Relationships between land use/cover (LUC) and stream water quality have been well-documented in many environments and at a range of spatial scales. From these analyses, reduced in-stream biological integrity and habitat quality are commonly associated with increasing amounts of anthropogenic LUC. However, very few studies have examined the influence of landscape condition, relative to studies using LUC, on water quality parameters. Landscape condition indices use remote sensing-based data to quantify biophysical land surface condition.

The primary objective of this study was to assess the relationships between LUC class proportions and indices of land surface condition (LSC) and macroinvertebrate-based water quality metrics. These relationships were examined at three spatial scales (reach, stream network, and catchment). Strong correlations were observed between both LUC class proportions and LSC indices with macroinvertebrate-based metrics, although there was almost twice the number of significant correlations for LUC as compared with LSC. Similar to previous research, relationships between landscape variables and macroinvertebrate metrics were not consistent across spatial scales. Overall, results suggest that LUC class proportions are better indicators of water quality conditions in the study area. Future work will expand this analysis and include additional water quality parameters and landscape variables with the goals of deepening the understanding of landscape impacts on stream water quality and providing resource managers with valuable information that will to help guide planning and assessment activities.

Introduction

Relationships between land use/cover (LUC) and stream water quality have been documented at a range of spatial scales (Sponseller et al., 2001; Potter et al., 2005) using a number of water quality indicators including nutrient and sediment loads (Busse et al., 2006), geomorphic factors (Roy et al., 2003; Pan et al., 2004), physical stream characteristics (Bolstad and Swank, 1997), fish (Roth et al., 1996) and macroinvertebrate assemblages (Alberti et al., 2007). These studies commonly report lower in-stream biological integrity and habitat quality with higher amounts of anthropogenic LUC. Agriculture and urban land use have been identified as major sources of impairment to stream ecosystems in the USA (USEPA, 2002). In addition to the relative area occupied by each LUC class, their spatial arrangement on the landscape (Goetz and Fiske, 2008) and LUC change (Harding et al., 1998) also affect stream ecosystem structure and function. Historically, there has been a near-exclusive focus on the relationship between LUC class proportions and stream ecosystem attributes. However, Griffith et al. (2002) found that the satellite image-derived Normalized Difference Vegetation Index (NDVI), a red and near infrared-based spectral index widely used in environmental remote sensing research for decades, was more correlated with stream ecosystem attributes than LUC class proportions because the NDVI integrates biophysical land surface conditions (LSC) which affect water quality (e.g. vegetation abundance and canopy moisture), while also capturing the within-class variability of individual LUC classes. Also, the satellite image-derived Kauth-Thomas (KT) Brightness, Greenness, and Wetness indices (Crist et al., 1986) have been successfully related to vegetation attributes indicative of LSC (Collins and Woodcock, 1996; Rogan et al., 2002; Healey et al., 2005). These indices are typically interpreted as characterizing landscape albedo, vegetation condition/vigor, and canopy and soil moisture, respectively.

The objective of this study was to assess the relationships between LUC class proportions, LSC indices (i.e. the NDVI and the KT Brightness, Greenness, and Wetness indices) and stream macroinvertebrate communities. Macroinvertebrates are commonly used in stream assessments given their documented sensitivity to a wide range of environmental variables, high species richness, and ubiquitous, yet sedentary, nature (Ormerond and Edwards, 1987; Hawkins et al., 1997). Based on Griffith et al. (2002), we hypothesized that stream macroinvertebrate metrics would be more correlated to LSC indicators than to LUC classes. These landscape-water quality relationships were examined at three spatial scales (reach, stream and catchment) as there is no clear consensus in the literature that any one particular scale is most important or useful in all environments and/or with some or all water quality parameters. While most of the previous landscape-water quality research has been conducted in areas dominated by urban development and agriculture, this study was carried out in a rugged area of Central Appalachia where anthropogenic LUC is largely confined to narrow stream valleys in primarily forested catchments.

Methods

The forest-dominated study area is located in the mixed mesophytic region of the Eastern Deciduous forest in Central Appalachia, and is characterized by a variety of winter-deciduous and evergreen tree species, mixed with smaller areas of hay pasture, agricultural crops and urban and residential development. Similar to previous research (Meyer et al., 2005; Tran et al., 2010), we quantified macroinvertebrate community structure and landscape-scale variables at 28 sites ranging in catchment size from 2 to 244 km2. While macroinvertebrate community structure can vary throughout the year (Johnson et al., 2012a), we sampled macroinvertebrates during a 2-week period in late June-early July (2006), which coincides with the sampling period established by the Kentucky Division of Water for these streams (KDOW, 2008), to minimize temporal variation in macroinvertebrate communities.

