Grantee Research Project Results
Final Report: A High Performance Analytic Element Model: GIS Interface, Calibration Tools, and Application to the Niagara Falls Region
EPA Grant Number: R827961Title: A High Performance Analytic Element Model: GIS Interface, Calibration Tools, and Application to the Niagara Falls Region
Investigators: Rabideau, Alan J.
Institution: The State University of New York at Buffalo
EPA Project Officer: Packard, Benjamin H
Project Period: March 1, 2000 through February 28, 2003 (Extended to April 1, 2004)
Project Amount: $996,545
RFA: Computing Technology for Ecosystem Modeling (1999) RFA Text | Recipients Lists
Research Category: Environmental Statistics
Objective:
The primary objectives of the research project were to:
(1) develop and test a groundwater flow model based on the Analytic Element Method (AEM) for very large-scale problems using massively parallel computers; (2) develop and compare two calibration/optimization tools for use with the newly expanded AEM software, based on two computational approaches—nonlinear regression and the genetic algorithm—each optimized for parallel processing; (3) develop a GIS-based graphical user interface and database for the above models; and (4) demonstrate the performance of the modeling system through case studies.
Summary/Accomplishments (Outputs/Outcomes):
Application of AEM to Very Large-Scale Problems: Overview
The motivation for this work was the development of groundwater models designed for applications at the super-regional scale (which we have defined as > 10,000 km2), an area consistent with the emerging concept of a hydrologic observatory. The AEM is an appealing technique because the computational demand is determined by the number of modeled hydrologic features, not by the size of the domain. Furthermore, the AEM provides very accurate solutions at any desired scale of resolution and does not require artificial model boundary conditions. Limitations to the common implementation of AEM include the assumption of steady-state DuPuit-Forchheimer flow. Although some recent research has addressed transient flow and leakage between multiple aquifers, these issues were not addressed in this project.
Although AEM models have been applied since the mid-1980’s, application to the super-regional problems of interest required a number of new developments, including: (1) the ability to access and process large amounts of digital data from a variety of sources; (2) the ability to map large numbers of hydrologic features to analytic elements, with a minimum of direct user involvement; (3) intelligent configuration of analytic elements to achieve a balance between hydrologic detail, accuracy, and computation; (4) efficient computational algorithms for the solution of the AEM simulation model; (5) efficient and robust calibration of models containing large numbers of parameters and/or observations; and (6) the ability to process and visualize a variety of model outputs over a range of scales.
The developments listed above required advances in the conceptualization, mathematics, and software implementation of AEM, which are described in the subsequent subsections.
Computational Improvements to AEM
The computational demand of an AEM model is divided into two tasks: (1) the initial calculation of site-specific element coefficients that enforce the boundary conditions (primarily surface water elevations and fluxes); and (2) the use of the AEM coefficients to calculate the groundwater potential (head) at the desired locations (e.g., generate flow nets). For super-regional models, the first task (initialization) comprises the vast majority of the required computation, which can extend to hundreds of CPU hours for very large models.
We have addressed this area in two ways: (1) development of new iterative solvers for calculating the AEM coefficients; and (2) development of specialized solvers for use with distributed memory parallel computers, using the Message Passing Interface (MPI) libraries. These developments are described in Craig, et al., 2005; and Bandilla, et al., 2005.
The successful computational improvements to the AEM resulted in: (1) over an order-of-magnitude speed-up for large AEM models using an iterative technique termed “nested superblocks” and (2) near-linear speedup for AEM models when implemented on parallel computing environments. Very long CPU times are still likely when implementing models containing thousands of surface water features, and efficient implementation on parallel workstations requires significant memory accessible by each processor. The recent availability of 64-bit processors has significantly extended the scope of tractable AEM models.
Integration With GIS
Prior to this work, implementation of very large AEM models required the user to manually process the model inputs on an element-by-element basis, a task that would be infeasible for applications containing thousands of elements. As we developed GIS-based AEM preprocessors, it became evident that additional utilities were needed to automatically process large numbers of elements. The details of these developments are summarized in Craig, et al., 2005; Rabideau, et al., 2005; and Flewelling, et al., 2005. In addition to the development of public domain GIS-based software, a significant component of this task was the development of algorithms for line simplification, polygon simplification, and the automated conversion of polygon elements to less computationally demanding circles and ellipses.
Calibration
An important premise of the original proposal was that super-regional flow modeling applications would benefit from the availability of calibration procedures based on global search procedures, rather than traditional nonlinear regression. To address this objective, we have implemented a model independent calibration software tool that includes a menu of global and gradient-based search procedures, optimized for parallel processing. Also, we have implemented a hybrid algorithm that combines the global capability of heuristic algorithms with the efficiency and diagnostic output of nonlinear regression. Application of these tools for calibrating AEM-based models is discussed in Rabideau, et al., 2005a, and Rabideau, et al., 2005b. At this time the most appropriate algorithm for calibrating super-regional AEM models remains an open question; our experience is that the calibration of large AEM models is quite sensitive to a number of issues, including the complexity of the model and the nature of the available data (Rabideau, et al., 2005).
