EXPLAINING BIOTIC INTEGRITY AND HABITAT ACROSS MULTIPLE SCALES

AN EMPIRICAL ANALYSIS OF LANDSCAPE, LAND USE, AND LAND COVER VARIABLES IN AN OHIO ECOREGION

Sanjeev Arya

Ph.D. Candidate, Department of City and Regional Planning

The Ohio State University, Columbus, Ohio 43210, USA.

Email: arya.7@osu.edu

http://www.ucgis.org/oregon/papers/arya.htm

Abstract

Stream habitat is widely recognized as the template on which community structures are shaped by landscape and anthropogenic factors across various spatial scales. Earlier studies have dealt with these ecological constructs, stream biotic integrity and stream habitat quality, separately.

The two major objectives of this study are 1) to explore and quantify variables which explain the variations in the stream habitat as measured by the Qualitative Habitat Evaluation Index, or QHEI, and 2) to relate stream habitat indices and other variables to the biotic integrity of the stream as measured by the Index of Biotic Integrity, or IBI. These numeric indices, unlike conventional chemical measures of stream water quality, encapsulate the impacts from multiple stressors and measure the ecological health of a stream.

The two-stage analysis uses linear regression models. First, the variation in QHEI is modeled as a function of local-scale natural and anthropogenic factors and stresses. At the second stage, the QHEI model developed in the first stage is integrated with the model for IBI which explains the variations in biotic integrity in terms of a comprehensive, multi-scale set of variables representing landscape, land use and land cover variables.

A detailed GIS database, including 1:24,000 scale roads and hypsography, 30m-resolution DEMs, streams, 30m-resolution land use classified from Landsat imagery, point sources, and tract-level census data, is compiled for the study area, comprising about 40 counties in an Ohio ecoregion. GIS is heavily used to tackle issues of collecting, storing, analyzing, and maintaining detailed region-wide database. GIS is also used to derive watershed boundaries, slopes, variable-width buffers, and reach sinuosity.

A pilot was conducted to address the first research question - regarding stream habitat quality. Linear regression models were used to explain the variations in QHEI at the subwatershed and riparian scales in Big Darby Creek and Great Miami River basins of west-central Ohio. Forest land cover, reach sinuosity, and number of point sources in the catchment were significant in one model. This model explained 63% of the variation in site-specific QHEI. The whole model was significant at the 1% level. Two other variables, watershed-scale road density and non-riparian agricultural land use, were not significant in explaining the linkage between landscape and stream habitat. Riparian-scale agriculture, roads, and steep slopes have significantly negative relationship with QHEI in a model. Riparian-scale moderate slopes has a significant positive impact. The pilot indicates QHEI may be better explained at the reach/riparian level rather than at the watershed level. Future modifications in this model may incorporate ordinal logit and probit modeling techniques, soil series data, local sinuosity instead of average reach sinuosity, and land use dataset of higher resolution.

This study will 1) provide a better understanding of the spatial processes shaping our stream habitats and biotic integrity, 2) provide justification for watershed- and ecoregion-based data collection, planning and management programs 3) indicate the feasibility, merits, and limitations of using biological and habitat indicators of water quality in state monitoring, assessment, and permit regulation programs.

Introduction

The concept of water quality has experienced a marked shift ever since the Clean Water Act of 1972 declared its objective of restoring and maintaining the ‘chemical, physical and biotic integrity of the nation’s waters’. Since then most of the water quality standards designed to protect designated uses for water bodies, have used chemical and physical criteria as surrogates for achieving the goal of biological integrity (Adler 1995, Yoder 1991, Yoder 1995).

A new class of measures have been proposed and implemented to monitor and evaluate the ecological health of streams (Karr 1981, Yoder and Rankin 1998). These indices have been conceived to encapsulate multiple stressors that may impact the biotic integrity and the habitat structure of freshwater streams. The exact specification of the components of an index depends on the particular regional location, geological structure, human intervention, and other factors defining the ecoregions of study areas. The indices are defined for certain reference sites within ecoregions and the performance of regional streams is calibrated and measured on this relative scale (Yoder 1991, Yoder 1995).

ECOREGIONS

Ecoregions provide the backdrop for collecting biological data, formulating biological criteria, and analyzing regional trends. The underlying basis for ecoregions is to define geographical boundaries such that watersheds within an ecoregion are more similar than those between ecoregions. Ecoregions are defined based on certain singular factors shared by the larger landscape as a whole such as soils, climate, and land use (Omernik, 1987). By association, therefore, the indices themselves may be reasonably expected to be differentiable between ecoregions. The defining factors of an ecoregion must also characterize an index in this reference system at the ecoregional level. Based on the first approximation of ecoregions defined by Omernik in 1987, five of the seventy six national ecoregions cover the state of Ohio – Eastern Corn Belt Plains (ECBP), Erie/Ontario Lake Plain (EOLP), Huron/Erie Lake Plain (HELP), Interior Plateau (IP), and Western Allegheny Plateau (WAP).

REFERENCE SITES

The basis for numerical biological criteria is to arbitrate biological performance relative to known regionally attainable biological criteria (Yoder and Rankin 1995). This is the raison d’être of reference sites. Reference sites are defined for each ecoregion based on the principle that they reflect the attainable biological potential of the regional waters. Expert biologists and ecologists choose the sites in ‘relatively undisturbed’ areas. Their selection suffers from two critical problems (Hughes 1995). First, excessive disturbance of the candidate reference site may imply the inclusion of reference sites from adjacent similar ecoregions. Second, anomalous reference sites might be chosen even as they are not representative of the surrounding region. For instance, ridgetop sites in plains, high gradient sites in low gradient area, or coldwater sites in characteristically warmwater areas, are all instances of such potential problems.

WATERSHEDS

Since topographical boundaries do not follow ecoregions some studies have recently attempted to address the need for watershed studies (Allan and others 1997, Johnson and others 1997, Lammert and Allan 1999, Richards and others 1997). Watersheds are defined by the general topographical patterns. USEPA has encouraged various watershed-based studies and programs because they take a more spatial and nonpoint source approach than site-specific point source pollution programs. However, the watershed approach suffers from some problems. First, watersheds do not depict similar ecosystems like ecoregions do. In themselves, watersheds are no more meaningless for comparisons of ecosystems than point source studies because they are not defined based on any geologic, soils, climatic, vegetative, or land use patterns. Second, it is not meaningful or feasible to define watersheds across regions of little relief, much aridity, or karst topography (with underground drainage patterns) as in the Mad River region of Ohio (Hughes and Omernik 1991). As stated earlier, watersheds within ecoregions are similar to each other compared to those across ecoregions. Therefore, watershed studies and other large region studies based on the watershed approach may benefit by using ecoregions as the spatial framework for ecosystem or water quality assessment.

Before proceeding further it may be useful to understand the different, overlapping, and sometimes confusing, terms related to watersheds. Watersheds are also variably called basins, catchment areas, contribution areas, or hydrologic units. Watersheds are regions on the surface of earth that, owing to the local topography, collect surface water from other areas. Watersheds are defined by the location of the pour-point of the watershed, and the topographic profile of the landscape. Pour-point is also known as outlet, or mouth, of the watershed. In summary, two different locations on the same topography will have different watersheds. Also, a point in space will have a different watershed if some major event altered the topography of its region. The term basin, as used in this project, depicts watersheds delineated for a large region based on the local topography without using a pour-point. Basins can be thought of as naturally occurring bowls in the region, which are delineated by ridge lines of highest local topography separating adjacent basins. On the other hand, locally-defined watersheds using specific locations as pour-points are often called subwatersheds.

STUDY AREA UNIT: INDEPENDENT SUBWATERSHEDS

In light of the objectives of ‘biological integrity of waters’ as stated in the 1972 CWA, the Ohio Environmental Protection Agency (OEPA) is at the forefront in the nation in terms of using biological criteria in water quality standards (Adler 1995). Ohio EPA periodically collects data on several indices, which capture the biological, chemical and physical health of the stream systems in Ohio. Because of the inherent complexities in defining the region of contribution to these individual sample points on the streams it is not always possible to accurately model the processes impacting a sample site for which data has been collected.

Watersheds are important units of study because they have the potential to define the boundary of the contributing effects upstream of the sampling location. As shown in Figure 1, one way is to define the contributing area based on the digital elevation model data for the study area using watershed defining areas in a Geographic Information System (GIS). We call such locally-defined watersheds as subwatersheds. The subwatershed is the unit of study for this project. The figure also clarifies some important terms regarding the different scales of measurement and delineation in watershed-based studies.

Figure 1. Basins and locally-defined watersheds (subwatersheds) for individual sites.

The definition of the study area boundary in watershed-level studies changes frequently based on the objective of the research. Some studies have used watersheds defined based on outlet points extraneous to the dataset (Gordon and Majumder 1999, Majumder 1998, and Mattson 1994). In other words, they have used whole basins instead of subwatersheds. This may be valid for specific objectives of the study. In the case of Gordon and Majumder the study was designed for studying biotic effects as fish and macroinvertebrate species are mobile over a considerable range of the stream habitat. Others have manually digitized watersheds upstream of the sample point (Richards 1994, Johnson 1997, Richards 1996, Wang 1997). Watershed delineation has also been performed using different software and algorithms (Lammert 1999, Wang and Yin 1997).

IBI AS A MEASURE OF STREAM BIOLOGICAL INTEGRITY

The Index of Biotic Integrity (IBI) is a method of assessing the biological quality of streams in the US. The structural and functional attributes of the fish community are measured as metrics or components of the index. Each metric is scored relative to the performance of minimally impacted communities in similar regions on the respective metric. The metrics are in three major categories – species richness and composition; trophic structure; and fish abundance and health. For the IBI used in Ohio, there are twelve metrics, each scored in the range of 1 to 5. Therefore, the IBI scores range from 12 to 60.

