Final Report: Ecological Classification of Rivers for Environmental Assessment: Demonstration, Validation, and Application to Regional Risk Assessment Across Illinois, Michigan, and Wisconsin

EPA Grant Number: R830596
Title: Ecological Classification of Rivers for Environmental Assessment: Demonstration, Validation, and Application to Regional Risk Assessment Across Illinois, Michigan, and Wisconsin
Investigators: Seelbach, Paul W. , Aichele, Stephen S. , Bissell, Ed , Brenden, Travis , Clark, Richard D. , Cooper, Arthur , Hinz, Leon , Holtrop, Ann , Lyons, John , Mitro, Matt , Pijanowski, Bryan , Steen, Paul , Stevenson, R. Jan , Stewart, Jana , Wang, Lizhu , Wehrly, Kevin E. , Wiley, Michael J. , Zorn, Troy
Institution: Michigan State University , Illinois Department of Natural Resources , Illinois Natural History Survey , Michigan Department of Natural Resources , Purdue University , United States Geological Survey [USGS] , University of Michigan , Wisconsin Department of Natural Resources
EPA Project Officer: Hiscock, Michael
Project Period: December 1, 2002 through December 31, 2006
Project Amount: $842,547
RFA: Development of Watershed Classification Systems for Diagnosis of Biological Impairment in Watersheds (2002) RFA Text |  Recipients Lists
Research Category: Water and Watersheds , Water

Objective:

Our goal is to couple landscape-based modeling from large, regional data sets and regional Land Transformation Models with a valley segment ecological classification approach already being employed in several Midwestern states. Objectives include completion of a GIS-based river segment classification and provision of a comprehensive status and risk assessment of river systems across the upper Midwestern states of Illinois, Michigan, and Wisconsin.

Approach:

We will build on existing pilot work to delineate and validate ecological valley segment units for all river systems in Illinois, Michigan, and Wisconsin. Using state resource agency survey databases we will build mathematical models for predicting riverine site habitats and biological reference conditions from mapped landscape and local variables. These models will be used to extrapolate results to unsampled river segments, producing region-wide summaries of current ecological status. Finally, we will couple this modeling system to a Land Transformation Model, and provide spatially explicit risk assessments for the river systems of the upper Midwest.

Summary/Accomplishments (Outputs/Outcomes):

The aims of this project were not changed from the original application. Our project work plan was organized under four sub-objectives (A through D). Summaries of our results are given below for each sub-objective.

Objective A: Delineate and Verify Ecological Segments for All River Systems in Illinois, Michigan, and Wisconsin

Develop the Geographic Information System (GIS) and Databases. We completed the GIS and landscape databases for all three states of Illinois, Michigan, and Wisconsin. This GIS was the foundation for modeling and analysis. We coordinated this work with the United States Geological Survey (USGS), Great Lakes Region, Riverine Aquatic Gap Analysis Program (GAP) Project. We consider this regionalpartnership to be a major accomplishment of the study. The partnership was synergistic in the sense that it influenced the thinking of both groups and ultimately improved products of both research projects. We expect the partnership and its benefits to continue well beyond this research project effort and to become a foundation for long-term river management cooperation within this region.

Early in planning, we decided to use confluence-to-confluence river reaches (smallest river unit or arc in the National Hydrography Dataset) as the base units for our analysis. This approach had many advantages but was more work than originally planned, and so it took longer to complete than expected and was the primary reason we requested a one-year, no-cost extension covering January 1 to December 31, 2006. We think the extra effort was worthwhile. We described the details of the GIS in three research papers (Brenden, et al., 2006; Stewart, et al., 2006; Roehl, et al., 2006) and in two presentations at professional conferences (Cooper, et al., 2003 and Brenden, et al., 2004). We have suggested that our approach is sound and broad enough to be adopted as a national standard for regional river studies. The GIS is one of the major products of this project and is maintained at the Institute for Fisheries Research, Ann Arbor, MI.

Delineate Valley Segments. We developed a clustering routine, ESSI-CAST (Ecological Stream Segment Identification by the Cluster Affinity Search Technique), for delineating valley segments. This routine combines the confluence-to-confluence river reaches into ecologically similar valley segments based on statistical measures of similarity among neighboring reaches. This new clustering routine provides more statistical rigor to the process of delineating valley segments than past methods as described by Seelbach and Wiley (1997) and Seelbach, et al. (2006). We described the details of the new clustering routine in a research paper (Brenden, et al., 2007a) and in four presentations at professional meetings (Brenden, et al., 2005a; 2005b; 2005c; 2005d).

