Creating Bayesian Belief Network Models in Ecology

Back to The Bayesian Network Models Page...

Talks on Bayesian Network
and Decision Support Modeling
Bruce G. Marcot
updated 6 November 2023
(a selected list of my presentations on the topic)


2023 -------------------
Burgess, T. I., B. G. Marcot, and P. Scott. 2023. Predicting invasiveness of Phytophthora species at a global scale. International Congress of Plant Pathology, 20-25 August 2023. Lyon, France.

Marcot, B. G. 2023. Moderator, Bayesian network modelling workshop. 3 November 2023 at Bayesian Network Modelling Association Annual Conference: Expanding Horizons - BNs and Beyond, 1-3 November 2023. Brisbane, Australia.

Marcot, B. G. 2023. Discussion panel anel moderator, Session 6: Alternative Methods. 2 November 2023 at Bayesian Network Modelling Association Annual Conference: Expanding Horizons - BNs and Beyond, 1-3 November 2023. Brisbane, Australia.

Marcot, B. G. 2023. Session chair, Session 6: Alternative Methods. 2 November 2023 at Bayesian Network Modelling Association Annual Conference: Expanding Horizons - BNs and Beyond, 1-3 November 2023. Brisbane, Australia.

Marcot, B. G., T. C. Atwood, D. C. Douglas, J. F. Bromaghin, A. M. Pagano, and S. C. Amstrup. 2023. Evolution of Bayesian network models of polar bears under changing Arctic conditions: Key roles of ongoing research, expert knowledge, and uncertainties. Presented 1 November 2023 at Bayesian Network Modelling Association Annual Conference: Expanding Horizons - BNs and Beyond, 1-3 November 2023. Brisbane, Australia.
Abstract:
Updating predictions of the response of high-profile, at-risk species to climate change and anthropogenic stressors is vital for informing effective conservation action. Here, we review the evolution of three generations of Bayesian network probability models from 2007, 2015, and 2023. The models predict changes in global polar bear (Ursus maritimus) population status based on research findings and projections of Arctic sea-ice. We explain how the model has evolved, we compare predictions of polar bear population response from all model generations, and highlight the importance of using contemporary research and climate change projections. We find that polar bears will continue to experience increasing probability of declining or greatly declining populations throughout the 21st century, varying by ecoregion and by climate change projections. Most of the influence, denoted by model sensitivity analysis, is from expected degradation and loss of sea ice access and associated declines in marine prey. Our findings continue to inform priorities for inventory, monitoring, and research needs, and suggest that similar updates to models of other at-risk species can capitalize on the comparison framework we present here.

Marcot, B. G., F. E. Rowland, C. J. Kotalik, J. E. Hinck, and D. M. Walters. 2023. Bayesian networks for assessing natural resource injury: a new framework for a major U.S. national program. Presented 1 November 2023 at Bayesian Network Modelling Association Annual Conference: Expanding Horizons - BNs and Beyond, 1-3 November 2023. Brisbane, Australia.
Abstract:
The U.S. Department of the Interior’s Natural Resource Damage Assessment and Restoration Program (NRDAR) aims to restore natural resources injured by oil spills and hazardous substance releases into the environment. Here, we present a new framework using Bayesian networks (BNs) to evaluate potential natural resource impairment needed to address legal claims, and in particular to estimate the influence and interrelatedness of abiotic and biotic environmental variables on environmental endpoints of interest. To demonstrate the potential procedures and application of a BN for injury assessment and proof of concept of the approach, we used a hypothetical case study by simulating data of acid mine drainage (AMD) affecting a fictional stream-dwelling bird species. The BN evaluated the plausible effects of the AMD outfall on water pH levels and concentration of heavy metals including copper in the stream sediments, and the model was structured with a combination of data and expert knowledge. We simulated spatial sampling of reference and contaminated locations, used an exponential decay model for Cu concentration and a logistic growth curve for water pH, with random error, to predict impacts on bird productivity and population density. We compared the BN-generated probability estimates for injury to a more traditional approach using toxicity thresholds for water and sediment chemistry. We found that advantages of BNs over traditional approaches include the use of expert knowledge, probabilistic estimates of injury using intermediate direct and indirect effects, and the incorporation of a more nuanced and ecologically relevant representation of effects. Our analysis and modeling framework will likely be useful to many U.S. Federal, State, and Tribal programs devoted to the evaluation, mitigation, remediation, and restoration of natural resources injured by releases or spills of contaminants. Our next phase of research is exploring a real-world case involving a Superfund site that has undergone remediation and restoration.

Marcot, B. G., P. Scott, and T. Burgess. 2023. Multivariate Bayesian analysis to identify traits associated with invasiveness of Phytophthora pathogens. 4th International Congress on Biological Invasions, 1-4 May 2023. Christchurch, New Zealand.

2022 -------------------
Marcot, B., P. Scott, and T. Burgess. 2022. Multivariate Bayesian analysis to identify traits associated with invasiveness of Phytophthora pathogens. Presented 21 June 2022 at Phytophthora in Forests and Natural Ecosystems, 10th Meeting of the International Union of Forest Research Organizations (IUFRO), Working Party S07.02.09, 19-25 June 2022 Berkeley CA.
Abstract:
The Phytophthora genus is associated with significant plant diseases in natural ecosystems, and in production and urban environments globally. Phytophthora pathogens pose formidable management and biosecurity challenges as they are increasingly spread globally and often cause major diseases within newly-invaded environments. Many significant new diseases, including Kauri dieback and Sudden Oak death, are caused by species of this pathogen which were only described after the discovery of the disease. Based on the rate of identifying new species, models suggest there may be up to four times more Phytophthora species than are currently described formally. These new species may have serious impacts, even if they are not currently associated with serious diseases, as their potential, yet unknown threat, may jeopardize market access. A multivariate Bayesian traits analysis was conducted to identify traits that may be associated with serious impacts and that can be easily assessed in newly-discovered species. The Bayesian approach proved more informative than logistic modeling approaches, as it was more tolerant of incomplete data, provided risk models and capacity to include management decisions and utility functions, and that proved robust using cross-validation approaches.

Marcot, B. G. 2022. Real world experience of using Bayesian decision networks. Presented 18 November 2022 at University of Sydney School of Public Health's Decision Making Symposium. University of Sydney, Sydney, Australia.
Abstract:
A review is presented on advances in integrated modeling techniques with Bayesian networks. Discussed are how Bayesian networks are being integrated with a variety of other modeling approaches, and expanded in scope, including with: geographic information systems, time-dynamic Bayesian networks, Bayesian decision networks, dynamic decision networks, structural equation modeling, neural networks, object-oriented Bayesian networks, agent-based Bayesian networks, state-and-transition Bayesian networks, quantum Bayesian networks, and more. Future potential applications are discussed, including real-time applications and updating, crowd-sourced Bayesian networks, big-data Bayesian networks, self-organizing Bayesian networks, and more.