Similar indices have been developed for other components of the aquatic system such as benthic macroinvertebrates, algae, and submerged aquatic vegetation (periphyton). Currently, the index is developed for only small rivers and streams but research is ongoing for development of the index for lakes, estuaries, and large rivers.

QHEI AS A MEASURE OF STREAM HABITAT QUALITY

In order to adhere to the CWA mandate of protecting biological integrity it is imperative to protect the natural physical habitat of streams. Until the advent of numerical biological criteria, chemical-specific criteria were ‘insufficient to address most habitat and development-related impairments to ecosystems’ (Rankin 1995). A number of conceptual studies have attempted to show the relationships between the landscape, habitat, and water quality factors (Vannote and others 1980, Frissell and others 1986). Anthropogenic intervention in a watershed alters the physical characteristics of a stream. In that sense, a stream’s habitat plays a crucial intermediary role between stages of human-induced modifications of watersheds and alteration of the biotic community patterns of the watershed streams.

In 1989, the Ohio EPA developed an index of macro-habitat quality, the Qualitative Habitat Evaluation Index (QHEI). The index is constructed to measure physical factors that influence fish communities and other aquatic life such as invertebrates. It is based on six interrelated metrics; substrate quality, instream cover, channel quality, riparian zone, pool/riffle quality, and gradient.

FACTORS AFFECTING THE STREAM HABITAT (AND BIOTIC INTEGRITY)

Before the advent of numerical biological criteria, chemical criteria was used to model chemical and biological changes in water quality. Studies have been performed to explain relationships between agricultural and urban land use and chemical indicators, such as dissolved oxygen, in streams (Gordon and Fromuth 1981, Townsend 1997, Johnson and others 1997).

Some studies point to the potential covariates of interest. These have been discussed in detail below. Intuitively, since QHEI components measure the site-specific physical habitat conditions, the explanatory factors must affect the stream and its adjoining features morphologically, directly, and at a relatively local scale.

Relatively few studies have hinted at using variables that potentially affect the habitat quality of a stream, to the biotic integrity of the stream. Habitat is widely recognized as the template on which community structures are shaped by landscape and anthropogenic factors across various scales (Poff 1997, Rankin 1995). These studies have mentioned the potential theoretical linkages between other causal factors, such as soils, sinuosity, and slopes, and the physical habitat of a stream in the study region (Mecklenburg 1998, Allan and others 1997, Rosgen 1994).

Land use

Trees in the riparian zone may contribute to the stream cover. Forests are known to be an intricate part of natural terrestrial ecosystems. Their relation to aquatic ecosystems has also been suggested in other studies. Generally, forest land use reduces runoff, sediments, and nutrients and stabilizes stream flow and channel morphology. Agricultural land use reverses most of these trends (Wang 1997). The presence of forests in the subwatershed may stabilize the soils in the upstream regions, reduce the uprooting of top soil by rainwater because of canopy intervention, and deplete particle content of surface runoff through infiltration. Forest may improve the quality of a stream’s overall physical habitat. Forest land use was positively correlated, and agricultural land use was negatively correlated, with both habitat quality and biotic integrity in Wisconsin streams (Wang and others 1997). It may be noted here that a greater percentage of land use in forest or woody areas may only be unplanned as lands in transition between abandoned agricultural land use and prospective urban land use usually end up being forested temporarily (Allan and others 1997).

The land-water interface near the riparian zone, also known as ecotone, is a potentially significant factor in the spatial processes defining the stream habitat and the linkages between the landscape and aquatic ecosystems (Johnson and others 1997). In order to capture the role of riparian zone vegetation the forest and agricultural land use in riparian areas may be used.

Riparian agricultural land use is also measured for its impact on stream habitat due to practices such as seasonal denuding of the land surface, and irrigation practices near the river that may add to greater erosion. Care has been taken to see the correlation between the two variables before adding both of them in the model. We may reasonably expect riparian forest to be positively correlated with QHEI and riparian agricultural land use to be negatively correlated with stream habitat quality.

A few recent studies have attempted to study the relationship between land use and the newer measures of water quality such as the indices for biotic and habitat integrity (Allan and others 1997, Lammert and Allan 1999, Richards and Host 1994).

For this study, riparian zone may be defined as one grid cell (30m) to each side of the stream. I may also study the variation of riparian land use in different buffer sizes. There may be a significant difference between the impact of variables in riparian buffers of 30m, 90m, and 150m. Intuitively, the smallest buffer of 30m (at the smallest scale possible with remote sensing data of 30m resolution) is expected to be the most significant to understand the effect of land use. Allan and others (1997) have suggested the use of different buffer widths, for different streams in the same watershed, based on the stream order or type. This relates to the different influence, of headwater streams and other streams near the mouth of a large-order basin, on the sediment transport and other similar hydraulic functions related to flow.

Reach sinuosity

Sinuosity is defined as the ratio of total stream arc length to the straight-line distance between the endpoints of the reach.[1] Sinuous streams may affect the stream habitat by altering the stream velocity, and bank stability. Sinuosity may also alter the pool and riffle structure of the stream (Rosgen 1994, Mecklenburg 1998). Sinuosity was used in a model to explain biotic integrity in a Michigan basin at different scales (Lammert and Allan 1999). Richards and Host (1994) also used visually-measured sinuosity in a GIS-based study of land use influences on stream habitats. According to them, sinuosity may also reflect upon the extent of flow modifications by different regional land uses. Channel geometry has also been mentioned as a potential reach-scale attribute of the landscape in future studies for understanding stream ecology.

The channel morphology metric of QHEI scores the sinuosity of the stream in the local area of survey. The variable used in this study, reach sinuosity, measures the sinuosity of the overall reach on which the water quality sample point is located. Sinuous streams might improve habitat conditions and increase the QHEI scores. Sinuosity may also be categorized to study the affect of different levels of sinuosity on the stream habitat. Very highly sinuous streams may slow down the natural transfer of stream materials and alter the habitat. Therefore, moderate sinuosity between 1.2 to about 1.8 should improve the QHEI values.

Roads and road-stream intersections

Another major anthropogenic contributor to this model is the presence of roads in the subwatershed. Roads are paved areas, therefore, they add to the surface runoff that may contribute to the stream upstream of the sample point in the subwatershed. Roads are also a surrogate measure for the extent of urban development in the watershed. Besides, the amount of road-covered areas, modeled as road density, the number of road-stream intersections may be strongly and inversely related to the stream habitat. Road-stream intersections may be in the form of bridges and culverts. Such events may result in direct drainage discharges into streams that may adversely affect the QHEI values. If deposited in reasonably large amounts, road salt may also damage riparian vegetation and thus indirectly affect the instream conditions (Mattson and Godfrey 1994).

The roads and road-stream intersections at the watershed level may not be measured at the appropriate scale to capture the significant impact that they might be expected to have on a site’s QHEI. To further the objectives of this study I may also derive data for road density and the number of road-stream intersections within a 500m buffer of the streams. This measure would indicate the extent of parcelization near streams as land subdivisions are usually greatly attracted by a stream’s aesthetic value in the riparian zone (Allan and others 1997). The roads near the periphery of the subwatershed may be adding surface runoff into neighboring subwatersheds rather than the subwatershed in which they have been measured as an attribute of the subwatershed. Riparian roads may also take this fact into account. Therefore, we would expect negative correlation between riparian road density and stream habitat quality. It is also hypothesized that more road intersections should translate into poorer QHEI score. However, depending upon the extent of agricultural land use in the area, rural nonpoint source pollution may still be more influential than urban nonpoint source pollution (Gordon and Fromuth 1981).

Point sources

Point sources may include industrial plants as well as other sources such as landfills. The presence of suspended particles in the discharge may directly influence the substrate structure of the receiving stream. The number of point sources in the subwatershed may be inversely related to the QHEI score at the outlet. Some studies have tried to include new explanatory variables by considering point source inputs as a stressor in the implied cause-and-effect relationship between chemical pollutants and stream biota (Gordon and Majumder 1999). It may be interesting to find whether there is a relationship between the discharges from point sources, and their relative location and numbers in the subwatershed, with significant impacts on the stream habitat.

Population density and number of housing units

These variables provide a measure of the urbanization of the subwatershed. Greater values of these variables may indicate heavier sewage loads on the local streams and damage to the stream habitat. Population density and housing units may also summarize other usual anthropogenic influences in the watershed such as urban development.

Studies have calculated housing density from 1:24,000 topographic maps (Richards 1994). The data on population density and number of housing units, for this project, will be generated from the database of greatest resolution available at this time – the block level data from STF1B files from the Bureau of the Census, 1990. The data will be processed for the ratio-proportioned population and housing units in blocks for only the area that was within the watershed boundary, assuming that the population is uniformly distributed within the watershed. This processing will be done using SAS statistical analysis software on the mainframe.

Population density and housing units may be used for impacts only within the buffer area. The reasoning is the same as that used in measuring riparian road density. Population near the periphery may affect the adjacent watersheds due to the divisioning of land into political units that do not follow hydrological unit boundaries. Riparian population or housing units are less likely to impact any stream other than the one in the buffer zone thereby making the variable more realistic.

Slopes

Slopes represent an important part of the landscape surrounding the subwatersheds containing the aquatic ecosystems of interest. Greater slopes may increase the surface runoff into streams. In riparian zones, the combination of loose soils and higher slopes my define the amount of sediments and the type of substrate as well as stream cover (woody debris) in the stream. Johnson and others (1997) used mean catchment slope and the standard deviation of catchment elevation as variables signifying topologic influences on stream chemistry. They also suggest using riparian slopes in further studies. Slopes causing bank erosion have been suggested to be a cause of increased sediment loads which may lead to excessive fines in the substrate and the consequent damage to microhabitats by filling of interstitial spaces (Mecklenburg 1998). Slopes have also been indicated to be a potential reach- or valley-scale variable in a hierarchical organization of landscape factors (Poff 1997). Statistically, valley slopes are dimensionless properties of a watershed and lend themselves to valid regression-based studies (Strahler 1957).

