Grantee Research Project Results
Final Report: Development of an Environment for Assembling Spatial Ecological Models Using Reusable Components
EPA Grant Number: R827958Title: Development of an Environment for Assembling Spatial Ecological Models Using Reusable Components
Investigators: Weinstein, David A. , Swaney, Dennis P. , Hong, Bongghi , Wenderholm, Elaine , Woodbury, Peter
Institution: Boyce Thompson Institute for Plant Research , The State University of New York
EPA Project Officer: Packard, Benjamin H
Project Period: January 1, 1997 through December 31, 1999 (Extended to May 1, 2004)
Project Amount: $308,120
RFA: High Performance Computing (1996) RFA Text | Recipients Lists
Research Category: Human Health , Aquatic Ecosystems , Environmental Statistics
Objective:
We have developed a modeling framework to support grid-based simulation of ecosystems at multiple spatial scales. This system is called the Ecological Component Library for Parallel Spatial Simulation (ECLPSS). ECLPSS separates chemical, physical, and biological concepts from simulation support functions. This allows biologists to readily construct new spatially explicit models without becoming computer scientists, while computer scientists can improve the modeling framework without being ecologists. ECLPSS is designed to allow ecologists to build robust spatially explicit simulations of ecological processes from a growing library of reusable interchangeable components. Some of these components represent specific ecological processes, such as how ozone alters the growth of individual trees. Others provide such simulation support as reading and writing files in different formats to allow interoperability with geographic information systems (GIS) and database management systems. Object-oriented techniques are used to assist in automating many modeling tasks, but users do not need to be computer experts to create models. ECLPSS improves the robustness of ecological models because individual components can be easily replaced without interfering with other aspects of model behavior. The framework is designed to operate on multiple platforms and be used across networks via a World Wide Web-based user interface. ECLPSS is designed for use with single processor machines for small models, but also on distributed networks in order to simulate large regions with complex interactions among many individuals or ecological compartments.
In order to test the system, we created a model to evaluate the effect of tropospheric ozone on forest ecosystem dynamics. This model is named OM and focuses on two tree species important in the Northeastern USA: red oak and sugar maple. The OM model is a reduced-form version of two existing models: TREGRO, which represents individual tree physiology (Weinstein, et al., 1992; Weinstein, et al., 1991), and Zelig, which represents forest stand growth and succession (Urban, 1990; Urban, et al., 1991; Urban, 1993). These models have previously been linked to examine ozone effects on individual tree growth and interspecies competition (Hogsett, et al., 1994; Laurence, et al., 2000). However, the linkage between these models was limited because the dynamics of forest ecosystems are dependent upon spatial interactions, but the models were not capable of considering spatial interactions across landscapes. To solve this problem, the OM model captures key aspects of TREGRO and Zelig in much less detail. Components within OM represent growth, development, recruitment, and mortality of up to 20 tree species. The shoot and root growth of individual trees in response to temperature, ozone, and drought stresses are modeled, as are the spatial dynamics of competition, gap development, and tree succession. Additionally, we produced components that perform debugging, input/output, and other supporting functions. The growth of all trees within approximately 1 hectare of forest near Ithaca, New York, USA, was modeled for 100 years with both above-ambient and subambient levels of ozone exposure.
Summary/Accomplishments (Outputs/Outcomes):
Platform Features
The ECLPSS modeling framework, implemented in JavaTM, was developed to ease the design, implementation, and debugging of models. In addition to supporting both bottom-up and top-down design and automatic documentation, biologists easily can rearrange and experiment with model structure, grids, grid cell size, and scale.
Models can be developed in ECLPSS through the use of the following attributes:
- Models are specified using a suite of Graphical User Interface (GUI)-based tools that facilitate and support the design practices that are natural to biologists. Additionally, the framework takes advantage of cutting-edge technology afforded by Java, which supports graphical Web-centric development, collaboration, and dissemination of models.
- The ease of developing models is largely because of the declarative nature provided by the GUI editors. Wherever applicable, framework users express the parts of a model at a high level of abstraction as specifications. Specifications relieve the user of the need to write (and rewrite) code that manages storage and other error-prone programming tasks. These tasks range from the mundane (explicit declaration of data structures and loops) to the esoteric (parallel fork/join threads for shared memory parallel model execution). The ECLPSS compiler, in turn, uses these specifications to generate (parallel) simulation code.
- All specifications are encoded and stored as XML documents. In addition to being text-based (and therefore application and platform independent), XML has become an industry standard for data exchange and Web publishing. XML tools are under constant development. For example, XSLT (XML style sheet language for transformations) may be used to translate XML documents into HTML, PDF, or Postscript files as an alternative way to display data. Such a tool may be used in the model documentation.