Macroinvertebrates were sampled from riffle habitats (1 m2 sampling area) using a kicknet (500 μm mesh size) during base flow and preserved in the field with 70% ethanol. In the lab, macroinvertebrates were sorted and identified to the lowest determinable taxonomic level (usually genus) by sequentially processing subsamples (1/16 of the entire sample) until at least 300 individuals had been identified. When 300 organisms had been reached, the rest of the subsample was processed so that macroinvertebrate taxon densities could still be estimated. After subsampling, the remainder of the sample was qualitatively processed to identify taxa not represented in sub-sampling. Eight macroinvertebrate metrics were calculated from these data to describe the species assemblage at each of the 28 sites based on Kentucky Division of Water standard protocol for assessing streams in this study area. Greater values of taxa richness, Ephemeroptera-Plecoptera-Trichoptera (EPT) richness, % EPT, % Ephemeroptera, and % Clingers indicate higher biotic integrity, while greater values of Hilsenhoff biotic index (HBI) and % Chironomidae and Oligochaete (% CO) indicate lower biotic integrity (KDOW, 2008). Only macroinvertebrates quantified in the sub-sampling process were used to estimate % EPT, % Ephemeroptera, % Clingers, HBI and % CO. However, the sum of taxa found in the subsampling procedure and the qualitative taxa search for each sample was used to estimate taxa richness and EPT richness.

Satellite image-based indicators of LUC and LSC were derived from 30 m Landsat satellite imagery for each sample site at three spatial scales including the entire catchment area above each sample site (catchment), a 100 m buffer around the entire stream system upstream of each site (stream), and a 100 m buffer around a 1000 m reach upstream of each site (reach). LUC percentages for each sampling unit (i.e. reach, stream and catchment) were computed from the Landsat 5 TM-based 2001 National Land Cover Database for Kentucky using ESRI's ArcGIS 9.0®. After Johnson et al. (2012b) and Tran et al. (2010), the individual developed (low-, medium-, and high-intensity developed and open space developed) and forest (deciduous, evergreen, and mixed) LUC classes were aggregated into two broad classes: developed and forest. However, the pasture and cropland classes were not combined into a single LUC class since it was expected that these areas would differentially impact water quality. Other non-anthropogenic LUC classes (i.e. wetlands, open water, etc.) were also found in these catchments, but, when present, represented <10% of total catchment area and were not correlated with stream macroinvertebrate community metrics.

The four LSC indices were derived from a 10 September 2005, cloud free terrain-corrected Landsat 7 (SLC-off) ETM+ image. Specifically, the VEGINDEX and TASSCAP models in IDRISI Andes® were used to calculate the NDVI and the KT Brightness, Greenness, and Wetness indices, respectively. The NDVI can range from −1.0 to 1.0, however, values typically fall between −0.1 (barren ground) and 0.6 (abundant green vegetation). While the absolute values of individual KT indices are meaningless, they can be used to assess relative differences in brightness, greenness, and wetness.

A fall 2005 satellite image was selected in order to characterize the potential leaf litter that was expected to contribute to the following growing season's macroinvertebrate assemblages, which were sampled in the summer of 2006. The ETM+ image was chosen due to the lack of a cloud-free TM image. As the study area was located in the center of the ETM+ image, the image striping caused by the SLC-off mode was of minimal concern and missing pixels were converted to null values. The zonal statistics tool in ArcGIS was used to calculate LUC class proportions and mean NDVI and KT Brightness, Greenness, and Wetness index values for each spatial scale at each sample site. Since the distributions of many of the landscape indicators and macroinvertebrate metrics were not normally distributed, relationships between the macroinvertebrate metrics and landscape indicators (both LUC and LSC) were evaluated for statistical significance using nonparametric Spearman's rank correlation analysis (Tran et al., 2010). This heuristic assessment (i.e. correlation analysis) can supply valuable insights into the relationships between landscape variables and water quality parameters (Griffith et al., 2002).

Results

Ninety-five taxa were identified from 112,953 individuals collected representing 19 orders and 51 families. Despite the high taxa richness, nine taxa represented over 90% of individuals indicating a high degree of dominance. Mean macroinvertebrate densities were 3,982 individuals m−2. Chironomidae was the most abundant taxon collected (mean density = 1,938 individuals m−2), and represented 48.9% of individuals collected in this study. Other dominant taxa in this study included Cheumatopsyche sp. (mean density = 444 individuals m−2), Baetis sp. (mean density = 211 individuals m−2), Stenelmis sp. (mean density = 295 individuals m−2), and Stenonema sp. (mean density = 199 individuals m−2).

Table 1 provides descriptive statistics for LUC, LSC, and the macroinvertebrate variables. Forest was the dominant LUC class at all three spatial scales, followed by hay pasture, developed land, and cultivated cropland. The variability in LUC class proportions (see SD; Table 1) was greatest at the reach scale for all four LUC classes, and was particularly high for the % Developed class. Values of mean ± SD NDVI for each spatial scale were similar (Table 1). In contrast, values of mean ± SD KT Brightness, Greenness, and Wetness were generally quite different among spatial scales (Table 1). Overall, the reach scale was characterized by higher mean Brightness values and lower mean Greenness and mean Wetness values than the stream and catchment scales. Mean Greenness and Wetness were greater at the catchment scale than at the stream scale, while mean Brightness was higher at the stream scale than the catchment scale. Variation was similar among spatial scales for the Greenness and Wetness indices, while they increased somewhat with spatial scale for the Brightness index (see SD; Table 1).

Table 1.

Descriptive statistics for LUC, LSC and macroinvertebrate variables.

