| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
,*,a
a Kansas Applied Remote Sensing Program and Dep. of Geography, Univ. of Kansas, Lawrence, KS 66045
b Kansas Applied Remote Sensing Program, Kansas Biological Survey and Dep. of Ecology and Evolutionary Biology, Univ. of Kansas, Lawrence, KS 66045
c Kansas Applied Remote Sensing Program, Univ. of Kansas, Lawrence, KS 66045
* Corresponding author (griffith{at}usgs.gov)
Received for publication February 19, 2001.
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: AVHRR, advanced very high resolution radiometer • HI, habitat index • IBI, index of biotic integrity • IJI, interspersion and juxataposition index • LPM, landscape pattern metric • LULC, land use–land cover • NDVI, normalized difference vegetation index • USGS, United States Geological Survey • VPM, vegetation phenological metric
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
To address problems arising from agricultural nonpoint-source pollution, aquatic resource managers and fish ecologists must pay greater attention to large-scale spatial and temporal heterogeneity (Schlosser, 1991; O'Neill et al., 1997; Labbe and Fausch, 2000; Marsh-Matthews and Matthews, 2000). Consensus is forming that stream condition assessments must include both stream reaches and whole catchments (Sidle and Hornbeck, 1991; Roth et al., 1996; Johnson and Gage, 1997; Wiley et al., 1997). As a result of the strong linkages between stream biotic communities, water quality, and the surrounding landscape, land cover information is used extensively to support water quality studies (Zelt et al., 1995), especially as advances in remote sensing and geographic information systems (GIS) have made regional-level studies more feasible (Johnson and Gage, 1997; Herlihy et al., 1998).
|
LANDSCAPE–WATER QUALITY APPROACHES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Landscape Pattern
Another new research avenue for studying LULC–water quality relationships focuses on landscape pattern effects on water quality (Sharpe, 1994; Cairns and Niederlehner, 1996; Johnson et al., 1997; Schuft et al., 1999). The LPMs potentially affecting stream conditions include measures of fragmentation and connectivity, patch size and density, and the number of cover types (Jones et al., 1996). Mixed results have been reported from the few studies that have used LPMs in water quality studies. Wear et al. (1998) simulated landscape changes along an urban–rural gradient in the Southern Appalachians and suggested that different landscape profiles, including various LPMs, can have important implications for water quality. In southern Illinois, Hunsaker and Levine (1995) found that two landscape pattern metrics, dominance and contagion, did not explain as much variation in water quality as did LULC when analyzing full watersheds. When using other analysis units, however, such as stream corridors or hydrologic contributing areas, contagion was found to explain significant variation in conductivity and nutrient levels. In a study of several landscape parameters and water quality in Michigan streams, Johnson et al. (1997) and Richards et al. (1996) found that patch density had some bearing on water quality, but other factors such as geology or slope had equal or greater effect in most cases. Sharpe (1994) found little correlation between LPMs in grid cells of a runoff model and their nutrient output.
|
OBJECTIVES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
The goal of this work is to seek methods that are able to characterize landscapes for regional assessment. Hence, we aim to explore screening indicators to identify watersheds that may be at risk of environmental degradation. This type of study is important for setting natural resource policy conducted at large scales. For example, Section 305(b) of the U.S. Clean Water Act mandates the assessment of water bodies, but a recent study found that about 80% of stream miles go unassessed (General Accounting Office, 2000). Thus, large-scale regional analyses are crucial. Our null hypothesis is that there is no difference between the relationships of NDVI and LPMs with the selected water quality parameters. Specifically, the focus of the research questions in this study are: (i) What are the relationships between landscape pattern metrics (LPMs) and selected water quality parameters? (ii) How much variation in the selected water quality parameters is explained by regression models using LULC and LPMs? (iii) How does the amount of variation explained by the combination of LULC and LPMs compare with that of the NDVI-derived metrics?
|
METHODS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Precipitation ranges from 380 to 450 mm in westernmost Kansas and Nebraska, to 900 to 1000 mm in eastern Kansas, and to nearly 1200 mm on the Mississippi River in southeastern Missouri (Schroeder, 1982; Goodin et al., 1995). Native vegetation in the region consists of shortgrass prairie in westernmost Kansas and Nebraska, tallgrass and mixed-grass prairie in the Nebraska Sand Hills and central Kansas, a mosaic of bluestem prairie and oak–hickory forest in both eastern Kansas and northern Missouri, and dense oak–hickory forests in the Ozark Highlands. The central human transformation of the Great Plains has been the conversion of native grasslands to cropland. Currently, 90% of the area is in farms or ranches and 75% of the land area is cultivated (Riebsame, 1990). Chapman et al. (2001) provide a synopsis of the physical geography of Kansas and Nebraska.
