Ecosystem Modeling Workshop Co Sponsored by CaRA GCOOS GOMA and

Document Sample
Ecosystem Modeling Workshop Co Sponsored by CaRA GCOOS GOMA and Powered By Docstoc
					         Ecosystem Modeling Workshop
Co-Sponsored by CaRA, GCOOS, GOMA, & SECOORA
     October 14-16, 2009, St. Petersburg, FL

A History and Evolution of Ecosystems Models
                        Claire B. Paris
                        Assistant Professor
         Rosenstiel School of Marine & Atmospheric Science
                        University of Miami

                         Nasseer Idrisi
         Assistant Professor, CaRA Subregional Coordinator
                          US Virgin Islands
           Center for Marine and Environmental Studies
                   University of The Virgin Islands

1.   what is an ecosystem model
2.   how are they developed
3.   where we were
4.   where we are
5.   where we want to be
1. what is an ecosystem model?

Definition: A simplified representation of complex
ecosystems (e.g. foodweb), aimed at characterizing their
major dynamics and predicting their behavior

- forced from the outside at the boundaries with input from data or
other models
- interact dynamically through coupling with physical models

-Temperature                                      P
-Interacting chemicals
-Currents                               N                    Z
2. how are ecosystem models developed?

  •   Holistic or reductionist approaches
      (bottom-up, POM)
  •   Predictive approach (hypothesis in terms
      of expected results)
  •   Inferential, deduct mechanism
      (hypothesis in terms of processes)
  •   Mechanistic or empirical methods
      deductive reasoning
  •   hypothesis generating (experimental
 Each has its own strengths and weaknesses. The development of a
 model, the type of model, the approach and method used, and the final
 analysis and synthesis should be suited to the study (e.g., start with a
 specific question, what observed patterns characterize the system
 dynamic, Grim 2005)
3. Where we were
Raymond Pearl (1920s): described population growth through
  mathematical representation using the logistic equation
  (Verhulst: 1845, 1847)) to forecast human population growth.
  Pearl also laid the foundations for developing empirically-
  derived models of demography that include information on
  fecundity and birth rates, death rates, immigration and
  emigration that statistically describe the dynamics of a
     dN =    rN(K-N)
     dt        K

   r and K are specific to a
   species and environment
   conditions, but can change
   depending on species and
3. Where we were
Alfred Lotka (1924) and Vito Volterra (1926) independently developed a
   system of coupled ordinary differential equations that describe
   predator-prey interactions over time. These are exponential in nature
   and the populations at different trophic levels are constrained by their
   coupled nature. The competitive model system, also attributed to
   Lotka and Volterra describe the interactions between two species
   through coupled logistic equations where each population is
   constrained by the other and a carrying capacity term in each

  where x and y are prey and
  α, is the prey’s growth rate
  β, is an interaction parameter
  (predation of Y on X)
  γ, the assimilation efficiency of Y
  δ is the mortality rate of the
3. Where we were: NPZ 3 compartments system
The next major step in ecosystem modeling came with Gordon Riley
  (1946) with a 3-level system of coupled ordinary differential
  equations that describes a planktonic marine ecosystem. These
  models increased in sophistication with John Steele (1974) and
  Steele and Frost (1977) that accommodate better means for
  parameterization, and more robust with regards to assumptions and
Franks and Chen (1996) coupled a Nutrient-Phytoplankton-
  Zooplankton (NPZ) model into a primitive equation model and
  applied it to examine the summertime plankton dynamics on GB.
  That was the first modeling effort to study the biological process
  under the “realistic” physical environment in the GoM/GB region.

   From Francisco Werner’s 1999 tutorial
   at UNC
3. Where we were NPZ 3 compartments system and
Franks and Chen (2000) carried out numerical experiments for 2-D
3-D cases on Georges Bank:

The 2-D experiments were conducted on a south-north transect across
the center of GB. The model is driven by tidal forcing only with an
assumption that the cross-bank distribution of temperature,
phytoplankton and nutrients on GB is mainly related closely to the tidal
mixing front.
The 3-D experiments were conducted with emphasis on the role of tidal
mixing and advection in the spatial and temporal distributions of
temperature, phytoplankton, zooplankton, and nutrients on GB. Similar
to the 2-D case, the initial fields of temperature and biological variables
are specified to be uniform in the horizontal with assumption that the
spatial and temporal variations of physical and biological variables are
caused by tidal mixing and advection.