ScaleaMin.25th%-ileMedian75th%-ileMax.MeanSD
LUC         
 % Developed 0.00 0.53 2.10 8.13 53.24 6.38 11.15 
 0.00 0.48 0.98 2.68 5.99 1.72 1.70 
 0.00 0.00 0.00 0.04 3.82 0.18 0.73 
 % Forest 1.52 23.13 41.96 51.93 100.00 42.20 25.78 
 6.43 65.93 77.14 82.43 94.40 70.53 21.21 
 5.58 74.52 82.32 86.10 96.83 74.72 21.77 
 % Pasture 0.00 14.92 39.45 56.15 84.73 37.73 25.03 
 0.00 7.00 10.06 19.60 86.77 17.70 19.52 
 0.00 5.07 8.27 14.91 84.16 14.72 18.66 
 % Cropland 0.00 0.00 0.00 0.00 35.46 1.85 7.00 
 0.00 0.00 0.02 0.56 15.57 0.88 2.95 
 0.00 0.00 0.04 0.51 22.52 1.09 4.24 
LSC         
 NDVI 0.44 0.47 0.53 0.56 0.64 0.52 0.05 
 0.43 0.48 0.52 0.54 0.64 0.52 0.05 
 0.36 0.47 0.52 0.57 0.65 0.52 0.07 
 Brightness 360.94 412.07 425.91 459.51 494.74 430.55 33.73 
 328.05 348.06 378.39 421.34 449.84 383.51 40.02 
 230.51 334.83 374.57 423.63 464.58 372.93 54.35 
 Greenness 32.13 48.82 57.04 68.21 91.02 58.04 15.60 
 49.91 70.19 82.39 91.07 104.86 80.58 13.97 
 4.01 78.34 89.06 97.94 123.18 85.91 21.70 
 Wetness −182.54 −148.89 −129.29 −108.25 −59.27 −127.38 30.44 
 −242.15 −178.12 −155.40 −127.56 −73.51 −152.25 35.81 
 −263.72 −190.64 −170.52 −142.39 −71.53 −167.53 39.61 
Macroinvertebrate         
 Taxa Richness N/A 20.00 22.50 26.00 29.00 42.00 26.07 4.95 
 EPT Richness N/A 6.00 7.00 11.00 12.00 14.00 9.71 2.64 
 mHBI N/A 4.41 4.85 5.23 5.76 6.43 5.29 0.58 
 % EPT N/A 9.43 24.67 40.24 55.90 78.57 40.97 19.46 
 % Ephem N/A 1.51 6.59 16.86 25.38 59.51 17.86 13.03 
 % CO N/A 7.81 38.71 49.43 60.71 81.29 48.62 16.76 
 % Clinger N/A 13.03 22.42 30.76 41.06 87.15 33.45 15.93 
ScaleaMin.25th%-ileMedian75th%-ileMax.MeanSD
LUC         
 % Developed 0.00 0.53 2.10 8.13 53.24 6.38 11.15 
 0.00 0.48 0.98 2.68 5.99 1.72 1.70 
 0.00 0.00 0.00 0.04 3.82 0.18 0.73 
 % Forest 1.52 23.13 41.96 51.93 100.00 42.20 25.78 
 6.43 65.93 77.14 82.43 94.40 70.53 21.21 
 5.58 74.52 82.32 86.10 96.83 74.72 21.77 
 % Pasture 0.00 14.92 39.45 56.15 84.73 37.73 25.03 
 0.00 7.00 10.06 19.60 86.77 17.70 19.52 
 0.00 5.07 8.27 14.91 84.16 14.72 18.66 
 % Cropland 0.00 0.00 0.00 0.00 35.46 1.85 7.00 
 0.00 0.00 0.02 0.56 15.57 0.88 2.95 
 0.00 0.00 0.04 0.51 22.52 1.09 4.24 
LSC         
 NDVI 0.44 0.47 0.53 0.56 0.64 0.52 0.05 
 0.43 0.48 0.52 0.54 0.64 0.52 0.05 
 0.36 0.47 0.52 0.57 0.65 0.52 0.07 
 Brightness 360.94 412.07 425.91 459.51 494.74 430.55 33.73 
 328.05 348.06 378.39 421.34 449.84 383.51 40.02 
 230.51 334.83 374.57 423.63 464.58 372.93 54.35 
 Greenness 32.13 48.82 57.04 68.21 91.02 58.04 15.60 
 49.91 70.19 82.39 91.07 104.86 80.58 13.97 
 4.01 78.34 89.06 97.94 123.18 85.91 21.70 
 Wetness −182.54 −148.89 −129.29 −108.25 −59.27 −127.38 30.44 
 −242.15 −178.12 −155.40 −127.56 −73.51 −152.25 35.81 
 −263.72 −190.64 −170.52 −142.39 −71.53 −167.53 39.61 
Macroinvertebrate         
 Taxa Richness N/A 20.00 22.50 26.00 29.00 42.00 26.07 4.95 
 EPT Richness N/A 6.00 7.00 11.00 12.00 14.00 9.71 2.64 
 mHBI N/A 4.41 4.85 5.23 5.76 6.43 5.29 0.58 
 % EPT N/A 9.43 24.67 40.24 55.90 78.57 40.97 19.46 
 % Ephem N/A 1.51 6.59 16.86 25.38 59.51 17.86 13.03 
 % CO N/A 7.81 38.71 49.43 60.71 81.29 48.62 16.76 
 % Clinger N/A 13.03 22.42 30.76 41.06 87.15 33.45 15.93 
a

R = Reach; S = stream; C = catchment.