Field Data
Water quality data were collected throughout the study area by USEPA Region VII during the late spring and summer of 1994 and 1995 as part of its Regional Environmental Monitoring and Assessment Program (REMAP) (USEPA, 1993). Stream sites (271) were randomly selected in Kansas, Nebraska, and Missouri to assess fisheries' health and stream condition, and to establish baseline data and methods usable for assessing long-term trends throughout the region (USEPA, 1994). At each stream sampling site, data on stream physical, biological, and habitat condition were collected. Field sampling was conducted once per site between June and September of 1994 or 1995 when flows were close to seasonal norms, which is generally low and when pollution stress is potentially high and the fish community is the most stable and sedentary (USEPA, 1994). Data collection techniques and water chemistry analytical methods followed USEPA Region VII Standard Operating Procedures (USEPA, 1994).
Four water quality parameters that are important determinants of water quality were included in this study: conductivity, turbidity, nitrite–nitrate nitrogen (NO2–NO3), and total phosphorus (TP). In addition, an index of biotic integrity (IBI) and habitat index (HI) were analyzed (Karr et al., 1986; USEPA, 1994; Karr and Chu, 1998; Kaufmann et al., 1999). Appendix 1 lists metrics used to calculate these indices.
Landscape Data
For each stream sampling point, the watershed area was delineated and digitized (Fig. 1)
. The LULC data for the region were obtained from the United States Geological Survey (USGS) LULC Composite Theme Grid data set (United States Geological Survey, 1990), which was derived from aerial photography from the mid- and late-1970s with a spatial resolution of 200 m. The Anderson Level I classification was used; eight LULC categories occurred in the study area. The LULC proportions for the watersheds are shown in Fig. 2
. Areas within the watersheds were "clipped" from the land cover data and processed with FRAGSTATS 2.0 (McGarigal and Marks, 1995) to calculate 10 landscape pattern metrics (Table 1). Although there exists more recent land cover data at a finer resolution (30 m from 1992) (Vogelmann et al., 2001), it was not available at the start of this project. However, Herlihy et al. (1998) found that land cover–water chemistry regressions using 30-m data from 1992 produced no better results than did the USGS LULC data from the 1970s. Also, our preliminary work using a 30-m 1990 LULC dataset for Kansas (Whistler et al., 1995) produced mixed results. Therefore, although the USGS LULC dataset does not model the current landscape precisely, we believe it is adequate for our study. The LULC change that has occurred since the date of the USGS LULC data include an increase in grasslands due to the Conservation Reserve Program, an increase in center-pivot irrigation agriculture in western Kansas and Nebraska, and urban growth around the major cities.
|
|
|
In addition to the NDVI values, a series of derived metrics describing vegetation phenology were developed using algorithms modified from Reed et al. (1994). The VPMs used were maximum NDVI, date of onset of greenness, NDVI value at onset of greenness, number of growing days, growing season duration, and rate of green-up (growth rate). We used NDVI values from biweekly Periods 7 to 22 (late March through early November). Loveland et al. (1995) found that VPMs were helpful in classifying LULC for the conterminous USA. For each watershed, GIS overlay techniques were used to extract LULC proportions and to calculate mean NDVI values for each biweekly period and mean and standard deviations of VPM values. The U-index (human use index) (USEPA, 1994), which equals the proportion of agriculture plus urban land, was also calculated to gauge the level of total anthropogenic disturbance in regional landscapes.
The data were stratified by USEPA ecoregions, or in some cases, groupings of ecoregions (Omernik, 1987). While using the entire data set without stratification may be useful, based on statistical analysis, we found it necessary to stratify by ecoregion, because differences in general agricultural crops and land cover between ecoregions tended to cancel out any relationships. We also stratified by watershed area when adequate sample size existed. Details of the watershed selection process and variable transformations are explained in Griffith (2000).
