Bed Stability and Sedimentation Associated With Human Disturbances in Pacific Northwest Streams¹

Wood and Riparian Measures.  A tally of large wood pieces ≥10 cm in diameter and ≥1.5 m long within the bankfull channel was recorded between each pair of sequential cross-sections along sample reaches. Pieces were tallied in three length and four diameter classes, and data were summarized as wood volume per m² of bankfull channel area. Vegetation cover and type were visually estimated for each of three vegetation layers (ground cover, mid-layer, and canopy) within each of the 11 pairs of streamside riparian plots. Mean areal cover values for these riparian vegetation layers were calculated as reach-wide summaries. At and beyond each riparian plot, the presence and proximity of 11 categories of streamside human influences were recorded (walls, dikes, revetments, and dams; buildings; pavement or cleared lots; roads and railroads; pipes; landfills and trash; parks and lawns; row crops; pasture or rangeland; logging; and mining). A proximity-weighted disturbance index (W1_HALL) was calculated by tallying the number of left and right bank riparian stations at which a particular type of disturbance was observed, weighting each observation according to its proximity to the stream, and then averaging over all 22 observation points on the reach (Kaufmann et al., 1999). Observations within the stream or on its banks were weighted by 1.5, those within the 10 × 10 m plots were weighted by 1.0, and those visible beyond the plots were weighted by 0.67. Scores of the disturbance index ranged from 0 to 5 in this dataset, reflecting a range from low to high riparian disturbance.

Slope and Bed Particle Size.  Stream reach slope (%) was calculated as the arithmetic mean of ten water surface gradient measurements from handheld clinometer sightings on survey rods between sequential pairs of the 11 equally spaced transects. The particle size distribution of the wetted streambed surface was quantified by means of a systematic “pebble count,” in which 55 particles (five from each of 11 channel cross-sections) were selected and classified according to size categories. The pebble count procedure differed from typical applications to reduce bias against fine particles (see Kellerhals and Bray, 1971; Whitman et al., 2003). For sand and smaller particles (<2 mm diameter), field crews determined and recorded the dominant size class (sand or silt) of particles in the “pinch” of fine particles between their fingers. Areal cover-based summaries of reach-wide bed surface particle size were calculated from this field data. They included the geometric mean diameter (D_gm) and the percentages of bedrock (>4,000 mm diameter), boulders (250-4,000 mm), cobbles (64-250 mm), coarse gravel (16-64 mm), fine gravel (2-16 mm), sand (0.06-2 mm), silt (<0.06 mm), and organic detritus. D_gm was calculated by nominally assigning to each particle the geometric mean diameter of the upper and lower bounds of its diameter class (e.g., 5.66 mm for fine gravel), and then calculating the geometric mean of those class midpoint values, where geometric mean were calculated as the antilog of the arithmetic mean of the logarithms of those frequency-weighted class midpoint values (Faustini and Kaufmann, 2007; Kaufmann et al., 2008). Bedrock and silt, respectively, were assigned class midpoint values of 5,660 and 0.0077 mm.

Channel Morphology.  Channel measurements on each sample reach included a longitudinal survey of maximum (thalweg) depth at 100 equally spaced points (150 on streams <2.5 m wide), wetted width at 21 equally spaced locations, bankfull height and width at 11 equally spaced cross-sections (using a stadia rod and handheld level), and channel depth cross-section profiles consisting of five equally spaced points at the same 11 cross-sections. Reach mean and standard deviations for thalweg depth, bankfull dimensions, wetted width, and other descriptors of channel morphology were calculated from these measurements. Bankfull height was estimated from field evidence at 11 separate cross sections on the basis of channel, bank and floodplain geometry, deposition features with sediments <2 mm diameter, riparian vegetation, and flood height evidence observed during the low flow sample visit. In channels where bankfull floods were very unlikely (incised channels), or very common (channels in the process of enlarging due to increases in flood magnitude), crews were instructed to give more weight to vegetation and field evidence of recent flooding. Mean bankfull thalweg depths were calculated by summing values for mean thalweg depth and bankfull height. Combined with mean reach slope, longitudinal profiles of thalweg depth were used to calculate mean residual depth according to procedures described by Robison and Kaufmann (1994). Mean residual depth is a flow-independent index of reach-wide pool volume and large-scale roughness (see Bathurst, 1981; Kaufmann, 1987), a reach-wide interpretation of the residual pool concept, where individual residual pools are defined as those portions of a stream that would contain water at zero discharge due to the damming effect of their downstream riffle crests (Lisle, 1982).

