Comparison of Modeling Tools

Document Sample
Comparison of Modeling Tools Powered By Docstoc
					Review of the Growing
  Modeling Toolkit

      Bruce G. Marcot
    USDA Forest Service
    Portland, Oregon USA
Marcot, B. G. 2006. Review of the growing modeling
toolkit: special session. Presented 5 December 2006
at: Habitat and Habitat Supply Modeling Practitioner's
Workshop, 5-7 December 2006. Ministry of Forests,
Research Branch, British Columbia, Canada. [Invited].
Chase, B.C. Canada.
Models, Models, Models
Models, Models, Models
Models, Models, Models
Models, Models, Models
Models, Models, Models
Lots of models to choose from !

Influence diagrams
               Influence diagrams –

              “Concept mapping”
              “Concept diagrams”
                “Cognitive Map”
            “Mental Map, Mind Map”
                        Influence diagrams

Marcot, B. G., et al.
2001. Forest Ecology
and Management
       Influence diagrams –

-   Mindjet MindManager Pro
-   Inspiration
-   Personal Brain
-   Netica
Mindjet MindManager Pro
   Personal Brian
IHMC CmapTools
        Building influence diagrams –

- empirical data
- expert judgment / opinion
    -   “knowledge engineering”
    -   peer review
    -   expert paneling (e.g., Delphi)
-   combination
From influence diagram …
   to models galore !
Path regression
 Quality Deer

   Woods, G. R., D. C.
   Guynn, W. E. Hammitt,
   and M. E. Patterson.
   1996. Determinants of
   participant satisfaction
   with quality deer
   management. Wildl.
   Soc. Bull. 24(2):318-
                              Hudson, R. J. 1995.
                              Paths to conservation.
                              Pp. 318-322 in: J. A.
                              Bissonette and P. R.
                              Krausman, ed.
                              Integrating people and
                              wildlife for a sustainable
                              future. The Wildlife
                              Society, Bethesda,
                              Maryland. 715 pp.

    Process model –
    Process model –
                 Types of Models

 Analytic   and numerical population models
     Leslie matrix life tables
     Genetic models of inbreeding, genetic drift
 Simulationmodels
 GIS-based models
     Spatially explicit, individual-based models
 Knowledge-based        (expert) models
     Expert systems
     Other expert-based models
                   Types of Models

 Statistical   empirical models
     Correlation, multivariate models
       • Regression tree, classification tree
Regression tree – 3 viability
        risk levels
                  Types of Models

 Statistical   empirical models
     Correlation, multivariate models
       • Structural equation models (SEMs) –
           a modeling procedure
Structural Equation Models (SEMs)

• A way to formalize and construct
  relationships among variables.
• Observational data
• A generalization of many statistical
      Regression, discriminant analysis, canonical
       correlation, factor analysis
• Differentiates among direct relationships,
  indirect causal relationships, spurious
  relationships, & association without
           Structural Equation Models (SEMs)

1.        Create the model structure as an influence diagram …
          including unexplained variance.
2.        Expand the latent variables into their components …
          e.g., “habitat” into measurable veg. variables.
3.        Compute regression weights for each variable.
     1.     Partial correlation analysis
     2.     Bayesian conditional probabilities
4.        Estimate measurement errors of each component
     1.     This depicts the amount of uncertainty in the habitat-species
            relations represented in the model.
         Structural Equation Models (SEMs)

       The final SEM model depicts:
         Specific variable relations
         Degree of uncertainty of those variables
         The relations among the variables
       SEM tests the hypothesized underlying causal
        relations among variables … by analyzing their
        covariance structure.
         Goodness-of-fit tests of congruence between the variance-
          covariance matrix derived from observational data … to that
          suggested by the hypothetical causal structure of the model
          (the predicted moment matrix).
         Structural Equation Models (SEMs)

       Methods of estimation for the goodness-of-fit tests:
         MLE (maximum likelihood estimation), for multivariate normal
          data & N>200 samples
         WLS (weighted least squares; asymptotically distribution free)
          methods, for continuous but nonnormal data
         Polychoric correlation analysis, for ordinal variables (computes
          correlation between unobserved normal variables & then uses
          WLS methods)
       Software for doing SEM:
 Statistical     empirical models
     Post-hoc pattern analysis
       •   Knowledge discovery
       •   Rule induction (problems w overfitting data)
       •   Data mining (association analysis)
       •   Text mining
                 Types of Models

 Statistical   empirical models
     Post-hoc pattern analysis
       • Knowledge discovery
       • Rule induction
       • Data mining
       • Text mining
                         Text Mining

   Biodiversity
       biocomplexity
       ecological complexity
       ecological functions
       disturbance regimes
       ecosystem resilience
       stability, resistance
       ecological integrity
       ecosystem services
       sustainability …. etc.
                         Text Mining