Pearson product–moment correlation coefficients (Davis, 1986) were calculated to quantify relationships between stream condition variables and the LPMs, NDVI, and VPMs. To examine the effect of watershed size on the LPMs, partial correlation analyses that controlled for the effect of watershed size were also performed. Multiple regression was performed using the stream condition parameters as dependent variables, and LPMs and LULC as independent variables. Input variables for the regression models were determined by checking for instances where both an LULC and an LPM were significantly correlated with a stream condition parameter, and where both retained this significant correlation even after watershed size was factored out. Regression models were built only for ecoregions in which there were significant separate bivariate correlations between water quality parameters and the LPMs or LULC. The rationale behind this was because maps of LULC are needed to calculate LPMs. If LULC proportion adds additional information to variation explained along with LPMs, we wondered whether using them in conjunction might better explain water quality variation than simply using NDVI, which does not require LULC data. In most cases, when comparing LPMs alone versus NDVI, the NDVI metrics were more strongly correlated to the selected parameters than were the LPMs. Only the AVHRR data are needed to produce the NDVI–VPMs, and we wished to compare LPMs and LULC together as a unit with the NDVI metrics alone.
To assess the robustness of the multiple regression models, the condition index (CI), and the variance inflation factor (VIF) were used. Condition indices are the square roots of the ratios of the largest eigenvalue to each successive eigenvalue (Montgomery and Peck, 1992). The VIFs measure how much the variances of the estimated regression coefficients are inflated as compared to when the predictor variables are not linearly related (Neter et al., 1996). Condition indices in most cases were kept at 10 or below, and if the index was higher, both it and the VIF were used to decide which variables to eliminate from the model. Thus, because some condition indices were higher than 10, variable collinearity could not be fully eliminated. Finally, histograms of the regression standardized residuals or plots of the regression standardized vs. standardized predicted values were used to assess the regression models.
|
RESULTS AND DISCUSSION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
|
|
The following discussion focuses on several notable issues and findings with respect to the use of LPMs for regional watershed monitoring: (i) problems in calculating a full suite of landscape pattern metrics, (ii) the lack of many significant correlations between the LPMs and stream condition, (iii) the sensitivity of the LPMs to watershed size, (iv) counterintuitive results, and (v) inconsistent patterns of correlations. The regression results will then be discussed, as will the comparison of these regressions with the amount of variation in water quality explained using NDVI metrics only.
Although a wider range of LPMs was initially tested, only a subset was able to be used to analyze all or most of the watersheds. This situation occurred because the 200-m resolution of the LULC data and small size of some watersheds resulted in several watersheds having only a single patch, or only one or two patches each of certain cover types. These circumstances prevented the calculation of some landscape-level metrics, or class-level metrics that focused on one land cover class such as grassland or forest. Metrics that focus on grassland in this region may be more useful to understanding water quality processes, but in parts of the study area some watersheds had very little grassland or none at all.
Most of the time an LPM was not correlated with water quality or stream condition parameters (Table 3). With 17 ecoregions or stratifications within ecoregions (by watershed size) in the study area, and six stream condition parameters, a total of 102 possible opportunities for a significant correlation with any of the LPMs existed. Not all stratifications are shown on Tables 3 and 4a,b because, as stated earlier, they had few or no significant correlations. For example, the Central Irregular Plains was considered one stratification, as were three watershed size groups for this ecoregion. There were only 48 times out of a possible 102 when a stream condition parameter had a significant relationship with any LPM.
Besides having relatively few significant correlations to the selected stream parameters, LPMs are affected by watershed size. This problem affected many correlations and was especially severe for the patch shape metrics. This situation resulted in part because the size and shape of LULC patches are constrained to some extent by watershed size. In particularly small watersheds, patches have only a limited number of shape configurations. While some of these metrics can be standardized, preliminary analysis including principal components analysis showed little effect on the outcome.