Marcot, B. G. 2022. Workshop for students on Bayesian network modeling. 15 November 2022 morning at University of Sydney, Sydney, Australia.
Abstract:
A workshop is held with attending doctoral graduate students of University of Sydney who are working with a probability and Bayesian causal and correlational networks, to review their progress and provide guidance and suggestions.
Marcot, B. G., and 5_others. 2022. Panel discussion. 18 November 2022 at University of Sydney School of Public Health's Decision Making Symposium. University of Sydney, Sydney, Australia.
Abstract:
A free-form discussion is provided a panel of speakers on a variety of topics pertaining to recent advances in quantitative aspects of risk analysis and decision science.
Marcot, B. G. 2022. Use of Bayesian decision networks for modeling risk analysis. Presented October 2022, US Geological Survey, Natural Resource Damage Assessment and Remediation Program, Columbia, Missouri. [invited].
Marcot, B. G., and M. Ormsby. 2022. Applying decision science in natural resource management: concepts and tools. Presented 29 September 2022, National Archives and Records Administration, College Park, Maryland. [invited].
Marcot, B. G., and T. Penman. 2022. The future for Bayesian network modelling. Presented 17 November 2022, Panel Discussion. Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 16-17 November 2022. Sydney, Australia.
Abstract:
A review is presented on advances in integrated modeling techniques with Bayesian networks. Discussed are how Bayesian networks are being integrated with a variety of other modeling approaches, and expanded in scope, including with: geographic information systems, time-dynamic Bayesian networks, Bayesian decision networks, dynamic decision networks, structural equation modeling, neural networks, object-oriented Bayesian networks, agent-based Bayesian networks, state-and-transition Bayesian networks, quantum Bayesian networks, and more. Future potential applications are discussed, including real-time applications and updating, crowd-sourced Bayesian networks, big-data Bayesian networks, self-organizing Bayesian networks, and more.
Marcot, B. G., P. Scott, and T. Burgess. 2022. Predicting invasiveness of a global pathogen. Presented 17 November 2022. In: Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 16-17 November 2022. Sydney, Australia.
Abstract:
The Phytophthora genus is associated with significant plant diseases in natural ecosystems, and in production and urban environments globally. Phytophthora pathogens pose formidable biosecurity challenges as they are increasingly spread globally and often cause major disease within newly invaded environments. Many significant new diseases, including Kauri dieback and sudden oak death, are caused by species which were identified only after the discovery of the disease. Based on the rate of identifying new species, models suggest there may be up to four times more Phytophthora species than are currently described. These new species may have serious impacts, even if they are not currently associated with serious diseases. Therefore, it is difficult to determine the management requirements of these new species when they are initially identified. We conducted a multivariate Bayesian analysis to determine if biological traits easily measured within the laboratory can usefully predict the impact of these newly identified species. Our Bayesian network model correctly predicts the known invasion risk of Phytophthora species and can effectively be used for new species to warn of their potential invasiveness. This approach can be used to develop risk models for other genera of plant pathogens.

Marcot, B. G., K. M. Thorne, J. A. Carr, and G. R. Guntenspergen. 2022. Predicting effects of climate change and sea-level rise on wetlands of the Pacific Coast, USA. Presented 16 November 2022. Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 16-17 November 2022. Sydney, Australia.
Abstract:
As climate change causes a rise in sea level, some of the most threatened ecosystems are coastal tidal saline wetlands (TSWs), including marshes and estuaries. Studies generally suggest significant losses over coming decades, but what are needed are projections of site-specific TSWs, particularly along the Pacific Coast USA. We present a Bayesian network (BN) model that we developed based on literature review and direct research experience, and parameterized from on a database we compiled of 26 sample TSWs, to predict changes (declines) in the resilience of specific TSWs to projected sea-level rise. We modeled resilience in two dimensions along wetland vertical elevation capital and wetland lateral migration capital. We found that all sites would lose at least 50% of their elevation capital resilience between 2060 and 2100, and 100% by 2070 to 2130, depending on the site. Under a 1.5-m sea-level rise scenario, nearly all sites in California will lose most or all of their lateral migration resilience. Our model can be useful in risk analysis and risk management to prioritize sites needing more immediate conservation action.

Mascaro, S., and O. Woodberry (B. G. Marcot, contributor). 2022. Introduction to Bayesian network modeling workshop. 15 November 2022 afternoon at University of Sydney, Sydney, Australia.

2021 -------------------
Marcot, B. G. 2021. Basics of Bayesian network modeling. Presented 24 June 2021 to: National Archives and Records Administration, Maryland [Virtual on-line presentation].

Marcot, B. G. 2021. Efficacy of Bayesian network modelling for capturing the complexity of the real world. Debate Panel, 17 November 2021 at: Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 16-17 November 2021, held virtually as video-conference presentations.

Marcot, B. G. 2021. Initial explorations of machine self-learning with generative adversarial Bayesian networks. Presented 17 November 2021 at: Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 16-17 November 2021, held virtually as video-conference presentations. In.
Abstract:
A popular A.I. machine-learning algorithm, known as generative adversarial networks (GANs), is used with neural network programming. BN modelling lends well to the self-learning and -updating procedures of GANs in the Bayesian context. This talk presents an initial framework, with an example, for using the approach in the context of Bayesian network (BN) modelling, here calling it generative adversarial Bayesian networks (GABNs.).

Marcot, B. G. 2021. The many faces of probability. Australasian Bayesian Network Modelling Society Webinar Series, presented 1 September 2021. https://www.youtube.com/watch?v=quIWe0bmuuA [Invited].

Meurisse, N., B. G. Marcot, O. Woodberry, B. Barratt, and J. Todd. 2021. BAIPA: A new ecologically-based, probabilistic risk assessment tool to support risk assessment for biological control agents. 16 November 2021 at: Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 16-17 November 2021, held virtually as video-conference presentations.

2020 -------------------
Marcot, B. G. 2020. To all the Bayesian Networks I’ve loved before. Presented 9 December 2020 at: Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 9-10 December 2020, held virtually as video-conference presentations. http://www.abnms.org/conferences/abnms2020/program?type=conference.
Abstract:
Like a living organism, Bayesian network (BN) modeling has evolved over time to become increasingly refined in its structures and functions, to grow new applications, and to become increasingly symbiotic with other fields of study. In this talk, I recount the way that my own work with BN models has progressed and matured along a set of 7 modeling-consideration themes of: the basis of Bayesian thinking, increased scientific rigor, ties to network theory, the value of uncertainty in decision advisories, methods of calibration and validation, emergence of machine learning, and integration with other modeling constructs. I briefly display and explain operational and applied examples or two for each theme, and suggest to where the field may evolve to next.

Marcot, B. G. 2020. Panelist, Discussion Panel, 10 December 2020 at: Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 9-10 December 2020, held virtually as video-conference presentations. http://www.abnms.org/conferences/abnms2020/program?type=conference.

Marcot, B. G. 2020. Session Chair, "Environmental Applications II", 9 December 2020 at: Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS), 9-10 December 2020, held virtually as video-conference presentations. http://www.abnms.org/conferences/abnms2020/program?type=conference.

McColl-Gausden, S., L. T. Bennett, D. Penman, B. G. Marcot, J. B. Fontaine, and T. Penman. 2020. Predicting extinction risk of fire-dependent plants under changing climate and fire regimes. Ecological Society of America Annual Meeting, 2-7 August 2020, Salt Lake City, Utah USA. pp.
Abstract:
Background/Question/Methods: Much of the world’s biodiversity is intimately linked to fire regimes. However, fire-dependent plants often require fire-free intervals to establish and grow to reproductive maturity. Warmer drier climates like those predicted for many temperate regions are likely to reduce rates of plant growth and reproduction and increase fire frequency. The “interval squeeze” hypothesis predicts the interaction of demographic change with more frequent fire could narrow the interval for reproduction and survival, thereby increasing the likelihood of local or regional extinction.
This research has two key questions: (1) What are the relative effects of time since fire, climate change and their interactions on the reproductive potential and productivity of different obligate seeder species? (2) Which groups of plant species are more at risk of extinction under future climate and fire regimes?
We use field measures of plant size and reproductive potential collected over the last 20 years in Mediterranean woodland and temperate forest ecosystems. Sites encompassed a chrono-sequence of time-since-wildfire and a rainfall gradient. We combined field data with demographic traits to conduct a population viability analysis using continuous Bayesian Networks, allowing us to predict how changing climates and fire regimes may impact populations across a plant species’ range.
Results/Conclusions: Climatic change alone presented a significant risk to populations of all of the obligate-seeder species. Increased aridity lowered reproductive rates such that populations were often no longer viable under current fire regimes. Some of these eàects were mediated by climate variability which increased over time, but this effect was relatively minor. Similarly, changed fire regimes resulted in significant population declines regardless of climatic shifts. Species that successfully recruited between fires were more resilient to climate change but only if the average inter-fire interval exceeded the age to reproductive maturity.
Climate change was the greatest risk to obligate seeder persistence with shifts in fire regimes exacerbating climate eàects. While we did not examine climate-fire impacts on resprouting species, we would expect that these species would be less likely to become extinct under future climates. Our modelled point-scale changes will not necessarily be indicative of population persistence across landscapes. Linking PVA models with spatial landscape-scale fire models will improve our estimation of population persistence and the ability of management actions to reduce extinction risk.