The map gradient component of QHEI measures the stream gradient. This may be very similar to the average slope of the subwatershed. I therefore hypothesize that relative degree of slopes in the watershed may impact the stream habitat. The slopes in the watersheds were categorized into 7 categories – from 0 to 6% in 1% intervals and greater than 6% slopes as the last category.

Similar to the reasoning used for justifying the inclusion of other riparian variables, and in the spirit of the objectives of this study, I will also generate slope variables for the riparian area only. Slopes within a riparian buffer of 500m along the streams are hypothesized to have a greater impact on the stream habitat than those in the whole watershed. The spatial separation of slopes in the farther confines of the watershed, combined with the interaction of erosional processes with soil permeability and other properties, may reduce the power of the variable in explaining the variation in QHEI. Therefore, riparian slope is preferred over the slopes in the whole subwatershed. Higher slopes may adversely affect the stream habitat because of the potential erosion of fine sediments into the stream.

Slopes within 2%-6% are not known to cause major erosion impacts. Higher than this range of slopes involves more expenses and is generally prohibitive for development purposes. Therefore, slopes between 2%-6% are expected to have a positive correlation and slopes greater than 6% are expected to have a negative correlation with QHEI. On the other hand, slopes less than 1% indicate totally flat land and may again involve some expenses for generating normal runoff from the site to avoid ponding and stagnation of water. The exact impact of such low slopes is not known and is not extensively studied for its effects on stream habitats in other similar studies.

Interaction between Road density and Slopes

Slopes might have multiplier effects on a variety of other factors in the landscape-land use interaction scenario. The landscape may be the template on which different human actions produce different effects based on the properties of the landscape in the neighborhood of the action. For instance, extending the same argument that riparian roads may have greater impact on stream habitat, I hypothesize that riparian roads on steep slopes may impact the stream habitat through greater and faster runoff than roads on relatively moderate slopes. Therefore, I will try to model this differential impact of roads, shaped by the slope of the land on which they have been built, by incorporating an interaction term. This interaction term is assumed to be multiplicative and is the product of a slope category variable as dummy, and the density of riparian roads. Slopes have been categorized into two categories – watersheds with 10% or less of their area with less than 6% slopes are coded as the base category of 0, and watersheds with 10% or more of their area with slopes greater than 6% are coded as 1. We might reasonably expect this interaction term to have a negative correlation with the stream habitat quality.

Soils

The soil series classification of the subwatershed may affect the manner of transfer of runoff over the surface. Soils may differ in their runoff potential, permeability, and erosion potential. If multiple soils are present in the subwatershed, the soil class in the majority of the subwatershed may be used as the contributing soil. If the difference between soil class areas is not significant then the soils in the neighboring subwatersheds may be visually inspected to assign a contributing soil class for the subwatershed of interest. Richards and others (1996) used quaternary geology (outwash sand, lacustrine clay) digitized from 1:500,000 topographic maps to study impacts on stream habitat and biota in east-central Michigan.

Similar to the interaction terms for the effect of roads in different slopes, soils may also be studied for their interaction with slopes. Loose soils may be eroded easily and may contribute sediment pollution. This, in turn, may harm stream habitat by fining or filing of interstitial spaces on streambed, lost pool depths, and increased turbidity (Mecklenburg 1998).

Dams

Dams are known to alter the natural flow regime of streams that may be linked to impacts on the stream habitats (Richards 1994). Impoundments may create distinctly different habitat and biotic groups upstream and downstream of the obstruction (Allan and others 1997). Even as the streams eventually might recover completely from the impoundment effect some distance downstream, the zone of negative influence a great distance based on the size and type of stream, size, type, and location of the dam, and the geologic nature of the stream’s parent basin.

Patches

A patch is a homogenous unit based on any parameter of interest such as substrate, land cover, land use, or topography (Johnson and Gage 1997). The relative location, shape, and dimensions of forest or agricultural patches in riparian zones may be of interest in studying the stream habitat. The support for including variables representing patch dynamics comes from literature in the recently developed field of landscape ecology. Patch dynamics seem to be a potentially important part of spatial modeling but most studies studying the effect of landscape on environmental quality have, however, ignored this factor (Graham and others 1991, Hunsaker and others 1990). Johnson and others (1997) used patch density, the number of forest polygons per unit area, in a study relating land use with stream chemistry. Variables of spatial heterogeneity may be in the form of patch size, patch edge-to-interior ratio, and inter-patch distance (Hunsaker and others 1990).

RESEARCH QUESTION I: EXPLAINING STREAM HABITAT QUALITY

Habitat and related sediment impacts on streams are among the major causes of damage to Ohio streams (Yoder 1995, p340). Therefore, one of the variables collected by Ohio EPA at multiple locations and over multiple time periods is the index for indicating the health of stream habitats – Qualitative Habitat Evaluation Index (QHEI). This study aims to build upon an earlier model (Gordon and Majumder 1999) that explains IBI as a function of watershed-scale stresses and QHEI metrics. The objective is to explain the variation in QHEI as a function of natural and anthropogenic factors and stresses, built in the earlier project. The independent effect, and not the cumulative effect, of the subwatershed upstream till the next sampling site will be analyzed for covariates in the study model.

It may also be noted at this stage that, according to Yoder and Rankin (1995, p140), there is greater natural heterogeneity in the ECBP ecoregion as compared to others such as the HELP ecoregion in Ohio. The spatio-temporal complexity and multi-scale functionality of natural processes implies that ‘prediction with the broad aim of generalization may be coarse-grained’ (Poff 1997). This is especially true if variables are aggregated at any scale.

It is clear from the earlier research by Gordon and Majumder (1999) that anthropogenic factors may negatively impact the stream biotic integrity, or the index for indicating the health of stream biota – Index of Biotic Integrity (IBI). However, as shown in Figure 3, in order to be used as a predictive tool by planners, the QHEI metrics in the set of explanatory variables must be explained themselves at the lower level by easily available real-world data. Therefore, a separate model for QHEI becomes imperative. This study aims to model the variation in QHEI as a function of natural and anthropogenic factors and stresses.

q What are the major causes for variations in QHEI at different spatial locations?

q Can these variations be reasonably explained by known parameters?

q What is the scale (see Figure 2) at which these variations may be appropriately explained?

q Does QHEI vary consistently across spatial scales or is it more pronounced at a particular scale?

q Are riparian zones (more) influential in determining the quality of stream habitat?

q Which real-world factors relate strongly to individual QHEI metric values?

Figure 2. A framework for defining study area boundaries for habitat analysis in a multi-scale quantitative approach.

RESEARCH QUESTION II: EXPLAINING STREAM BIOTIC INTEGRITY

The agenda at this stage is to fit the QHEI model developed in the first stage into the model for IBI which may explain the variations in biotic integrity in terms of a comprehensive, multi-scale set of variables representing landscape, land use and land cover variables. It may not make much theoretical sense to aggregate QHEI values for a basin and then model for basin-scale IBI because most morphological characterizations may be applicable to reaches only (Rosgen 1994). Please see Figure 3 again for the devised strategy.

Rankin in 1989 developed a linear relationship between IBI and QHEI ‘sufficient to forecast nonattainment due to habitat degradation with a reasonable degree of certainty’ (Yoder 1995, p336). Thus, a comprehensive IBI-QHEI-Landscape/Land-use/Land-cover model can be applied to monitor, assess, or regulate proposed activities along the riparian corridor or in the parent subwatershed. A proposed activity’s impacts on the present QHEI may be estimated using the QHEI-Landscape/Land-use/Land-cover model in the first-stage and the projected impacts on the biological potential may be estimated using the second-stage comprehensive IBI-QHEI-Landscape/Land-use/Land-cover model.

Lammert and Allan (1999) have modeled IBI at the subcatchment level even as a mention is made regarding another study that indicates the relation of fish to variation at larger spatial areas. They found that habitat variables in the model explained stream biotic integrity better than land use. This result matches the expectations by Rankin’s study in 1989 relating QHEI to IBI through a linear regression model (Yoder 1995). Urban land use was not a significant part of the study area therefore probably only the impacts of agricultural land use are explained.

It might be interesting to model IBI at the subwatershed scale and study the meso-scale effects on spatial process through a hierarchical design of the study (Poff 1997). This will also probably add to the ease of integrating the QHEI model without the need for aggregating variables from subwatershed- to basin-scale, as both models will measure landscape variables at the same subwatershed scale. Because predictive models may improve our ability to assess and regulate water quality programs, some researchers have attempted different statistical techniques to relate stressors with biotic integrity. Logistic regression was used in a research to study macroinvertebrate species traits (Richards 1997). It may be noted that some studies have focussed on separate species traits rather than composite indices (Richards 1997, Poff 1997). Others have earlier criticized such attempts on account of making the index formulations redundant (Suter 1993). The IBI is, however, a combination of individual species traits as well as taxonomic classifications. Therefore, it may be fairly powerful, flexible and robust (Poff 1997).

DATABASE DESIGN

Both vector and raster digital data formats will be used in the project. The data will be analyzed on a variety of GIS for different analyses as different GIS have different specialization. Consequently, the project will be performed on a variety of computer operating systems. See Figure 2 for an overview of the main operations performed in different GIS and operating system environments. File Transfer Protocol (FTP) software, WS_FTP95, is frequently used to move data and results between different workspaces.

GIS SOFTWARE

The research questions clearly entail a detailed analysis of spatial distribution of different explanatory and dependent variables. Therefore, at the very outset, I have decided to use different GIS because of the different levels of expertise most GIS have with raster and vector data. The DEMs for the six major watersheds in the ECBP ecoregion are more than 250MB in size and may require up to 4 times this space for some intermediate processes while analyzing the composite study area. Therefore, my choice of GIS, and the extent of the study area are guided at some stages by database size and computation space available in the UNIX and PC environments.

