Final Report: Linking Land Use Change, Stream Geomorphology, and Aquatic Biodiversity in a Hierarchical Classification Scheme

EPA Grant Number: R830595
Title: Linking Land Use Change, Stream Geomorphology, and Aquatic Biodiversity in a Hierarchical Classification Scheme
Investigators: Watzin, Mary C. , Hession, W. C. , Keeton, William S. , Troy, Austin R.
Institution: University of Vermont
EPA Project Officer: Hiscock, Michael
Project Period: December 1, 2002 through November 30, 2005 (Extended to November 30, 2006)
Project Amount: $664,339
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:

The goal of our project was to develop and evaluate a watershed and stream reach classification system based on the relationships among land-use change, river geomorphic condition, riparian habitat fragmentation, and riverine ecological condition. Our objectives were as follows: (1) beginning at the finer scale, determine which geomorphic classes can be consistently related to the ecological condition of a stream reach; (2) expanding to a coarser scale, determine which land-use metrics best predict those geomorphic classes that consistently relate to ecological condition; (3) evaluate the ability of our classification system to target sites in greatest need of watershed management and stream restoration based on current land use and geomorphology; (4) develop and evaluate the ability of our classification system to identify sites in need of conservation based on predicted land-use change and resulting effects on geomorphology and aquatic ecology; and (5) develop a general framework for the Vermont Department of Environmental Conservation and others to use the resulting classification system as a foundation for statewide watershed protection, management, restoration, and education.

Field Data Collection and Analysis of Reach Characteristics

Twenty-five independent stream reaches and their associated watersheds were chosen within the Lake Champlain Basin for assessment during the 2003 and 2004 summer field seasons. At each site, reach lengths of 10–20 bankfull widths (about 200–300 m for most sites) were selected for geomorphic assessment, habitat assessment, riparian vegetation analysis, and biological assessment.

The geomorphic assessment included a longitudinal profile and six cross sections where erosion was also measured using erosion pins. Water depth, flow velocity, embeddedness of cobble and gravel materials, bed compaction, and deposition of fine silt were also recorded at each of these six cross section locations. The presence or absence of large woody debris (LWD) within the bankfull channel was also noted.

A Rapid Geomorphic Assessment (RGA) form was completed for each reach according to Vermont Agency of Natural Resources Phase 2 Protocols. Each reach was rated on a scale of 0–20 (0 = poor, 20 = optimal/reference) in four categories: degradation, aggradation, channel widening, and planform change. These four scores were then combined for a possible total of 80 points. The purpose of this assessment was to determine the overall condition of the stream and identify the dominant adjustment process occurring within the stream. A Vermont Rapid Habitat Assessment (RHA), a semi-quantitative offshoot of the U.S. Environmental Protection Agency’s (EPA) rapid bioassessment protocol, was also completed for each reach.

Riparian vegetation was evaluated using a stratified plot design. Vegetation was stratified as upland forest, floodplain forest, floodplain non-forest, agricultural, or agricultural buffer (e.g., unmowed grass). Vegetation sampling plots were randomly distributed as a proportionate sample of each vegetation type based on its relative abundance within the buffered area along each reach. Upland forest vegetation composition and structure were sampled within variable radius plots. The percent cover of herbaceous vegetation and substrate within 1-m2 quadrats was also estimated, and LWD volume was sampled using a line intercept method.

Structural variables were calculated independently for each vegetation type. Reach means were calculated by weighting type-specific averages by the proportional representation of types within each site to reflect the relative degree of influence of each vegetation type on river reach condition. Classification and regression tree (CART) analysis was used to model RGA scores as a function of multiple vegetation structural characteristics. GIS habitat maps were used to examine three scales of potential linkage: a 50-meter wide reach (50 mR), a 100-meter wide reach (100 mR), a 50-meter wide upstream watershed (50 mWS), a 100-meter wide upstream watershed (100 mWS), and the entire upstream watershed (WS). Area by land-use/land-cover type was classified as agriculture land, urban land, forested land, and other.

The biological assessment included sampling of macroinvertebrates, fish, and riparian birds. Macroinvertebrate samples were collected from mid-June through mid-August. Subsamples were collected at 6 regularly spaced intervals along the length of each stream reach using a 500-µm mesh Surber sampler, alternately assigning subsamples to the mid-channel, towards the left bank or towards the right bank. Taxon richness and evenness were calculated using Hurlbert’s PIE (Probability of an Interspecific Encounter). This index calculates the chances that two randomly sampled individuals from an assemblage represent different taxa. PIE and taxon richness were calculated for the entire assemblage and for the Ephemeroptera, Plecoptera, and Trichoptera (the EPT).