2019 -------------------

Marcot, B. G. 2019. Dilemmas in dealing with peer reviews in publishing Bayesian networks. Presented 14 November 2019 at: Joint Conference on Risk and Decision-Making. Society for Risk Analysis Australia and New Zealand (SRA-ANZ) with Australasian Bayesian Network Modelling Society (ABNMS). 13-14 November 2019, Rutherford House, Victoria University of Wellington, New Zealand.
Abstract:
Publishing articles on concepts and applications of Bayesian network (BN) modeling can be met with editorial inconsistencies and misunderstandings of the approach, particularly when submitting manuscripts for publication in specialty fields other than in statistics and mathematical modeling. Journal and book editors generally rely on a blind peer-review process to determine the publication worthiness of submitted manuscripts. However, subject-matter editors and their selected peer reviewers often misunderstand the assumptions, mechanics, and appropriate interpretation of BN-based analyses and models, and often conflate those with traditional frequentist and parametric statistical approaches, thereby often rejecting submissions for publication. The more common errors made by peer reviewers include: not understanding that BN models are essentially first-order Markovian processes and are not multivariate models of dimension-reduction of variables such as with factor analysis frequentist methods; requiring greater sample sizes more aligned with frequentist multivariate methods; assuming that BN modeling also must adhere to strict assumptions of multivariate normality and of avoiding homoscedasticity of covariates, as required in frequentist approaches; confusing classification of BN response variable states with frequentist ordination and discriminant analysis methods; assuming that BN models are essentially based on expert opinion; misunderstanding the appropriate roles of prior and uniform probability distributions in dealing with uncertainty and missing data; misunderstanding the form and purpose of naive Bayes model structures; lack of experience with model performance metrics such as logarithmic loss, quadratic loss, and spherical payoff; and misunderstanding the form and role of BN sensitivity analysis, particularly as indexed by measures of variance- and entropy-reduction. I suggest fixes for each of these misconceptions by clarifying BN modeling assumptions and approaches in the methods sections of manuscripts, particularly for non-statistical audiences. The good news is that application of BN models is becoming more mainstream, so that competency and experience in the peer review and editorial processes may be increasing. I base this talk on my > 25 years of BN modeling experience and publishing over 30 peer-reviewed journal articles, book chapters, and a co-edited textbook on BN modeling concepts and applications.

Marcot, B. G. 2019. Risk analysis and risk management in natural resource conservation in the U.S. government. Presentation made, with Bayesian network modeling workshop, 15 November 2019 to Ministry for Primary Industries, Wellington, New Zealand. [Invited].
Abstract:
Covered will be a brief review of regulatory mandates under (1) the National Environmental Policy Act (NEPA), Environmental Protection Agency; (2) the Planning Rule under the National Forest Management Act (NFMA), U.S. Forest Service; and (3) the Endangered Species Act (ESA), U.S. Fish and Wildlife Service. Informing these mandates is a suite of approaches to structured decision-making, such as use of the PROACT (problem, objectives, alternatives, consequences, trade-offs) approach. I present several case studies to illustrate modeling of risk analysis and risk management for natural resource conservation across these regulatory mandates, including an ongoing NEPA analysis of engineering access to an outflow tunnel at Mount St. Helens, listing and analyzing viability of Species of Conservation Concern under NFMA, and monitoring a threatened coastal shorebird under ESA.

Marcot, B. G. 2019. Risk analysis and structured decision making in U.S. Forest Service Research and Management. Presented 20 October 2019 at session "Applications of Decision Analysis for Federal, State, Provincial, and Tribal Natural Resource Management" for The Decision Analysis Society (DAS), INFORMS Conference 20-23 October 2019, Seattle, WA.
Abstract:
USFS has been applying risk analysis and decision science to a wide variety of research and management issues, including: monitoring and threats assessments of federally-listed and at-risk species, impacts of engineering projects on research priorities, and analysis of potential injuriousness of introduced and invasive species. I review the SDM framework and the suite of tools useful for each stage in the SDM process, and suggest topics of study to aid applying SDM to natural resource risk management in a changing world.

Marcot, B. G., M. H. Hoff, C. D. Martin, S. D. Jewell, and C. E. Givens. 2019. Catching them early: A decision support system to predict invasive and injurious fish. Presented 14 November 2019 at: Joint Conference on Risk and Decision-Making. Society for Risk Analysis Australia and New Zealand (SRA-ANZ) with Australasian Bayesian Network Modelling Society (ABNMS). 13-14 November 2019, Rutherford House, Victoria University of Wellington, New Zealand.
Abstract:
We developed a decision-support, risk-assessment system to aid identification of potentially invasive and injurious fish species. Our system consists of a semi-quantitative, rapid-assessment procedure called the Ecological Risk Screening Summary (ERSS) and a quantitative Bayesian probability network model called the Freshwater Fish Invasive Species Risk Assessment Model (FISRAM). ERSS provides information on a species’ invasiveness history elsewhere in the world and on its biology and ecology, potential or known effects of introduction, global and domestic distribution, and climate associations, and provides conclusions on potential risk of invasiveness. FISRAM is used to assess risk probability when ERSS categorizes invasion risk as uncertain. FISRAM calculates expected probability of invasiveness as a function of species potential establishment, spread, and harm, based on probable effects on native species and ecosystems, suitability of climate and habitat in introduced areas, ease of dispersal and transport, and harm to humans. We developed both models with peer review, and calibrated and updated the probability structure of FISRAM using a data set of 50 species with known invasiveness outcomes. We demonstrate the use of these two tools for risk assessment and decision-support in identifying and documenting invasive species for potential risk management actions, such as listing wildlife species under the U.S. Lacey Act as potentially injurious and to be prohibited for import.

McColl-Gausden, S., B. G. Marcot, J. Fontaine, and T. Penman. 2019. A tale of two factors: Extinction risk of fire dependent plants under changing climates and fire regimes. Presented 14 November 2019 at: Joint Conference on Risk and Decision-Making. Society for Risk Analysis Australia and New Zealand (SRA-ANZ) with Australasian Bayesian Network Modelling Society (ABNMS). 13-14 November 2019, Rutherford House, Victoria University of Wellington, New Zealand.
Abstract:
Australia’s biodiversity is intimately linked to fire regimes. However, fire-dependent plants require fire-free intervals to establish and grow to reproductive maturity. Such species face many challenges under future climates. A warmer drier climate is likely to reduce rates of growth and reproduction, while fire frequency or intensity is known and predicted to increase. The “interval squeeze” hypothesis predicts the interaction of demographic change with more frequent fire could narrow the interval for reproduction and survival, leading to increased likelihood of local or regional extinction. Here we use a continuous Bayesian network to test the interval squeeze hypothesis on serotinous obligate seeding species. This approach allows us to examine the impact of a drier, more fire-prone system on a single plant population. We found the optimal fire interval varied between the current climate and a drier future climate. However, these results need to be spatialised to understand the context for population viability across landscapes. We plan to do this by coupling the BN with a geographically-based fire regime model, rather than with just probabilistic fire intervals. We will then use this technique to better identify areas where population level extinctions are more likely in the future due to large changes in the fire regime coupled with a drier climate.