ARC/INFO 7.2.1 is the GIS I will probably use for the most part in data preparation. It will initially be used for working with raw data such as for roads and contours in DLG format. It has useful features for vector data manipulation such as conversion from DLG to arc coverage format, clipping, and dissolving. Its TABLE and INFO modules are useful for vector and raster attribute data management.

The GRID module is particularly useful for displaying raster data, and clipping raster data with vector data boundary. Also, most spatial analysis tools for hydrological functions are available in the GRID module. ERDAS Imagine 8.3.1 is used for working with the satellite scene covering the study area. It has useful functions for image rectification and classification that work on the composite multispectral scene such as the one on the study area - Landsat TM image from August 29, 1994 (seven bands). Similar procedures in ARC/INFO GRID module would have been required on each of the different bands separately.

Raster layers can be exported from ERDAS and converted into grid in ARC/INFO. Similarly, arc coverages from ARC/INFO can be exported to ARCVIEW in the arc interchange file format (.e00). ARCVIEW 3.1 is used because it is extremely flexible and fast in terms of data visualization, exploratory analysis, and high quality map output. Also, it is compatible with ARC/INFO coverages and grids through extensions. Most GIS software packages will be used on the UNIX platform.

OPERATING SYSTEMS

The choice of the operating system is simplified by factors such as choice of GIS software and operating system accessibility. Owing to the large databases involved at the initial stages and the complex spatial analysis functionality desired, ARC/INFO software is used on the UNIX platform. For similar reasons, the ERDAS software will be used on the UNIX platform. ARCVIEW software will be mostly on a Personal Computer (PC) with Windows NT operating system. The statistical analysis for block-level Census data is performed using SAS software on the WYLBUR operating system for the mainframe at the Ohio State University campus. Microsoft WORD 97 is used for reporting of the project procedures on both Windows NT and Windows 95.

RASTER DATA

The remote sensing data for the study area is subset from the larger scene available for the Big Darby Creek Watershed in Ohio. The data is a Landsat Thematic Mapper (TM) scene from August 29, 1994. There are seven layers, one from each of the seven sensors. The scene is planimetrically correct but not rectified for projection.

The DEMs are in raster format, therefore, various coverages derived from them, such as subwatersheds and slopes, will also be in raster format. Slopes may also be generated for the riparian buffers through GRID module’s buffering and overlay operations. The subwatersheds will also be generated in raster format and then converted to vector data using Arc/Info’s GRID module. Stream buffers will also be generated by converting streams to grid coverages and then using GRID module’s buffering functionality. This will facilitate overlaying operations with land use and slope layers.

VECTOR DATA

The basic vector data is the United States Geological Survey (USGS) data. The data files are in the Digital Line Graph (DLG-3) format. The scale of these files is 1:24,000. The files cover two layers - Roads, and Hypsography data. Data for these layers is for each of the many USGS quadrangles such as Galloway, Harrisburg, Hilliard, Shawnee Hills, and Commercial Point in Big Darby Creek basin, that cover the study area.

Table 1. Some USGS files for study area in DLG format

Number	USGS Quadrangle	Quadrangle Name
1	Cw107	Galloway
2	Cw115	Harrisburg
3	Cw116	Commercial Point
4	My323	Shawnee Hills
5	My331	Hilliard

The vector data will frequently be used with raster data for data extraction or display. It may also be required during the course of this project to move the vector (and raster) data between the different GIS software, depending on the most suitable functionality available in the software. Table 2 lists the different file formats I might have to work with during the course of this project.

Table 2. Different file formats used in the project

Format	Description
ARC/INFO export (.e00)	A proprietary vector format intended to transfer data, including attributes, between ESRI and third party GIS.
ArcView shapefile	An openly published vector format for use in ArcView, consisting of main files (.shp), index files (.shx), and dBase tables (.dbf).
ARC/INFO coverage	A vector format for storing point, line, and area features along with attributes in INFO database format.
ARC/INFO grid	A raster format for cell-based geoprocessing compatible with ARC/INFO
ERDAS IMAGINE (.img)	A proprietary raster format suitable for multispectral satellite images or layers
Digital Line Graph (DLG)	A vector format used by USGS for exchanging cartographic data files. It only supports integer attribute information.

GIS METHODOLOGY

Once the raw data has been prepared for research, the spatial modeling will use GIS technology extensively. The objective of using GIS is to collect a database of geographic variables that might be used in a statistical regression model. This model may explain the variations in the indices’ values, such as for QHEI, as a function of these spatial variables processes and derived using spatial analysis in GIS. The sources of most spatial data layers are shown in Table 3.

Table 3. Potential data layers used in the project

DATA LAYER	SOURCE
Digital elevation model (DEM)	From USGS 1:24,000 DLG hypsography files in 7.5min quadrangles.
Water quality sample points	Generated using longitude/latitude in USEPA ECOS database.
Subwatershed boundary	Derived from DEMs and Water Quality sample point coverage using GIS functionality.
Streams	From USEPA PEMSO database or USGS DLG hydrography files
Land use	Unsupervised classification of Landsat TM imagery of August 1994, classified into major categories such as agricultural, urban, forest. 30m resolution.
Roads	From USGS 1:24,000 DLG transportation files in 7.5min quadrangles.
Soils	From NRCS 1:250,000 STATSGO database.
Point sources	From USEPA database of 1993 NPDES permittees.
Population	From Bureau of Census TIGER/Line95 files and block-level CENSUS database in STF1B files.
Housing units	From Bureau of Census TIGER/Line95 files and block-level CENSUS database in STF1B files.
Sinuosity	Derived programmatically from Streams coverage.
Slopes	Derived from DEMs using GIS.
Water quality parameters	From USEPA STORET database

In order to do the analysis as suggested by literature review and the relationships hypothesized in the previous sections, I may need to derive various data layers from the raw coverages. Slopes may be directly derived from the DEM grid. Subwatershed boundaries may use the DEM grid and say, QHEI sample point coverage, for deriving subwatersheds upstream of each sample point. The cell values in the resulting grid tell how many cells contribute to the accumulation of water at the point defined as the mouth of the watershed. Only those subwatersheds that have a size within a reasonable range (from comparable studies available in the literature review) will be included for further study. I already have the data layers for land use grid classified from satellite imagery, streams, point sources, population density, housing unit density, and roads coverages. Soils will need to be included in this model from the STATSGO database available from NRCS.

Various issues arise during preparation of the data. There are positional errors in the location of sample points. Multiple samples have been taken during different time-periods at the same location. Some points in the database are at different locations but have the same River mile identification in the database. Some points do not fall exactly on the streams as the stream database is from a different database. Soils are available in different scales and classification levels. There may not be enough variation in soils inside some smaller subwatersheds. The land use is derived from satellite scene that does not cover a few subwatersheds. Different quadrangles, and other raw data are in different projections. All these issues have been considered in preparing the database. The projection will most probably be Transverse Mercator, with a central meridian passing through the study area, for all these layers.

OTHER MAJOR ISSUES

A number of problems may arise during the course of the project because of the characteristics of the spatial processes, the software limitations, database resolution, and other factors. Johnson and Gage (1997) indicate common statistical problems of skewness, multicollinearity, and autocorrelation in most environmental studies. If land use categories add up to 100% then it is safer to drop at least one category to avoid problems of collinearity (Mattson and Godfrey 1994). Some true effects may also be masked due to the underlying correlation between land use and subwatershed geology (Richards 1997). The problem of heteroscedasticity, indicated by fan-shaped residual plots, is also common in environmental studies (Mattson and Godfrey 1994). A few major anticipated problems are briefly mentioned below.

FACTORS AFFECTING WATER RESOURCE INTEGRITY

Biotic factors (intra- and inter-species competition, predation), energy source, and flow regime (groundwater, surface runoff). Poff (1997) mentions the relevance of biotic factors such as species interactions in the understanding and prediction of stream ecology.

MULTIPLE SCALES

It is not clear at which scale, the microhabitat, channel, reach, subwatershed, or basin, the various variables in the model have the most meaningful impact (Allan 1997). The spatial factors affecting biotic communities and instream habitat are not completely understood (Johnson 1997). The variables may also be correlated across multiple scales (Poff 1997).

SMALL NUMBERS

The range of the study area unit, the subwatershed, may be relatively large. Therefore, the effect of percentage values or ratios may be disproportionate and distorted due to the small number problem. In principle, the effect of riparian agricultural land use in an area covering 1 sq.km., that is in a subwatershed of 10 sq.km. area is not the same as that of a riparian zone with 100 sq.km. in agricultural land use, in a large subwatershed with a drainage area of 1,000 sq.km., (even as both subwatersheds have 10% of their area in riparian agriculture land use) on the habitat quality at the drainage outlet of these subwatersheds.

SPATIAL AUTOCORRELATION

Spatial autocorrelation is based on the principle that locations closer in space have similar attributes because the attribute value at one location is dependent on that at the neighboring location(s). This is a second-order property or spatial dependency of spatial processes, as distinct from the first-order properties or heterogeneity (Bailey and Gatrell 1995, p. 4, 32, 247-290). First-order properties measure variations in the mean value of the process. It is a global trend. Second-order properties measure the tendency of the deviations, in values of the process from mean, to correlate with nearby values. This is a local trend.

The design of this study is expected to facilitate autocorrelation measurements because the study units are relatively small and localized. By considering relatively small areas as local watersheds contributing to the QHEI values at a sampled site, we can minimize the first-order spatial dependency in the reference environment and analyze the second-order effects in the spatial process. Care will be taken in recognizing the presence of any such spatial effects.

RESOLUTION

The available data is of relatively coarser resolution than that ideally desired for a detailed riparian zone evaluation. Research has indicated the potential efficacy of studying riparian areas up to a detail of 5m from the stream (Richards 1994). Also, a more detailed classification of agricultural land use into crops, or forest land use into deciduous and conifers, may provide further insights into the understanding of the spatial process (Johnson and others 1997). For instance, conifers may acidify and lower the pH of waters. In that sense, loss of conifers may be positively correlated to the chemical quality of aquatic systems in some cases (Graham and others 1991).