The NDVI was most significantly correlated with % Forest (all spatial scales) and % Pasture (stream and catchment scales), while Greenness was not significantly correlated with LUC classes at any spatial scale (Table 2). Significant correlations were observed between Wetness and all LUC classes at most scales, with the strongest relationships found at the stream and catchment scales (Table 2). Reach-scale Brightness was most significantly correlated with reach- and stream-scale % Forest and % Pasture (Table 2).

Table 2.

Spearman's rank correlation coefficients between LUC and LSC variables.

% Developed% Forest% Pasture% Crop
ScaleaRSCRSCRSCRSC
NDVI −0.50c −0.24 −0.40b 0.51c 0.52c 0.48c −0.33 −0.43b −0.45b −0.25 −0.41b −0.40b 
 −0.18 −0.24 −0.44b 0.49c 0.59c 0.72c −0.41b −0.54c −0.62c −0.15 −0.35 −0.35 
 −0.34 −0.16 −0.36 0.40b 0.54c 0.51c −0.26 −0.50c −0.44b 0.03 −0.27 −0.32 
Brightness 0.41b 0.39b 0.28 −0.85c −0.71c −0.45b 0.69c 0.61c 0.45b 0.32 0.41b 0.30 
 0.03 0.10 0.27 0.02 −0.24 −0.25 −0.36 0.16 0.18 0.01 −0.09 −0.14 
 0.22 0.28 0.36 −0.07 −0.31 −0.33 −0.31 0.16 0.11 0.05 −0.11 −0.07 
Greenness −0.06 0.16 −0.05 −0.05 0.21 0.35 0.00 −0.23 −0.37 −0.05 −0.02 −0.09 
 0.08 0.20 0.13 −0.06 0.11 0.15 −0.04 0.00 −0.06 0.17 0.20 0.09 
 −0.18 −0.01 −0.04 0.05 0.14 0.11 0.18 0.08 0.10 0.20 0.16 0.05 
Wetness −0.36 −0.24 −0.38b 0.75c 0.62c 0.52c −0.67c −0.53c −0.52c −0.36 −0.53c −0.46b 
 −0.39b −0.47b −0.58c 0.46b 0.54c 0.62c −0.25 −0.63c −0.65c −0.50c −0.59c −0.49c 
 −0.31 −0.45b −0.63c 0.42b 0.60c 0.75c −0.22 −0.68c −0.69c −0.43b −0.52c −0.54c 
% Developed% Forest% Pasture% Crop
ScaleaRSCRSCRSCRSC
NDVI −0.50c −0.24 −0.40b 0.51c 0.52c 0.48c −0.33 −0.43b −0.45b −0.25 −0.41b −0.40b 
 −0.18 −0.24 −0.44b 0.49c 0.59c 0.72c −0.41b −0.54c −0.62c −0.15 −0.35 −0.35 
 −0.34 −0.16 −0.36 0.40b 0.54c 0.51c −0.26 −0.50c −0.44b 0.03 −0.27 −0.32 
Brightness 0.41b 0.39b 0.28 −0.85c −0.71c −0.45b 0.69c 0.61c 0.45b 0.32 0.41b 0.30 
 0.03 0.10 0.27 0.02 −0.24 −0.25 −0.36 0.16 0.18 0.01 −0.09 −0.14 
 0.22 0.28 0.36 −0.07 −0.31 −0.33 −0.31 0.16 0.11 0.05 −0.11 −0.07 
Greenness −0.06 0.16 −0.05 −0.05 0.21 0.35 0.00 −0.23 −0.37 −0.05 −0.02 −0.09 
 0.08 0.20 0.13 −0.06 0.11 0.15 −0.04 0.00 −0.06 0.17 0.20 0.09 
 −0.18 −0.01 −0.04 0.05 0.14 0.11 0.18 0.08 0.10 0.20 0.16 0.05 
Wetness −0.36 −0.24 −0.38b 0.75c 0.62c 0.52c −0.67c −0.53c −0.52c −0.36 −0.53c −0.46b 
 −0.39b −0.47b −0.58c 0.46b 0.54c 0.62c −0.25 −0.63c −0.65c −0.50c −0.59c −0.49c 
 −0.31 −0.45b −0.63c 0.42b 0.60c 0.75c −0.22 −0.68c −0.69c −0.43b −0.52c −0.54c 
a

R = Reach, S = stream; C = catchment; b0.05 > p ≳ 0.0; cp < 0.01.

