In today's complex world, just about everything is interrelated. UIUC Geography Professor Bruce Hannon and U.S. Army Construction Engineering Research Laboratory (CERL) Senior Scientist James Westervelt are working on computational methods to associate interrelationships in many fields.
From ecology to economics, Hannon's classes in dynamic modeling help students understand the underlying principles of modeling. Last spring, his Advanced Ecological Modeling class simulated the plight of the Desert Tortoise (gopherus agassizii) for Fort Irwin, CA, site of army tank training in the Western Mojave Desert. The Desert Tortoise is both an endangered species and a species considered indicative of desert ecological conditions.
Fort Irwin receives only 7 inches of rain yearly, with temperatures reaching up to 110 degrees Fahrenheit during the summer. Its environment is stable. "General Patton's tank tracks [from World War II maneuvers] are still visible in this area," says Westervelt. To best depict the area on a computer, the fort was divided into a grid of 56 x 56 cells with each measuring 1 square kilometer. Changes in this ecosystem over the course of a century were simulated using a time-step of 1 month.
"The tortoise is a long-lived species, with a life span similar to ours," says Hannon. "So we must run the model for a long time to accurately depict any impact on their lifecycle."
"STELLA lets us bring people together who are experts in science, but who may have little or no programming and math experience," Hannon explains. "In our case, 'experts' are represented by teams of students who routinely link out to the pertinent experts around the campus and the nation. The basics of STELLA are so easy to learn that previous programming expertise isn't essential."
Model-building is simplified through STELLA's three-step method in which equations that formulate simulations are automatically generated. In the first stage, called mapping, building-block icons construct a graphical representation of input parameters. In the second phase, called the model phase, the software automatically creates equations that are needed to simulate processes. The simulation, or third step, can be viewed as graphs, tables, or animations.
"The key is to have STELLA as the common language among the people who are putting together the basic cellular model," says Hannon. "You have two choices in modeling software: either everyone can understand and use it, or you must have a modeling guru who works along with the group. The old way was to use a modeling guru, and that leads to what I call a 'cult of the modeler.' I don't care for that approach, because people who participate in the project lose track of their contribution and how it fits into the model. Eventually they lose ownership in the whole model and cease to believe in its results. This new process provides them with an understanding of what they are doing every step of the way, and they believe in the final output. The experts stay with the process at all times."
Hannon and Westervelt divided the class into small working groups, each building a sector of a larger model that could be run independent of the whole. Group members constructed and tested their part during the week, assembling with the class once weekly to build the full cell model.
Westervelt's Geographical Resources Analysis Support System (GRASS) has been used in models developed in past courses. GRASS is a geographic information interface system. "It is not a database, but a process," says Hannon. "You can make databases with it. It is raster-based, rather than vector- based. While GRASS is public-domain, thanks to Jim, both can be used, depending on the situation." Westervelt is finishing his doctorate in Urban and Regional Planning on the construction of a general modeling environment that allows a seamless connection from desktop computers to an HPCC environment.
Initialization of cells in the Desert Tortoise model required generating a vegetation map with the existing Geographic Information System (GIS) maps of Fort Irwin (including elevation, slope, watershed, roadways, etc.) and 200 random transects of 100 meter length (noting shrubs, grass, cover, etc.). These two types of information were then combined in a back-calculated neural network used to describe the vegetation distribution, by type, across the entire area.
Other collaborators in the Desert Tortoise modeling project are Scott Isard, UIUC Department of Geography; Tony Krzysik, CERL Environmental Sustainability Laboratory; Kevin Seel, Oak Ridge Associated Universities; and Eric Lambert, Illinois State Historical Survey.
"It was excellent to see the interrelationships in modeling," says Safia Aggarwal, UIUC geography graduate student. Aggarwal will use Hannon's methodology in her thesis on zebra mussels. "The techniques in dealing with the zebra mussel so far have been crude and mostly inadequate. I would like to look at other forms of control, perhaps biological. For example, can a natural predator be prodded into affecting the zebra mussel population?"
Michelle Duffield, English major working towards a degree in Environmental Law, felt the class helped her understand how environmental factors interrelate. "I hope to be able to apply what I have learned in this class to understanding EPA laws that may soon be based on environmental modeling."
"By properly designing the model, we can use it to test ecological theory. In essence, we have created a virtual ecosystem that can also be used as a testbed for ecological theories."
One theory Hannon and Westervelt desire to test involves the mixing of tortoises. "The turtles we are modeling live in clumps scattered across the desert. They get along in their areas for a while, but occasionally they must mix to sustain the genetic makeup."
Hannon says that roads endanger turtles and disturb the mixing process. There are likely no solutions to this problem without bringing harm to the species. "You cannot just run in and take one turtle from Group A to Group B, because you might pick up a diseased one that will kill off the population of Group B. We used this mixing idea in forming our model. We will be able to show turtles in their natural conditions, then add roads and other aspects of human activity to see how each impacts the population level of turtles in the long run."
The past summer Hannon and Westervelt worked with Thomas Maxwell, University of Maryland at College Park, to convert the STELLA II code to C+ or Fortran for the tortoise model. Hannon and Westervelt then plan to run it on a distributed network of Sun workstations in CERL's laboratory on campus.
The class ended with a complete model of the tortoise and its ecosystem for the generic Mojave cell that was 1 kilometer square together with a full set of initializing GIS "maps" for the fort area. Currently Hannon and Westervelt are transforming the STELLA equations to C+ and arranging the necessary networks of workstations to run the multicellular modeling conglomerate. Their results will be posted on the WWW along with the Sage Grouse model.
The Desert Tortoise model will predict increasing vehicular effects on green vegetative cover due to soil compaction. Vegetation could shift from shrubs to annuals, meaning less food for the tortoise. In addition, there may already be observable impact from the army's training exercises. The Desert Tortoise now congregates in certain areas of the range, rather than the more random widespread distribution expected.
Hannon and Westervelt are well on their way to helping the army understand how its use of the Mojave Desert impacts the Desert Tortoise. Perhaps more importantly, they are helping students understand many interrelated factors that control and affect their environment.