A number of authors have expressed the need for assembling spatial databases with higher resolution (Richards 1994, Graham 1991, Johnson and Gage 1997). However, this project will use the best available multispectral imagery for 1994, the Landsat TM digital data, with a resolution of 30m. This is still more accurate than some other studies that have attempted a riparian-zone analysis with aerial photography data with resolution of 60m, or USGS Land-use/Land-cover data with resolution of 200m to 400m (Richards 1996, Wang 1997).

ACCURACY AND PRECISION

Accuracy is the truthfulness of the measurement relative to the ground reality (Robinson and others 1995, p248). There is some inaccuracy in a few data layers. For instance, the coordinates of some sample points in the ECOS database for IBI and QHEI do not lie on any stream. The accuracy assessment of the unsupervised classification, of Landsat imagery, for land use categories has not been performed. In case of using percentage of areas for variables such as land use, road density, or slopes, the planimetric area has been used which may be different from the true surface area (Strahler 1957).

Precision is the closeness of repeated measurements (Robinson and others 1995, p248). It depends on the tolerance level of instruments and techniques of measurements. Some sample points in the ECOS water quality database lie on the same river mile location because the river mile precision is not very high. The precision of the digitizing method for arcs representing the PEMSO streams is not known. This may affect the values of sinuosity for different reaches. The data will be approximated, when necessary, within reasonable limits and the change will be reported clearly.

CONCLUSION

I hope that this study will provide a better understanding of the spatial processes shaping present-day stream habitats and biotic integrity. It may also provide further insight and justification for watershed-based planning and management programs which may coordinate between regional and local political agencies for regulating point and nonpoint sources of pollution. The study may also indicate the feasibility, merits, and limitations of using biological and physical habitat indictors of water quality in the state monitoring, assessment, and probably permit regulation programs.

ACKNOWLEDGMENTS

I wish to thank Dr. Steve Gordon (Assistant Director, Ohio Supercomputer Center, and Professor of City and Regional Planning, Ohio State University) for letting me use the data available on the study area for the pilot study as well as providing invaluable guidance and technical support at all crucial times. This research proposal will from a part of a comprehensive watershed-based modeling and planning project funded by a grant from USEPA/NSF Water and Watersheds Program R824769.

I also thank Kang-Ping Shen for assistance with TIGER data and the report on pilot study, Tracy Douglass for help on the UNIX system at the Center For Mapping, and Arnold Engelmann for sharing expertise on extracting hydrologically-correct DEMs.

APPENDIX

This pilot study aims to build a linear regression model for explaining the variation in QHEI as a function of natural and anthropogenic factors and stresses. A number of preliminary models were tested. It was clear from these preliminary models that anthropogenic factors may negatively impact the stream habitat, or the index for indicating the health of stream habitats – Qualitative Habitat Evaluation Index (QHEI). Logging, agricultural land use and urban development may all reduce the forest cover that was significantly and positively related to QHEI. I found forest land cover (both riparian and non-riparian), reach sinuosity, and number of point sources in the catchment to be significant in the model. The best preliminary model explained 63% of the variation in site-specific QHEI. This model as a whole was significant at the 1% level. Two other variables, road density in the watershed and non-riparian agricultural land use, were not significant in explaining the linkage between landscape and stream habitat.

The preliminary models provided invaluable insight into the stream physical, chemical, and biotic processes defining the land-water interface. I need to evaluate the relative significance of riparian zones vs. the variables measured at the whole watershed or localshed scale. Road density may be correlated with QHEI at the riparian scale but not at the catchment scale as the surface runoff from peripheral roads may be going into neighboring watersheds instead of the watershed in which they have been measured as an attribute of the sampling site. Similarly, roads between riparian and peripheral zones may not be impacting the stream habitat as much as riparian roads because of greater distance and probably some other underlying soil interactions that have not been measured in this study.

Since any valid regression analysis for this project may require a significant number of subwatersheds. It must be pointed out that there is published research in this field with similar sample sizes. Richards and others (1994) studied 11 sample points along Lake Superior’s North Shore. Lammert and others (1999) sampled 18 sites, six reaches on each of three tributaries, of River Raisin basin in southeastern Michigan. In a study by Allan and others (1997), the same basin, River Raisin, was sampled at 23 different places across 7 regional tributaries.

DATASET

In this study, a watershed has been defined for each sample point in the QHEI dataset. These locally-defined watersheds, localsheds, are generated using raster GIS operations on a base Digital Elevation Model (DEM) grid of the study area. A broader scope of research has been sacrificed, at this time, in favor of a simpler pilot study because of two main reasons. First, the DEMs usually have enormous digital size for any watershed of reasonable area. The DEMs for this project come from 7.5 minute (1:24,000 scale) USGS quadrangle topographical maps. Second, the watershed generation functions in raster GIS software, the Arc/Info GRID module, are memory- and space-intensive. Therefore, only two watershed regions, Big Darby Creek and Great Miami River, have been selected for testing the hypothesis.[2] Big Darby Creek watershed is to the west of Columbus metropolitan area. The Great Miami River watershed covers the region around Springfield, Ohio.

Considering the different resolution of the various datasets in the study only those localsheds have been selected which have an area of more than 0.45 km² (or 500 cells). There are a total of 18 subwatersheds. For areas below this threshold the variation in some dataset layers is lost. On the other hand, no upper threshold is chosen at this time. This may generate an interesting debate related to the small-number problem.[3] The areas of the localsheds occur over a considerable range, from a minimum of 3 km² to a maximum of 900 km².

Riparian and non-riparian landuse

For this study, riparian zone is defined as one grid cell (30m) to each side of the stream. I also studied the variation of riparian land use in different buffer sizes. There was no significant difference between the variables in riparian buffers of 30m, 90m, and 150m. Therefore, the smallest buffer of 30m is used in this study to understand the effect of land use at the smallest scale possible with remote sensing data of 30m resolution.

The average percentage of riparian agricultural land for Big Darby Creek and Great Miami River regions is about 51% and 53% respectively. The average percentage of riparian forest land for Big Darby Creek and Great Miami River regions is about 28% and 21% respectively. The data suggests that riparian landuse is very similar in both the Big Darby Creek and Great Miami River regions.

In non-riparian areas, the average percentage of agricultural land for Big Darby Creek and Great Miami River regions is about 78% and 65% respectively, while the average percentages of forest land for Big Darby Creek and Great Miami River regions is about 7% and 17% respectively. Interestingly, the landuse pattern in non-riparian areas is opposite to that in riparian areas. In general, Big Darby Creek region displays a more agricultural-based non-riparian land use pattern while the Great Miami River region has greater forest-oriented land use in the non-riparian areas.

Reach sinuosity

There is not much difference in both regions in terms of sinuosity. The average sinuosity for Big Darby Creek and Great Miami River regions is 1.25 and 1.19 respectively. It ranges from about 1.05 to 1.65.

Roads and road-stream intersections

Road density in Big Darby Creek region (1674 meters per kilometer square) is lower than that in Great Miami River region (2450 meters per kilometer square). Since road density is employed as an indicator of urban development, this result may imply that Great Miami River region is more developed than Big Darby Creek region is.

The number of road-stream intersections in Big Darby Creek area range from 1 to 114, while the number of intersections for Great Miami River region range from 0 to 306. The average number of intersections for Big Darby Creek and Great Miami River regions are 34 and 47 respectively. According to my hypothesis more road intersections should translate into poorer QHEI score.

Point sources

Population density and number of housing units

The data was processed for the ratio- proportioned population and housing units in blocks for only the area that was within the watershed boundary, assuming that the population was uniformly distributed within the watershed. This processing was done using SAS statistical analysis software on the mainframe.

Slopes

Slopes were processed from the DEMs. The map gradient component of QHEI measures the stream gradient. This may be very similar to the average slope of the localshed. I, therefore, hypothesize that relative degree of slopes in the watershed may impact the stream habitat. The slopes in the watersheds were categorized into 7 categories – from 0 to 6% in 1% intervals and greater than 6% slopes as the last category.

Soils

Soils have not been used at this stage but may be significantly related to stream habitat. I intend to analyze these variables for this model in the future. Similar to the interaction between roads and slopes, I may study the relationship between different types of soils at different slope degrees in the watershed and in the riparian zone, in future studies.

METHODOLOGY AND RESULTS

This study focussed on the research questions mentioned previously in the main section. Correlation analysis was performed to study the relationships between the explanatory variables. Then regression models were built to explain the variation in stream habitat quality in the study area.

In some preliminary models, the adjusted R-square of 0.62 indicated that more than 62% of the variation in dependent variable, the QHEI, could be accounted for by all independent variables. Compared with adjust R-square, R-square was relatively high (0.78). This indicated that there is multicollinearity among independent variables in my model. A study of the residuals indicated some heteroscadasticity in a few variables. However, since theoretical and intuitive reasoning did not rule out a linear relationship, I have decided not to transform many variables at this stage. Some variables, such as riparian road density and riparian slopes within 2-3% indicated non-normality, therefore, models were also tested with a log-transformed form of these variables.

Table C-1. Pearson product-moment correlation matrix for all the variables of interest, in sorted order, for 18 localsheds in Big Darby Creek and Great Miami River watersheds. Only the correlation with stream habitat quality is shown.