   Biodiversity
       biocomplexity             >13,000    references
       ecological complexity
                                      EndNote biblio.
       ecological functions
       disturbance regimes
       ecosystem resilience
                                      “concept proximity
       stability, resistance
       ecological integrity
       ecosystem services
       sustainability …. etc.
                            Marcot, B. G. In revision.
                            Biodiversity and the lexicon zoo.
Text mining – Concept map   Forest Ecology and Management
Data mining
Information mapping
 - topographical maps
 - closeness maps
 - interactive trees
 - concept clustering
 - (many others)
Decision-Support Models
    Decision-Support Models
   Many tools
    •   Bayesian statistics, Bayesian belief networks
    •   Data and text mining
    •   Decision tree analysis
    •   Expert systems
    •   Fuzzy logic, fuzzy set theory
    •   Genetic algorithms
    •   Rule and network induction
    •   Neural networks
    •   Reliability analyses
    •   Landscape simulators
Bayesian Belief Network Model
Influence Diagrams as Bayesian
      Belief Network Model
Influence Diagrams as Bayesian
      Belief Network Model
 sensitivity   analysis
     identifies most influential factors
     identifies degree of influence
Influence Diagrams as Bayesian
      Belief Network Model
Node                 Mutual    Variance of
   Influence Diagrams as Bayesian
----                  Info      Beliefs
          Belief Network Model
Caves or mines     0.02902 0.0069284
Lg snags or trees    0.00953   0.0023053
Cliffs               0.00599   0.0014514
Forest edges         0.00599   0.0014514
Bridges, buildings   0.00063   0.0001543
Boulders             0.00002   0.0000038
      Fuzzy logic model –
        Penn St. Univ.

fuzzy logic
Penn St.

     Fuzzy arithmetic

     Fuzzy arithmetic
                  EMDS –
Ecosystem Management Decision Support
                  EMDS –
Ecosystem Management Decision Support

         Stream condition depends
             reach condition
                riparian vegetation
                bank stability
                w/d ratio
                pool frequency
                large wood
             spawning fines
             water temperature
 Decision-Support Models
• inputs expressed as management
activities ... or environmental variables
potentially affected by management
• outcomes expressed as probabilities ...
risk management
• explicitly show management decisions
and values of outcomes
decision tree – (DecisionPro, Vanguard
            Software Corp.)
Influence Diagrams as Bayesian
      Belief Network Model
Influence Diagrams as Bayesian
      Belief Network Model
      Decision-Support Models
 Many decision-analysis techniques are
     Value of perfect information
     Value of additional or sample information
     Credibility of information
     Quantitative measures of the state of
      Decision-Support Models
 Many    analysis methods
     MAUT (Multi-Attribute Utility Theory)
     Goal Hierarchy
     AHP (Analytic Hierarchy Process)
     MCDM (Multiple Criteria Decision Making)
     EPA’s Quantitative Risk Analysis
       Decision-Support Models --
        Useful Model Attributes
   probability-based
   accounts for missing data
   provides for sensitivity testing
   provides management hypothesis (adaptive
   incorporates new data to update model
    functions, probabilities, structure
   allows rapid prototyping
   combines expert judgment w/ empirical data;
    multiple experts
Population Models
              Population Models
   Populus – general popn dynamics
   Vortex (, Nemesis – PVA,
   RAMAS (
         • RAMAS Red List, RAMAS GIS, RAMAS Metapop
           RAMAS Landscape, RAMAS Multispecies Assessment
           RAMAS Ecotoxicology, RAMAS Ecosystem, RAMAS Stage
   Outbreak – disease (
   SELES – spatially explicit landscape event simulator
   Biomapper – habitat suitability mapping, GIS
   PATCH – individual, spatially-explicit simulator
    (Nathan Schumaker, U.S. EPA)
  Models of Vegetation Processes
 Vegetationgrowth, ecological succession,
  ecosystem disturbance, other processes
 Techniques
     Markov chain
     Transition matrix analysis
     Loop analysis, graph theory
 Models:
Models of Optimizing Land Allocations

 Many    techniques
     e.g., genetic algorithms, neural networks,
      adaptive kernel
     Multi-objective integer programming (MOIP)
     Nonlinear integer programming
 Models:SITES, BioMapper
 Approaches: reserve complementarity,
 redundancy, representativeness
        Managing Under Uncertainty

 many   tools in the toolkit
 make explicit the inference from habitat to
  biodiversity parameter
 display uncertainty of parameters and
 useful for risk management:
     clearly show risk attitude
     clearly articulate decision criteria
     very helpful for monitoring, adaptive research