Penman, T., B. G. Marcot, S. McColl-Gausden, and D. Ababei. 2019. Modelling population viability using Bayesian Networks. Presented 14 November 2019 at: Joint Conference on Risk and Decision-Making. Society for Risk Analysis Australia and New Zealand (SRA-ANZ) with Australasian Bayesian Network Modelling Society (ABNMS). 13-14 November 2019, Rutherford House, Victoria University of Wellington, New Zealand.
Abstract:
Conservation managers need to estimate the return for investment actions across the life cycle of each species. Population viability analysis (PVA) is a useful tool for the quantitative projection of a biological population under scenarios of specified survivorship and reproduction vital rates, to determine probabilities of extinction over a specified time horizon. At each time step (usually one year), PVA modeling considers development, reproduction, and survivorship of each stage class. A key challenge is with complex life stages, for example a frog may be an egg, tadpole, metamorphling and juvenile within a single year. Traditional PVA approaches would usually merge survival rates across these four stages to a single annual figure. This means that it is impossible to compare management actions that are of benefit or detriment to individual life stages.
Bayesian network (BN) modelling holds promise for addressing these considerations in an efficient manner. The structure of BNs is flexible and lends to modelling the population stage classes to track cohort strength over time, mimicking and extending traditional PVA constructs. We modelled population dynamics for three species using traditional methods and continuous Bayesian Networks. BN models were able to replicate the key values out of a PVA such as population size and time to extinction for the simple models. BN models were also able to consider species with complex life stages and identify the sensitivity of the model to changes in these values. We argue that these models provide greater flexibility than traditional PVA approaches for species with complex life stages.

2018 -------------------

Hanea, A., V. Hemming, B. G. Marcot, and A. Christophersen. 2018. The use and abuse of expert judgement. Panel 7 December 2018 presentation at the 10th Annual Conference of the Australasian Bayesian Network Modelling Society, Adelaide, Australia.
Abstract:
     Environmental decision-making requires an understanding of complex interacting systems across scales of space and time. Few, if any, systems have perfect knowledge and managers are required to make decisions in the face of high uncertainty. One construct useful for addressing uncertainties in environmental decision-making that can provide an intuitive and relatively simple structure is that of Bayesian decision networks (BDNs). In this paper, we provide a BDN approach to wildfire management and decision-making, an area typically wrought with uncertainties in the effects of wildfire, changing environmental conditions, fire control, fuels management, and related activities, on biological, social, cultural, economic, and future conditions. The study was set in the east central highlands of Victoria (~950,000ha) within and to the north east of the city of Melbourne in south-eastern Australia. We developed a BDN model to examine the risk of trade-offs for varying prescribed burning effort in the landscape and at the interface across four assets considered in the study: houses, powerlines, carbon loss and biodiversity conservation. Asset loss values showed differing responses to the prescribed burning treatment combinations. The lowest costs were seen in treatments with 15% of edges treated per annum and 0-3% of the landscape treated per annum, but these are unlikely to be socially acceptable. Treating landscapes with either 2 or 3 % per annum and 0-5% of edges per annum were the next most effective treatment options. These options had lower risk values for houses which are considered the priority asset for fire management. In this case study, BDN provided a useful tool for environmental decision making by explicitly calculating expected values of costs and benefits of alternative decisions, and clearly depicting the implications of uncertainties of parameters and decision utilities.

Hanea, A., B. G. Marcot, T. Penman, and S. Mascaro. 2018. The many faces of Bayesian networks. Panel 6 December 2018 presentation at the 10th Annual Conference of the Australasian Bayesian Network Modelling Society, Adelaide, Australia.
Abstract:
     Environmental decision-making requires an understanding of complex interacting systems across scales of space and time. Few, if any, systems have perfect knowledge and managers are required to make decisions in the face of high uncertainty. One construct useful for addressing uncertainties in environmental decision-making that can provide an intuitive and relatively simple structure is that of Bayesian decision networks (BDNs). In this paper, we provide a BDN approach to wildfire management and decision-making, an area typically wrought with uncertainties in the effects of wildfire, changing environmental conditions, fire control, fuels management, and related activities, on biological, social, cultural, economic, and future conditions. The study was set in the east central highlands of Victoria (~950,000ha) within and to the north east of the city of Melbourne in south-eastern Australia. We developed a BDN model to examine the risk of trade-offs for varying prescribed burning effort in the landscape and at the interface across four assets considered in the study: houses, powerlines, carbon loss and biodiversity conservation. Asset loss values showed differing responses to the prescribed burning treatment combinations. The lowest costs were seen in treatments with 15% of edges treated per annum and 0-3% of the landscape treated per annum, but these are unlikely to be socially acceptable. Treating landscapes with either 2 or 3 % per annum and 0-5% of edges per annum were the next most effective treatment options. These options had lower risk values for houses which are considered the priority asset for fire management. In this case study, BDN provided a useful tool for environmental decision making by explicitly calculating expected values of costs and benefits of alternative decisions, and clearly depicting the implications of uncertainties of parameters and decision utilities.

Hanea, A. M., and B. G. Marcot. 2018. What is an optimal value of k in k-fold cross-validation in discrete Bayesian network analysis? Presented 6 December 2018 at the 10th Annual Conference of the Australasian Bayesian Network Modelling Society, Adelaide, Australia.
Abstract:
     One of the important steps in model-building is ensuring the credibility of the model. Validation entails testing a model against an independent data set not used to initially parameterize the model. One such procedure uses k-fold cross-validation The question we address here is, what is the best value of k to help ensure optimal model evaluation? Also, is there a lowest value of k, for least computation time, for which bias, variance, and model accuracy might stabilize? These questions have been largely ignored in the literature, particularly with discrete Bayesian networks (BNs). This study focuses on k-fold cross-validation with discrete BN models. We expect the best selection of k for a given data set to depend on several attributes of the data such as the degree of collinearity and variability, and attributes of the constructed model, such as the degree of data discretization and the presence and form of interaction terms among the covariates. For this project, we develop a BN model and sample an artificial data set from it. In this way, we can fix the BN structure, and control for the parameter uncertainties. We create 6 model variants to account for 3 different degrees of variation of the values of the variables and 2 different multivariate dependence strengths. We vary the data set size, and the value of k and for each fold we fit parameters using the maximum likelihood estimation procedure. We then use 3 measures of model performance and investigate their relationship with k. We will present these relationships and comment on the appropriateness of the conclusions for a real-life application modelled with a BN sharing similar properties with the ones used in the artificial example.

Marcot, B. G. 2018. EcoQBNs: Toward a new framework of ecological quantum Bayesian networks. Presented 7 December 2018 at the 10th Annual Conference of the Australasian Bayesian Network Modelling Society, Adelaide, Australia.
Abstract:
     A recent advancement in modeling is development of quantum Bayesian networks (QBNs). QBNs generally differ from BNs by substituting traditional Bayes calculus in probability tables with quantum amplification wave functions. QBNs can solve a variety of problems that are unsolvable by, or are too complex for, traditional BNs. These include problems: with feedback loops and temporal expansions; with non-commutative dependencies where the order of specifying priors affects posterior outcomes; with intransitive dependencies constituting circular dominance of outcomes; where input variables can affect each other even if not causally linked (entanglement); where there may be >1 dominant probability outcome dependent on small variations in inputs (superpositioning); and where outcomes are nonintuitive and defy traditional probability calculus (Parrondo's paradox and violation of the Sure Thing Principle). I present simple examples of these situations illustrating problems in prediction and diagnosis, and I demonstrate how BN solutions are infeasible or at best require overly-complex latent variable structures. I then argue that many problems in ecology and evolution can be better depicted with ecological QBN (EcoQBN) modeling. I present such situations that fit these kinds of problems, such as noncommutative and intransitive ecosystems responding to suites of disturbance regimes with no specific or single climax condition, or that respond differently depending on the specific sequence of the disturbances (priors); entrainment of signals among duetting birds, flashing fireflies, chirping crickets, and predator-prey cycles; and mutual conditioning in the evolution of species mimicry systems. I argue that many current ecological analysis structures such as state-and-transition models, individual-based movement simulations, and ecological disturbance models could greatly benefit from a QBN approach. I conclude with presenting EcoQBNs as a nascent field needing much development of the quantum mathematical structures and, eventually, either adjuncts to existing BN modeling shells or entirely new software programs to facilitate model development and application.