- Most implementation details are (rightfully) invisible to the user. As a consequence, the user does not need a deep understanding of most of the generated simulation code. This is particularly important for the code that is generated for parallel execution. The (parallel) simulation code, because it is generated by the Eclpss compiler, is correct and does not require debugging. This invisibility also permits framework developers to add independent enhancements (such as optimization and additional functionality) to the compiler without the need for framework users to modify existing models.
- Both the framework and framework-generated simulation code are platform independent. These have been tested on a variety of platforms: Apple Powerbook (under OS X); Sun/Solaris uniprocessors and an 8-processor SunFire shared-memory parallel machine; Intel architectures under several Windows operating systems (98, 2000, NT, XP) and several versions of Linux (SuSE, RedHat).
- The structure of Eclpss components allows them to be shared easily. There is no restriction on the internal structure of component code (i.e., users may write component code that contains loops), but only on access to the state variables on the grid via methods that get and set their values.
In addition to simplicity for the user, the ECLPSS compiler generates efficient (parallel) simulation code. The framework-imposed restricted access to state variables not only eases the conceptual task of the user, but also the analytical task of the compiler. Thus, by rigorously enforcing just enough structure, a model is relatively easy for the ECLPSS compiler to analyze. By not imposing too much structure, users have a high degree of modeling freedom.
Because grid access methods are compiler-generated, simulation code may be generated that performs boundary checking for each read and write to the grid. Users then are assured that all grid accesses conform to the grid accesses that have been specified for the model.
Demonstration of the ECLPSS System
To demonstrate the usefulness of the ECLPSS system, we developed a model of N dynamics, Simple Nitrogen Cycle (SINIC) within this system, examining the interaction of N and ozone deposition on nitrate (NO3) retention.
From 1964 through 1994, the pattern of NO3 export from Watershed 6 at Hubbard Brook Experimental Forest (HBEF) in New Hampshire exhibited 10 years of high export (1968-1977) followed by 12 years of low export (1978-1989), with four “spikes” in 1970, 1973, 1976, and 1990. Disruptions of N cycling by soil freezing, insect defoliation, or drought have been suggested to explain this pattern. With our model, we demonstrated that most of the long-term pattern can be reproduced without explicit consideration of these events. Comparisons of simulated N fluxes between high- and low-export years suggested that inorganic N input to the soil, from both atmospheric N deposition and N mineralization, was significantly higher during periods of high-streamflow NO3 flux than in low periods. Simulated inorganic N pools (ammonium and NO3) and fluxes (nitrification, plant uptake, denitrification, and ammonia volatilization) also were significantly higher in these periods. By swapping the time sequences of inorganic N input between high- and low-export years, it was shown that N mineralization, not atmospheric N deposition, drives the simulated long-term pattern. Although simulated nitrification showed a stronger relationship with measured streamflow NO3 flux than N mineralization, nitrification rate depended on the availability of soil ammonium supplied from N mineralization. Because N mineralization in the model varies only with soil temperature and moisture, we concluded that shifts in the interaction of these two variables over time produced the shifts in NO3 stream exports.
It also was relatively easy to construct a spatially explicit version of this model within the Eclpss system. We then demonstrated that available information on spatial heterogeneity in biotic, topographic, and climatic variables within a forested watershed—HBEF Watershed 6- was necessary to reproduce the observed elevational pattern in stream NO3 concentration during the 1982-1992 period. Five gridded maps (N mineralization factor, N uptake factor, precipitation, elevation, and soil depth factor) were created from spatial data sets and successively added to the spatially explicit model SINIC-S as spatially varying input parameters. Adding more spatial information generally improved model predictions, with the exception of the soil depth factor. Ninety percent of the variation in the observed stream NO3 concentration was explained by the combination of the spatial variation of the N mineralization and N uptake factors. Simulated streamflow NO3 flux at the outlet point was improved slightly by introducing spatial variability in the model parameters. The model exhibited substantial cell-to-cell variation in soil N dynamics and NO3 loss within the watershed during the simulation period. The simulation results suggest that the spatial distributions of forest floor organic mass and standing biomass are mostly responsible for creating the elevational pattern in stream NO3 concentration within this watershed.