Riparian Disturbance and Condition.  We used a composite riparian condition index (RCOND) defined by Kaufmann and Hughes (2006) from field data tallying evidence of streamside human activities and the cover and structure of riparian vegetation. The index decreases with increases in streamside human activities (W1_HALL), and increases with increasing large diameter tree cover (XCL) and riparian vegetation complexity (XCMGW, the sum of areal cover of woody vegetation in the tree, shrub, and ground cover layers). These riparian habitat measures are described in detail by Kaufmann et al. (1999). The index RCOND was defined as follows

(1)

Basin Disturbances.  As a surrogate index of the level of human activity and disturbance within the contributing drainages of our sample sites, we used road densities expressed as km of roads per km² of stream basin area (RdDenkm) calculated from readily available digital road coverages provided by the U.S. Census Bureau (1990). Although it is only one possible measure of total basin disturbance, road density has been shown to be negatively correlated with basin percent areal cover of forests with trees having medium to very large diameter (e.g., Burnett et al., 2006). Furthermore, road density is known to play the dominant role in anthropogenic augmentation of sediment supply by erosion and mass-wasting in upland forested landscapes in the PNW region (Cederholm et al., 1981; Furniss et al., 1991; Madej, 2001). Hughes et al. (2004) and Kaufmann and Hughes (2006) showed road density to be consistently negatively related to an index of stream vertebrate biotic integrity (IBI) in the PNW Coast Range Ecoregion. In turn, they also showed that IBI was consistently negatively related to percent fine sediment (<2 mm), excess fine sediment with diameter <0.06 mm, and a formulation of RBS defined by Kaufmann et al. (1999), suggesting the importance of roads as a likely source of sediments that limit biotic integrity of streams.

Combined Basin + Riparian Condition.  We used an index of basin + riparian condition that combines the RCOND with road density (scaled by its regional maximum) as defined by Kaufmann and Hughes (2006):

(2)

In addition, we defined the ordinal basin + riparian disturbance class variable DISTLEV as shown in Table 2, with low, moderate, and high levels of anthropogenic disturbance based on RCOND with additional high and low threshold values of basin road density. The low and high DISTLEV classes separate out the lower and upper quartiles of basin + riparian disturbance in the study region.

Measurement Precision.  We evaluated the precision of LRBS* measures by analysis of variance, partitioning variance within sites (repeat measurements) and among sites across the region (see Kaufmann et al., 1999, 2008; Larsen et al., 2004). The root mean square error from this partitioning of variance (RMS_rep), termed RMSE by Kaufmann et al. (1999), is analogous to the standard deviation of measurements at each individual site, averaged over the sites where repeat measurements were made. We also calculated a relative measure of precision, a “signal-to-noise ratio” (SN), as the ratio of among-site variance to the variance of within-year repeat measurements to sites (see Kaufmann et al., 1999; Faustini and Kaufmann, 2007).

Regression Modeling.  We examined the relationship of LRBS* to riparian and basin human disturbances and natural geomorphic controls using multiple linear regression (MLR). We analyzed a comprehensive suite of geomorphic and hydraulic variables as potential predictors, including basin area, channel gradient, unit stream power, and lithology. This enabled us to account for natural variation in LRBS* that may occur among relatively undisturbed streams in different geologic and geomorphic settings. We also included a comprehensive suite of potential interactions among variables, allowing us to test whether the response of LRBS* to disturbance differs among streams of differing size, gradient, stream power, or lithology. With the exception of channel gradient, we did not consider other variables that were used to calculate D*_cbf, and subsequently LRBS* (bankfull depth, in-channel wood volume, residual pool depth, particle Reynolds number, and estimated Shields parameter).