Of the eight macroinvertebrate metrics examined, only three (Taxa Richness, EPT Richness and the HBI) exhibited significant correlations with LUC class proportions (Table 3). EPT Richness was the sole macroinvertebrate metric correlated (negatively) with % Developed. % Forest was positively correlated with Taxa Richness at the reach and stream scales, with EPT Richness at the stream and catchment scales, and negatively with the HBI at all three spatial scales. Negative correlations were observed between % Pasture and Taxa Richness at the stream scale and EPT Richness at the stream and catchment scales; positive correlations between % Pasture and the HBI were found at the stream and catchment scales. Finally, % Cropland was positively correlated with the HBI at the reach scale and negatively correlated with Taxa Richness (stream and catchment scales) and EPT Richness at all three spatial scales.

Table 3.

Spearman's rank correlation coefficients for LUC and LSC with macroinvertebrate metrics.

TaxaEPT
ScaleaRichnessRichnessHBI% EPT% Ephem% CO% Clinger
Catchment Area (km2N/A −0.29 −0.15 0.01 0.21 −0.16 −0.01 0.41b 
 % Developed −0.14 0.03 0.16 0.31 0.23 −0.21 0.14 
 −0.22 −0.21 0.20 0.19 0.09 −0.26 0.34 
 −0.37 −0.43b 0.24 0.13 0.02 −0.36 0.14 
 % Forest 0.46b 0.31 −0.41b −0.17 −0.13 0.05 0.15 
 0.42b 0.53b −0.55c −0.07 −0.12 0.24 0.00 
 0.36 0.47b −0.46b 0.04 −0.02 0.19 0.04 
 % Pasture −0.27 −0.22 0.22 −0.06 −0.03 0.01 −0.04 
 −0.47b −0.53c 0.46b 0.03 0.03 −0.08 0.02 
 −0.26 −0.43b 0.50c −0.05 −0.01 −0.16 0.08 
 % Crops −0.24 −0.47b 0.40b −0.22 −0.21 0.08 0.00 
 −0.45b −0.47b 0.33 −0.07 −0.13 −0.13 0.04 
 −0.60c −0.53c 0.22 −0.14 −0.18 −0.02 −0.10 
NDVI 0.32 0.24 −0.46b −0.15 −0.08 0.13 −0.04 
 0.16 0.33 −0.42b −0.03 −0.08 0.36 −0.12 
 0.31 0.16 −0.41b −0.24 −0.37 0.22 0.03 
Brightness −0.32 −0.31 0.60c 0.24 0.23 −0.07 −0.03 
 −0.12 −0.09 0.35 0.46b 0.39b −0.17 0.08 
 −0.23 −0.20 0.28 0.44b 0.27 −0.15 0.07 
Greenness 0.14 0.14 −0.30 −0.01 −0.18 0.10 0.06 
 −0.12 −0.10 −0.16 −0.05 −0.38b 0.34 0.12 
 −0.01 −0.05 −0.28 −0.13 −0.39b 0.32 0.18 
Wetness 0.25 0.32 −0.52 −0.11 −0.05 0.18 −0.05 
 0.36 0.35 −0.34 −0.21 0.04 0.07 −0.25 
 0.40b 0.35 −0.25 −0.21 0.05 0.07 −0.20 
TaxaEPT
ScaleaRichnessRichnessHBI% EPT% Ephem% CO% Clinger
Catchment Area (km2N/A −0.29 −0.15 0.01 0.21 −0.16 −0.01 0.41b 
 % Developed −0.14 0.03 0.16 0.31 0.23 −0.21 0.14 
 −0.22 −0.21 0.20 0.19 0.09 −0.26 0.34 
 −0.37 −0.43b 0.24 0.13 0.02 −0.36 0.14 
 % Forest 0.46b 0.31 −0.41b −0.17 −0.13 0.05 0.15 
 0.42b 0.53b −0.55c −0.07 −0.12 0.24 0.00 
 0.36 0.47b −0.46b 0.04 −0.02 0.19 0.04 
 % Pasture −0.27 −0.22 0.22 −0.06 −0.03 0.01 −0.04 
 −0.47b −0.53c 0.46b 0.03 0.03 −0.08 0.02 
 −0.26 −0.43b 0.50c −0.05 −0.01 −0.16 0.08 
 % Crops −0.24 −0.47b 0.40b −0.22 −0.21 0.08 0.00 
 −0.45b −0.47b 0.33 −0.07 −0.13 −0.13 0.04 
 −0.60c −0.53c 0.22 −0.14 −0.18 −0.02 −0.10 
NDVI 0.32 0.24 −0.46b −0.15 −0.08 0.13 −0.04 
 0.16 0.33 −0.42b −0.03 −0.08 0.36 −0.12 
 0.31 0.16 −0.41b −0.24 −0.37 0.22 0.03 
Brightness −0.32 −0.31 0.60c 0.24 0.23 −0.07 −0.03 
 −0.12 −0.09 0.35 0.46b 0.39b −0.17 0.08 
 −0.23 −0.20 0.28 0.44b 0.27 −0.15 0.07 
Greenness 0.14 0.14 −0.30 −0.01 −0.18 0.10 0.06 
 −0.12 −0.10 −0.16 −0.05 −0.38b 0.34 0.12 
 −0.01 −0.05 −0.28 −0.13 −0.39b 0.32 0.18 
Wetness 0.25 0.32 −0.52 −0.11 −0.05 0.18 −0.05 
 0.36 0.35 −0.34 −0.21 0.04 0.07 −0.25 
 0.40b 0.35 −0.25 −0.21 0.05 0.07 −0.20 

aR = Reach; S = stream; C = catchment; b0.05 > p ≳ 0.0; c p < 0.01.