Validation of Valley Segments as Ecological Units. We developed a protocol to sample 57 sites within five tributary basins, such that each valley segment unit was sampled at multiple sites. Identification of macroinvertebrates and chemical analyses were performed in the laboratory. Analysis of the empirical data collected supported the homogeneity of valley segments as ecological units. The valley segment structure generally did a good job at partitioning the variability of channel characteristics, chemistry, and invertebrate and fish assemblages. Sites within a valley segment unit were more similar to each other than to sites in adjacent units. Boundaries between units could be identified at tributary junctures and dams. However, it appears that upstream-most reaches with intermittent flows might better be considered as separate valley segments. The details of the validation of valley segments will be described in a future University of Michigan Ph.D. dissertation by Sparks-Jackson.

Objective B: Build Statistical Models for the River Systems of IL, MI, and WI That Predict Riverine Site Conditions and Biota from Variables Representing Natural, Climatic, Landscape, Valley, and Channel Influences, as well as Human Influences

We developed models to predict water temperature, flow, fish, macroinvertebrates, and periphyton conditions across the landscape. Models for water temperature, flow, and fish were applied across all three states. Models for macroinvertebrates and periphyton were applied to broad regions to demonstrate their utility. These models were all similar in the sense that they developed statistical relationships between data collected in the field and the landscape data for the sites where data were collected. Then, because we had landscape data for all the reaches and valley segments, we could predict conditions at other sites across the landscape where field data were never collected (Seelbach, et al., 2001).

Water Temperature. We completed temperature models for all three states. For Illinois, we described the details in a technical report (Holtrop, et al., 2006). For Michigan and Wisconsin, we described the details of the temperature modeling work in a research paper (Brenden, et al., 2007b).

Stream Flow. We completed stream flow models for all three states. We described the details of the flow modeling work in a technical report and a research paper (Holtrop, et al., 2006; Seelbach, et al., 2007).

Fish. We completed fish modeling work in all three states. First, we used cluster analysis to group fishes into assemblages that shared similar abundance patterns. Then, we used Classification and Regression Tree (CART) analysis to predict the occurrence (presence /absence) and relative abundance of the fish assemblages across each state. In Michigan, we also predicted occurrence and relative abundance for the most common fish species. For Illinois, we described the details of the fish modeling work in a research report (Holtrop, et al., 2006). For Michigan, we described the details of the fish modeling work in three research papers (Brenden, et al., 2007d; Steen, et al., 2006; and Steen, et al., 2007), two oral presentations at professional meetings (Steen, et al., 2005; and Steen, et al., 2006a), and one poster presentation (Steen, et al., 2006b). For Wisconsin, we described the details of the fish modeling work in a poster presentation (Lyons and Stewart, 2006) and a draft research report (Lyons, 2004). In addition, three other scientific papers specific to Wisconsin are planned, one describing the construction, structure, and performance of the predictive fish models, a second covering the results of the different land-use scenarios and describing how fish distribution has and will likely continue to change in the state, and a third using the models to calibrate the observed/expected ratio of species as a bioassessment tool.

Macroinvertebrates. We completed macroinvertebrate models for Illinois. We described the details of the models in two research reports (Holtrop, et al., 2006; Sparks-Jackson, 2007). We have also developed macroinvertebrate models for Michigan and Wisconsin. We are currently working on a research paper describing the macroinvertebrate models for all three states.

Periphyton. We did not have enough data to develop models for all three states, but we did develop a new indicator of diatom trophic status in streams and illustrated its use in Michigan. Diatom indicators of trophic status describe nutrient conditions in streams and whether nutrients have affected species composition of streams. These indicators have been shown to be more precise and accurate indicators of nutrient conditions in streams than one time measurements of nutrient concentrations. They can be also used to infer probability of nuisance algal growths. These indicators are particularly valuable for developing and implementing nutrient criteria by showing effects of nutrients on streams and assessing nutrient conditions in streams.