Applications
The original research proposal emphasized a case study located in the Niagara Falls, New York region. As discussed in the Year 1 project report, this application was deemed infeasible because of the inability of existing AEM models to handle some of the three-dimensional flow features. A new case study was identified based on Wisconsin’s Northern Highland Lakes Region (NHLR), which was selected because of the presence of numerous surface water features (e.g. ~ 4,000 lakes). Development of AEM models for this has been an ongoing process, which has identified many practical issues related to AEM implementation, which in turn has motivated changes to mathematical algorithms and software.
Our approach to the case study has been to develop a suite of NHLR models of varying degrees of complexity, accompanied by a comparison of model results in terms of computational demand and numerous measures of model prediction (flow contours, water budgets, predicted travel times, etc). Preliminary results from this effort are described in Rabideau, et al., 2005; and Frederick, et al., 2005. A more comprehensive analysis will be submitted during the summer of 2005 ( Frederick, et al., 2005). Results to date indicate that the hydrologic output of NHLR models is very sensitive to the degree of geometric complexity. These results will likely motivate the development of new automated element simplification algorithms that incorporate hydrologic criteria in addition to geometric criteria. Application of the models to a smaller case study located in Western New York is discussed in Frederick, et al., 2004; and Silavisesrith, et al., 2005
Software Implementation
The concepts developed through this project have been implemented in five software tools, all of which are available for public download:
- The AEM flow simulators SPLIT (Bandilla, et al., 2005) and BLUEBIRD (Craig, et al., 2005).
- The GIS-based modeling support systems Visual BLUEBIRD (Craig, et al., 2005) and ArcAEM (Silavisesrith, et al., 2005).
- The multi-algorithm calibration tool OSTRICH (Matott, 2004), which is fully integrated into Visual Bluebird and ArcAEM, but can also be implemented as a stand-alone model-independent calibration/optimization utility.
The above software tools differ from other available AEM systems in several important respects, including the manner of GIS integration, the use of automated routines for element assembly and simplification, robust model integrity checking, efficient iterative solvers, implementation on distributed computing environments (SPLIT, OSTRICH), and direct integration with a contaminant transport model (BLUEBIRD). The two modeling support systems ¾ArcAEM and Visual BLUEBIRD ¾embed special programs for identifying and configuring elements within a GIS, but the actual flow simulations are performed by the separate stand-alone models. ArcAEM was developed as an extension for the ESRI program ArcGIS, and configures and executes the SPLIT flow simulator from within this environment. Visual Bluebird (VBB) provides much of the functionality of ArcAEM in a non-proprietary product, but without the standardization and extensive set of spatial tools afforded by ArcGIS. Both programs are fully integrated with the OSTRICH calibration tool.
The two flow simulators developed through this work ¾ SPLIT and BUEBIRD ¾utilize similar element formulations and solution procedures, but have several important differences. SPLIT, written in FORTRAN, can be implemented on a distributed memory computing environment using the Message Passing Interface (MPI) protocol, which requires special solvers. For very large-scale applications, the combination of ArcAEM and SPLIT provides both computational efficiency and powerful data management tools to users of the ArcGIS system. The Bluebird simulator, written in C++ and currently configured only for single-processor architecture, offers the advantage of object-oriented (OO) programming, which facilitates the extension to include new elements and/or multiple aquifers, and provides a simple vehicle for exporting information to other process models.
All of the software tools have extensive documentation and are available for download from . Beginning in November 2004, users were requested to provide some basic information at the time of the software download. At this time approximately 400 different individuals from over 10 countries have downloaded one or more of the software products. A short course held in June 2004 was attended by 11 users of the software (from 5 states), including Dr. Stephen Kraemer from the U.S. Environmental Protection Agency.
Extensions
Over the 5-year duration of the project, the researchers associated with the project have produced a number of new AEM developments not directly identified in the initial research proposal, including but not limited to:
- Integration of the AEM technique for groundwater flow modeling with reactive contaminant transport models, the first research of this type (Craig, et al., 2005a; Craig, et al., 2005b).
- Application of AEM modeling to the problem of hydraulic optimization of pump-and-treat systems (Matott, et al., 2005).
- Integration of the AEM with the DRASTIC method for aquifer vulnerability indexing ( Frederick, et al., 2004).