Of the eight macroinvertebrate metrics examined, only four (Taxa Richness, HBI, % EPT, and % Ephemeroptera) exhibited significant correlations with LSC indices (Table 3). The NDVI was negatively correlated with the HBI metric at all spatial scales. The Brightness index was positively correlated with HBI at the reach scale, with % EPT at the stream and catchment scales, and with % Ephemeroptera at the stream scale. The Greenness index was negatively correlated with % Ephemeroptera at the stream and catchment scales. The Wetness index was positively correlated with Taxa Richness at the catchment scale and negatively correlated with the HBI at the reach scale. Except for % Clinger, the macroinvertebrate variables used in this analysis were not statistically related to differences in catchment area (Table 3).

Discussion

Despite multicollinearity between some LUC and LSC variable pairs at some spatial scales, there was a sufficient lack of collinearity between other variable pairs at other scales to warrant a correlation analysis between all landscape and macroinvertebrate variables. Forest was the dominant LUC at all three spatial scales of analysis, however the relative prevalence of forest varied with spatial scale, with the reach scale comprising only about half of the % Forest as the other two spatial scales (which had approximately equal proportions). While pasture was the next most dominant LUC class at each scale, % Pasture at the reach scale was over twice that found at the stream and catchment scales. These patterns reflect both historic and current farming practices in the area, where hay pastures and cropland are commonly observed on the flat areas and gentler slopes of the narrow stream valleys. The amount of Developed LUC was three times larger at the reach scale than at the stream or catchment scales as building sites are typically confined to cleared, flat areas in the valleys.

While there was no difference in mean NDVI values between spatial scales, the values themselves were moderately high indicating the presence of abundant vegetation. It is important to remember that the NDVI index (like the KT indices) reflects, or integrates, the mixture of individual LUC types present on the landscape. Thus, it is likely that different combinations, relative proportions, and biophysical conditions of LUC classes may have resulted in similar NDVIs. In contrast, there was clear differentiation in mean KT Brightness, Greenness, and Wetness values at the reach, stream and catchment scales. As expected, Brightness decreased from reach to stream to catchment scales, while Greenness and Wetness followed the opposite trend. These patterns clearly reflect the observed dominance of forest LUC at the stream and catchment scales and developed and pasture LUC at the reach scale. The apparently higher sensitivity to variation in landscape condition of the KT indices (relative to the NDVI) may be a function of the fact that their calculation involves all six optical Landsat spectral bands, whereas the NDVI utilizes only two of these bands (red and near-infrared).

In this study, Taxa Richness and EPT Richness were correlated with increasing % Forest at multiple spatial scales. This finding supports previous studies which demonstrated that water quality indicators frequently improve as the amount of forested land increases (Roy et al., 2003; Alberti et al., 2007), as greater forest cover acts to reduce sedimentation and slow overland flow, as well as serving as a physical and biological filter for many pollutants. Degraded water quality is often associated with agricultural LUC (Wang et al., 1997; Pan et al., 2004), which contributes to increased sedimentation, nutrient enrichment, and pesticide use. In this study, increasing % Pasture and % Cropland were both significantly correlated with HBI (positively) and Taxa Richness and EPT Richness (negatively). The HBI index is particularly sensitive to organic pollution, which may increase as forests are cleared for agricultural production.

With one exception, % Developed was not statistically linked to macroinvertebrate variables at any spatial scale. This is a surprising finding as others (Wang et al., 1997; Roy et al., 2003; Meyer et al., 2005; Busse et al., 2006; Alberti et al., 2007; Tran et al., 2010) have found strong associations between developed LUC and water quality, given the many negative impacts of impervious surfaces (e.g. reduced infiltration and warmer surface runoff), even at very low percent cover (e.g. 0.5%; King et al., 2011). While the lack of correlation in this study could be related to the relatively low amount of developed land found in the area, it may be more a function of the density, type, and/or arrangement of this LUC class on the landscape compared with more urbanized areas.

Griffith et al. (2002) found that the NDVI was much more strongly and frequently correlated with water quality parameters than LUC class proportions. They concluded that this was likely because LSC indices better represent the biophysical conditions of the vegetation and soil cover that are actually affecting water quality conditions. Results from this study, however, do not support their findings. In fact, there were twice the number of significant correlations between LUC class proportions and macroinvertebrate metrics than between LSC indices and macroinvertebrate metrics. While LUC and LSC were each significantly correlated with four of the eight macroinvertebrate metrics, only two of these metrics were significantly correlated with both LUC and LSC (Taxa Richness and HBI). Again, despite some multicollinearity between LUC and LSC variables, each appears to provide some unique information with respect to water quality conditions.