We sampled the fish community at all sites using seine nets. At each site, three to four representative locations were sampled that reflected the flow habitat composition of the reach, distributed at approximately equal intervals along the reach length. From each reach, 150 fish of each subsampling effort were randomly selected, identified to species, weighed to the nearest 0.1 g, and measured to the nearest 1.0 mm; all others were enumerated.

Bird communities were surveyed using the double-observer method at all sites. At each reach, centered 250-meter long transects were established along the top of both right and left banks. The width of each transect ended at the center of the channel on one side and at 25 m into the riparian zone on the other. Twenty-five-meter intervals along each transect were flag-marked to serve as reference points. In order to increase the chances of encountering birds that are active in both the morning and the evening, we modified our proposed protocol to visit each site twice between May 15th and June 15th, once in the morning (sunrise to 3 hours after sunrise) and once in the evening (3 hours before sunset to dark). The species, gender, and location of each individual were recorded, as well as the vegetative community in which it was sighted. Twenty minutes were allocated to survey each transect (i.e., 40 minutes per reach).

A variety of multivariate methods were used to explore the relationships between geomorphic condition and measures of habitat and biodiversity across functional and taxonomic ranges. These methods included stepwise linear regression, principal components analysis, Akaike’s information theoretic methodology (AICc) to model relationships between the biotic community and key habitat variables, and semi-variogram analyses to look for spatial patterns in the data.

Land-Use Change Analysis and Watershed Classification

A 2002 land-cover/land-use layer was created by classifying three 2002 Landsat-7 ETM+ scenes, supplemented by ancillary data sources. An accuracy assessment of the map was conducted using stratified random sampling. Two hundred samples were selected, including 50 samples from each of the four categories: urban (impervious surface), forest, water, and other (agriculture). The overall accuracy of the classification was 88%. The producer’s accuracies of urban, forest, water, and other were 97.67%, 81.03%, 100%, and 78.96%, respectively. The user’s accuracies of urban, forest, water and other, which measure commission errors, were 84%, 94%, 84%, and 90%.

Watersheds were classified at the Hydrologic Unit Code (HUC) 14 scale based on both their current level of impairment and their estimated vulnerability to future impairment from urbanization. We classified as impaired any watershed with higher than average levels of urbanization or agriculture. Percent urban, agricultural, and forest cover was summarized at three scales: (1) the entire watershed; (2) the 100-meter buffer around all watercourses in the watershed as represented in the National Hydrography Dataset (NHD), which includes all watercourses mapped at the 1:20,000 scale; and (3) the 100-meter buffer around only major rivers and streams for the watershed mapped at the 1:100,000 scale, as represented by U.S. Geological Survey (USGS) surface water Digital Line Graphs. Stream networks were buffered and then intersected with watersheds to identify them by their HUC code. Watersheds were then classified as “impaired,” and “non-impaired.” Because land-use composition in the 100 meter buffer around the reach best predicted variation in RGA score, we classified watershed condition based on land-cover composition in the NHD 100 meter buffer. The impaired category consisted of watersheds with either greater than 40% agriculture, less than 25% forest, or greater than 20% urban within the 100-meter NHD buffer.

Vulnerability of watersheds to urbanization was determined using a set of GIS algorithms based on supply and demand factors. Supply reflected site suitability for development, and demand was based on drive time to the employment center. Supply of developable land was calculated based on site factors that could be easily quantified in GIS. Many of these factors were based on the review criteria for Vermont’s Land Use and Development Law, Act 250. Three network drive time service areas were generated for each employment center by factoring in distance and speed limit by road segment. The demand and supply layers were then combined. The demand layer, ranked 0–5, was multiplied by the 1/0 site suitability layer, yielding a value only for pixels that could be built upon and that were within a service area. The mean score was then summarized by HUC 14 watershed using the Zonal Statistics tool. Because so many pixels within any watershed were zeros, resulting in low mean scores, the vulnerability scores were rescaled to five classes using quantile breaks. Once watersheds were classified by impairment and vulnerability, queries were run to classify them based on needed action.

Summary/Accomplishments (Outputs/Outcomes):

Relationships Between Stream Geomorphology and Ecological Integrity

We documented a range of geomorphic conditions in our 25 field sites, from poor to near-reference condition. There were different patterns of association between stream geomorphic condition and stream ecological condition using our three groups of organisms. For the macroinvertebrates, we found that both RGA and RHA scores were associated with community composition. Although stream reaches with the highest RGA scores did not generally support higher densities of macroinvertebrates than reaches with lower scores, the percent of the macroinvertebrate community in the Ephemeroptera, Plecoptera, and Trichoptera (EPT) taxa was significantly and positively correlated with both RGA and RHA scores.

