Final Report to the National Science Foundation on Grant BIR-9318160 to The University of Tennessee, Knoxville Parallel Processing for Individual-Based Ecological Models Submitted by: Louis J. Gross (gross@tiem.utk.edu) The Institute for Environmental Modeling and Departments of Ecology and Evolutionary Biology and Mathematics University of Tennessee, Knoxville, TN 37996-1610 July 24, 1997 Part II - Summary of Completed Project Individual-based models in ecology track the behavior, growth and reproduction of many individuals in a population or community, requiring substantial computational resources for which parallelization methods can be appropriate. This project dealt with three types of ecological models. For a spatially-explicit rule-based model of white-tailed deer and Florida panther in South Florida, which tracks the behavior of individual organisms on realistic landscapes, we developed parallel implementations using a 32 processor Connection Machine CM-5, a 4000 processor Maspar MP-2, and a network of workstations using the Message Passing Interface MPI. A second model investigated was for a structured population tracking the dynamics of many cohorts of individuals within the populations making up a community. As an example, a Daphnia model was linked to a fish model, and parallel implementations were developed using the CM-5, Parallel Virtual Machine (PVM) run over a network of workstations, and a 40 processor IBM SP-2. A third model involved a community of mussel species along linear arrays of cells representing different spatial cells of a river, implemented on the MP-2. In all cases, a general result was that parallel implementations require different underlying biological assumptions than those used in serial versions of the models, and thus do not necessarily produce exactly the same model behavior as a serial implementation. Part III - Technical Information Investigators and their affiliations: L. J. Gross, T.G. Hallam, H-K. Luh and S. Ramachandramurthi, The Institute for Environmental Modeling, UTK; D. DeAngelis, Department of Biology, University of Miami; M. Leuze, Joint Institute for Computational Science, UTK; M. Berry, Computer Science Department, UTK; M. McKinney, Mathematics Department, NC State Univ.; C. Abbott and L. Mellott, Grad Students, Computer Science Department, UTK; J. Nichols, Grad Student, Mathematics Department, UTK; H-L. Lee, Grad Student, Ecology Program, UTK. The award period for this project was initially 5/94-4/96 and was extended to 4/97. Over the time period of the award, it has supported in part the following individuals on post-doctoral or student research assistant positions: Dr. Hang-Kwang Luh, Ph.D. Zoology. Research Assistant Professor, August 1994-September 1996 (currently on staff of the Entymology Department, Oregon State University) Dr. Siddharthan Ramachandramurthi, Ph.D. Computer Scinece. Research Assistant Professor, August 1994-September 1996 (currently on the staff of LSI Logic, Inc., Boston, MA) Dr. Hooi-Ling Lee, Ph.D. Ecology, Graduate Research Assistant, August 1994 - August 1996 (currently in a visiting position at University Sains Malaysia) Catherine Abbott, M.S. Computer Science, Graduate Research Assistant, December 1994 - August 1995 (currently working for Resource Technology, Longwood, FL) John Dempsey, Graduate Research Assistant, August 1994 - November 1994 (currently pursuing M.S. in Computer Science at Univ. of Albany) Linda Mellott, M.S. Computer Science, Graduate Research Assistant, August 1995 - December 1996 (currently employed in Washington, D.C.) Jeffrey Nichols, Graduate Research Assistant, August 1994 - December 1996 (currently completing Ph.D. in Mathematics, degree expected December 1997) Publications acknowledging support from this award: Abbott, C. A., M.W. Berry, E. J. Comiskey, L. J. Gross and H.-K. Luh. Computational models of white-tailed deer in the Florida Everglades. IEEE Computational Science and Engineering (accepted, to appear 1997) Lee, H. L. and D. L. DeAngelis. A simulation study of the spatio-temporal dynamics of the Unionid mussels. Ecological Modeling (accepted, to appear 1997) Luh, H.-K., C. Abbott, M. Berry, E.J. Comiskey, J. Dempsey, and L. J. Gross. 1997. Parallelization in a spatially-explicit individual-based model (I) Spatial data interpolation. Computers and Geosciences 1997. 23:293-304 Luh, H.-K., J. J. Mineskey and H. Bozdogan. Choosing the best predictors in regression analysis via the genetic algorithm with informational complexity as the fitness function. Computational Statistics & Data Analysis (submitted). Luh, H.