A number of factors may have contributed to the contrasting findings between this study and Griffith et al. (2002). Possibly the biggest difference was related to the areas studied, since Griffith et al. (2002) worked at the catchment scale in a largely flat, agriculturally-dominated four state area, while this research was carried out at three spatial scales in a much smaller, more rugged forested region. As a result, LUC and LSC gradients were most likely much larger (i.e. ranges of LUC and LSC variables were greater) in Griffith et al. (2002). Furthermore, different response variables were analyzed in each study. Griffith et al. (2002) utilized a fish-based Index of Biotic Integrity, a habitat index, and water chemistry parameters, while this study calculated macroinvertebrate-based metrics; these variables may respond differently to landscape perturbations. Finally, Griffith et al. (2002) analyzed correlations between water quality variables and NDVI values obtained for multiple dates during the year, which allowed them to investigate how the NDVI related to water quality at several time points. In this study, the NDVI and KT indices were calculated for a single date, but multi-date indices may be more effective assessment tools since stream parameters could be more correlated with particular land surface indicators in different seasons.

Many authors have noted the importance of studying landscape-water quality relationships at multiple spatial scales (Roy et al., 2003; Tran et al., 2010), as was done in this study. However, there is little consensus in the literature regarding the relative influence of spatial scale on these relationships. For example, Tran et al. (2010) found stronger relationships between LUC and water quality parameters at finer scales, whereas other studies have demonstrated stronger associations at broader spatial scales (Roth et al., 1996; Wang et al., 1997; Alberti et al., 2007). In addition, the strength of the LUC-water quality relationship has been observed to vary with scale depending upon the pairing of LUC and water quality variables (Sponseller et al., 2001; Potter et al., 2005). In this study, there were no clear patterns between LUC class proportions or LSC indices and macroinvertebrate metrics with spatial scale. While LUC exhibited nearly twice the number of significant correlations with macroinvertebrate metrics at the stream and catchment scales compared with the reach scale, there were approximately the same number of significant correlations between LSC indices and macroinvertebrate metrics at all three spatial scales.

These mixed findings may be related to a number of factors. First, at finer spatial scales (reach), contaminants in close proximity to a stream may be more likely to enter it. Second, certain types of LUC found at the reach scale (e.g. agriculture, development) may dominate water quality at the sample site even in primarily forested catchments (Sponseller et al., 2001; Tran et al., 2010). Third, at moderate spatial scales (stream), upstream pollutant sources outside of the reach area that may be contributing to water quality conditions at the sample point have been accounted for (Busse et al., 2006). And fourth, the greater area encompassed at the broadest spatial scale (catchment) has the potential to overwhelm or mitigate landscape influences at finer scales (Roth et al., 1996).

Conclusions

As illustrated in this study, appropriate scales for landscape analysis of stream ecosystems should be selected based on multi-scale analyses that quantify multiple aspects of landscape condition and structure. Furthermore, our results suggest that LSC metrics can potentially add useful information to these analyses, which commonly only incorporate LUC class composition. Strong, statistically significant correlations were found in this study between both LUC class proportions and LSC indices and macroinvertebrate metrics, though there was almost twice the number for LUC compared with LSC. Moreover, LUC and LSC indicators were generally significantly correlated with different macroinvertebrate metrics. These results suggest that LUC class proportions may be better indicators of water quality conditions in this area; however it is probably premature to conclude that LSC indices are simply ineffective water quality indicators in this region. While no clear patterns were observed between landscape variables and macroinvertebrate metrics with spatial scale, the larger number of significant correlations found at the stream and catchment scales suggests that future analyses in this area should be focused at these spatial scales.

Future work could examine temporally-varying landscape indicators as a compliment to the existing (multi-scale) geographic dimension. For example, a number of authors have found seasonally-varying relationships between LUC and stream water quality variables (Boldstad and Swank, 1997; Griffith et al., 2002; Pan et al., 2004). However, while it is fairly simple to produce multi-temporal LSC indices for a particular study area, LUC mapping is typically more difficult and time-consuming. Thus, seasonal or annual LSC maps may possibly serve as useful indicators of changing landscape conditions that impact water quality between LUC map updates. Finally, the use of additional stream parameters related to nutrient and sediment loads, habitat quality, and fish assemblages could provide a more comprehensive assessment of land surface impacts on water quality.

Acknowledgements

Financial support was provided by the Institute for Regional Analysis and Public Policy at Morehead State University. Juli Taylor, Brittany Moody and Alan Grubb assisted with data collection and processing.