For the fish community, we also found that stream reaches in good geomorphic condition supported assemblages with higher diversity than those in poorer geomorphic condition. Using principal component analysis (PCA) and linear regression to build exploratory models that linked stream geomorphic condition to fish community characteristics, we found that geomorphic condition explained the greatest amount of the total variance in species diversity of fish communities. In conjunction with additional reach characteristics, geomorphic condition explained up to 31% of the total variance observed in models for species diversity, 44% of the variance in assemblage biomass, and 45% of the variance in a regional index of biotic integrity.

We also looked at fish diversity relative to channel morphology to determine whether fish community diversity differed between pool-riffle and intermediate channel morphologies (i.e., pool-riffle/cascade, pool-riffle/mix, and forced pool-riffle). We found that fish diversity was influenced by channel morphology, with pool-riffle/cascade morphology consistently exhibiting the greatest fish diversity and forced pool-riffle the lowest. Because intermediate morphologies differed in their capacity to support fish diversity, these morphologies should be added to stream classification systems used for ecological applications. Our results also suggest that sampling strategies for fish that focus on homogeneous reaches may underestimate diversity.

To determine which physical characteristics might be most important for fish, we further investigated the relationships between RGA and RHA scores and modeled usable habitat area at the reach scale for selected fish species. For these analyses, we used a two-dimensional hydrodynamic model (River2D), and collected detailed discharge, velocity, bed substrate, and topographic data for a 100-m reach at six of our stream sites. After calibration, weighted useable areas (WUAs) were calculated for each stream at three flows (minimum, median, and bankfull) using modeled velocity and depth estimates, channel bed data, and habitat suitability curves for blacknose dace (Rhinichthys atratulus), brown trout (Salmo trutta), common shiner (Notropis cornutus), and white sucker (Catostomus commersoni) in both the adult and spawn stages.

The predicted reach scale WUAs that resulted were negatively correlated or not correlated with RGA scores, suggesting that rapid assessment protocols may not completely capture habitat conditions important for fish. In addition, reach-averaged WUA values did not accurately predict measures of fish biodiversity in these streams. Because the areas of high WUA were patchy and all streams exhibited different spatial distributions of habitat, we used spatial analysis techniques (semi-variogram analyses) to characterize patch size at selected discharge intensities. Streams with more distinctly separated patches of higher WUA had lower fish biodiversity than streams with a more uniform distribution of WUA, and in fact, the distribution of usable habitats may be a determining factor for fish communities. These results suggest that additional research should focus on identifying the appropriate spatial scales to capture the connections between usable patches of stream channel habitat.

Because we suspected that large woody debris is particularly important in providing smaller scale habitat heterogeneity for both macroinvertebrates and fish, we explored relationships between large woody debris (LWD) patches and fish and macroinvertebrate communities in a subset of nine of our stream reaches. Using paired t-tests, we found that fish densities and percent Chironomidae were both higher around LWD patches than in the general reach samples. Our results also indicated that fish and macroinvertebrates were using patches in different ways; while fish preferred patches around larger pieces of LWD, macroinvertebrates were found in higher abundance around more moderately sized pieces. We concluded that sampling at a reach level may not accurately portray the community composition of a stream environment because the unique biodiversity in patches such as LWD may be overlooked.

For the birds, we took a larger riverscape perspective of streams, recognizing that the heterogeneous habitat types within the stream corridor were a single, integrated ecological unit operating across spatial scales. Using the information theoretic methodology (AIC) to model relationships between bird community attributes and key habitat variables, we found that birds responded to a variety of habitat characteristics at the in-stream, floodplain, and riparian corridor scales. Although channel slope was the variable selected in the most models, 16 of a potential 18 habitat variables were used in at least one model. We found that some bird guilds, particularly piscivores, were potentially important indicators of overall riverscape condition, responding to a host of variables that reflect both geomorphic and riparian corridor condition.

Relationships Between Riparian Vegetation Condition and Stream Geomorphology

Four structural variables emerged as important in explaining variance in RGA scores among sites. These were the standard deviation of total (live and dead tree) basal area (representing spatial variation within individual sites), mean total basal area, dead tree stem density, and shrub density. Basal area standard deviation was most predictive of RGA score. RGA scores were highest at sites with highly variable basal area and low to moderate shrub densities. In general, greater levels of forest stand structural complexity (e.g., dead tree density and basal area) were associated with higher RGA scores.