*Variables*	*QHEI*
QHEI	1.00
Riparian Forest (%)	0.72
Sinuosity (exp)	0.29
Sinuosity	0.28
Slope 0-1 (%)	0.21
Non-Riparian Agriculture(%)	0.17
No of Point Sources	0.15
Total Agricultural land use (%)	0.15
No of Road-Stream Intersections	0.11
Watershed Area (km^2)	0.09
Riparian Slope 0-1 (%) in 500m buffer	0.08
Riparian Slope 4-5 (%) in 500m buffer	0.08
Riparian road length (km) in 500m buffer	0.08
Riparian Slope 5-6 (%) in 500m buffer	0.07
Road length (km)	0.06
Riparian road density in 500m buffer (log)	0.04
Population density (persons per km^2)	0.02
Housing Unit density (units per km^2)	0.00
Population count	-0.02
Population count (log)	-0.02
Riparian Slope 3-4(%) in 500m buffer	-0.02
Slope 1-2 (%)	-0.03
Total Forest (%)	-0.03
Road length (log)	-0.04
Housing Unit count	-0.05
Housing unit count (log)	-0.06
Riparian Slope 1-2 (%) in 500m buffer	-0.07
Riparian Slope >6 (%) in 500m buffer	-0.09
Slope 4-5 (%)	-0.09
Riparian Slope 2-3 (%) in 500m buffer	-0.11
Slope 3-4 (%)	-0.12
Slope 5-6 (%)	-0.12
Non-Riparian Forest (%)	-0.14
Road Density (km/km^2)	-0.19
Slope >6 (%)	-0.20
Slope 2-3 (%)	-0.22
Riparian road density (km/km2) in 500m buffer	-0.27
Riparian Agriculture (%)	-0.48

Correlation analysis showed that most of the highly interrelated variables involve riparian or non-riparian agriculture (Table C-1). Therefore, the formulation of another set of variables, excluding these two variables, may be necessary. Another thing to note is the weak relationship of individual parameters with QHEI.

The coefficient of riparian forest land use is 0.72, indicating that it is highly positively correlated with QHEI in a watershed. On the other hand, riparian agricultural is the variable with the highest negative correlation with QHEI. Riparian road density in a 500m buffer is also negatively correlated with QHEI. Sinuosity is positively correlated with stream habitat in the given range of the sinuosity in my dataset that ranges from low to moderate sinuosity. Many variables, such as population density and housing density, have almost no correlation with stream habitat. Probably the scale of measurement of these variables is not appropriate or their effect on QHEI might change in the presence of other interacting variables.

Linear Regression Models

Two types of models were produced. One is for the QHEI itself, and the other is for QHEI components. The reason I want to have a close look at QHEI components is that the QHEI is an highly aggregated index. One variable might not be significant in the QHEI model, but that variable may possibly be very significant in the QHEI component models. Because of highly aggregated index, other explanatory variables might mask some more obviously important variables. However, only two models for QHEI as a whole are discussed here. Models for QHEI components are not discussed here.

The following table shows the result of my best initial regression model. In this model, five independent variables, percentage of riparian agricultural land, road density within 500-meter buffer zone along the stream, reach sinuosity, percentage of slope ranging above 6% within 500-meter buffer zone along the stream, and percentage of slope ranging between 2% and 3% within 500- meter buffer zone along the stream, are used to explain the dependent variable, qualitative habitat evaluation index (QHEI).

Table R-1 . The summary of the result of Linear Regression Model

Variable	Coefficient	Std. Error	t Value	Prob > \|t\|	Sign	Significant
Intercept	103.85	23.37	4.44	0.0008
Riparian Agricultural Land	-1.20	0.17	-6.96	0.0000	X	X
Road Density (500 m)	-10.28	2.69	-3.82	0.0024	X	X
Reach Sinuosity	10.82	15.86	0.68	0.5081	X
Slope > 6 (500 m)	-0.33	0.18	-1.86	0.0882	X	X
Slope 2-3 (500 m)	4.04	0.82	4.92	0.0004	X	X

N = 18 Residual standard error: 640.38 on 12 degrees of freedom

F-statistic: 12.19 on 5 and 12 degrees of freedom Prob > F = 0.0002

R-Squared: 0.84 Adj. R-square: 0.77

The result shows that except the variable of sinuosity (t-statistic = 0.68, p-value = 0.5081), all other explanatory variables are statistically significantly related to the dependent variable at a significance level of either 0.05 or 0.10. The signs of all the explanatory variables confirm my expectations. The overall model is statistically significant (F-statistic = 12.19, p-value = 0.0002).

Since the coefficients for such variables as sinuosity and percentage of slope ranging between 2% and 3% within 500-meter buffer zone along the stream are 10.82 and 4.04 (> 0) respectively, the relationship between independent and dependent variables is positive. On the other hand, because the coefficients for the rest variables are less than zero, the relationship between independent and dependent variables is negative. The value of coefficient means that for each unit increase in explanatory variable, the QHEI score is increased (+ sign) or reduced (- sign) by the amount of coefficient value, depending on the sign of the regression coefficient for each explanatory variable. It should be noted that such a causal relationship is true only when one explanatory variable changes, and the others remain constant.

The coefficient of percentage of riparian agricultural land (-1.20) indicates that for each unit increase in independent variable (one percent increase in riparian agricultural land in a localshed), the QHEI score will be decreased by about 1.2. The result follows my earlier expectation. Because of the current agricultural practices, the higher percentage of riparian agricultural land in a given localshed, the lower QHEI score.

The coefficient of road density within 500-meter buffer along the stream (-10.28) indicates that for each unit increase in independent variable (one percent increase in road density within the 500-meter buffer zone), the QHEI score will be decreased by about 10.28. In other words, for each kilometer of new roads per square kilometers, the stream habitat index drops by a value of about 10. Since the range of the QHEI is from 0 to 100, this variable plays a very crucial role in the model. Moreover, because the road density is used as an indicator of urban development, the relationship between QHEI and road density should be negative, based on the common sense that urban development tends to damage the natural habitats. The result not only shows the significance of the variable in the model, but also confirms my general hypothesis that the higher road density, the lower the stream habitat quality.

The coefficient of reach sinuosity is 10.82, indicating that for each unit increase in reach sinuosity, the QHEI score will be increased by about 11. Such a high increase in QHEI may be very strong, since QHEI ranges from 0 to 100. However, studies reveal that the general range of reach sinuosity is between 1 and 2.5. The interpretation that for each 0.1 unit increase in reach sinuosity, the QHEI score will be increased by about 1.1 should be more appropriate here. Although the sign for this variable is the same as my expectation, the variable is not statistically significant in my model. The previous studies show that the sinuosity is one important factor to the water habitat (Rosgen 1994, Mecklenburg 1998). I decided to keep this explanatory variable in spite of the poor performance in the statistical model.

The coefficient of percentage of slope ranging above 6% within 500-meter buffer zone along the stream is –0.33, indicating that for each unit increase in independent variable (one percent increase in the area with slope greater than 6% within the 500-meter buffer zone), the QHEI score will be decreased by 0.33. The sign confirms my earlier hypothesis that the higher percentage of slope ranging above 6%, the lower the QHEI score. This is because the higher slope, the higher possibility of erosion might occur.

The coefficient of percentage of slope ranging between 2% and 3% within 500-meter buffer zone along the stream is 4.04, indicating that for each unit increase in independent variable (one percent increase in the area with slope ranging between 2% to 3% within the 500-meter buffer zone), the QHEI score will be increased by 4.04. The sign confirms my earlier hypothesis that the higher percentage of gentle slope within the buffer zone, the higher the QHEI score. In terms of magnitude, this variable has more influential power in the model than previous variable, percentage of slope ranging above 6% within 500-meter buffer zone along the stream.

R-square is one of the most important indicators to assess the model. The adjusted R-square (= 0.77) indicates that more than 77% of the variation in dependent variable, the QHEI, can be accounted for by all independent variables. Compared with adjust R-square, R-square is a little bit high (0.84). This may indicate that there is multicollineraity among independent variables in my model. A study of the residuals indicates some heteroscadasticity in a few variables. However, since theoretical and intuitive reasoning does not rule out a linear relationship, I have decided not to transform any variables at this stage. However, such variable as percentage of slope ranging between 2% and 3% within 500-meter buffer zone along the stream will use log-transformation to do further analysis at a later stage. Similarly, soils may be incorporated in the model as a dummy (good, moderate, and poor) variable and modeled for interaction with landuse variables such as agricultural land use in the non-riparian region.

Now, I am interested in more detailed effects of the variable of slope in the model. To understand such effects, a dummy variable and an interaction term were introduced to my model. The dummy variable, SlopeHigh, is defined as that SlopeHigh is equal to the 1 if there are more than nine percent of the area with slope greater than 6% in the whole localshed. Otherwise, the dummy variable is 0. The interaction term is defined as SlopeHigh *RdDen500 (Road Density with 500-meter buffer zone). The following table shows the result of the regression model with interaction term. In this model, four independent variables, percentage of riparian agricultural land, reach sinuosity, the percentage of slope ranging between 2% and 3% within 500- meter buffer zone along the stream, and road density within 500-meter buffer zone along the stream as well as the dummy variable, SlopeHigh, and an interaction term of SlopeHigh*RdDen500 are used to explain the dependent variable, qualitative habitat evaluation index (QHEI).

Table R-2 . The summary of the result of Linear Regression Model

Variable	Coefficient	Std. Error	t Value	Prob > \|t\|	Sign	Significant
Intercept	97.37	21.85	4.46	0.0010
Riparian Agricultural Land	-1.24	0.16	-7.61	0.0000	X	X
Reach Sinuosity	13.81	14.67	0.94	0.3668	X
Slope 2-3 (500 m)	4.13	0.76	5.44	0.0002	X	X
Road Density (500 m)	-8.06	3.19	-2.53	0.0282	X	X
SlopeHigh (dummy)	-7.47	9.67	-0.77	0.4562	X
SlopeHigh*RdDen500	-1.03	4.34	-0.24	0.8175	X

N = 18 Residual standard error: 496.35 on 11 degrees of freedom

F-statistic: 12.55 on 6 and 11 degrees of freedom Prob > F = 0.0002

R-Squared: 087 Adj. R-square: 0.80

The table R-2 shows that except the variables of sinuosity, SlopeHigh, and Slope*RdDen500, all other explanatory variables are statistically significantly related to the dependent variable at a significance level of either 0.05 or 0.10. The signs of all the explanatory variables confirm my expectations. The overall model is statistically significant (F-statistic = 12.55, p-value = 0.0002). Compared to my best model with no dummy variable and interaction term, this model has some improvement, according to the F-statistic and R-square measures.