Some of the correlations between LPMs and stream parameters in larger watersheds (>25 km2) of the Western Corn Belt Plains revealed counterintuitive relationships, especially to those persons not familiar with the study area. In this predominantly agricultural landscape, lower edge density, lower landscape diversity, and a landscape with land cover patches in larger aggregations (higher contagion) were all associated with increased habitat quality (Table 3). Figures 3 and 4 show graphs of these conditions. Considering that these same landscape patterns are also associated with greater amounts of agriculture in the watersheds (Table 5), these results seemed surprising, initially. One might expect that patches of forest or grassland in an agricultural matrix would add edge amount and increase landscape diversity with land covers that are intuitively associated with better stream conditions. Probable reasons for this arise from the importance of other factors besides landscape pattern or LULC that influence water quality. A closer examination of the five watersheds with the lowest HI scores and the five with the highest scores proved instructive. Two of the watersheds with the lowest scores straddled the loess-derived bluffs that flank the Missouri River floodplain. This environmental setting may be significant because of the enhanced erosion resulting from the higher slopes and erodible material found there. These bluffs are partially forested, which helps to explain the negative (albeit not statistically significant) relationship between percent forest and habitat index (Table 6). One of the subcomponents comprising the habitat index is substrate quality. Figure 4 shows that in the watersheds having the lowest HI scores, the substrate quality index is extremely low, supporting the occurrence of increased erosion from these bluffs. To test this hypothesis, we examined soils on eight watersheds having low HI scores that straddled the bluffs of the Loess Hills, and compared them with the five upland watersheds having the highest HI scores. We used digital STATSGO soils data (USDA, 1994) to estimate the percentage of watersheds covered by silt loam, which is highly erodible when wet due to lack of an adequate number of clay particles, which normally bond soils together. The mean percentage of silt loam in the watersheds on the bluffs was 59.5%, compared with 12.4% for the upland watersheds having high habitat index scores and substantially more silty clay loam, loam, silty clay, and clay soils. These results support the claim that erosion of fine silt sediments (especially given the increased slopes on the bluffs) might be contributing to the observed relationships in Fig. 4.
|
|
|
|
The relationship described above contrasts with that when using a VPM. The mean date of onset of greenness had a negative relationship with HI (Table 4b; r = -0.55). In other words, a later onset date (indicative of watersheds having a greater percentage of late-season crops such as corn or soybean) was associated with poorer habitat conditions, which might be expected in watersheds with more intensive agriculture. Although watersheds having higher HI scores had more agriculture than the watersheds in worse condition, the difference was small. Moreover, in the USGS LULC data, pasture land is classified as agriculture as opposed to grassland. With a dataset that did not have a more detailed classification, there was no means to assess how much of the agricultural area was row crop and how much was pasture. Thus, in contrast to LULC, the VPM is reflective of biophysical data and is not categorized, and it seemed to provide a more intuitive relationship in this example. Comparing Tables 3 and 4a,b, there were fewer moderate or strong correlations between LPMs and water quality than there were between NDVI and water quality. Finally, examining the correlations for all ecoregions (Table 3) does not show any consistent patterns across stream measures or across ecoregions with respect to which LPMs were correlated with water quality. We believe that this result derives from basic differences in landscape pattern and composition that occur across the different ecoregions.
Multiple Regression Models
Based on the condition requirements for creating regression models described at the end of the Methods section, there were a limited number of times a regression model was built because there were few times when both an LPM and an LULC proportion were significantly correlated to a stream parameter (Table 3). The condition indices (CIs) or variance inflation factors did not indicate severe violations of assumptions for the regressions, although in certain instances the CI was higher than ideal. Although for two of the models, scatterplots of standardized residuals and predicted values were not perfectly randomly distributed, histograms of the standardized residuals did not reveal severe departures from normality. Midsized watersheds in the Ozark Highlands (50–500 km2) provided one instance where a combination of LULC and landscape pattern performed better than the NDVI variables (Tables 4a,b, 7) at explaining variation in water quality. Using percent forest, IJI, and patch density in regression models produced coefficients of variation (adjusted R2) of 0.64 and 0.80 using NO2–NO3 and conductivity, respectively, as dependant variables. In a highly forested environment, one might expect higher patch density to be associated with poorer water quality conditions because the other LULC types in this area, urban or agriculture, are typically detrimental to water quality conditions. In a more simple landscape, mechanisms may be easier to surmise. Forest is associated with less erosion in these cases, and more patches would indicate that the forest cover had been fragmented, leading to situations where increased erosion could occur. Although there is some degree of correlation between the LPMs (especially the diversity indices) and the LULC (Table 8), the IJI appears to explain unique information, as it is not correlated with LULC.
|
|
Landscape Pattern and Processes Affecting Water Quality
An analysis of significant correlations between an LPM and a water quality parameter leads to the question of the mechanisms by which landscape pattern affects water quality. For example, in the Western Corn Belt Plains, a higher IJI value was associated with higher (better) IBI scores (r = 0.74; Table 3). This correlation coefficient was higher than that for any of the VPMs (Table 4b), but how interpretable is this? What is the process, if any, behind the connection of interspersion and juxtaposition of LULC patches to IBI? Complicating matters is a contrasting situation in the Central Irregular Plains, where higher IJI values represented the opposite condition, in other words, a more degraded condition (higher turbidity) (Table 3). Several criteria of good ecological indicators (Griffith, 1998) are not met by several landscape pattern metrics studied here including sensitivity to changes in environmental variables, reliability in response, and ease of understanding.