Marcot, B. G. 2018. Eliciting expert knowledge for Bayesian network models. Presented 12 September 2018, to Norwegian Polar Institute, Tromsø, Norway.
Abstract:
     I discuss the role of expert knowledge in constructing and using Bayesian network models, and rigorous procedures for holding expert panels to elicit knowledge for use in modeling. I also discuss the value and use of local and traditional knowledge for model use and validation.

Marcot, B. G. 2018. Use of Bayesian networks to model plants and wildlife for conservation planning under uncertainty. Presented 10 September 2018, to Norwegian Polar Institute, Tromsø, Norway.
Abstract:
     I cover the basics of the structure of Bayesian network models and present examples of their use in a variety of wildlife and resource management analyses. I discuss the use in decision modeling for depicting knowedge and for interpreting uncertainty.

Marcot, B. G., and T. D. Penman. 2018. Integration of Bayesian network modelling technologies: a review. Presented 6 December 2018 at the 10th Annual Conference of the Australasian Bayesian Network Modelling Society, Adelaide, Australia.
Abstract:
     Bayesian networks (BNs) are becoming integrated into a suite of other technologies and model frameworks. We review this widening field of BN applications. Advances include algorithms for Bayesian classification; machine-learning for structuring networks and parameterizing probability functions; and time-expansion of networks to avoid path cycles. BNs are being applied to management and decision science, and used in complement with structured equation modelling and neural networks for identifying causality. Other areas of integration include object-oriented and agent-based modelling; state-and-transition models; geographic information system mapping; and calculations of quantum probability. Integrated BNs (IBNs) are being used increasingly in areas of risk analysis, risk management, and structured decision-making, for a variety of applications in resource planning and environmental management. We foresee that IBNs will become increasingly self-structuring and self-learning from information sources of real-time monitoring, big data, and social media. This near-future expansion of IBNs holds great promise but also great difficulty for model validation and ensuring model credibility, particularly when based on uncertain or undisclosed knowledge sources and systems.

Penman, T. D., B. Cirulis, and B. G. Marcot. 2018. Advances in development and application of Bayesian decision network modeling: a wildfire management case study. Presented 6 December 2018 at the 10th Annual Conference of the Australasian Bayesian Network Modelling Society, Adelaide, Australia.
Abstract:
     Environmental decision-making requires an understanding of complex interacting systems across scales of space and time. Few, if any, systems have perfect knowledge and managers are required to make decisions in the face of high uncertainty. One construct useful for addressing uncertainties in environmental decision-making that can provide an intuitive and relatively simple structure is that of Bayesian decision networks (BDNs). In this paper, we provide a BDN approach to wildfire management and decision-making, an area typically wrought with uncertainties in the effects of wildfire, changing environmental conditions, fire control, fuels management, and related activities, on biological, social, cultural, economic, and future conditions. The study was set in the east central highlands of Victoria (~950,000ha) within and to the north east of the city of Melbourne in south-eastern Australia. We developed a BDN model to examine the risk of trade-offs for varying prescribed burning effort in the landscape and at the interface across four assets considered in the study: houses, powerlines, carbon loss and biodiversity conservation. Asset loss values showed differing responses to the prescribed burning treatment combinations. The lowest costs were seen in treatments with 15% of edges treated per annum and 0-3% of the landscape treated per annum, but these are unlikely to be socially acceptable. Treating landscapes with either 2 or 3 % per annum and 0-5% of edges per annum were the next most effective treatment options. These options had lower risk values for houses which are considered the priority asset for fire management. In this case study, BDN provided a useful tool for environmental decision making by explicitly calculating expected values of costs and benefits of alternative decisions, and clearly depicting the implications of uncertainties of parameters and decision utilities.


2017 -------------------
Marcot, B. G. 2017. Experto crede: crafting Bayesian networks from expert knowledge. Presented 21 November 2017 at the combined meeting of the Tenth Annual Conference of the Australasian Bayesian Network Modelling Society and the Annual Conference of the Society for Risk Analysis, University of Melbourne, Melbourne, Australia. [Invited keynote address].
Abstract:
     This talk explores how expert elicitation and expert paneling are, and can be, used to devise and test Bayesian network models. Discussed are step-wise procedures, "tips and tricks," cautions and caveats, and ideas for future research on the role of expert knowledge in devising and using Bayesian network models and Bayesian decision network models. Specific examples and a general framework are presented.
Marcot, B. G. 2017. A decision support system for identifying potentially invasive and injurious freshwater fishes. Presented 30 November 2017 to Scion Forestry Crown Research Institute, Research Seminar, Rotorua, New Zealand [invited presentation]. Abstract:
     An increasing threat to U.S. waterways is the establishment and spread of invasive and injurious fishes. A species may be designated by the U.S. Fish and Wildlife Service as federally injurious under the national Lacey Act, on the basis of a species either causing harm without establishing and spreading, or causing harm and establishing and spreading. Species designated as injurious are prohibited from being imported, which is a highly effective way of preventing invasions by nonnative species. We developed a decision-support, risk-assessment system to predict a species’ potential injuriousness. The system -- consisting of a rapid assessment and a more detailed Bayesian network model -- calculates the expected probability of injuriousness as a function of species potential establishment, spread, and harm, based on probable effects on native species and ecosystems, suitability of climate and habitat in introduced areas, ease of dispersal and transport, and harm to humans. Potentially injurious invasive species can then be included in listings as injurious under the Lacey Act to prohibit their import.

2016 -------------------
Fortin, J. K., K. D. Rode, G. V. Hilderbrand, J. Wilder, S. Farley, C. Jorgensen, and B. G. Marcot. 2016. The impacts of human recreation on brown bears (Ursus arctos): a review and new management tool. Presentation at: 24th International Conference on Bear Research and Management, Anchorage, Alaska.
Marcot, B. G. 2016. Bayesian network modeling workshop. 12 June 2016 at 24th International Conference on Bear Research and Management, Anchorage, Alaska.
Marcot, B. G. 2016. Building a common vocabulary for modeling. 28 June 2016, Workshop on Restoration Prioritization Models, Tools, Frameworks, and Products, US Forest Service, Pacific Northwest Region and Pacific Northwest Research Station, Portland, Oregon.
Penman, T., and B. G. Marcot. 2016. Single Bayesian network society seeking same. Poster presented at The International Society for Ecological Modelling Global Conference, 8-12 May 2016. Towson University, MD, USA.
Abstract:
     The Australasian Bayesian Network Modelling Society (ABNMS) established in 2009 to bring together Bayesian Network researchers and practitioners (www.abnms.org). Its purpose is to promote the understanding and use of Bayesian Network models in scientific, industrial and research applications. It aims to provide opportunities for modelers to exchange ideas and socialize by organizing conferences and regular events, as well as maintaining mailing lists and online forums. Our society holds annual conferences and training courses. In 2014, Bruce Marcot was our invited guest at the annual conference in Rotorua, New Zealand. During the course of this conference, the potential for linkages with researchers in North America was recognised. Our society is keen to establish links with the diversity of Bayesian Network modellers in North America. We are seeking interested individuals or research groups to get involved in discussions about Bayesian Network modelling. Initially, we are keen to talk to researchers and practitioners who would be interested in participating in online forums, but we are also scoping interest in the establishment of a North American society that could share conferences and experiences with members of ABNMS. If you are interested, come along and talk to us and let’s see what we can do.

2015 -------------------
Marcot, B. G. 2015. Application of new tools, uncertainty, and risk in species assessment and conservation planning. Presented 19 November 2015 at Ecological Process and Function Program Review, USDA Forest Service, Corvallis, Oregon.

Marcot, B. G. 2015. Basics of Bayesian network modeling. Presented 7 December 2015 at National Institute of Water and Atmospheric Research Crown Research Institute, Hamilton, New Zealand.

Marcot, B. G. 2015. Bayesian networks and expert elicitation. Presented 1 December 2015 at Scion Crown Research Institute, Rotorua, New Zealand.