We followed the MLR procedure described by Kaufmann and Hughes (2006), which includes explicit measures to avoid over-fitting, and employs procedures to build mechanistically meaningful models. The MLR predictor variable selection procedure was stepwise (forward-backward), including and retaining only variables with p < 0.15, and confirming best-model selection by examining all possible MLR models with less than or equal the number of resultant predictor variables. The final model with five variables described in our results was significant at p < 0.0001. The highest p value for individual predictor variables actually selected in the MLR model using our variable selection criteria was p = 0.0275, and the other four had p < 0.0001. To avoid over-fitting (over-parameterization), we did not build models with more than n/10 predictor variables, where n is the number of sample sites in a particular modeled stratum. To further avoid over-fitting, we also constrained the number of predictor variables so that the RMSE of the regression model (0.70) was larger than 0.43, which is the pooled RMS_rep reported by Kaufmann et al. (2008) for same-stream repeat measurements of LRBS*. RMS_rep is equivalent to the pooled standard deviation of site revisits, and represents a practical limit of an MLR model to associate variation of LRBS* in sites across the region with ancillary site data (Kaufmann et al., 1999). A regression model with RMSE substantially less than the RMS_rep for measurement variation would be suspected of over-fitting.

Analysis of Associations.  We conducted correlation analyses comparing the sign and strength of associations (Pearson’s r) between LRBS* and measures of riparian, basin, and combined riparian + basin condition. We then conducted two analyses to assess the degree to which lower LRBS* associated with greater land disturbance was due to decreased D_gm, or to increased D*_cbf or D_cbf, where D_cbf is the critical diameter calculated exactly as for D*_cbf, but without the adjustment for woody debris and bedform resistance. In the first analysis, we examined plots and correlations (Pearson's r) relating regional variation in LRBS* to variation in D_gm, D*_cbf, and D_cbf, where all three diameters were log-transformed. Stronger correlations between LRBS* and D_gm than between LRBS* and D*_cbf or D_cbf would indicate that variation in the region is due to differences in the numerator (D_gm) of the LRBS* ratio rather than the denominator (D*_cbf). Strong correlations between LRBS* and D_cbf (but not between LRBS* and D*_cbf) would be consistent with an interpretation that effects of flow increases on LRBS* are negated by differences in hydraulic roughness that are positively correlated with total shear stress (e.g., deeper pools). In the second analysis, we examined correlations of LRBS*, its subcomponents D_gm and D*_cbf, and the unadjusted critical diameter value D_cbf (log-transformed) with the combined basin + riparian condition variable WRCOND to examine the weight-of-evidence for anthropogenic influence on these attributes.

Influence of Lithology.  To examine the influence of basin lithology on the response of streams to anthropogenic basin + riparian disturbances, we grouped the sample streams into two groups based on the dominant bedrock lithology within their drainage basins: volcanic vs. sedimentary. These groupings were based on classifications and discussion by Pater et al. (1998), from which we interpreted basin lithology to be “erodible” where dominated by sedimentary sandstones or siltstones, and “resistant” where largely composed of competent volcanics, mainly basalt. Streams with a large amount of basin + riparian anthropogenic disturbance (star symbols) were found throughout the range of D*_cbf, as was displacement below the 1:1 line in Figures 2A and 2B. Textural fining is suggested in both lithologies, as D_gm was substantially smaller than D*_cbf in the most anthropogenically disturbed streams of both lithologies.