Regression analysis showed RGA to be positively and linearly correlated with percentage forest cover (t = 3.19, p < 0.001) and negatively related to percentage agricultural land (t = -3.409, p = < 0.001) at the 50-mR scale. RGA was also positively related to the percentage of each reach in forest and negatively related to the percentage in agricultural land at the 100-mR buffer level.

There were no statistically significant relationships between land cover and RGA at any of the larger (upstream watershed) spatial scales.

Riparian forest structure and land cover were associated with stream geomorphic condition. Streams running through riparian forests of greater structural complexity (e.g., greater amounts of, and more spatial variation in, basal area) are likely to have higher RGA scores. These results are consistent with the limited previous research in the eastern United States documenting linkages between the structure of deciduous and mixed deciduous-coniferous forests and in-stream habitat characteristics. The relationship between land cover and RGA was scale-dependent. RGA scores primarily reflect fine scale (50-mR and 100-mR) relationships between forest cover and river geomorphology. At these scales, interactions between forest cover and structure and geomorphic processes seem to be important in maintaining the integrity and dynamics of river channels. While previous research has demonstrated relationships between land cover and stream condition at larger (e.g., watershed) scales, these were not evident or detected in our dataset.

Watershed Classification and a Hierarchical System for Management

We classified watersheds as “impaired” if they included greater than 40% agriculture, less than 25% forest, or greater than 20% urban area within the 100-meter NHD buffer. This classification was based on our strong results linking riparian land use with stream geomorphic condition. We also believe there is a relationship between overall land use in the watershed and geomorphic condition, but our regression relationships to support that hypothesis were not as strong. We did find that when the overall watershed contained greater than 7% urban area, or greater than 25% agriculture, stream geomorphic condition declined.

Vulnerability of watersheds to urbanization was determined based on site suitability for development (supply of land) and drive time to employment centers (demand for land). Once watersheds were classified by impairment and vulnerability, queries were run to classify them based on needed action. Unimpaired watersheds with no or low vulnerability were classed as “low priority for action.” Unimpaired watersheds with medium or high vulnerability were classified as “high priority for conservation.” Impaired watersheds with no or low vulnerability were classified as “high priority for restoration,” and impaired watersheds with medium or high vulnerability were classified as “high priority for conservation and restoration.”

Conclusions:

Our results clearly show that there are links between watershed and river corridor condition, stream geomorphic condition and aquatic ecosystem health. Although the complexity of these linkages is not fully captured in land-use characterization or rapid geomorphic and habitat assessments, these planning and general assessment tools can provide a reasonable first approximation of stream ecological integrity. This is positive news for the large number of state agencies and other organizations that are using geomorphic approaches as a regular part of watershed planning and risk assessment efforts designed to target streams and watersheds in the greatest need of restoration and conservation. The State of Vermont is using the results of our research in their basin planning process and to refine their impaired waters assessment and river corridor management efforts.

We found that macroinvertebrate, fish, and water-dependent bird diversity were all associated with stream geomorphic and habitat condition, but the nature and strength of the association varied significantly. Overall, bed aggradation was particularly important in determining both macroinvertebrate and fish distribution. This result is particularly important because sediment loading is the greatest cause of stream and river impairment throughout the United States. Fish were also negatively affected by changes in water flow and isolation of the river channel from its floodplain through incision. In all of our modeling, however, the strongest relationships with reach-level geomorphic and habitat assessment scores were found with fish community metrics, not macroinvertebrate metrics. These results suggest that macroinvertebrates may not be the best biological indicators of in-stream ecological integrity at the larger reach scale, despite their widespread use as bioindicators of water quality and stream impairment.

We also found that some bird guilds, particularly piscivores, were potentially important indicators of ecological integrity, responding to a number of variables that reflect both in-stream geomorphic condition and riparian corridor vegetation and structural complexity. Although inventories of bird distribution and abundance are not regularly conducted as part of stream assessments, this taxonomic group may provide an integrated indicator of riverscape condition and should be given more consideration as an appropriate bioindicator.

Although no hard and fast thresholds were identified, we did find that when a watershed contained greater than 7% urban area or greater than 25% agricultural area, stream geomorphic condition declined. In general, streams in better geomorphic condition, and thus ecological condition, were associated with riparian corridors that had more forested area and greater levels of forest stand structural complexity (e.g., dead tree density and basal area). Thus watersheds that included less than 25% forested area, greater than 40% agriculture, or greater than 20% urban area within the 100-meter stream corridor (GIS buffer) were generally impaired. In our hierarchical classification system, watersheds with these characteristics were classified as impaired and were considered a priority for stream restoration.

