Grantee Research Project Results
2005 Progress Report: Development of Risk Propagation Model for Estimating Ecological Responses of Streams to Anthropogenic Watershed Stresses and Stream Modifications
EPA Grant Number: R830885Title: Development of Risk Propagation Model for Estimating Ecological Responses of Streams to Anthropogenic Watershed Stresses and Stream Modifications
Investigators: Novotny, Vladimir , Bartosova, Alena , Manolakos, Elias , Ehlinger, Timothy
Institution: Northeastern University , Illinois State Water Survey , University of Wisconsin - Milwaukee
EPA Project Officer: Packard, Benjamin H
Project Period: June 1, 2003 through May 31, 2006 (Extended to May 31, 2007)
Project Period Covered by this Report: June 1, 2005 through May 31, 2006
Project Amount: $747,760
RFA: Developing Regional-Scale Stressor-Response Models for Use in Environmental Decision-making (2002) RFA Text | Recipients Lists
Research Category: Aquatic Ecosystems , Ecological Indicators/Assessment/Restoration
Objective:
The goal of this research is the development of a regionalized watershed-scale model to determine aquatic ecosystem vulnerability to anthropogenic watershed changes, pollutant loads, and stream modifications (such as impoundments and riverine navigation). The model will assist watershed managers in their decisions on methods to mitigate stream degradation and biological impairment, assess potential watershed impacts, and identify watershed restoration opportunities. The layered hierarchical model system, developed by Artificial Neural Network (ANN) modeling and analysis, will be based on probabilistic risk propagation and linking the stresses with ecologic endpoints, from physical attributes of the watershed and water body and pollutant loadings at the lowest level to measures of biotic integrity, such as the Index of Biotic Integrity (IBI), at the highest level.
The main objectives and outcomes of the research are: (1) developing a model that would consider pollutant effects of impoundments for navigation and other purposes, channelization, watershed modification, and riparian corridor and land use changes as the key root stressors, using primarily data obtained from Midwest streams; (2) developing a layered hierarchical progression of risks from the basic root stressors to the biotic endpoints (fish and macroinvertebrate IBIs); (3) using the model to study the possibility of mitigating the stressors in a way that would have the most beneficial impact on the biotic endpoints; (4) developing a manual for watershed managers and other users; and (5) investigating adaptability and transferability of the model to a stressed New England stream.
Progress Summary:
In the first phase of our research (first two years) we established and demonstrated that a particular ANN structure, Self Organizing Maps (SOMs), can be used to pattern and profile the distribution of stressors in large stream ecosystems, and discriminate sampling sites according to multi-stressor impact. SOMs were used to analyze the biological integrity of the streams in the States of Ohio and Maryland. This type of ANN analysis is called unsupervised learning. Using Canonical Correspondence Analysis (CCA) and Principal Component Analysis (PCA) the research teams are identifying ranking of stressors as to their impact on IBI and Cluster Dominating Parameters (CDP).
In the second phase of the research (third year) we capitalized on the very promising results of the first phase. We added supervised ANN based prediction capabilities as a step following the hierarchical unsupervised nonlinear clustering of sampling sites according to fish IBI metrics vectors distribution. Our objective is to be able to build simple yet powerful models that could be used to predict the IBI and its metrics (both fish and macroinvertebrate).
The team at the University of Wisconsin (Milwaukee) has assembled and is using extensive fish, habitat, and land cover databases for the State of Wisconsin and has developed a GIS based system to be used for analyzing impacts of stream habitat and fragmentation, hydrological and hydraulic parameters, and watershed land use on stream biological integrity, also using SOMs and correlating it to the various “stressor” metrics calculated from the GIS Database.
Because habitat parameters have been identified as the stressors that have the greatest impact, significant effort is now devoted by both teams to synthesize and analyze the habitat metrics. Most of the current metrics identified, for example, in the Rapid Biotic Assessment Protocols, are observational, that is, they cannot be predicted. Our teams are striving towards developing predictive habitat indices and measures.