Marcot, B. G. 2015. Bayesian networks and expert paneling: tools for conservation planning under uncertainty. Presented 26 November 2015 at Centre of Excellence for Biosecurity Risk Analysis, University of Melbourne, Australia.

Marcot, B. G. 2015. Common quandaries and their practical solutions in Bayesian network modeling. Presented 23 November 2015 at the Seventh Annual Conference of the Australasian Bayesian Network Modelling Society, Monash University, Melbourne, Australia. [Invited keynote address].
Abstract:
     The ease with which Bayesian network (BN) models can be built had led to some common problems in their construction and interpretation. In this keynote, I address common quandaries and their solutions in BN modeling. Common problems of BN model construction include: assuming that every conditional probability table (CPT) must span [0,100]; use of vague node names and unmeasurable node states; too many parent nodes and little consideration for latent variables; ignoring outlier CPT values; not linking correlated covariates; no peer review of expert-structured BNs; no tests of model calibration or validation; confusing calibration with validation; and "holes" left in CPTs from machine-learning algorithms. Common problems of BN model interpretation include: unclear initial objectives and "model creep;" assuming sensitivity structure from model depth; conflating correlation with causation; conflating proportion with probability; and conflating expectation with probability. Solutions to these various problems pertain to learning correct BN model construction techniques, understanding basics of Bayesian statistical inference, experience with expert knowledge elicitation and necessity of expert peer review and reconciliation, and just being clear with model objectives and variables.
     Published as: Marcot, B. G. 2017. Common quandaries and their practical solutions in Bayesian network modeling. Ecological Modelling 358:1-9. PDF (1.3MB)

Marcot, B. G. 2015. Conservation modeling under uncertainty. Presented 9 December 2015 at Department of Conservation, Hamilton, New Zealand.

Marcot, B. G. 2015. Workshop on Bayesian network modeling. Presented 8 December 2015 at National Institute of Water and Atmospheric Research Crown Research Institute, Hamilton, New Zealand.

Pawson, S. M., B. G. Marcot, and O. Woodberry. 2015. Using Bayesian networks for predicting the risk of forest insect flight activity. Presented 23 November 2015 at the Seventh Annual Conference of the Australasian Bayesian Network Modelling Society, Monash University, Melbourne, Australia.
Abstract:
     Insect development and activity are weather dependent processes. Flight behaviour, in particular, is strongly mediated by meteorological conditions and the form of this relationship is normally species specific. To predict the activity of forest insects, their flight patterns, and the subsequent likelihood that they may establish a colony on a recently harvested log during a particular time period, requires an understanding of the relationship between key meteorological conditions and flight behaviour. We test key parameters (temperature, humidity, wind speed, rainfall) and their influence on the flight abilities of key forest insect pests (Hylurgus ligniperda, Hylastes ater, and Arhopalus ferus) over a twelve week period at a temporal resolution of 1 hour. We will demonstrate a Bayesian network developed to predict the likelihood of flight activity on the basis of available weather forecast data. We structured Bayesian network models from field data using a TAN naive Bayes approach, and parameterised the probability tables by using case-file training algorithms. Best-performing models were calibrated against the training data with 5.19% overall confusion error, broken down as 10.8% Type I (false positive) error and 4.1% Type II (false negative) error. We discuss some of the theoretical challenges of using Bayesian networks to analyse flight activity data.

Pawson, S. M., B. G. Marcot, and O. Woodberry. 2015. Using Bayesian networks to predict forest insect flight activity. Presented at the New Zealand Ecological Society Conference, 16-19 November 2015, University of Canterbury, Christchurch, New Zealand.


2014 -------------------
Havron, A., C. Goldfinger, S. Henkel, B. G. Marcot, C. Romsos, and L. Gilbane. 2014. Bayesian inference of benthic infauna habitat suitability along the U.S. West Coast. Presented at the International Marine Conservation Congress, Glasgow, Scotland, August 14-18.

Marcot, B. G. 2014. Applying modelling tools to improve decision making in land management and for the conservation of biodiversity in a climate of uncertainty. Presented 21 November 2014 to Ministry of Primary Industries, Wellington, New Zealand. [Invited]
Abstract:
     Important decisions that affect the management of New Zealand’s primary and conservation resources must be made despite uncertainty, i.e., all the facts are not yet apparent. These decisions must balance the competing interests of stakeholders and must be objective, transparent, and accountable. Dr Marcot will present case studies of Bayesian network modelling approaches that have been used by several U.S. state and federal agencies to A) identify invasive species, B) provide a framework for evaluating sustainability of management of primary land and forest resources, C) determine effects of climate change on threatened species, D) provide a structured approach to conservation management decision making. Dr Marcot will discuss the pros and cons of different approaches and demonstrate how such decision-aiding tools fit into a broader decisionmaking framework.

Marcot, B. G. 2014. Of confidence, control, and cause: using Bayesian networks for management decisions. Presented 26 November 2014 at the Annual Conference of the Australasian Bayesian Network Modelling Society, Rotorua, New Zealand. [Invited keynote address] Available with annotations as PDF (23.3MB)
Abstract:
     Several examples are presented from North America of using Bayesian networks for natural resource management, showing the evolution of their use in a structured decision-making context. The presentation raises practical questions of denoting confidence in expert judgment used to develop probability structures, of identifying management control and influence of decisions, and of determining causality.

Marcot, B. G. 2014. Using Bayesian networks to model plants and wildlife for conservation planning under uncertainty. Presented 21 November 2014 to Department of Conservation, Wellington, New Zealand. [Invited]
Abstract:
     Bayesian network models are useful tools for projecting plant and animal responses to habitat conditions, environmental disturbances, and anthropogenic stressors in a risk analysis and risk management framework. We will explore examples of how such models are being developed and used in the U.S. Forest Service and other resource management agencies to help advise on conservation decisions for rare and little-known species, to project habitat conditions for species, and to evaluate species persistence under climate change and associated stressors. Brief presentations will be complemented by open discussions of the potential applicability of such approaches to local conservation planning and management needs.

Marcot, B. G. 2014. Structured decision-making process. Presented 23 January 2014 at Federal Mitigation Fund - Strategic Restoration Plan Development Meeting, Umpqua National Forest, U.S. Forest Service. Roseburg, Oregon. [Invited]


2013 -------------------
Marcot, B. G., L. A. Fisher, M. P. Thompson, and M. Tomosy. 2013. Structured decision making, NEPA, and the National Forest System. NEPA Knowledge Cafe (online webinar series of USDA Forest Service). [Invited] http://fsteams.fs.fed.us/sites/cop/nepacafe/default.aspx.
Abstract:
     US Forest Service, Research & Development, initiated this research team project to provide guidelines for use of structured decision-making (SDM) approaches and tools for the National Forest System (NFS). We provide a brief overview of SDM steps; compare SDM to NEPA procedures; present results of a national poll on use of SDM we posed to NFS decision-makers and analysts; and conclude with suggestions for use of, and training in, SDM in management and planning on national forests and grasslands.

