Multi-scale Integrated Models of Ecosystem Services

Dear Dr Costanza,

I am the National Program Director for the Ecological Research Program within the Environmental Protection Agency’s Office of Research and Development. My role is to develop and implement an Ecological Research Program that meets the needs of EPA’s Program Offices having responsibilities for managing air, water, waste, and toxic materials; for providing technical support to EPA’s Regional Offices; as well as tools and support for EPA’s clients in State and Tribal natural resource agencies, and in local communities.

I believe the common thread connecting these diverse client needs is the overarching need to conserve ecosystem services through proactive decision-making. As you know, the processes and functions of ecosystems – the very foundation of our health, livelihoods, and well-being – are now at risk world wide. Last year, I chose to set EPA’s Ecological Research Program on a new strategic course to focus solely on ecosystem services. Our Program goal is to transform the way decision-makers understand and respond to environmental issues by making clear the ways in which their management choices affect the type, quality, and magnitude of the services we receive from ecosystems – such as clean air, clean water, productive soils, and generation of food and fiber. We want to accomplish this at every level of governance, from local to national. We believe this will enable the Agency to better secure the long-term integrity and productivity of our ecological systems without necessarily imposing a greater regulatory burden.

I believe our success will depend in large measure on creating and disseminating a wholly new generation of modeling and decision-support tools that not only embody the best in ecological science, but also capture people’s attention so that they can explore options and make informed choices.

2

This is no small task. EPA’s Ecological Research Program is staffed by 300 scientists and administrative support personnel located at ten different laboratories and centers across the U.S. For 2007, the Program budget is $65 million dollars per year to support in-house research; at present, I have essentially no discretionary funds to support academic research. However, I am seeking to partner with the best scientists I can find who work in this field. To that end, I initiated a Memorandum of Understanding with the Gund Institute for Ecological Economics, which I believe is working at the forefront of trans-disciplinary research. We share many common goals and believe that by working together we can advance this important field. One of our many opportunities for interaction is with the Gund’s Multi-scale Integrated Modeling of Ecosystems Services (MIMES) platform. For example, our Ecological Research Program is conducting four place-based studies for mapping and managing ecosystem services, principally through developing scenarios that examine the effect of human and natural stressors on ecosystem services and to portray alternative management options. These studies vary in scale from fairly localized coastal cities to the entire landscape of the U.S. Midwest. EPA’s ecologists will be using MIMES in these place-based studies. In addition, we will be providing our place-based study data to the Gund Institute for model building, as well as providing our staff time and resources to help develop, refine, and test the MIMES platform.

We believe MIMES could well be a focal point for collecting and analyzing EPA’s ecosystem service data and scenarios. We have even had discussions about potential long-term housing of MIMES platform in EPA’s Office of Environmental Information. Several of our Program’s best ecological modelers will be visiting the Gund Institute next month to learn more about MIMES. When fully developed, the product your Foundation is supporting could well be put to immediate use within the Agency.

EPA’s Ecological Research Program has already benefited from the cutting edge, trans-disciplinary research on ecosystem services underway at the Gund Institute. We look forward to continued and expanded collaboration with you and your research team.

Sincerely,

/s/ Rick A. Linthurst
Rick A. Linthurst, Ph.D.
National Program Director for Ecology

Name email Location | Division

1 Rajbir Parmar parmar.rajbir@epa.gov “EPA, Athens, GA”
2 Dan MC Garvey mcgarvey.dan@epa.gov “EPA, Athens, GA”
3 Shawn Mattot Mattot.shawn@epa.gov “EPA, Athens, GA”
4 Tom Purucker purucker.tom@epa.gov “EPA, Athens, GA”
5 John Johnston johnston.johnm@epa.gov “EPA, Athens, GA”
6 Nicolas Sayavedra Sayavn@rpi.edu “c/o EPA, Corvallis, OR”
7 Denis White white.denis@epa.gov “EPA, Corvallis, OR”
8 Don Phillips phillips.donald@epa.gov “EPA, Corvallis, OR”
9 Kurt Wolfe wolfe.kurt@epa.gov “EPA, ERD Athens, GA”
10 Darius Semmens semmens.darius@epa.gov “EPA, Las Vegas, NV”
11 Karim Chachakly kchichakly@kua.org UVM-Gund
12 Roelof Boumans rboumans@uvm.edu UVM-Gund
13 Eric Garza egarza@uvm.edu UVM-Gund
14 Amos G. Baehr abaehr@uvm.edu UVM-Gund
15 Azur Moulaert amoulaer@uvm.edu UVM-Gund

From L. Shawn Matott, Ph.D.
U.S. Environmental Protection Agency
Ecosystems Research Division

Modeling frameworks such as MIMES and FRAMES can be leveraged to conduct integrated environmental modeling in support of the ERP core focus on ecosystems services. For such tools to be useful in a policy-relevant context, elucidating model uncertainty is a challenging but vital task.

To start off the discussion, let’s consider a few definitions and related concepts:

1. Uncertainty may be classified into two ‘types’: aleatory, i.e. irreducible and stemming from natural variability, and epistemic, i.e. reducible and stemming from lack of knowledge. Aleatory and epistemic uncertainties have different ramifications for policy decisions.