We have also completed our database for storing and querying vast amounts of data from several states (Ohio, Maryland, Massachusetts, Wisconsin, and Minnesota).
An extensive effort was devoted in the first 6 months of 2007 to the development methodology and execution of the supervised ANN modeling that was finding the best relationship between the IBI, its metrics, and environmental variables (habitat metrics, land use, and water quality). Supervised ANN modeling is a powerful tool to predict fish IBI or fish metrics from the watershed stressors. In this extensive effort, datasets from the Ohio Environmental Protection Agency (OEPA) were used.
OEPA datasets had 1848 records of 34 watershed stressor parameters, IBI, and 12 fish metrics that were used to construct IBI for 1193 stations. This set included missing values for the watershed stressor parameters at a few stations. For the optimal use of the available data, missing values were substituted either with the average values (if the number of missing values was small) or with the kriged GIS values (if missing values were large in number).
Maryland data were extracted from Maryland Biological Stream Survey (MBSS) data at 955 sampling stations for the period of 1995–1997. The sampling stations were spread throughout western (Appalachian Plateau), central (Piedmont), and eastern (Coastal Plains) regions. The dataset contained quantitative and qualitative data of chemical, habitat, and land use parameters. In addition, Benthic Index of Biological Integrity (BIBI) and Physical Habitat Index (PHI) were also available.
Three approaches were used to predict fish IBI: single supervised ANN for the whole preprocessed Ohio data, three supervised ANNs for the clustered data, and multivariate regression models. The best set of input watershed stressors was selected from CCA and Principal Component based analyses. A three layer ANN (input layer-hidden layer-output layer) was employed for prediction. Sixty percent of the data was used for training, and the rest was used in validation and testing.
For developing prediction models from data, four approaches were tested:
- Traditional multivariate regression model.
- Supervised ANN models with whole dataset.
- Supervised ANN models for clusters, and SOM based prediction models.
Over 80 IBI models were developed and analyzed from the Ohio data by supervised learning. The best models can account for more than 50% of the IBI variability.
The University of Wisconsin at Madison (UW-M) team demonstrated in a similar analysis that the SOM clusters are detecting and patterning differently than the traditional IBI developed and calibrated for Wisconsin. What the SOMs are showing is that each of the clusters is defining a unique class of fish community each of which may have its own combination of natural and anthropogenic stressors. This may help one rethink the standard concept of degraded versus non-degraded sites on a large geographic level.
Future Activities:
The time between now and May 2007 will be devoted to completion of the entire project. It is expected that subcontractors will finish their work and submit technical reports by the end of 2006. This work will include the results of the biological studies in Wisconsin and the risk model by the University of Illinois.
The Northeastern University team will complete the following studies:
- Analysis of habitat clustering for Ohio, Maryland, Minnesota, and Wisconsin.
- Macroinvertebrate clustering and modeling for Ohio, Maryland, and Massachusetts (test state).
- Impact of impoundments.
- Manual for SOM analyses and modeling.
- The final report will be compiled and written in 2007.
A workshop proposal will be sent to the U.S. Environmental Protection Agency (EPA) in July–August 2006.
Journal Articles:
No journal articles submitted with this report: View all 18 publications for this projectSupplemental Keywords:
RFA, Scientific Discipline, Water, ECOSYSTEMS, Ecosystem Protection/Environmental Exposure & Risk, Water & Watershed, Aquatic Ecosystems & Estuarine Research, Monitoring/Modeling, Aquatic Ecosystem, Regional/Scaling, Terrestrial Ecosystems, Environmental Monitoring, Ecological Risk Assessment, Ecology and Ecosystems, Watersheds, neural net modeling, risk assessment, ecosystem modeling, anthropogenic stress, ecology, ecosystem assessment, watershed, ecosystem monitoring, decision making, ecological variation, risk propagation model, regional scale impacts, water quality, aquatic ecosystems, environmental stress, ecological indicators, ecology assessment models, ecosystem stress, watershed assessment, ecological models, water monitoring, adaptive implementation modeling, stress responseRelevant Websites:
http://www.coe.neu.edu/environment 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.