorithm relies upon graphical interpretation of signed digraph models.
• For large or complex systems (i.e. high connectance), signed digraph analysis grows factorially.
• Hence use of simulations to evaluate probabilities of responses.

Take home messages

• Loop analysis can be a useful tool for informing EBM especially in situations with limited data.
• May best be viewed as a screening tool before testing hypotheses with
• more detailed models,
• field or laboratory experiments,
• long-term monitoring.
• Forces simplification – can lead to model predictability and accessibility at the expense of resolution
• Ecosystem simulation models (e.g. EwE) require extensive data on species and interactions.
• comparisons among systems may be limited because similar levels of data are not collected in all systems.
• Predictions from loop analysis can be monitored and used in an adaptive management framework, and suggest research gaps (Lester et al., 2010; Link, 2010).

Additional resources

• Qpress R package
• http://github.com/SWotherspoon/QPress

​

• ‘Loop Group’ (Oregon State Univ)
• http://ipmnet.org/loop/
• Lots of example foodweb matrices
• Java software

​

Questions?