2. Uncertainty may also be classified in terms of its ’sources’, which include uncertain model input (e.g. amount of rainfall), uncertain model parameters (e.g. various rate constants), and uncertain model structure (e.g. the degree to which the model equations represent the real world).

3. Formal uncertainty analysis (UA) methods propagate sources of uncertainty through the model to generate statistical moments or complete probability distributions for various model outputs. Traditional UA schemes wrap a stochastic Monte Carlo shell around one or more linked (and possibly calibrated) models. More recently, there has been increased emphasis on Hierarchical Bayesian approaches, which utilize sampling procedures that explicitly condition model output on available data.

4. Less formal methods, such as scenario analysis and alternative futures, can also be important tools for uncertainty analysis; albeit in a more qualitative sense. Along these same lines, the NUSAP (Numeral, Unit, Spread, Assessment, Pedigree) system of Funtowicz and Ravetz can serve as an important tool for assessing and assuring quality throughout the modeling process.

5. Sensitivity analysis (SA) studies the degree to which model output is influenced by changes in model input. By measuring model input importance, SA methods enable identification of critical areas where knowledge or data are lacking. In this way, SA can assist decision-makers in prioritizing future research and data collection activities.

Given these definitions and concepts, we can discuss what (if any) infrastructure is available within the MIMES and FRAMES systems to support the UA/SA needs of the ERP.

I am speaking for myself here as a member of the Willamette project who has been following the DSS and valuation issues for the project. This past summer, at the encouragement of our division director, we ran a Stella exercise, in which about half a dozen of the project members were variously involved, and that resulted in an initial model combining hydrology, carbon, and nitrogen processing for the basin (attached below). It is only conceptual at this point because there are no data entered except for Corps dam operational curves. We were much aided in this effort by Nico Sayavedra, an undergrad at RPI, who did all the Stella programming, and who will be in Burlington. Part of the inspiration for this was GUMBO; some of us have read the Boumans, Costanza et al. paper. Nico got familiar with GUMBO, and he and another person gave a demo of GUMBO to some 8th graders in Corvallis as a world futures gaming experiment. The question is where to go with this kind of activity, whether to maybe use the Stella modeling as a planning tool for designing field experiments, or to actually attempt to build simulation capability.

On the other hand, an alternative approach of interest here is an integrated model that could be inspired by the Evoland modeling system (http://evoland.bioe.orst.edu/) of John Bolte and colleagues at Oregon State. Evoland is a stimulating example of a plug and play design into which valuation and policy modules as well as environmental process modules can be inserted.

So what I was hoping would happen at Burlington is a thorough examination of the philosophy, capabilities, development environment, and transferability of MIMES and its host language SIMILE, as either an alternative to the other directions that have been explored for the Willamette, or as another example of ideas that could be adapted to the needs here.

Denis White
US EPA
Corvallis, Oregon

From Daniel J. McGarvey, Ph.D.
U.S. Environmental Protection Agency
Ecosystems Research Division

It seems to me that one potential hurdle to a large-scale implementation of MIMES could be its spatially-explicit structure. As I understand it (and please clarify if I am wrong), MIMES is designed to calculate any number of endpoints for each cell within a gridded landscape (much like its ‘precursor’, the Spatial Modeling Environment (SME) of the Landscape Modeling Framework (LMF)). Since we’re looking to be able to make predictions at the national scale, we’ll obviously need to be able to adjust model parameters (e.g. species lists) within distinct cells, depending upon the locations of those cells. This is certainly do-able (though it will likely be a huge time investment).

I’m wondering if a more systemic problem might be the use of
fundamentally different models within cells. Imagine, for example, that fish assemblage structure in one region (i.e. cluster of cells) is
mediated by deterministic (or quasi-deterministic) mechanisms, such as critical habitat availability; BASS might be a good way to model and ultimately predict the structure and productivity of such assemblages.

Now, imagine that fish assemblage structure in another region has more to do with stochastic immigration-extinction dyamics (re-colonization following flood events, for example); in this case, we might want to implement a literal Island Biogeography model. Given the radically different structures of these two (hypothetical) models, I’m skeptical that we could get away with simply adjusting model parameters; I think we’d need a routine that is capable of calling independent sub-procedures/models on demand (regardless of whether the call is automated or manual). This might be a trival issue to the high-powered comp. sci. folks, but it seems like a potentially signficant challenge to a lowly ecologist, such as myself. . .

Also, I’m wondering if MIMES can synchronously run model procedures at two different scales? That is, can we use model endpoints calculated for one size of grid cell to feed or parameterize model procedures within smaller, nested cells? Here’s an example of what I mean: there’s a lot of good evidence to suggest that species-area relationships (or species-discharge relationships in aquatic systems), though entirely correlative and empirical, are a robust means of predicting total community-level richness. Noting this, I can imagine using a species-area relationship to predict total species richness on a per-basin or per-region basis, then using a process model (again, I use BASS as an example) to predict particular species occurances and densities within each of the region’s cells. This “top-down THEN bottom-up” approach might provide robust predictions - I’m just not sure how easy it would be to automate.

Hopefully, the GUND meeting will provide answers. . .