2012 -------------------
Marcot, B. G. 2012. Introduction to structured decision-making. Presented 30 April 2012 at: Alaska Science and Decision-Making Workshop. Department of the Interior, Washington, D.C. [Invited]

2011 -------------------
Marcot, B. G. 2011. A new future for polar bears. Presented at: 2011 US-IALE (U.S. Regional Association of the International Association for Landscape Ecology) Annual Symposium. Portland, Oregon.
Abstract:
     As part of a larger Polar Bear Science Team effort with USDI Geological Survey, analyses were made of historic, current, and future (next century) habitat and populations of polar bears in four ecoregions throughout their global distribution. Habitat carrying capacity was modeled using a simple analytic framework, projecting future capacity as a function of present polar bear crude and ecological densities. Population outcomes were modeled based on potential effects of environmental conditions, prey and foraging habitat availability, and anthropogenic stressors, in a Bayesian network. All future projections were made using minimum, maximum, and ensemble mean values of sea ice conditions summarized from a suite of 10 general circulation models under 4 greenhouse gas (GHG) concentration scenarios.
     Polar bear populations were projected to decline during the 21st century throughout their range, and severity of decline depended on future ice floe availability. In 2 ecoregions, under the "business as usual" GHG scenario, the most likely population outcomes were extirpation within 45-75 years, resulting in overall potential loss of about two-thirds of the world's current polar bear population by mid-century. The main factor accounting for such declines is loss of sea ice habitat.
However, under other GHG mitigation scenarios, future declines in sea ice habitat and polar bear populations might not be nearly as severe, and up to a point could be reversible without "tipping point" thresholds otherwise pushing the species to imperiled or extirpated status.
Marcot, B. G. 2011. A future for polar bears under climate change in the 21st century. Presented 10 January 2011 to World Forest Institute and The Oregon Zoo [invited].   World Forestry Center, Portland, Oregon.
Marcot, B. G., and J. N. Pauli. 2011. Development and testing of empirical probability network models to predict age of martens (Martes americana and M. caurina) based on DNA telomere analysis. Presented at: 91st Annual Meeting of the American Society of Mammalogists - Joint Meeting with the Australian Mammal Society. Portland State University, Portland, Oregon.
Abstract:
     In this project, with others we developed and reported on a means of predicting age of free-ranging and live-capture martens from DNA data and population and environmental covariates (J. Mamm. 92(3)). Predictions were made by use of probability (Bayesian) network models based on telomere length, sex, species, population density, body size, and other variables. The final models assigned free-ranging martens to juvenile or adult stage classes with 75-88% accuracy, and live-captured martens to five discrete age classes with 90-93% accuracy. Here, we report on how we innovated these new ways to create alternative empirical probability network models, and how we tested each model with training data using a variety of existing and new performance criteria.
     We developed and tested a total of 35 models with alternative combinations of 11 covariates (prediction variables), 3 response variables (stage or age class categories), 2 levels of resolution of the covariates (coarse and fine), and 3 strata of geographic occurrence of the martens (island, continental, and both). For each model, all values of prior and conditional probabilities were calculated from empirical data on 399 cases of martens and covariates by use of the expectation maximization algorithm, using the Bayesian network modeling shell Netica (Norsys, Inc.), using different subsets of the data for training and testing. We evaluated the accuracy of the models for predicting marten stage or age class by comparing model confusion error rates, spherical payoff values, Schwarz' Bayesian information criterion indices, use of k-fold cross validation, model complexity indices (number of covariates and conditional probabilities), and new indices we devised that weight confusion error rates by model complexity. Results suggested a final set of models that provide lowest prediction error and most parsimonious model structure. Our methods could be used to develop and test similar models predicting stage or age class of other species based on telomere and other covariates.

2010 -------------------
Marcot, B. G. 2010. Decision modeling under uncertainty - the many tools in the toolbox. Presented 21 April 2010 to USGS National Wetlands Research Center, Lafayette, Louisiana.

Marcot, B. G. 2010. Modeling with Bayesian networks. Workshop held 21 April 2010 at USGS National Wetlands Research Center, Lafayette, Louisiana.
Marcot, B. G. 2010. Approaches to structured decision assessment and expert-based models. Presented 27 January 2009 at: Biological Reviews for Bearded and Ringed Seals Structured Decision Making Workshop, National Marine Mammal Laboratory, NOAA, Seattle WA. [Invited]

2009 -------------------
Jay, C. V., B. G. Marcot, and A. S. Fischbach. 2009. Forecasting Pacific walrus status: modeling key factors with Bayesian networks. For presentation at: Society for Marine Mammalogy - Biennial Conference on the Biology of Marine Mammals, 12-16 October 2009.
Abstract:
    Sea ice over the Chukchi and Bering Seas is important to many Arctic marine mammals, including the Pacific walrus (Odobenus rosmarus divergens). Sea ice provides walruses with substrates to rest, molt, and give birth in relative seclusion of humans and predators. Sea ice also provides walruses access to large areas over the continental shelf to forage on productive patches of benthic invertebrates. Changes in seasonal sea ice conditions from climate warming will likely impact walruses in complex ways. For example, in the Chukchi Sea, walruses responded to recent extreme reductions in summer ice extent by occupying previously unused coastal haul-outs, which led to increased(?) calf mortalities from crowding and may have imposed increased physical stress on adults from reduced foraging opportunities. However, reduced sea ice in the Arctic also has led to increased levels of primary production by creating a longer growing season and larger areas of open water, which could potentially provide more food for walrus prey on the seafloor. The combined impact of these and other factors on the Pacific walrus population is difficult to forecast. To evaluate the implications of climate change on walruses, we are developing a Bayesian network model that will integrate key environmental correlates and potential anthropogenic stressors to the walrus population. With the model, we will estimate probabilities of future persistence of Pacific walrus populations, quantify uncertainties, and evaluate the sensitivity of model outcomes to climate model uncertainties, environmental conditions, and stressors. Results will provide policy makers and regulatory agencies with new information to address emerging concerns about the species’ status under changing environmental conditions in the Arctic.

Jay, C., B. G. Marcot, and D. Douglas. 2009. Bayesian network modeling of Pacific walrus population status. Briefing 30 June 2009 by USDI Geological Survey Walrus Science Team to USDI Fish and Wildlife Service, Anchorage, Alaska.
Abstract:
     Purpose of this project: to develop a Bayesian network model, to a level that has undergone peer review of selected walrus experts, that may be useful to DOI in its listing decision for the Pacific walrus. The end product of this work will be a peer-reviewed USGS report detailing the model and results of the study.
Marcot, B. G., and S. C. Amstrup. 2009. Polar bears in the greenhouse: global populations under stress. Presented 16 November 2009 at: 2009 Carnivore Conference: Carnivore Conservation in a Changing World. Defenders of Wildlife. Denver, Colorado.
Abstract:
We have modeled global populations of polar bears in four major ecoregions under six time periods (out to a century) and range and mean values of ice habitat amount and distribution projected with 10 global circulation models (GCM) running the A1B “business as usual” greenhouse gas climate change scenario of the Intergovernmental Panel on Climate Change (IPCC). We used a deterministic projection of polar bear carrying capacity based on known crude and ecological densities in each ecoregion and GCM projections of future ice habitat, and a probabilistic Bayesian network model to evaluate the effects of multiple anthropogenic stressors and environmental conditions. Our findings suggested that approximately two-thirds of extant polar bears will most likely trend toward extirpation by mid-century, and the rest will occur in smaller distributions than at present. These findings contributed to US Department of the Interior’s 2008 listing of the species as globally threatened.

Marcot, B. G., and S. C. Amstrup. 2009. Warm times ahead: modeling the future of polar bear global populations. In: Annual Meeting, Oregon Chapter of The Wildlife Society. 12 February 2009. Glen Eden, Oregon.
Abstract:
    As part of a larger Polar Bear Science Team effort with USDI Geological Survey, analyses were made of historic, current, and future (next century) habitat and populations of polar bears in four ecoregions throughout their global distribution. Habitat carrying capacity was modeled using a simple analytic framework, projecting future capacity as a function of present polar bear crude and ecological densities. Population outcomes were modeled based on potential effects of environmental conditions, prey and foraging habitat availability, and anthropogenic stressors, in a Bayesian network. All future projections were made using minimum, maximum, and ensemble mean values of sea ice conditions summarized from a suite of 10 general circulation models. Polar bear populations were projected to decline during the 21st century throughout their range, and severity of decline depended on future ice floe availability. In 2 ecoregions, most likely outcomes were extirpation within 45-75 years, resulting in overall potential loss of about two-thirds of the world’s current polar bear population by mid-century. The main factor accounting for declines is loss of ice sea habitat.

2008 -------------------
Amstrup, S. C., B. G. Marcot, and D. C. Douglas. 2008. Forecasting the 21st century world-wide status of polar bears using a Bayesian network modeling approach. Second USGS Modeling Conference. 11-15 February 2008, Orange Beach, Alabama.