Since the coefficients for such variables as sinuosity and percentage of slope ranging between 2% and 3% within 500-meter buffer zone along the stream are 13.81 and 4.13 (> 0) respectively, the relationship between independent and dependent variables is positive. On the other hand, because the coefficients for the rest variables are less than zero, the relationship between independent and dependent variables is negative. The value of coefficient means that for each unit increase in explanatory variable, the QHEI score is increased (+ sign) or reduced (- sign) by the amount of coefficient value, depending on the sign of the regression coefficient for each explanatory variable. It should be noted that such a causal relationship is true only when one explanatory variable changes, and the others remain constant.

The coefficient of percentage of riparian agricultural land (-1.24) indicates that for each unit increase in independent variable (one percent increase in riparian agricultural land in a localshed), the QHEI score will be decreased by about 1.24. The result follows my earlier expectation. Compared to my best model without interaction term, the explanatory power of this variable is interestingly increased (regardless of sign, from 1.20 to 1.24). This may suggest that some problems related to the multicollinearity in my best model without interaction term have been corrected in this new model.

The coefficient of reach sinuosity is 13.81, indicating that for each unit increase in reach sinuosity, the QHEI score will be increased by about 13.81. Such a high increase in QHEI may be very strong, since QHEI ranges from 0 to 100. However, studies reveal that the general range of reach sinuosity is between 1 and 2.5. The interpretation that for each 0.1 unit increase in reach sinuosity, the QHEI score will be increased by about 1.3 should be more appropriate here. Although the sign for this variable is the same as my expectation, the variable is not statistically significant in my model. Again, the previous studies show that the sinuosity is one important factor to the water habitat (Rosgen 1994, Mecklenburg 1998). I decided to keep this explanatory variable in spite of the poor performance in the statistical model. Moreover, the explanatory power of this variable also goes up (from 10.82 to 13.81).

The coefficient of percentage of slope ranging between 2% and 3% within 500-meter buffer zone along the stream is 4.13, indicating that for each unit increase in independent variable (1 percent increase in the area with slope ranging between 2% to 3% within the 500-meter buffer zone), the QHEI score will be increased by 4.13. The sign confirms my earlier hypothesis that the higher percentage of gentle slope within the buffer zone, the higher the QHEI score. Compared to my best model, the variable becomes more statistically significant, and its explanatory power slightly increases (from 4.04 to 4.13).

The coefficient of road density within 500-meter buffer along the stream (-8.06) indicates that for each unit increase in independent variable (one percent increase in road density within the 500-meter buffer zone), the QHEI score will be decreased by about 8.1. In other words, for each kilometer of new roads per square kilometers, the stream habitat index drops by a value of about 8.1. Because the road density is used as an indicator of urban development, the relationship between QHEI and road density should be negative, based on the common sense that urban development tends to damage the natural habitats. The result not only shows the significance of the variable in the model, but also confirms my general hypothesis that the higher road density, the lower the stream habitat quality. Moreover, the explanatory power of this variable also decreases (regardless of sign, from 10.28 to 8.06).

The coefficient of dummy variable, SlopeHigh, is –7.47. Dummy variable itself does not change the coefficients in the model. However, it will change the value of intercept when the value of the dummy variable is equal to 1. So, when the dummy variable, SlopeHigh, is equal to 1, the intercept becomes 89.9.

The interaction term is used to examine the effects of road density in the high slope area (slope > 6%) within 500-meter buffer zone. The coefficient of this interaction term is –1.03) indicates that for each unit increase in independent variable (one percent increase in road density in the high slope area within the 500-meter buffer zone), the QHEI score will be decreased by about 1.03. In other words, for each kilometer of new roads per square kilometers in the high slope area, the stream habitat index drops by a value of about 1.03. Because the road density is used as an indicator of urban development, the relationship between QHEI and road density should be negative, based on the common sense that urban development tends to damage the natural habitats. The result not only shows the significance of the variable in the model, but also confirms my general hypothesis that the higher road density, the lower the stream habitat quality. However, this variable is not statistically significant in the model.

The adjusted R-square (= 0.84) indicates that more than 84% of the variation in dependent variable, the QHEI, can be accounted for by all independent variables. Compared with adjust R-square, R-square is a little bit high (0.90). This may indicate that there is still multicollineraity among independent variables in my model.

DISCUSSION

In conclusion, I have reason to believe that QHEI is better explained by variables measured at the riparian or local scale, relative to those measured at the watershed scale. Three riparian variables, agriculture, roads, and steep slopes (above 6%), have significantly negative relationship with QHEI. Another riparian variable, moderate slopes between 2 to 3%, has a significant positive impact on the stream habitat quality. The sinuosity of a stream may need more accurate description before its effect might become significant in the model.

The resolution of the dataset may be a matter of concern in relating the stream habitat quality to variables that measure the extent of urban or suburban development on the physical landscape. The land use in the riparian zone is measured to a buffer of 30m, 90m, and 150m, because the resolution of satellite data, from which the land use data is generated, is only 30m. A buffer of lesser width may result in little variation in the dataset. Ideally, the stream habitat quality is measured, and may be related, to only the nearest 4-5 meters on either side of the stream.

Similarly, this pilot project has used average reach sinuosity as a proxy for actual stream sinuosity in the vicinity of the sample point. The actual sinuosity may be an effective measure for only a distance of 200-500m upstream of the sampling site. Moreover, the reach sinuosity, as measured in this project, may be influenced by the accuracy with which the streams have been digitized in the first place. Road-stream intersections might be more strongly correlated with the QHEI at a sampling location if the location of intersections is distance-weighted relative to the sampling site. A bridge 5 km upstream of a sampling site may not affect a location’s habitat as another located just 100m above the point of sampling. Future modifications in this model may incorporate actual empirical analysis of soil series data in the localsheds.

Statistically, I can use different forms to model the data other than the linear function used in this study. I can also model the QHEI in terms of probabilities of the stream habitat attaining a ‘quality status’ rather than an absolute value for the index. Such models might be based on the ordinal logit and probit modeling techniques. Also, linear equations may be used to arrive at statistical estimates of QHEI for unsampled areas. Theoretically, the landscape and land use/cover variables may be modeled to estimate the individual metrics that may then be combined simultaneously to produce an estimate for the composite QHEI.

References

Adler, R.W., 1995. Filling the gaps in water quality standards: Legal perspectives on biocriteria. Chapter 22 in Biological Assessment and Criteria: Tools for Water Resource Planning and Decision Making. Edited by Wayne S. Davis and Thomas P. Simon. Lewis Publishers, Boca Raton, FL.

Allan, J.D., and Johnson, L.B., 1997. Catchment-scale analysis of aquatic ecosystems. Freshwater Biology. 37: 107-111. Special applied issues section.

Allan, J.D., Erickson, D.L., and Fay, J., 1997. The influence of catchment land use on stream integrity across multiple spatial scales. Freshwater Biology. 37: 149-161. Special applied issues section.

Bailey, T.C., and Gatrell, A.C., 1995. Interactive spatial data analysis. Addison Wesley Longman Limited, Essex CM20 2JE, England.

Bretschko, G., 1995. River/land ecotones: scales and patterns. Hydrobiologia. 303: 83-91.

Cooper, S.D., Diehl, S., Kratz, K., and Sarnelle, O., 1998. Implications of scale for patterns and processes in stream ecology. Australian Journal of Ecology. 23: 27-40.

Frissell, C.A., Liss, W.L., Warren, C.E., and Hurley, M.C., R.K., 1986. A hierarchical framework for stream habitat classification, viewing streams in a watershed context. Environmental Management. 10: 199-214.

Gordon, S.I., and Fromuth, R.K., 1981. A point, non-point source model of dissolved oxygen for the Great Miami River. Journal of Environmental Systems. 10: (3) 185-199.

Gordon, S.I., and Majumder, S., 1999. Empirical stressor-response relationships for prospective risk analysis. Environmental Toxicology and Chemistry. (accepted article for forthcoming publication).

Graham, R.L., Hunsaker, C.T., O’Neill, R.V., and Jackson, B.L., 1991. Ecological risk assessment at the regional scale. Ecological Applications. 1: (2) 196-206.

Hunsaker, C.T., Graham, R.L., Suter II, G.W., O’Neill, R.V., Barnthouse, L.W., and Gardner, R.H., 1990. Assessing ecological risk on a regional scale. Environmental Management. 14: (3) 325-332.

Hughes, R.M., 1995. Defining acceptable biological status by comparing with reference conditions. Chapter 4 in Biological Assessment and Criteria: Tools for Water Resource Planning and Decision Making. Edited by Wayne S. Davis and Thomas P. Simon. Lewis Publishers, Boca Raton, FL.

Hughes, R.M., Larsen, D.P., and Omernik, J.M., 1986. Regional reference sites: a method for assessing stream potential. Environmental Management. 10: 629-635.

Hughes, R.M., and Omernik, J.M., 1981. Use and misuse of the terms watershed and stream order. Proceedings of the warmwater streams symposium. American Fisheries Society, Bethesda, Maryland. Edited by L.A. Krumholz. 320-326.

Hutchinson, M.F., and Dowling, T.I., 1991. A continental hydrological assessment of a new grid-based digital elevation model of Australia. Hydrological Processes. 5: 45-58.

Johnson, B.L., Richardson, W.B., and Naimo, T.J., 1995. Past, present, and future concepts in large river ecology: How rivers function and how human activities influence river processes. Bioscience. 45: (3) 134-141.