References

Alberti, M., Booth, D., Hill, K., Coburn, B., Avolio, C., Coe, S. and Spirandelli, D.
2007
.
The impact of urban patterns on aquatic ecosystems: An empirical analysis in Puget lowland sub-basins
.
Landscape Urban Plan.
,
80
:
345
361
.
Bolstad, P. V. and Swank, W. T.
1997
.
Cumulative impacts of landuse on water quality in a southern Appalachian watershed
.
J. Am. Water Resources Assn.
,
33
:
519
533
.
Busse, L. B., Simpson, J. C. and Cooper, S. D.
2006
.
Relationships among nutrients, algae, and land use in urbanized southern California streams
.
Can. J. Fish. Aquat. Sci.
,
63
:
2621
2638
.
Collins, J. B. and Woodcock, C. E.
1996
.
An assessment of several linear change detection techniques for mapping forest mortality using multitemporal Landsat TM data
.
Remote Sens. Env.
,
56
:
66
77
.
Crist, E. P., Lauren, R. and Cicone, R. C.
1986
. “
Vegetation and soils information contained in transformed thematic mapper data
”. In
Final Proceedings: IGARRS 1986 Symposium
,
Noordwijk, , The Netherlands
:
ESA Publication Division SP-254
.
Goetz, S. and Fiske, G.
2008
.
Linking the diversity and abundance of stream biota to landscapes in the mid-Atlantic USA
.
Remote Sens. Env.
,
112
:
4075
4085
.
Griffith, J., Martinko, E., Whistler, J. and Price, K.
2002
.
Interrelationships among landscapes, NDVI, and stream water quality in the U.S
.
Central Plains. Ecol. Appl.
,
12
:
1702
1718
.
Harding, J. S., Benfield, E. F., Bolstad, P. V., Helfman, G. S. and Jones, E. B.D.
1998
.
Stream biodiversity: The ghost of land use past
.
Proc. Natl. Acad. Sci.
,
95
:
14843
14847
.
Hawkins, C. P., Hogue, J. N., Decker, L. M. and Feminella, J. W.
1997
.
Channel morphology, water temperature, and assemblage structure of stream insects
.
J. N. Am. Benthol. Soc.
,
16
:
728
749
.
Healey, S. P., Cohen, W. B., Zhiqiang, Y. and Krankina, O. N.
2005
.
Comparison of Tasseled Cap-based Landsat data structures for use in forest disturbance detection
.
Remote Sens. Env.
,
97
:
301
310
.
Johnson, R. C., Carreiro, M. M., Jin, H. and Jack, J. D.
2012a
.
Within-year temporal variation and life-cycle seasonality affect stream macroinvertebrate community structure and biotic metrics
.
Ecol.
,
13
:
206
214
.
Indic.
Johnson, R. C., Smith, D. P. and McMichael, C. E.
2012b
.
Scale dependence in relating land use/cover to stream macroinvertebrate communities in the Central Appalachian Mountains, USA
.
GISci. Remote Sens.
,
49
:
53
70
.
Kentucky Division of Water (KDOW)
.
2008
.
Methods for Assessing Biological Integrity of Surface Waters in Kentucky, Revision 3
,
Frankfort, Kentucky: Kentucky Department of Environmental Protection, Division of Water
.
King, R. S., Baker, M. E., Kazyak, P. F. and Weller, D. E.
2011
.
How novel is too novel? Stream community thresholds at exceptionally low levels of catchment urbanization
.
Ecol. Appl.
,
21
:
1659
1678
.
Meyer, J. L., Paul, M. J. and Taulbee, W. K.
2005
.
Stream ecosystem function in urbanizing landscapes
.
J. N. Am. Benthol. Soc.
,
24
:
602
612
.
Ormerond, S. J. and Edwards, R. W.
1987
.
The ordination and classification of macroinvertebrate assemblages in the catchment of the River Wye in relation to environmental factors
.
Freshwater Biol.
,
17
:
533
546
.
Pan, Y., Herlihy, A., Kaufmann, P., Wigington, J., Van Sickle, J. and Moser, T.
2004
.
Linkages among land-use, water quality, physical habitat conditions and lotic diatom assemblages: A multi-spatial scale assessment
.
Hydrobiologia
,
515
:
59
73
.
Potter, K. M., Cubbage, F. W. and Schaberg, R. H.
2005
.
Multiple-scale landscape predictors of benthic macroinvertebrate community structure in North Carolina
.
Landscape Urban Plan.
,
71
:
77
90
.
Rogan, J., Franklin, J. and Roberts, D. A.
2002
.
A comparison of methods for monitoring multitemporal vegetation change using Thematic Mapper imagery
.
Remote Sens. Env.
,
80
:
143
156
.
Roth, N. E., Allan, J. D. and Erickson, D. L.
1996
.
Landscape influences on stream biotic integrity assessed at multiple spatial scales
.
Landscape Ecol.
,
11
:
141
156
.
Roy, A. H., Rosemond, A. D., Paul, M. J., Leigh, D. S. and Wallace, J. B.
2003
.
Stream macroinvertebrate response to catchment urbanization (Georgia, U.S.A.)
.
Freshwater Biol.
,
48
:
329
346
.
Sponseller, R. A., Benfield, E. F. and Valett, H. M.
2001
.
Relationships between land use, spatial scale and stream macroinvertebrate communities
.
Freshwater Biol.
,
46
:
1409
1424
.
Tran, C. P., Bode, R. W., Smith, A. J. and Kleppel, G. S.
2010
.
Land-use proximity as a basis for assessing stream water quality in New York State (USA)
.
Ecol.
,
10
:
727
733
.
Indic.
United States Environmental Protection Agency (USEPA)
.
2002
. “
National Water Quality Inventory: 2000 Report
”.
EPA-841-R-02-001
Wang, L., 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
12
.