The magnitude of LRBS* can be graphically interpreted as the deviation above or below the 1:1 line in Figures 2A and 2B, with negative LRBS* below the 1:1 line. The estimated degree of uncertainty in D_gm and D*_cbf in individual sites is shown by the error bars (each 2 × RMS_rep) in Figures 2A and 2B. An ellipse surrounding these error bars describes the average uncertainty of the numerator and the denominator of LRBS* for any individual site. Because the magnitude of LRBS* at each site in this figure is shown by its displacement above or below the 1:1 line, the error bounds also give a visual indication of the magnitude of individual LRBS* differences that can be considered significant. For example, in sedimentary lithology (Figure 2A), individual site values of D_gm that are two orders of magnitude lower than D*_cbf (i.e., those that have LRBS*≤−2.0) are displaced >4 × RMS_rep from the 1:1 line. They can be clearly interpreted to be lower than those near the 1:1 line with LRBS*∼0.0. Similarly, in volcanic lithology (Figure 2B), individual site values of D_gm that are greater than first order of magnitude lower than D*_cbf (i.e., those with LRBS*≤−1.0) are displaced >5 × RMS_rep from the 1:1 line, and can be clearly interpreted to be lower than those with LRBS*≥0.0.

In all sedimentary lithology streams with low anthropogenic disturbance (DISTLEV = low), D_gm was within two orders of magnitude of D*_cbf (solid dots in Figure 2A). Conversely, D_gm was more than two orders of magnitude smaller than D*_cbf in all but four highly disturbed streams in that lithology (star symbols in Figure 2A). Similarly, there was no overlap in displacement between low and high disturbance streams in volcanic lithology. D_gm was within 1.3 orders of magnitude of D*_cbf in the least disturbed streams (solid dots in Figure 2B), and more than 1.3 orders of magnitude smaller in the most disturbed streams (star symbols in Figure 2B). Although all the most disturbed streams in both lithologies were substantially below the 1:1 line in Figures 2A and 2B, the offset was generally much greater for sedimentary streams.

There was a wide range of LRBS* at intermediate disturbance (including very high values), but LRBS* in the least-disturbed streams (DISTLEV = low) ranged from −1.9 to +0.5, with most (>50%) values between −1.1 and −0.2. LRBS* in highly disturbed streams (DISTLEV = high) ranged from −4.2 to −1.1, with most values between −2.9 and −1.8. The mean LRBS* in low disturbance streams was very similar in the two lithologies (0.28 lower in sedimentary lithology; p = 0.33). By contrast, the mean LRBS* among highly disturbed streams draining sedimentary lithology was 0.99 log units lower than for those draining volcanics (p = 0.04). The difference in apparent bed textural fining between low and high disturbance streams in both lithologies strongly suggests that anthropogenic disturbance may cause these textural differences.

Combined Basin + Riparian Disturbance. LRBS* showed a pattern of progressive or threshold declines associated with increasing anthropogenic basin + riparian disturbance (DISTLEV) in various subgroups of sample streams across the PNW Coast Range Ecoregion (Figure 3). In all groupings (lithology, basin area, and channel gradient), mean LRBS* in streams with high disturbance was substantially lower than in those with low disturbance (p < 0.0001 for all, except p = 0.0016 for resistant volcanics). The same was true for the contrast between moderate and high disturbance classes (p < 0.0001, except for p = 0.0002 in volcanics and p = 0.0155 in small basin areas). Only in small streams (Figure 3C) were all three DISTLEV classes distinct (considering moderate and high classes distinct at p = 0.0155); in the other groupings, LRBS* in the low and intermediate disturbance classes overlapped. Although the mean LRBS* in low disturbance streams did not differ greatly between the two lithologies (see previous paragraph), streams draining sedimentary lithology showed a markedly steeper decline of LRBS* with increasing human disturbance (Figure 3A) than did those draining volcanic lithology (Figure 3B). In both cases highly disturbed streams had significantly lower LRBS* than those with low or moderate anthropogenic disturbance, but the pattern was progressive in the erodible sedimentary group and more of an apparent threshold response in the volcanic group. Similarly, when streams were grouped by basin area, those with smaller drainages showed a more progressive pattern of decline with disturbance (Figure 3C), compared with the apparent threshold response seen in larger streams (Figure 3D). There was a similar apparent threshold response to disturbance for both low-gradient (≤1%) and steeper streams (Figures 3E and 3F).