-K., J. L. Gittleman, H. Bozdogan, and C. G. Anderson. Phylogeny and multivariate correlated traits: an information-based approach. Evolution (submitted). Mellott, L. E., M. W. Berry, E. J. Comiskey and L. J. Gross. The design and implementation of an individual-based predator-prey model for a distributed computing environment. J. of Supercomputing (submitted) Ramachandramurthi, S., T. G. Hallam, and J. A. Nichols. Parallel simulation of individual-based, physiologically structured populations. Mathematical and Computer Modelling 25 (12) (to appear 1997). Ramachandramurthi, S., J. A. Nichols and T. G. Hallam. Ecological Assessment in watersheds:Individual-based modeling and parallel simulations. Mission Earth Proceedings. . Proceedings, Mission Earth:Modeling and Simulation for a Sustainable Global System, Society for Computer Simulation 1996, 193-198. Theses and Dissertations: Abbott, Catherine Ann M.S. (Computer Science) A Parallel Individual-Based Model of White-Tailed Deer in the Florida Everglades. August 1995. Lee, Hooi-Ling Ph.D. (Ecology) Use of a Spatially-Explicit Model to Study the Distributional Dynamics of Unionid Mussels. December 1996. Mellott, Linda M.S. (Computer Science) A Distributed Implementation of an Individual-Based Predator-Prey Model. May 1997. Presentations made related to this project: An Invited Minisymposium on Environmental Modeling and Computation was organized at the Annual Meeting of the Society for Industrial and Applied Mathematics in Charlote, NC in October 1995. This Symposium focused on results from this project and included talks on ecotoxicology models utilizing partial differential equations for population structure and landscape-scale models using spatially-explicit simulation of individual organisms. For each application, one speaker described the underlying modeling problems, and the second speaker discussed the related computation issues, with emphasis on parallelization methods. The talks were: Dr. Thomas Hallam: Risk Assessment for Chemically Stressed Populations and Communities Dr. Siddharthan Ramachandramurthi: Parallel Computation in Structured Population and Community Models Dr. Louis J. Gross: Modeling the Everglades: Integrating Alternative Methodologies across Scales Dr. Hang-Kwang Luh: Parallel Computation for Individual-based Ecological Models at Landscape-scale Other presentations (not all of which required financial support from this award) include: 1995 Logan, UT. Conference on Mathematical Models in Population Dynamics. L. Gross and T. Hallam (2 presentations) 1995 San Diego, CA. San Diego Supercomputer Center Workshop on Computational Ecology. L. Gross 1996 Orlando, FL. Sixth Symposium on Environmental Toxicology and Risk Assessment: Modeling and Risk Assessment. L. Gross and T. Hallam (2 presentations) 1996 San Diego, CA. Mission Earth: Modeling and Simulation for a Sustainable Global System. S. Ramachandramurthy 1996 Lubbock, TX. Departments of Biology and Mathematics, Texas Tech University. L. Gross 1996 Providence, RI. Session on Advanced Technologies in Ecological Science, Annual Meeting of the Ecological Society of America. L. Gross 1996. Providence, RI. Symposium on The Interface between Theoretical Ecology and Conservation Biology, Annual Meeting of the Society for Conservation Biology. L. Gross 1996 Hamburg, Germany. Prodynamics: A workshop on progress in dynamics of ecological models. T. Hallam 1996 Trieste, Italy. Third Autumn Workshop on Mathematical Ecology. L. Gross and T. Hallam (2 presentations) 1996 Athens, Greece. International Society of Nonlinear Analysts. T. Hallam 1997 Lafayette, LA. Mathematics Department, University of Southwestern Louisiana. T. Hallam 1997 Lafayette, LA. National Wetlands Laboratory. T. Hallam 1997 Amsterdam, Holland. Department of Theoretical Biology, Vrije University. T. Hallam 1997 London, UK. Zeneca Chemical Company. T. Hallam 1997 Winrock, AR. Workshop on Aquatic Ecosystem Modeling and Assessment Techniques, Army Corps of Engineers, Waterways Experiment Station. L. Gross and T. Hallam (2 presentations) Brief description of research carried out during this project: There were three major subprojects included in this project, all of which involve the development of parallelization schemes for ecological models which track the behavior of individual organisms, making use of a variety of architectures and platforms. The three components (spatially-explicit rule-based models, structured population models derived from partial differential equations, and spatially-explicit community models) are each discussed briefly below, though much more of the details of each project component are included in the various papers which are available upon request from L. Gross, the Lead P.I. for this award. Spatially-explicit rule-based models track the behavior of individual organisms on realistic landscapes, including rules for foraging, movement, growth, mortality, and reproduction. As an example of this type of model, this project supported the development of parallel implementations of a model for white-tailed deer and Florida panther in South Florida, coupled to spatially-varying water levels and vegetation models. Parallelization was developed using a 32 processor Connection Machine CM-5, a 4000 processor Maspar MP-2, and a network of workstations using the Message Passing Interface MPI. In