The model-predicted distribution of tidal mixing front is in excellent agreement with previous
3. Where we were

Model formulations and assumptions: Up to this point, it is understood
  that the models used to represent ecosystems require many
  assumptions regarding the model boundaries and how these systems
  are forced and the nature of the behavior of the modeled components
  to forcing from outside the system and from within the system. How do
  we constrain a model variable and how do we constrain the variable’s
  behavior? These are important questions that need to be asked when
  developing the model system for a particular study.
3. Where we were
 Population modeling: At the base of ecosystem models is the population,
   and throughout the 20th century, population model development
   evolved with the Leslie Matrix Model (P.H. Leslie, 1945) that uses
   demographics, information that Pearl used to describe human
   populations. Beverton and Holt (1957) developed models for fisheries
   to hindcast fish cohorts to understand fish population growth, and this
   evolved into Virtual Population and Multispecies Virtual Population
   Analysis that are The for fisheries management.
Food web modeling: used following step was to extend to food web
   modeling (Robert May, Joel Cohen, Robert Ulanowicz, Neo Martinez,
   and many more) with increased complexity with regards to populations
   and communities within ecosystems.
Individual-Based Models: Other than food webs, population models also
   became more sophisticated with the increase in computer power. This
   advancement in computational power has led to the development of
   individual-based and agent-based models (IBMs) of populations and

    From Francisco Werner’s 1999 tutorial
    at UNC
                   spatially-explicit Lagrangian
4. Where we are now:          models
Biophysical coupled ecosystem models (e.g., P. Franks, C. Miller) include
   what are termed NPZ/NPZD ecosystem models that simulate
   interactions among pelagic state variables of nutrients, phytoplankton,
   zooplankton and detritus in the Eulerian field. These Eulerian models
   are coupled to IBMs that are modeled as Lagrangian particles.
It has been argued that advances in coupled NPZ models have not moved
    forward as much as they should (Franks, in press), the reason has
    been attributed to a fact that many of these studies are not
    hypothesis-driven, or if they are, the study design does not allow for
    the rejection of the main hypothesis in favor of an alternative
    hypothesis. Rather, model parameters are tweaked endlessly until the
    model output fits the data and the desired outcome is achieved.
5. Where we are now 3D OGCM-7 compartments NP
Tsiaras and Kourafalou (2006)explored the main physical and biological
  processes that control the seasonal cycle of the plankton dynamics
  over the Western Black Sea by means of a 3-D, 7-compartment, on-
  line coupled biophysical model - high frequency forcing indicated that
  seasonal production was linked to the Danube river’s discharge.
4. Where we are now
Coupled physical-biological Lagrangian particle models have followed a
  more productive path than NPZ models as evidenced by the nature and
  quality of publications produced over the past several years.
    coupling NPZD with
    Lagrangian models
Idrisi et al. (2004) use the physiologically-
     explicit NPZD model and a simple
     behavior for the Lagrangian model that
     is also physiologically constrained to
     simulate emergence from diapause of
     an Arabian Sea copepod
     species.2.71% of individuals were
     physically upwelled into favorable
     conditions and 23.4% of individuals
     swam into bloom conditions.
 5. Where we are now
 Olascoaga et al. (2005) coupled an NPZD to MICOM for the Arabian Sea
    using a growth function that is temperature dependent and
    parameterized from JGOFS data from the 1995-1996 Arabian Sea

Physiologically-explicit temperature-dependent growth can be global and
applicable across latitudes

Coupling to ocean circulation:            Coupling an NPZD to
                                           ocean circulation
4. Where we are now: static ecological model
Ecopath, a static mass-balance system, ecosim – time-dynamic
  simulation, and ecospace – spatially explicit simulation (Christensen
  and Walters).
Ecopath introduces prey behavior with the predator-prey transfer
   functions of a Holling type II or III equation. The system has been
   criticized such as – that the general predictions on predator declines
   can be understood from basic life-history information, and that all
   energy is cycled within the system and the species diet is inflexible.
   The authors of ecopath make valid counter-arguments, but it is yet to
3 Components:
   be seen as to the success of this system.
  ▪ Ecopath – a static, mass-balanced
snapshot of the system
 ▪ Ecosim – a time dynamic simulation
module for policy exploration
  ▪ Ecospace – a spatial and temporal
dynamic module primarily designed for
exploring impact and placement of protected
5. Where we want to be
The path forward is to reflect on the past – triumphs and failures,
  evaluate past mistakes and apply corrective measures. An ecosystem
  model should be treated as an hypothesis, and should be rejected if the
  results do not represent the phenomenon being tested.
How then will we know if, given a certain scenario, a Michaelis-Menton
  function is adequate to represent phytoplankton growth, or do we need
  a temperature-dependent growth function, or PAR absorption function,
  or maybe something else?
Good models use experiments and field data to develop parameters,
  modeling studies need to work hand in hand with laboratory/field
  experiments and field data collection, and to take advantage of
  advances in assimilation techniques. These advances are necessary,
  especially in the context of this workshop for IOOS, whose mission is
  to develop operational models for use by managers, policy makers, and
  other product users. We cannot provide an operational ecosystem
  model if we cannot get the science of Field observations &
                                         the ecosystem model correct.
                                          Lab experiments
                                          Data Assimilation
5. Where we want to be
deYound, Heath, Werner, Chai, Megrey, Monfray (2004 Science)
   Relationship between trophic level and functional complexity within
   marine ecosystem models. The rhomboids indicate the conceptual
   characteristics for models with different species and differing areas of
   primary focus.
Key Modeling Aspects & Challenges
•   physical model in which the biological
    representation is embedded should have
    appropriate resolution & complexity
•   modeling with uncertainties: probabilistic
    vs deterministic simulations (endemic
    lack of knowledge of processes &
    structures at those scales), i.e. ensemble
    forecast, iterative stochastic approach -
    less precise but more accurate!
•   The response of marine food webs to
    environmental changes cannot be based
    on the predictions of static models whose
    parameters are chosen based on the
    goodness of fit of model output to
    currently observed phenomena. The
    reason is that communities of organisms
    are adaptive. In order to understand how
    biological communities adapt, it is first
    necessary to understand the principles
    that drive the organization of those
5. Where we want to be
The European Regional Seas Ecosystem Model (ERSEM, Baretta et al.,
1995) consists of five modules (conceptual units): Primary producer
module; Microbial loop module; Mesozooplankton module; Benthic
nutrients module and Benthic biology module.
5. Where we want to be
Chen (UMass) and Beardsley (WHOI), have developed an integrated model system for the Gulf
of Maine (GoM)/Georges Bank (GB)/New England shelf (NES) region. The major components of
this system include:
1) the modified fifth-generation community mesoscale atmospheric model (MM5);
2) the unstructured grid Finite-Volume Coastal Ocean circulation Model (FVCOM);
3) a 3-D suspended sediment transport model;
4) a generalized lower trophic level food web model [called Flexible Biological Module
 5) a multi-stage zooplankton models (developed by Cabell Davis at WHOI).
4. Where we want to be:
                     Multi-scale modeling