We then employed a Bayesian parameter estimation technique for improving estimates of simulation model input parameters and applied the technique to the N-cycle model SINIC, which has been used to simulate the streamflow and streamflow NO3 flux at HBEF Watershed 6 during the 1964-1994 period. Uncertainty in initial estimates of model input parameters was incorporated by replacing each estimate with a probability distribution of values, or “prior” distribution, usually centered at the initial estimate and having a large variance. These prior distributions were then “updated” by incorporating available data on model output variables, producing a “posterior” probability distribution of parameter values. Applying this technique, the level of uncertainty in input parameters was reduced substantially by incorporating the observations on streamflow and streamflow NO3 flux. Several key parameters used for calculating the rates of N mineralization and N uptake were identified as controlling the predicted NO3 export from this watershed. Strong interdependencies, measured using posterior correlations, existed among input parameters describing N mineralization and N uptake. The posterior distribution of predicted yearly streamflow NO3 flux shifted from year to year, with relatively large uncertainties in years with high-streamflow NO3 flux.
Finally, the effects of air pollutant ozone on soil N dynamics and spatial and temporal patterns of streamflow NO3 flux at HBEF Watershed 6 during the 1964-1994 period were assessed using the N-cycle model SINIC. The ozone effects were assumed to occur via two distinct mechanisms; reduction in surface conductance and reduction in plant N demand. Aggregated (one-celled) and spatially explicit (208-cell) versions of the SINIC N flux model were run under various ozone exposure scenarios to investigate changes in the temporal and spatial patterns of streamflow N flux in response to altered ozone exposure. The uncertainty in model prediction was evaluated using the variability in predicted outputs in 5,000 model runs made with 5,000 randomly chosen sets of parameters. The ozone was estimated to cause an additional 0.042 gN/m2/year to be added to the mean annual streamflow NO3 flux, which is about 12 percent of the mean annual streamflow NO3 flux simulated under the ambient level of ozone. The range of the 95 percent credible interval of this estimate was 0.002-0.083 gN/m2/year, or 0.72-27.3 percent of the mean annual flux. Thus, large uncertainties existed in this estimate, suggesting that it may be difficult to identify the ambient ozone effect on streamflow NO3 flux from field studies. There was little change in the predicted elevational pattern of stream NO3 concentration in response to the elevated level of ozone, and ozone induced an increase in the stream NO3 concentration at all elevations. Analysis of the simulation result demonstrated the importance of the plant buffering capacity in controlling the extra N released into the soil through N mineralization and preventing the NO3 loss from the watershed.
Conclusions:
The ECLPSS framework is very useful for producing simple models for investigating multimedia pollutant transport problems. The model created to demonstrate the use of this system, SINIC, identified key processes regulating the ability of a forested watershed to retain N in the presence of excess N deposition and ozone deposition. We also demonstrated how a model constructed in this system could be analyzed thoroughly for its uncertainty properties to identify the parameters most affecting the ability of the model to produce accurate predictions.
Journal Articles on this Report : 4 Displayed | Download in RIS Format
Other project views: | All 6 publications | 4 publications in selected types | All 4 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Hong BG, Strawderman RL, Swaney DP, Weinstein DA. Bayesian estimation of input parameters of a nitrogen cycle model applied to a forested reference watershed, Hubbard Brook Watershed Six. Water Resource Research. 2005;41(3):Art. No. W03007 |
R827958 (Final) |
not available |
|
Hong BG, Swaney DP, Woodbury PB, Weinstein DA. Long-term nitrate export pattern from Hubbard Brook Watershed 6 driven by climatic variation. Water, Air, and Soil Pollution. 2005;160(1-4)293-326 |
R827958 (Final) |
not available |
|
Hong B, Weinstein DA, Swaney DP. Assessment of ozone effects on nitrate export from Hubbard Brook Watershed Six. Environmental Pollution 2006;141(1):8-21. |
R827958 (Final) |
not available |
|
Hong BG, Swaney DP, Weinstein DA. Simulating spatial nitrogen dynamics in a forested reference watershed, Hubbard Brook Watershed 6, New Hampshire, USA. Landscape Ecology. 2006;21(2):195-211 |
R827958 (Final) |
not available |
Supplemental Keywords:
Bayesian, uncertainty, parallel, simulation, Java, object oriented, biogeochemistry, forested watershed, Hubbard Brook, streamflow nitrate flux, nitrogen dynamics, modeling, spatial heterogeneity, model building, modeling methodology, model reuse, model construction tools, process model, spatially explicit model, component-based modeling, reusable code, ecological model,, RFA, Scientific Discipline, Ecosystem Protection/Environmental Exposure & Risk, Geography, Forestry, computing technology, Ecology and Ecosystems, ecosystem modeling, geospatial data, environmental decision making, HPCC, supercomputing, computer science, geographical information systems, reusable components, computer simulation modeling, component-based software, data analysis, GIS, information technology, spatial modeling, TREGRORelevant Websites:
http://www.cs.oswego.edu/~wender 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.