all cases, extensive reworking of a serial version of the model was necessary to the parallelization, with somewhat different assumptions about individual behavior needed relative to the randomized sequential procedure used in the serial version. One published paper from this project focused on data-parallel procedures for spatial interpolation of the hydrology and vegetation components, comparing performance statistics for the CM-5 and the MP-2 and indicated that speed improvements were very much a function of the architecture as well as the particular method used for spatial data interpolation. Another paper focused on alternative methods to handle hydrology, vegetation, and the coupling to deer foraging, with a parallel code developed using spatial grid partitioning on the CM-5 showing significant speedups (ranging from 9 to 27) achievable relative to a sequential version on a single processor. An additional paper discusses a parallelization of the full deer/panther model using a distributed network of workstations and the Message Passing Interface. The resulting model behavior was similar, though not identical to that from the sequential version of the model, due to somewhat different assumptions about individual movements needed in the parallel version. Moderate speedups (3 to 4) were obtainable on a network of up to 12 processors, with a major speedup limitation being the communication costs due to the need for synchronization across processors. Structured population models derived from partial differential equations track the behavior of many cohorts of individuals within a population, where each cohort may include many individuals. This project supported the development of parallel implementations of a Daphnia model, designed in part to be applied to analyze the effects of toxicants on population levels and structure, as well as the linkage of this model to a structured population model for fish, thus making a simple community model. The focus of these efforts were on the CM-5 and using PVM over a network of workstations, though the Intel iPSC/860 and the IBM SP-2 have been utilized as well. The papers published on this aspect of the project describe a general scheme for parallel simulation of individual-based, structured population models, with the development of algorithms to simulate such models. The simulation model consists of an individual model and a population model that incorporates the individual dynamics. The individual model is a continuous time representation of organism life history for growth with discrete allocations for reproductive processes. The population model is a continuous time simulation of a nonlinear partial differential equation of extended McKendrick-von Foerster type. As a prototypical example, the Daphnia population model results indicate that individual-based, physiologically structured models are is well suited for parallelization and significant speed-ups can be obtained by using efficient algorithms. Because the parallel algorithms are applicable to generic structured populations, and these serve as the building blocks for more detailed community-level or food-web models, parallel computation appears to be a valuable tool for ecological modeling and simulation. A spatially-explicit community model for freshwater mussels was developed as part of this project to analyze changes in community composition through time along various reaches of a river/lake system. The model includes age-structured populations of 35 species of mussels along linear arrays of cells representing different spatial cells of a river. This has been implemented on the MP-2, in which each river cell is handled by a single processor element. During dispersal phase, the front-end of the computer is used to combine dispersing larvae from many cells and then redistributing them back to respective processors as appropriate. The model used in the study is based on the concept of a single open marine population with space-limited recruitment, with each species of mussel in each cell having up to a maximum of 15 different age classes, and the physiological growth of the mussels is described by a deterministic function that is parameterized individually for each species. As the mussel species involved are transported by fish hosts for part of their life cycle, movements of fish need to be taken into consideration to analyze museel community dynamics. Parallelization involved the MP-2 splitting up the spatial region into cells with dispersal between neighboring cells. A typical speedup was 4 to 6 for the MP-2 relative to a single processor version run on a SUN Sparcstation 5. The key difficulty in setting up a parallel structure for this model was the tree-like structure of branching rivers. This project supported one dissertation which describes one set of methods to deal with parallelization of spatial branching patterns for models with populations or communities moving within the branching structure.