The model for the trophic diatom index (D) was:

D = 4.275 + 0.703 × Ln(L + 1) − 0.656Ln(O + 1),

where

L = the proportion of the riparian zone of the watershed that was occupied by any type of agricultural and/or urban land use, and

O = the proportion of the watershed with outwash geology.

The model had an R2 value of 0.443. Predicted values of the trophic diatom index varied from 4.0 to 4.75 as Ln(L+1) varied from 0.0-0.7. Predicted values of the trophic diatom index varied from 4.5 to 3.75 as Ln(O+1) varied from 0.0-0.7. Thus, natural variation in outwash geology was an important determinant of trophic condition in streams, or it covaries with important determinants of streams. Whereas human activities are important sources of nutrients (thus the positive effect of land use), high proportions of outwash geology probably increase hydrologic conductivity and groundwater discharge into streams, thereby diluting nutrient concentrations in streams in cases where groundwater has less phosphorus in it than surface water. This indicator represents a major step forward in algal bioassessment methods and would probably be transferable to other regions of the Great L akes. We described more details of the indicator in an oral presentation (Stevenson, 2006), and we are preparing a scientific paper for publication in the near future.

Objective C: Use Statistical Models and LTM Maps To Predict Ecological Attributes Under Various Scenarios

Land Transformation Model. We used the land transformation model of Pijanowski, et al. (2002a; 2002b) to parameterize a model that covered all three states. We provided an overview of the land transformation modeling in a draft research report (Pijanowski, 2007). We described the details of the land transformation modeling work in the following publications and research reports (Wiley, et al., 2004; Alexandridis, et al., 2005; Lei, et al., 2005; Pijanowski, et al., 2005; Tang, et al., 2005a and 2005b; Pijanowski, 2006; Pijanowski, et al., 2006; Alexandridis and Pijanowski, 2007; and Alexandridis, et al., 2007).

Risk Assessment. For Illinois, we demonstrated the use of our GIS for risk assessment by conducting a regional normalization approach on the macroinvertebrate assemblages (Sparks-Jackson, 2007). We also used the Land Transformation model for risk assessment by linking the flow, temperature, macroinvertebrate, and fish models to the 2025 output of the land transformation model for the Kaskaskia River Basin (Holtrop, et al., 2006). For Michigan and Wisconsin, we demonstrated the use of our GIS for risk assessment by identifying human disturbance gradients and references for streams. We described and documented this risk assessment work in two oral presentations (Wang, et al., 2004; Wang, et al., 2005) and two research papers (Wang, et al., 2006a; Wang, et al.,2006b ). In addition, we have completed a statewide streams risk assessment for Michigan using both fish and macroinvertebrate metrics in a regional normalization approach (Riseng, et al., 2007).

Objective D: Use Resulting Data System To Develop Series of Region-Wide Classifications

Valley Segment Classification. Under a previous task, we combined adjacent river reaches if they were ecologically similar to create valley segments. Besides creating ecologically meaningful river units, this work reduced the number of river units by about an order of magnitude (Brenden, et al., 2007a). Yet the number of river valley segment units in a region the size of a state continues to be in the tens of thousands, perhaps too many to help guide some important river management activities, such as designing sampling or monitoring programs. Under this task, we sought to combine non-adjacent, but ecologically similar, valley segments into a more manageable number of classes or types. We used our GIS to develop a classification system for valley segments in Michigan in which the classes were identified using an objective, multivariate approach. Our work resulted in two valley segment classification schemes, one with 10 classes and one with 26 classes. Each scheme has advantages and disadvantages, depending on the desired management application. We described the details of classifying valley segments in a draft research paper (Brenden, et al., 2007c).

Conclusions:

We satisfied the objectives of our study. We developed a GIS for rivers in Illinois, Michigan, and Wisconsin with help from our cooperators from USGS. We concluded that our approach was scientifically sound and broad enough to be used as a national standard data model for river management. We used the GIS to cluster ecologically similar river segments into larger valley segments. We used the GIS to classify valley segments into a smaller number of groups which might be more useful for certain river management activities. We used approaches to delineate and classify valley segments that were more statistically rigorous than previous approaches, and we validated the homogeneity of a sample of the valley segments. We developed models relating available field data describing the ecological conditions at specific sites (as the dependent variables) to the landscape variables in the GIS (as the independent variables). The field data were from the files of the cooperating agencies and included water temperature, water flow, fish presence/absence, fish abundance, macroinvertebrate metrics, and periphyton metrics. Because we had all the landscape variables for every river segment in the region in our GIS, we were able to use the models to predict the present ecological reference conditions for every river segment in the region. We included both natural features and human land-use factors in our landscape variables, so we could examine alternative scenarios and conduct ecological risk assessments. We demonstrated the use of our GIS and models for risk assessment by comparing the expected ecological conditions based on natural features to the actual conditions under current land use. We also developed a companion land transformation model that predicts future human land-use changes across the three states, which in turn allows prediction of future ecological conditions. We concluded that these predictions can be useful to help identify river segments which are expected to be the most impacted by future land-use practices. Such information could allow management agencies to be preemptive in taking protective action or to efficiently concentrate limited resources to the areas of greatest need.

We expect the GIS, models, and other products of this research to be applied in a number of future management settings. These include state Wildlife Action Plans, state stream environmental monitoring programs, state stream fish monitoring programs, state stream fisheries management programs, state development of legal standards for groundwater protection, USGS Great Lakes Regional Aquatic GAP analyses, the U.S. Environmental Protection Agency (EPA) Regional Vulnerability Assessments, and National Fish Habitat Initiative regional prioritizations and plans. Our research products are likewise influencing similar regional-scale river management efforts in other regions and states.

References:

Pijanowski BC, Brown DG, Manik G, Shellito B. Using neural nets and GIS to forecast land use changes: a land transformation model. Computers, Environment and Urban Systems 2002a;26:553-575.

Pijanowski BC, Shellito B, Pithadia S. Using artificial neural networks, geographic information systems and remote sensing to model urban sprawl in coastal watersheds along eastern Lake Michigan. Lakes and Reservoirs 2002b;7:271-285.

Seelbach PW, Wiley MJ. The Michigan Rivers Inventory: project description. Michigan Department of Natural Resources, Fisheries Technical Report 2036, Ann Arbor, 1997.

Seelbach PW, Wiley MJ, Soranno PA, Bremigan MT. Aquatic conservation planning: using landscape maps to predict ecological reference conditions for specific waters. In: Gutzwiller KJ, ed. Applying Landscape Ecology to Biological Conservation. New York, NY: Springer, 2001, pp. 454-478.

Seelbach PW, Wiley MJ, Baker ME, Wehrly KE. Initial classification of river valley segments across Michigan’s Lower Peninsula. In: Hughes RM, Wang L, Seelbach PW, eds. Landscape Influences on Stream Habitat and Biological Assemblages. Bethesda, MD: American Fisheries Society, Symposium 48, 2006, pp. 25-48.

Brenden TO, Wang L, Seelbach PW, Clark RD Jr., Wiley MJ, Sparks-Jackson BL. Spatially-constrained clustering program for river valley segment delineation from GIS river networks. Environmental Modelling & Software (submitted, 2007a).

Brenden TO, Wang L, Wehrly KE. Comparison of several statistical approaches and methods for predicting stream temperature for spatially extensive regions for scarce measurements. Canadian Journal of Fisheries and Aquatic Sciences (submitted, 2007b).

Brenden TO, Wang L, Seelbach PW. A landscape-based river classification system for Michigan. (in preparation, 2007c).

Tang Z, Engel BA, Pijanowski BC, Lim KJ. Forecasting land use change and its environmental impact at a watershed scale. Journal of Environmental Management 2005a;76(1):35-45.

Tang Z, Engel BA, Lim KJ, Pijanowski BC, Harbor J. Minimizing the impact of urbanization on long term runoff. Journal of the American Water Resources Association 2005b;41(6) 1347-1359.

Brenden TO, Wang L, Clark R Jr., Seelbach PW, Wiley MJ. A clustering algorithm for ecological stream segment identification from spatially extensive digital databases. Presented at the Annual Meeting of the North American Benthological Society, New Orleans, LA, July 13-17, 2005a.

Brenden TO, Wang L, Clark R Jr., Seelbach PW, Wiley MJ. A clustering algorithm for ecological stream segment identification from spatially extensive digital databases. Presented at the International Association of Great Lake Research Conference, Ann Arbor, MI, May 27, 2005b.