As Haines-Young (1999) states, landscape pattern has little intrinsic meaning or significance until it is placed in the context of problems or processes. Regarding water quality, the mechanisms connecting many landscape pattern metrics and stream conditions have yet to be established. For ecoregions that have more diverse LULC patterns or highly human-impacted landscapes, it may not be clearly associated with mechanisms that have deleterious effects on water quality. Contrasting this is the simpler landscapes, such as the forested Ozarks. In this case, increased interspersion and patch density can logically be associated with the fragmentation of the forest, which would potentially be associated with increased erosion or point sources of pollution. The NDVI (e.g., Period 15 NDVI with IBI in Table 4a,b, Western Corn Belt Plains) provides a more intuitive feel for what that metric represents. Higher NDVI values at this time of year (late July) probably represent greater amounts of corn-based agriculture, which would strongly influence water quality through runoff containing fertilizers and through increased vulnerability to erosion resulting in increased sedimentation of streams, which in turn would affect biotic communities.
Implications of Findings for Regional Watershed Monitoring
Jones et al. (1996) stated that "correlations between landscape pattern and certain levels of ecological process are generally lacking." This study provides further direction to LPM–water quality studies by providing an account of the relationships between LPMs and empirical stream data across a multistate region. Although LPMs have frequently been suggested as tools to study water quality, the few studies that examined them have had mixed results. Hunsaker et al. (1992) found that contagion explained 20% of conductivity levels in southern Illinois watersheds, but in a later study determined that land cover proportions explained more variance. Johnson et al. (1997) and Richards et al. (1996) found that patch density explained variation in water quality in Michigan in some seasons, but other landscape factors (geology, LULC, slope) generally were more important. Sharpe (1994) found no correlation between LPMs and water quality in a nutrient runoff model.
Despite the limitations of LPMs demonstrated here, a few significant relationships may be helpful in monitoring watershed conditions. The LPMs were more understandable in "simpler" landscapes or where a strong urban–rural gradient existed. Examples are the IJI in the Ozark Highlands, and diversity indices in the Sand Hills. In the Sand Hills, the mechanism behind the relationship with land cover diversity is easy to understand. In this contiguous grassland region, the presence of even small agricultural areas (especially on the fringes of the ecoregion) increases land cover diversity and thus probably negatively affects stream condition due to enhanced erosion potential and/or chemical applications. In the Mississippi River Lowlands, higher patch density was strongly correlated with higher habitat index scores (r = 0.92; Table 3) because several mid-sized watersheds in the St. Louis area had relatively large, but few, patches of urban land. Based on this research, it is recommended that, if analyzing relatively small watersheds (about <50 km2), LULC data resolution should be at least 30 m to allow a high probability that a full range of land cover types would occur, and so that class-level metrics can be used. This research also demonstrated the need to further refine the use of LPMs with respect to water quality applications. Basic differences in landscape structure probably caused different landscape metrics to be related to different parameters in different ecoregions. The same metric will probably not work for every ecoregion or for every water quality parameter. Due to this result, using a suite of metrics to evaluate conditions is appropriate (Qi and Wu, 1996; Jones et al., 1996, 2000). For most ecoregions, there were stronger correlations with NDVI or VPMs than there were with LPMs. When using LPMs, it may be useful to stratify watersheds into size classes so as to reduce the effect that size of the watershed or other unit has on patch shape variables (Turner et al., 1989; O'Neill et al., 1996). One must also be aware of site-specific factors when interpreting LPMs. As demonstrated, other factors influencing water quality besides LPMs can cause counterintuitive relationships. Implications of our results indicate that, for the purposes of watershed condition monitoring, simpler metrics such as patch density or diversity may be more useful than the more esoteric metrics such as fractal dimension or shape indices.