Srinivasan and Paris (RSMAS) have developed a coupled physical-
   biological IBM, the Connectivity Modeling System (CMS) with particle
   swimming behavior and trophic levels interactions (NPZ), featuring 1)
   on-the-fly access of ocean model data using OPeNDAP, 2) the Earth
   Modeling System Framework (EMSF) that channels multi-scale data
   from model and observation, 3) statistical interpolation of
   observations, 4) Lagrangian and Eulerian data assimilation, 5) nesting
   capabilities, 6) a flexible biological module, a 7) a GIS-based habitat
   module, and 8) a genetic matrix-based module. Partners providing
   high-resolution ocean models for the CMS are Kourafalou (SoFL-
   HYCOM, FLKey-HYCOM) and Cherubin (ROMS).
1. Raymond Pearl human population growth models (logistic).
2. Lotka-Volterra predator-prey and competition models.
3. Wiegert predator-prey models.
4. Leslie matrix population models.
5. Virtual Population Analysis of commercially important fish
6. Multi-virtual population analysis of commercially important fish
7. Nutrient-phytoplankton-zooplankton-detritus numerical difference
8. Food web models: networks, links, and energy flows.
9. Individual-based models.
10. Agent-based models.
11. Ecopath.
12. Coupled biophysical NPZD and higher complexity models in the
Eulerian field.
13. Coupled biophysical individual-based models in the Lagrangian field.

The first three models (1-3) are either individual or a set of coupled
differential equations. Models 4-9 and 12-13 are individual or sets of
coupled linear difference equations. Models 12 and 13 are coupled to a
physical model and be spatially explicit, rather than being a ‘box’ model
What type of ecosystem models we want for the region?

Objectives of Ecosystem Modeling for CaRA, GCOOS, GOMA, & SECOORA:

-Identify existing weaknesses, theoretical constraints, and needed
 advances in ecosystem modeling
-what are the issues/problems that ecosystem models are expected to
-review ecosystem modeling activities underway in the Gulf of Mexico,
 Caribbean Sea, and southeastern U.S. coastal waters, including estuaries
 and bays
-Unified, coordinated program of ecosystem modeling for the region
  Ecosystem Modeling Workshop
       14-16 October 2009
        St. Petersburg, FL
5. Where we want to be
   Conceptual Model: hypothesis
From Idrisi et al. (2001) Impact of an invasive grazer on pelagic lower
  food web (N-P-Z).

                         Factors                   Hypothesized
   Zebra mussel invasion=>    Phosph        InorganicP

                                  Phyto       Biomass

                                  Zoopl     Biomass
Contrary to the hypothesized decrease in primary production, increase in
   water clarity
led to the system maintaining constant productivity levels that led to an
Lagrangian particle tracking algorithm
                                     Adlandsvik et al.

 Advantages with particle tracking
  • Good numerical properties:
         Avoids numerical diffusion and dispersion
         Permit longer time steps and offline tracking
  • Individual based (IBM): Physical basis for individual based biological models
  • Relatively easy to implement
  Scientific status for particle tracking
  •  Theoretical basis known for at least 20 years
  •  Good algorithms are available
  •  Some details regarding vertical boundary conditions are not understood (Ross and Sharples
 • Turbulence parameterization beyond eddy diffusivity – random walk formulation
 Practical issues with particle tracking
  •   Numerical method
  –   Euler forward
  –   Higher order (e.g. RK4, adaptative)
  •   Random number generator
  –   Quality of Rn generator
  –   shape of distribution (Hunter et al. 1993, Ross & Sharples 2004)
  •   Land boundaries
  –   Important practial problem, not discussed, no good solution
  •   Artificial 2D convergence
  •   Spatially varying diffusivity
  –   ‘naïve’ RW does not work, but never addressed in larval transport
  –   Need correction velocity from low to high diffusivity (Visser 1997)