Brenden TO, Wang L, Clark R Jr., Seelbach PW, Wiley MJ. A variable selection process for identification of ecological stream segments. Presented at the International Association of Great Lake Research Conference, Ann Arbor, MI, July 13-17, 2005c.

Brenden TO, Wang L, R. Clark R Jr., Seelbach PW, Wiley MJ. Ecological stream segment identification for spatially extensive digital databases. Presented at the Midwest Fish and Wildlife Conference, Grand Rapids, MI December 12, 2005d.

Steen PJ, Wiley MJ, Schaeffer JS, Stewart JS. The distribution of Michigan river fish: model development, results, and analysis. Presented at the Annual Meeting of the American Fisheries Society, Lake Placid, NY, September 12, 2006a.

Steen PJ, Wiley MJ, Schaeffer JS, Stewart JS. Development of distribution models of Michigan riverine fish: does land use affect fish assemblages? Poster presented at the International Association of Great Lakes Research, Windsor, ON, May 23, 2006b.

Holtrop AM, Hinz LC Jr, Epifanio J. Ecological classification of rivers for environmental assessment and management: model development and risk assessment. Champaign, IL: Illinois Natural History Survey, Division of Ecology and Conservation Science, December 2006.

Seelbach PW, Hinz LC Jr., Wiley MJ, Cooper AR. Using multiple linear regression to estimate flow regimes for all rivers in Illinois, Michigan, and Wisconsin. Draft manuscript, 2007.

Sparks-Jackson B. Assessment of Illinois streams using macroinvertebrates and a regional normalization approach. Draft research report, 2007.

Riseng C, Sparks-Jackson B, Wang L, Wiley MJ, Holtrop A, Hinz L, Seelbach P, Cooper J. Biological assessment of Midwestern streams using standardized invertebrate metrics: a landscape analysis. Draft manuscript, 2007.

Pijanowski B. Land transformation modeling. Draft research report, 2007.

Steen PJ, Passino-Reader DR, Wiley MJ. Modeling brook trout presence and absence from landscape variables using four different analytical methods. In: Hughes RM, Wang L, Seelbach PW, eds. Landscape Influences on Stream Habitats and Biological Assemblages. Herndon, VA: The American Fisheries Society, 2006, pp. 513-531.

Steen PJ, Zorn TG, Seelbach PW, Schaeffer JS. Classification tree models for predicting distributions of Michigan stream fish from landscape variables, (in preparation, 2007).

Wang L, Seelbach PW, Lyons J. Effects of levels of human disturbance on the influence of catchment, riparian, and reach-scale factors on fish assemblages. In: Hughes RM, Wang L, Seelbach PW, eds. Landscape Influences on Stream Habitats and Biological Assemblages. Herndon, VA: The American Fisheries Society, 2006a, pp. 199-219.

Wang L, Brenden T, Seelbach P, Cooper A, Allan D, Clark R Jr., Wiley M. Landscape based identification of human disturbance gradients and reference conditions for Michigan streams. Environmental Monitoring and Assessment 2006b Dec 14 [Epub ahead of print] doi:10.1007/s10661-006-9510-4.

Expected Results:

Our ultimate products will be: 1) a GIS based river classification and modeling system, developed in cooperation with each state resource agency (and coordinated among states) that contains a series of standard landscape maps and a map of ecological river. Also, associated data tables containing attributes linked to segments: raw data, attribute classes, and risk assessment classes; and 2) the illustration of a landscape based approach to modeling, classification, and status/risk assessment of rivers that would be transferable to other regions.