Relative Influence of Basin vs. Riparian Disturbance.  In the region overall, LRBS* was more strongly correlated with riparian condition than with basin road density (Table 3). This was especially true of larger streams (basin area ≥15 km²) in sedimentary drainages. LRBS* in steeper streams (channel gradient >1.0%) in smaller basins draining both lithologies tended to have nearly equal correlation (but opposite sign) with riparian condition and basin road-density (i.e., same sign for riparian disturbance and road density). The pattern was mixed or unclear in large streams draining volcanic lithology, where only road density was significantly (p < 0.1) correlated (negatively) with LRBS*, and only in lower gradient streams. Correlations of LRBS* with combined basin + riparian condition were higher than with either riparian or basin condition alone in most subclasses of streams (Table 3), suggesting that the combined effect of basin and riparian disturbances was greater than that of either by itself.

Associations of LRBS* With Geomorphic Setting.  One might question whether correlations between LRBS* and land use disturbances result from covariance of both of these variables with other natural geomorphic variables that actually cause the observed associations. For example, one might expect lower bed stability in low gradient streams further down the drainage network where anthropogenic disturbance is typically greater. To the contrary, LRBS* was unrelated to channel slope overall (r = −0.03, p = 0.76), and negatively related to channel slope for streams in volcanic lithology (r = −0.69, p < 0.0001; Figure 4A). Also contrary to our expectations, LRBS* had moderate positive correlations with drainage area in streams draining both lithologies (Figure 4B). Although LRBS* was not clearly related to elevation overall (Figure 4C), it exhibited a weak negative correlation with elevation in volcanic streams (r = −0.36, p = 0.049) that might reflect its stronger association with drainage area. Interestingly, the widest range in LRBS* among streams was observed at low elevation. The positive correlation (r = +0.51, p < 0.0001 for combined lithologies) of LRBS* with basin + riparian condition (WRCOND), shown in Figure 4D was stronger than its associations with geomorphic variables in the region overall, though its correlation with basin area was comparable (r = +0.44, p < 0.0001). When viewed by separate lithologies, the correlations between LRBS* and WRCOND (Figure 4D) were of approximately equal strength as its correlation with basin area (Figure 4B) in both lithologies. In volcanic terrain, LRBS* was more strongly (negatively) correlated with channel gradient than with human disturbances (Figure 4A). Nonetheless, Table 3 and Figure 3 show that LRBS* was negatively associated with human disturbance in most drainage area and slope subsets of streams. Small streams draining <15 km² in both lithologies showed LRBS* depressions associated with land-use disturbance, but only in sedimentary lithology did larger streams draining ≥15 km² show the same pattern. These findings suggest that, particularly for streams of volcanic lithology, anthropogenic influences on LRBS* would be clarified by MLR modeling, which included channel gradient and basin area as potential controls.

Associations of Human Disturbances With Geomorphic Setting.  Contrary to common expectations in other regions, basin + riparian condition WRCOND was generally not strongly associated with channel gradient (Figure 5A), basin area (Figure 5B), elevation (Figure 5C), or the unit stream power surrogate log[A_ws^0.5S] in this region overall (Figure 5D), where A_ws is basin area in km² and S is water surface slope (%). In streams of volcanic lithology, however, there was moderate evidence of negative associations of WRCOND with channel gradient (r = −0.37, p = 0.04; Figure 5A), elevation (r = −0.36, p = 0.07; Figure 5C) and stream power (r = −0.45, p = 0.01; Figure 5D). These findings are contrary to the typical pattern of decreasing human disturbance with elevation and gradient in most mountainous regions of the U.S. As observed for LRBS*, low elevation sites exhibited the widest range in basin + riparian condition in this region. There were clear positive correlations of LRBS* with WRCOND in both lithologies (Figure 4D). However, in streams of volcanic lithology, negative relationships of WRCOND with channel gradient and elevation increase the uncertainty of causal interpretation of controls on LRBS* for streams in that lithology, and suggest MLR modeling to include both anthropogenic disturbances as well as potential landscape and geomorphic controls.