To identify watersheds in need of conservation, we also assessed each watershed’s vulnerability to urbanization. Watersheds that were currently unimpaired and also had a medium or high vulnerability were considered high priorities for conservation.

We found our greatest challenge in developing strong predictive models of future growth in watersheds. Because the necessary econometric data to drive these models are so difficult to collect, additional development is necessary to refine a model to identifying areas where future risks are greatest. However, additional work in this area is warranted because it would allow managers to envision and evaluate potential future risks in a spatial context. Because we were quite successful in documenting linkages among watershed and riparian land use, stream geomorphic condition, and stream ecological condition, a more refined predictive model of future growth would be highly useful in watershed planning and evaluation of relative risk.


Journal Articles on this Report : 6 Displayed | Download in RIS Format

Other project views: All 46 publications 6 publications in selected types All 6 journal articles
Type Citation Project Document Sources
Journal Article Cianfrani CM, Hession WC, Rizzo DM. Watershed imperviousness impacts on stream channel condition in Southeastern Pennsylvania. Journal of the American Water Resources Association 2006;42(4):941-956. R830595 (Final)
  • Full-text: University of Vermont PDF
    Exit
  • Abstract: Blackwell-Synergy Abstract
    Exit
  • Journal Article Sullivan SMP, Watzin MC, Hession WC. Understanding stream geomorphic state in relation to ecological integrity: evidence using habitat assessments and macroinvertebrates. Environmental Management 2004;34(5):669-683. R830595 (Final)
  • Abstract from PubMed
  • Full-text: University of Vermont-Full Text PDF
    Exit
  • Abstract: SpringerLink
    Exit
  • Journal Article Sullivan SMP, Watzin MC, Hession WC. Influence of stream geomorphic condition on fish communities in Vermont, U.S.A. Freshwater Biology 2006;51(10):1811-1826. R830595 (Final)
  • Full-text: Oregon State University-Full Text PDF
    Exit
  • Abstract: Blackwell-Synergy Abstract
    Exit
  • Journal Article Sullivan SMP, Watzin MC, Hession WC. Differences in the reproductive ecology of belted kingfishers (Ceryle alcyon) across streams with varying geomorphology and habitat quality. Waterbirds 2006;29(3):258-270. R830595 (Final)
  • Full-text: University of Vermont-Full Text PDF
    Exit
  • Abstract: BioOne Abstract
    Exit
  • Journal Article Sullivan SMP, Watzin MC, Keeton WS. A riverscape perspective on habitat associations among riverine bird assemblages in the Lake Champlain Basin, USA. Landscape Ecology 2007;22(8):1169-1186. R830595 (Final)
  • Abstract: Springerlink Abstract
    Exit
  • Other: Landscape Ecology
    Exit
  • Journal Article Sullivan SMP, Watzin MC. Relating stream physical habitat condition and concordance of biotic productivity across multiple taxa. Canadian Journal of Fisheries and Aquatic Sciences 2008;65(12):2667-2677. R830595 (Final)
  • Full-text: Oregon State University-Full Text PDF
    Exit
  • Abstract: Canadian Journal of Fisheries and Aquatic Sciences
    Exit
  • Supplemental Keywords:

    hydrologic units, fluvial morphology, restoration, risk assessment riparian zones, Northeast, cumulative effects, ecological integrity, land-use change,, RFA, Scientific Discipline, INTERNATIONAL COOPERATION, Water, ECOSYSTEMS, Ecosystem Protection/Environmental Exposure & Risk, Aquatic Ecosystems & Estuarine Research, Water & Watershed, Aquatic Ecosystem, Monitoring/Modeling, Water Quality Monitoring, Terrestrial Ecosystems, Ecology and Ecosystems, Watersheds, anthropogenic stress, bioassessment, anthropogenic processes, watershed classification, nutrient transport, ecosystem monitoring, watershed management, biodiversity, nutrient flux, conservation, diagnostic indicators, ecosystem indicators, stream geomorphology, aquatic ecosystems, bioindicators, watershed sustainablility, water quality, biological indicators, ecosystem stress, watershed assessment, conservation planning, nitrogen uptake, ecosystem response, aquatic biota, land use, restoration planning, watershed restoration

    Relevant Websites:

    http://www.uvm.edu/envnr/rubenstein/ Exit

    Progress and Final Reports:

    Original Abstract
  • 2003
  • 2004 Progress Report
  • 2005