Beever, E. A., and B. G. Marcot. 2008. Bayesian network models as a framework for forecasting wildlife response to GCC. Presented 18 November 2008 at the WildREACH Workshop, USDI Fish and Wildlife Service, Fairbanks, Alaska.

Marcot, B. G. 2008. Bayesian network modeling. Presented 20 February 2008 to USDI Fish and Wildlife Service, Regional Office, Portland, Oregon. [Invited]

Marcot, B. G. 2008. Hosted workshop, presented talk on use of Bayesian networks for modeling species-habitat relationships of Mardon skipper (Polites mardon) in the Washington Cascades. 01 April 2008, Gifford Pinchot National Forest, Vancouver, Washington.

Marcot, B. G. 2008. I believe, therefore I model: Evaluating species at risk with with Bayesian belief networks and other tools. Presented 10 September 2008 to Seminar Series, Alaska Science Center, USDI Geological Survey, Anchorage, Alaska. [Invited]
Marcot, B. G. 2008. Of polar bears and spotted owls: modeling imperiled species for conservation decisions. Presented 17 July 2008 to USDI Fish and Wildlife Service, Spotlight on Science lecture series, Portland, Oregon. [Invited]

2007 -------------------
Marcot, B. G. 2007. Review of the decision modeling toolkit. Presented 4 April 2007, Oregon Department of Forestry, Salem, Oregon. [Invited]
Marcot, B. G. 2007. Workshop on building Bayesian belief network models. 22 March 2007 for the Landscape Level Wildlife Assessment Project, Washington Department of Fish and Wildlife, Olympia, Washington. [Invited]

2006 -------------------
Marcot, B. G. 2006. Modeling with Bayesian belief networks. Presented 27 October 2006 to USDI Fish and Wildlife Service shortcourse on Principles of Modeling for Conservation Planning and Analysis, NCTC Course No. ECS 3149, Portland Oregon [Invited]

2003 -------------------
Marcot, B. G. 2003. Characterizing species at risk: experience in species and decision modeling under the Northwest Forest Plan. In: September 2003 Annual National Conference of The Wildlife Society, Session on "Assessing Risks to Wildlife Populations From Multiple Stressors". Burlington VT.
Abstract:
    A "Survey and Manage Species Program" has been established under the Northwest Forest Plan (USFS, BLM) that, in part, entails site surveys and annual reviews of rare and little-known species of fungi, lichens, bryophytes, vascular plants, mollusks, and vertebrates in late-successional and old-growth forest reserves. Under this program, colleagues and I have developed 12 species-habitat models using Bayesian belief networks (BBN) in a rigorous procedure involving expert knowledge, peer review and model revision, testing with known site data, and validation and updating with new, random-site data. These species-habitat risk analysis models are intended to help managers predict likelihoods of species presence based on microsite characteristics, and to prioritize sites for expensive and intensive pre-disturbance survey. Thus, their sensitivity to site conditions, and their accuracy of predicting species presence (more so than absence), are paramount and being tested. A second set of BBN models represents risk management in the decision framework for annual species reviews conducted by managers and specialists under the program. These decision models were crafted strictly from the guidelines in the Record of Decision and are helping to ensure consistency in the species reviews, as well as helping to identify hidden inconsistencies in the guidelines. Overall, the species risk analysis models and the decision risk management models are best used to guide thinking and evaluation, not to substitute for human decision-making.

Marcot, B. G. 2003. ROD criteria as Bayesian belief network decision models. In: Step 3 Workshop, 2003 Survey & Manage Annual Species Review. Pacific Northwest Research Station Director's Office, USDA Forest Service, Portland, Oregon, 29 May 2003. [ Invited.]

2002 -------------------
Marcot, B. G. 2002. I believe, therefore I model: Modeling wildlife-habitat relations with Bayesian belief networks. Presented 20 January 2002 to Faculty and Graduate Seminar, Department of Wildlife, Utah State University, Logan, Utah.

Marcot, B. G. 2002. Modeling rare species with Bayesian belief networks. Seminar presented 31 January 2002 to graduate class on species modeling, Department of Wildlife, Utah State University, Logan, Utah.

Marcot, B. G. 2002. Modeling species and decision with Bayesian belief networks. Presented 17 April 2002 at: Workshop on species modeling. Ministry of Forests, Victoria, British Columbia, Canada.

Marcot, B. G., and K. Mellen. 2002. Ways to locate, map, and model high priority sites, including use of GIS and Bayesian belief network modeling. Presented 05 April 2002 at: Taxa team work session, high priority site selection. Survey and Manage Species Program, Portland, Oregon.

2001 -------------------
Marcot, B. G. 2001. Decision support models for plant and animal conservation. In: Restoration and recovery: beyond good intentions. Society for Ecological Restoration Northwest Chapter Conference, 2-6 April 2001. Bellevue WA.
Abstract:
    A major challenge in conservation is managing species and ecosystems with scant scientific knowledge and great uncertainty. Decision support models (DSMs) can aid this by (1) evaluating the implications of uncertainty in meeting management goals, (2) combining empirical data with expert judgment, and (3) through sensitivity testing and validation steps, identifying key habitat elements as a basis for prioritizing inventory and monitoring. DSMs include a wide range of tools and include Bayesian analyses and belief network modeling, data and text mining, decision modeling such as decision tree analysis, expert systems, fuzzy logic and fuzzy set theory models, genetic algorithms, rule and network induction, neural networks, reliability analyses, quantitative (environmental) risk analysis, simulation and scenario modeling, and other approaches. Successful use of DSMs for plant and animal conservation depends largely on the availability of data or experts, and the willingness of decision-makers to articulate their risk attitudes and decision criteria. These are no small hurdles. Many DSMs can aid in merging scientific data with expert knowledge, although no model can replace empirical field studies. Several examples of DSMs are demonstrated to illustrate the 3 objectives listed above.

Marcot, B. G. 2001. Functional assessments and Bayesian belief network modeling. Presented 9 May 2001 to the INLAS modeling team, USDA Forest Service and others, The Dalles, Oregon. Invited presentation.

1999 -------------------
Marcot, B. G. 1999. The DecAID advisory system for managing snags and down wood for wildlife habitat; the Old Forest Remnants Study; and use of Bayesian belief networks for modeling species viability for the Interior Columbia Basin Ecosystem Management Project. Presented 27 October 1999 to the Survey and Manage Committee, USDA Forest Service, Corvallis Forestry Sciences Lab (invited presentation).

Marcot, B. G. 1999. Use of Bayesian belief network models for evaluating Final EIS alternatives for wildlife viability. Presented 10 March 1999 at Annual Northwest Section Conference, The Wildlife Society, Bozeman MT. (invited presentation).
Abstract:
    The Terrestrial Staff of the Science Advisory Group has developed “causal web” models relating key environmental correlates (KECs) of wildlife species, to potential population response under several Final EIS alternatives for the Interior Columbia Basin Ecosystem Management Project. The models involve use of Bayesian belief networks (BBNs), which represent conditional probabilities of population response given environmental conditions at two scales of spatial resolution. The KECs were identified by use of literature and expert panels and formalized into a Species-Environment Relations database. The probabilities and BBN model structures were derived from literature and, where needed, expert judgment. The BBN models provide a consistent, testable framework by which to represent simple habitat relations of a wide array of species. Sensitivity analyses using entropy-reduction metrics identify controlling KECs that may be worthy of further study or monitoring. BBN species modeling represents a major step beyond using expert panels to evaluate population viability; it opens the “black box” of expert opinion by formally modeling the subjacent ecological relations.

Marcot, B. G. 1999. Use of Bayesian belief networks for modeling wildlife population viability: a sing-along. Presented 27 October 1999 to Wildlife Seminar, Department of Fisheries and Wildlife, Oregon State University. (invited presentation). Corvallis OR.



Back to The Bayesian Network Models Page...