High Performance Computing and Geographical Information Systems (GIS)

Several high performance computing projects are part of the Grid Computing project at UTK.


Computational ecology is an emerging multidisciplinary field, similar in concept to the cell and molecular emphasis of bioinformatics, which applies modern computational methodology to key problems at higher levels of biological organization. The goal of computational ecology is to combine realistic models of ecological systems with the often large data sets available to aid in analyzing these systems, utilizing techniques of modern computational science to manage the data, visualize model behavior, and statistically examine the complex dynamics which arises. A major factor which differentiates computational ecology from much of what is currently considered computational biology is the inherent parallel nature of ecological systems. Organisms interact concurrently across space. This is particularly a problem for models which follow individual organisms, called individual-based models (IBMs), which are a form of agent-based models developed to be applied specifically to ecological questions.

Due to the variety of scales involved in natural resource management, the use of a multimodeling approach is appropriate. In this, different models with potentially different mathematical/computational forms are chosen for different components of the system, based upon the necessary levels of detail and available data for different components. These models are linked into a multimodel to evaluate the interactions of biotic and abiotic components. Grid computing provides a natural framework to develop toolsets for ecological multimodels due to the dispersed nature of database, software and computational resources, in addition to the general lack of appropriate hardware at many natural resource agencies. The challenges go well beyond the current state-of-the-art in geographic information systems (GIS), requiring spatially-explicit dynamic models linked to spatio/temporal optimization schemes.


We developed a prototype system of integrating geographic information systems (GIS), fire spread model, and spatial control problem. The GIS software environment used in this prototype include the Environmental Systems Research Institute's (ESRI) ArcGIS 9, ArcObjects, and Visual Basic programming. A simple fire spread model based on a Cellular Automata (CA) approach is implemented as a custom tool in ArcGIS Desktop. Users can interactively specify fire starting location(s) and fire break location(s) on the map display and choose wind direction and wind speed via the user interface. The CA model will then generate a simulated outcome that can be visualized as an animated image of the fire spread process. The CA-based model incorporates several probabilistic rules and a stochastic component to simulate fire spreads. In addition, the model allows fire to jump to non-adjacent cells based on a distance decay function and the fuel load level at each cell. It takes about 2 seconds to run this CA-based fire spread model on a 100*100 grid, about 10 seconds on a 500*500 grid, and less than 1 minute on a 1000*1000 grid. Figure 1 illustrates the user interface of this CA-based fire spread model. Figure 2 shows the outcome of a simulation run.

Figure 1
figure 1
Figure 2
figure 2

This prototype system also includes a Genetic Algorithm (GA) module, which is coupled with the CA-based fire spread model, to find the optimal location of a fire break. Due to the combinatorial nature of this optimal spatial control problem, the current implementation is limited to one fire break with a fixed length and a fixed orientation. Users specify the number of generations and the population size for the GA-based module (see Figure 3). The analysis module then reports the best fire break location. However, this optimal location search module is computationally intensive and is not practical for solving problems of large data size. The project team is currently testing a strategy for coupling the GIS environment with a high performance computing (HPC) design that can steer the GIS-based optimal location search module by significantly reducing the search space size.

Figure 3
figure 1

Related Papers

Testing the robustness of management decisions to uncertainty: space, relativity, and uncertainty in Everglades restoration scenarios. (Manuscript).

A grid service module for natural-resource managers. 2005 IEEE Internet Computing.

Design and implementation of a parallel fish model for south florida. (Manuscript).

On parallelization of a spatially-explicit structured ecological model for integrated ecosystem simulation. (Manuscript).

A parallel structured ecological model for high end shared memory computers. (Manuscript).

A parallel fish landscape model for ecosystem modeling. (Manuscript).

A parallel implementation of ALFISH: simulating hydrological compartmentalization effects on fish dynamics in the Florida Everglades. (Manuscript).