Journal Articles on this Report : 10 Displayed | Download in RIS Format

Other project views: All 34 publications 15 publications in selected types All 10 journal articles
Type Citation Project Document Sources
Journal Article Alexandridis K, Pijanowski BC. Assessing multiagent parcelization performance in the MABEL simulation model using Monte Carlo replication experiments. Environment and Planning B: Planning and Design 2007;34(2):223-244. R830596 (Final)
  • Full-text: EPB PDF
    Exit
  • Abstract: EPB Abstract
    Exit
  • Other: Advance Online PDF
    Exit
  • Journal Article Brenden TO, Wang L, Clark RD Jr, Seelbach PW, Lyons J. Comparison between model-predicted and field-measured stream habitat features for evaluating fish assemblage-habitat relationships. Transactions of the American Fisheries Society 2007;136(3):580-592. R830596 (Final)
  • Abstract: AFS Abstract
    Exit
  • Journal Article Brenden TO, Wang L, Seelbach PW, Clark Jr RD, Wiley MJ, Sparks-Jackson BL. A spatially constrained clustering program for river valley segment delineation from GIS digital river networks. Environmental Modelling & Software 2008;23(5):638-649. R830596 (Final)
  • Full-text: ScienceDirect-Full Text HTML
    Exit
  • Abstract: ScienceDirect
    Exit
  • Other: ScienceDirect-PDF
    Exit
  • Journal Article Lei Z, Pijanowski BC, Alexandridis KT, Olson J. Distributed modeling architecture of a multi-agent-based behavioral economic landscape (MABEL) model. Simulation 2005;81(7):503-515. R830596 (Final)
  • Full-text: Purdue University PDF
    Exit
  • Abstract: SAGE Abstract
    Exit
  • Journal Article Pijanowski BC, Pithadia S, Shellito BA, Alexandridis K. Calibrating a neural network-based urban change model for two metropolitan areas of the Upper Midwest of the United States. International Journal of Geographic Information Science 2005;19(2):197-215. R830596 (Final)
  • Full-text: Purdue University PDF
    Exit
  • Abstract: InformaWorld Abstract
    Exit
  • Journal Article Pijanowski BC, Alexandridis KT, Muller D. Modelling urbanization patterns in two diverse regions of the world. Journal of Land Use Science 2006;1(2-4):83-108. R830596 (Final)
  • Full-text: Humboldt-Universitat zu Berlin PDF
    Exit
  • Abstract: InformaWorld Abstract
    Exit
  • Journal Article Steen PJ, Zorn TG, Seelbach PW, Schaeffer JS. Classification tree models for predicting distributions of Michigan stream fish from landscape variables. Transactions of the American Fisheries Society 2008;137(4):976-996. R830596 (Final)
  • Abstract: Taylor & Francis-Abstract
    Exit
  • Journal Article Tang Z, Engel BA, Lim KJ, Pijanowski BC, Harbor J. Minimizing the impact of urbanization on long term runoff. Journal of the American Water Resources Association 2005;41(6):1347-1359. R830596 (Final)
  • Full-text: Purdue University PDF
    Exit
  • Abstract: Wiley
    Exit
  • Journal Article Tang Z, Engel BA, Pijanowski BC, Lim KJ. Forecasting land use change and its environmental impact at a watershed scale. Journal of Environmental Management 2005;76(1):35-45. R830596 (Final)
  • Abstract from PubMed
  • Full-text: Science Direct Full Text
    Exit
  • Abstract: ScienceDirect
    Exit
  • Other: Science Direct PDF
    Exit
  • Journal Article Wang L, Brenden T, Seelbach P, Cooper A, Allan D, Clark Jr. R, Wiley M. Landscape based identification of human disturbance gradients and reference conditions for Michigan streams. Environmental Monitoring and Assessment 2008;141(1-3):1-17. R830596 (Final)
  • Abstract from PubMed
  • Abstract: SpringerLink-Abstract
    Exit
  • Supplemental Keywords:

    RFA, Scientific Discipline, INTERNATIONAL COOPERATION, Geographic Area, Waste, Water, ECOSYSTEMS, Ecosystem Protection/Environmental Exposure & Risk, Bioavailability, Aquatic Ecosystems & Estuarine Research, Water & Watershed, State, Aquatic Ecosystem, Water Quality Monitoring, Environmental Monitoring, Terrestrial Ecosystems, Ecology and Ecosystems, Watersheds, anthropogenic processes, fate and transport, model, nutrient transport, anthropogenic stress, bioassessment, watershed classification, biodiversity, watershed management, ecosystem monitoring, conservation, diagnostic indicators, ecosystem indicators, Illinois (IL), aquatic ecosystems, water quality, Wisconsin (WI), bioindicators, watershed sustainablility, biological indicators, ecosystem stress, watershed assessment, transport modeling, nitrogen uptake, conservation planning, bioavailable phosphorus, agricultural community, aquatic biota, land use, restoration planning, watershed restoration, Michigan (MI), ecosystem response

    Progress and Final Reports:

    Original Abstract
  • 2003 Progress Report
  • 2004 Progress Report
  • 2005
  • 2006