Study Limitations and Future Research
We have provided possible and reasonable explanations of the results from a landscape perspective, but the explanations are not meant to be exhaustive. Certainly, we acknowledge that some instream processes not discussed here could be affecting the results and would be interpreted differently by aquatic biologists or water chemists. Moreover, there are other important factors that determine water quality besides landscape pattern or vegetation condition as represented by NDVI. This fact is reflected in some cases by the relatively low or moderate r and R2 values. Using the random sampling framework by which stream sites in this study were located, there was no control on geology, soils, slope, ground water hydrology, or point sources of pollution. Additionally, hydrometeorological conditions may not have been ideal at the time of summer sampling. During low flow periods, ground water typically supplies most of the flow. In these situations, water quality may be less affected by landscape surface features (Wang, 1997). Taylor et al. (1996) and Frenzel and Swanson (1996) have also stressed the importance of hydrologic events to stream biotic assemblages in the Central Plains. They found that discharge-related disturbances and other changes in environmental parameters were associated with varying fish assemblages (Taylor et al., 1996). Because the time from a rainfall or flow disturbance event was not necessarily controlled for during sampling, results pertaining to the IBI could have been affected. There can also be bio–physico–chemical differences between headwater and downstream sites (Harding et al., 1999). Other factors potentially affecting results include the use of 1995 AVHRR data, which in some cases did not match the 1994 sampling of streams. Additionally, for many small watersheds, besides containing only a few pixels (especially the NDVI data), a small amount of positional error in drawing or digitizing watersheds may have resulted in a large variation in the land cover proportions and landscape pattern.
For future research, it would be helpful to isolate specific watersheds, particularly in the dynamic urban–rural fringe of metropolitan areas, and examine how LPMs change over time in correspondence with changes in stream conditions. Manipulation of experimental watersheds to understand the effects of landscape pattern might also shed insight into the mechanisms by which landscape pattern affects stream conditions. Other potentially useful research might involve pattern analysis of NDVI as opposed to LULC (Keane et al., 1999). Finally, we attempted to simplify the analysis in this study by using many individual correlations and regression analyses. Some multivariate statistical procedures may have also shed insight into some of the problems, such as discriminant analysis or regression tree analysis, but practical constraints precluded their use here.
|
CONCLUSIONS |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
The fact that in several cases a VPM explained the largest proportion of variance in IBI scores was surprising since there is no direct connection between vegetation indices and fish communities. Indirect correlations are still helpful, however, because NDVI and VPMs reflect conditions to which factors that negatively affect stream water quality are associated (Frenzel and Swanson, 1996). The VPMs are in some cases very strongly correlated with land cover, but apparently reveal additional information as well, such as crop type and vegetation condition. There also may be a connection between NDVI and agricultural intensity, which Harding et al. (1999) say has been ignored, but could be an indicator of agricultural effects in streams. The LPMs have been very useful in terrestrial applications and have been shown to have some bearing on water quality. However, compared with 200-m LULC data and LPMs derived from them, NDVI-derived metrics showed more promise in monitoring stream conditions in the U.S. Central Plains. Smith et al. (2000) discussed the development of environmental indicators to estimate environmental trends, conditions, and the sustainability of agroecosystems. Findings presented here are important in the quest to identify broad-scale indicators of watershed condition for use in monitoring and assessment programs.
Component Indices or Variables for the Index of Biotic Integrity and Habitat Index
Index of Biotic Integrity
Total number of fish species
Number and identity of darter species
Number and identity of sunfish species
Number and identity of sucker species
Number and identity of intolerant species
Proportion of individuals as green sunfish, carp, bullheads, goldfish
Proportion of individuals as omnivores
Proportion of individuals as insectivorous cyprinids
Proportion of individuals as piscivores (top carnivores)
Number of individuals in sample
Proportion of individuals with anomalies
Habitat Index, Comprised of Eight Subindices
Riparian vegetation quality
Lack of riparian human disturbance
Substrate quality
In-channel disturbance and deviance from expected chan- nel morphology and substrate
Habitat volume
Spatial complexity
Instream fish cover
Stream power and velocity
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Current address: U.S. Geological Survey, EROS Data Center, Sioux Falls, SD 57198. 
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONLANDSCAPE-WATER QUALITY...OBJECTIVESMETHODSRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
= 0.05) between selected landscape pattern metrics and stream variables. These metrics were significant even after controlling for watershed size using partial correlation analysis. Additional stratifications of the data included nine extra categories in the Western Corn Belt Plains, Central Irregular Plains, Ozark Highlands, and Central Great Plains. These were not shown because there were either few significant correlations or very low correlations. The number of samples sites n is denoted as (n) for the interspersion and juxtaposition index (IJI), which could not be calculated for some watersheds due a lack of adequate patch numbers. Also, n varies among parameters, because in some cases a specific parameter measurement was not collected.