- Development of analytic elements for elliptical geometry (Suribhatla, et al., 2004), with application to modeling lakes (Rabideau, et al., 2005) and permeable reactive barriers (Bandilla, et al., 2005; Rabideau, et al., 2005).
Future of the AEM
The AEM method for modeling groundwater flow is a viable technique for modeling groundwater flow at the “super-regional” scale, and is an appealing alternative to the popular finite difference (FD) method implemented in the widely used MODFLOW. Unlike its FD counterpart, for which significant computational improvements are largely driven by ongoing advances in computing hardware, there is considerable potential for fundamental expansion to the performance and applicability of the AEM. Promising areas for future research include, but are not limited to, the following:
- Continued improvement in computational efficiency. Although AEM models benefit from specialized iterative solvers formulated to exploit the unique features of AEM (e.g., superblocks), AEM developers have yet to fully explore the use of the type of sophisticated linear solvers (a component of the process) widely used for FD modeling.
- Further development of GIS-based methods for automated model processing and simplification. For the most part, our work has relied on principles derived the cartographic simplification literature, which should be modified to more directly incorporate hydrologic concepts.
- Continued integration of AEM models with models for other media, including vadose zone flow and conjunctive groundwater/surface water modeling. In this regard, our success in integrating AEM with contaminant transport modeling is encouraging.
- Continued development of large scale AEM applications.
Journal Articles on this Report : 11 Displayed | Download in RIS Format
Other project views: | All 46 publications | 15 publications in selected types | All 15 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Bandilla KW, Jankovic I, Rabideau AJ. A new algorithm for analytic element modeling of large-scale groundwater flow. Advances in Water Resources 2007;30(3):446–454. |
R827961 (Final) |
Exit Exit |
|
Becker MW, Georgian T, Ambrose H, Siniscalchi J, Fredrick K. Estimating flow and flux of ground-water discharge using water temperature and velocity. Journal of Hydrology 2004;296(1-4):221-233. |
R827961 (2000) R827961 (2001) R827961 (2002) R827961 (Final) |
Exit Exit |
|
Craig JR, Rabideau AJ, Suribhatla R. Analytical expressions for the hydraulic design of continuous permeable reactive barriers. Advances in Water Resources 2006;29(1):99-111. |
R827961 (Final) |
Exit Exit |
|
Craig JR, Jankovic, I, Barnes R. The nested superblock approach for regional-scale analytic element models. Groundwater 2006;44(1):76-80. |
R827961 (2000) R827961 (2001) R827961 (2002) R827961 (Final) |
Exit Exit Exit |
|
Craig JR, Rabideau AJ. Finite difference modeling of contaminant transport using analytic element flow solutions. Advances in Water Resources 2006;29(7):1075-1087. |
R827961 (Final) |
Exit Exit |
|
Fredrick KC, Becker MW, Flewelling DM, Silavisesrith W, Hart ER. Enhancement of aquifer vulnerability indexing using the analytic-element method. Environmental Geology 2004;45(8):1054-1061. |
R827961 (2000) R827961 (2001) R827961 (2002) R827961 (Final) |
Exit Exit |
|
Matott LS, Rabideau AJ, Craig JR. Pump-and-treat optimization using analytic element method flow models. Advances in Water Resources 2006;29(5):760-775. |
R827961 (Final) |
Exit Exit Exit |
|
Rabideau AJ, Matott LS, Jankovic I, Craig JR, Becker MW. Influence of numerical precision on the calibration of AEM-based groundwater flow models Environmental Geology 2005;48(1):57-67. |
R827961 (Final) |
Exit Exit |
|
Rabideau AJ, Suribhatla R, Craig JR. Analytical models for the design of iron-based permeable reactive barriers. Journal of Environmental Engineering 2005;131(11):1589-1597. |
R827961 (Final) |
Exit Exit |
|
Rabideau A, Craig J, Silavisesrith W, Fredrick D, Flewelling D, Jankovic I, Becker M, Bandilla K, Matott L. Analytic-element modelling of superaregional groundwater flow: concepts and tools for automated model configuration. Journal of Hydrologic Engineering 2007;12(1):83-96. |
R827961 (Final) |
Exit |
|
Suribhatla R, Bakker M, Bandilla K, Jankovic I. Steady two-dimensional groundwater flow through many elliptical inhomogeneities. Water Resources Research 2004;40(4):Art. No. W04202. |
R827961 (Final) |
Exit Exit |
Supplemental Keywords:
water, drinking water, watersheds, groundwater, exposure, engineering, social science, hydrology, geology, modeling, analytical, northeast, central, Great Lakes, Wisconsin, GIS, parallel processing, cyberinfrastructure, uncertainty analysis,,Relevant Websites:
http://www.groundwater.buffalo.edu Exit
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.