Johnson, L.B., and Gage, S.H., 1997. Landscape approaches to the analysis of aquatic ecosystems. Freshwater Biology. 37: 113-132. Special applied issues section.

Johnson, L.B., Richards, C., Host, G.E., and Arthur, J.W., 1997. Landscape influences on water chemistry in Midwestern stream ecosystems. Freshwater Biology. 37: 193-208. Special applied issues section.

Johnston, C.A., 1993. Introduction to quantitative methods and modeling in community, population, and landscape ecology. Chapter 25 in Environmental modeling with GIS. Edited by Michael F. Goodchild, Bradley O. Parks, and Louis T. Steyaert. Oxford University Press, New York, NY 10016-4314.

Karr, James R., 1981. Assessment of biotic integrity using fish communities. Fisheries. 6: (6) 21-27.

Karr, J.R., and Dudley, D.R., 1981. Ecological perspective on water quality goals. Environmental Management. 5: (1) 55-68.

Lammert, M., and Allan, J.D., 1999. Assessing biotic integrity of streams: Effects of scale in measuring the influence of land use/cover and habitat structure on fish and macroinvertebrates. Environmental Management. 23: (2) 257-270 FEB.

Lanfear, K.J., 1990. A fast algorithm for automatically computing Strahler stream order. Water Resources Bulletin. 26: (6) 977-981.

Lillesand, T.,M., and Kiefer, R.W., 1994. Remote Sensing and Image Interpretation. Third edition. John Wiley & Sons, Inc., New York, NY.

Maasdam, R., and Smith, D.G., 1994. New Zealand’s National River Water Quality Network: Relationships between physico-chemical data and environmental factors. New Zealand Journal of Marine and Freshwater Research. 28: 37-54.

Majumder, S., 1998. A spatial empirical analysis of stressor-response relationships for prospective ecological risk assessment in the Eastern Cornbelt Plains ecoregion of Ohio. Unpublished Ph.D. dissertation. Ohio State University.

Mattson, M.D., and Godfrey P.J., 1994. Identification of road salt contamination using multiple-regression and GIS. Environmental Management. 18: (5) 767-773 SEP-OCT.

Mecklenburg, D.E., 1998. Channel pattern’s influence on sediment pollution: A case study of Salt Creek, a previously straightened channel redeveloping meanders. Presented at ASAE annual international meeting at Orlando, Florida. July 11-16. Paper No. 982084.

Meybeck, M., 1998. Man and river interface: multiple impacts on water and particulates chemistry illustrated in the Seine river basin. Hydrobiologia. 373/374: 1-20.

Milne, J.A., and Sear, D.A., 1997. Modeling river channel topography using GIS. International Journal of Geographical Information Science. 11: (5) 499-519.

Miltner, R.J., and Rankin, E.T., 1998. Primary nutrients and the biotic integrity of rivers and streams. Freshwater Biology. 40: 145-158. Applied issues.

Omernik, James M., 1995. Ecoregions: A spatial framework for environmental management. Chapter 5 in Biological Assessment and Criteria: Tools for Water Resource Planning and Decision Making. Edited by Wayne S. Davis and Thomas P. Simon. Lewis Publishers, Boca Raton, FL.

Omernik, James M., 1987. Ecoregions of the coterminous United States. Annals of the Association of American Geographers. 77: (1) 118-125.

Poff, N.L., 1997. Landscape filters and species traits: towards mechanistic understanding and prediction in stream ecology. Journal of North American Benthological Society. 16: (2) 391-409.

Rankin, Edward T., 1995. Habitat indices in water resource quality assessments. Chapter 13 in Biological Assessment and Criteria: Tools for Water Resource Planning and Decision Making. Edited by Wayne S. Davis and Thomas P. Simon. Lewis Publishers, Boca Raton, FL.

Richards, C., and Host, G.E., 1994. Examining land-use influences on stream habitats and macroinvertebrates – a GIS approach. Water Resources Bulletin. 30: (4) 729-738 JUL-AUG.

Richards, C., Haro, R.J., Johnson, L.B., and Host, G.E., 1997. Catchment and reach-scale properties as indicators of macroinvertebrate species traits. Freshwater Biology. 37: 219-230. Special applied issues section.

Richards, C., Johnson, L.B., and Host, G.E., 1996. Landscape-scale influences on stream habitats and biota. Canadian Journal of Fisheries and Aquatic Science. 53: (Suppl. 1) 295-311.

Ricklefs, R.E., 1997. The economy of nature. Fourth edition. W.H. Freeman and Company, New York, NY 10010.

Rosgen, D.L., 1994. A classification of natural rivers. Catena. 22: 169-199.

Strahler, A.N., 1957. Quantitative analysis of watershed geomorphology. American Geophysical Union Transactions. 38: 913-920.

Suter, G.W., 1993. A critique of ecosystem health concepts and indexes. Environmental Toxicology and Chemistry. 12: 1533-1539.

Tarboton, D.G., Bras, R.L., and Rodriguez-Iturbe, I., 1991. On the extraction of channel networks from digital elevation data. Hydrological Processes. 5: 81-100.

Townsend, C.T., Arbuckle, C.J., Crowl, T.A., and Scarsbrook, M.R., 1997. The relationship between land use and physicochemistry, food resources and macroinvertebrate communities in tributaries of the Taieri River, New Zealand: a hierarchically scaled approach. Freshwater Biology. 37: 177-191. Special applied issues section.

Vannote, R.L., Minshall, W.W., Cummins, K.W., Sedell, J.R., and Cushing, C.E., 1980. The River Continuum Concept. Canadian Journal of Fisheries and Aquatic Science. 37: 130-137.

Wang, L.Z., Lyons, J., Kanehl P., and Gatti R. 1997. Influences of watershed land use on habitat quality and biotic integrity in Wisconsin streams. Fisheries. 22: (6) 6-12 JUN.

Wang, X., and Yin, Z., 1997. Using GIS to assess the relationship between land use and water quality at a watershed level. Environment International. 23: (1) 103-114.

White, Dale A., Smith, R.A., Price, C.V., Alexander, R.B., and Robinson, K.W., 1992. A spatial model to aggregate point-source and nonpoint-source water-quality data for large areas.. Computers and Geosciences. 18: (8) 1055-1073.

Wiley, M.J., Osborne, L.L., and Larimore, R.W., 1990. Longitudinal structure of an agricultural prairie river system and its relationship to current stream ecosystem theory. Canadian Journal of Fisheries and Aquatic Science. 47: 373-384.

Yoder, Chris O., 1991. Answering some concerns about biological criteria based on experiences in Ohio. Water Quality Standards for the 21^st Century, Proceedings of a National Conference. U.S. Environmental Protection Agency, Office of Water, Washington, D.C.

Yoder, Chris O., 1995. Policy issues and management applications for biological criteria. Chapter 21 in Biological Assessment and Criteria: Tools for Water Resource Planning and Decision Making. Edited by Wayne S. Davis and Thomas P. Simon. Lewis Publishers, Boca Raton, FL.

Yoder, C.O., Albeit, P.S., and Smith, M.A., 1981. The distribution and abundance of fishes in the mainsteram Scioto River as affected by pollutant loadings. Technical report OEPA 81/3. State of Ohio Environmental Protection Agency. Office of Wastewater Pollution control. Division of Surveillance and Standards. Columbus, Ohio. JUNE.

Yoder, C.O., and Rankin, E.T., 1995. Biological criteria program development and implementation in Ohio. Chapter 9 in Biological Assessment and Criteria: Tools for Water Resource Planning and Decision Making. Edited by Wayne S. Davis and Thomas P. Simon. Lewis Publishers, Boca Raton, FL.

Yoder, C.O., and Rankin, E.T., 1998. The role of biological indicators in a state water quality management process. Environmental Monitoring and Assessment. 51: 61-88.

[1] This definition is sufficient for small areas. For larger regions, the stream length in valleys is used instead of distance between endpoints. This takes into account the presence of mountainous or similar obstructions. Some people define sinuosity as the inverse of the ratio mentioned above.

[2] These regions have been defined, and named as such, by the Ohio EPA based on the watersheds for the seventh-order (Strahler) streams of Ohio. A total of 93 seventh-order watersheds have been defined for Ohio. This study comprises of regions in Ohio EPA’s seventh-order watershed 39 (Big Darby) and parts of watersheds 55 and 56 (Great Miami).

[3] The small-number problem relates to using very small numbers as denominators in calculating ratios or rates such as mortality rates in epidemiological studies. Simply stated, a road density of 10% in a 100 km² watershed is very different from the same ratio in a 0.5 km² watershed. On the other hand, if road length is used, instead of road density, then the issue of ‘modifiable areal unit problem’ becomes a concern.

http://www.ucgis.org/oregon/papers/arya.htm

Abstract

Introduction

STUDY AREA UNIT: INDEPENDENT SUBWATERSHEDS

IBI AS A MEASURE OF STREAM BIOLOGICAL INTEGRITY

QHEI AS A MEASURE OF STREAM HABITAT QUALITY

Land use

Reach sinuosity

Roads and road-stream intersections

Point sources

Population density and number of housing units

Slopes

Interaction between Road density and Slopes

Soils

Dams

Patches

RESEARCH QUESTION I: EXPLAINING STREAM HABITAT QUALITY

RESEARCH QUESTION II: EXPLAINING STREAM BIOTIC INTEGRITY

DATABASE DESIGN

VECTOR DATA

OTHER MAJOR ISSUES

CONCLUSION

ACKNOWLEDGMENTS

APPENDIX

DATASET

Riparian and non-riparian landuse

Reach sinuosity

Roads and road-stream intersections

Point sources

Population density and number of housing units

Slopes

Soils

METHODOLOGY AND RESULTS

Table R-1 . The summary of the result of Linear Regression Model

Intercept

Table R-2 . The summary of the result of Linear Regression Model

Intercept

DISCUSSION

References