Model Simulation. Figure 6 illustrates this complex model behavior in a plot of predicted (modeled) LRBS*vs. basin + riparian condition (WRCOND). Road density in this simulation was set to values from 0 to 6.0 km/km² as WRCOND varied from 1.0 to 0 (pristine to very disturbed). Modeled LRBS* is plotted separately for hypothetical small and large basins (A_ws = 3 and 55 km²) in volcanic and sedimentary lithology. For streams with these two drainage areas in each lithology, we plotted modeled LRBS* for hypothetical streams with low and high surrogate unit stream power values (A_ws^0.5S = 2.4 and 5.8 km). The example basin areas and A_ws^0.5S values are the 25th and 75th percentiles of the stream sample distribution in the regional survey. Median-sized streams and median-power streams (not shown) would plot midway between those shown in the figure.

Implications of the Model Simulation.  The model suggests that in pristine sites, LRBS* varies with basin size and unit stream power (therefore channel gradient), but does not differ greatly between sedimentary and volcanic basins (compare Figures 6A and 6B at WRCOND = 1.0). Although the apparently minor effect of lithology alone in pristine streams is somewhat surprising, the differences in stream power that control LRBS* variation at any given basin size and disturbance level are associated with differences in topography and runoff that are affected or even controlled by lithology. As basin and riparian human disturbances increase in these streams, the influence of lithology becomes progressively more apparent. The LRBS* decline with basin + riparian disturbance is more than twice as steep in streams draining the more erodible sedimentary basins than in those draining less erodible volcanic basins. For example, a decline in WRCOND from 1.0 (pristine) to 0 (highly disturbed), is predicted to result in a stream bed LRBS* decline of 3.3 log units in sedimentary lithology, compared with a decline of 1.4 log units for volcanic lithology under the same conditions. The controlling influence of stream basin size was consistent between lithologies and across the disturbance gradient: larger streams are predicted to always have somewhat higher LRBS* than small streams flowing at the same gradient with the same level of disturbance (compare gray and black lines in Figure 6). However, the combined influence of basin size and channel gradient (as expressed by the unit stream power surrogate, A_ws^0.5S) was opposite between the two lithologies, and was slightly smaller than the influence of basin size alone. In sedimentary basins, high power streams have higher LRBS* than low power streams, while in volcanic basins, high power streams have lower LRBS* than low power streams (compare dotted and solid lines in Figures 6A and 6B).

To summarize, MLR modeling suggests that there is a characteristic level of LRBS* in relatively undisturbed streams of this region that does not differ greatly from 0, but varies between −0.7 and +0.5 with stream size and power, regardless of lithology. As basins and riparian areas are disturbed by human activities, LRBS* typically declines, the decline being more pronounced in sedimentary than in volcanic basins. With severe basin and riparian disturbance, LRBS* declines to between −3 and −4 in streams draining sedimentary lithology, in contrast to a decline to between −1 and −2 in volcanic basins. There is a rather large amount of uncertainty in the exact details of these model results, as reflected in its RMSE (0.70). Therefore, we suggest that these model results should be treated more as an aid to understanding regional patterns and differences in the relative magnitude of LRBS* and proposing testable hypotheses concerning its potential controlling factors, rather than as a tool for precisely predicting individual channel responses. However, in combination with previous analyses, these modeling results strongly link LRBS* declines to anthropogenic disturbances in streams draining both sedimentary and volcanic lithologies. They also show the influence of slope and basin size to be minor in relation to human influences in sedimentary lithology, but somewhat greater in relative terms in volcanic lithology.

Rationale for Interpreting Likely Responses.  We have shown that LRBS* depressions in this region are associated with human disturbance. Up to this point, we have not shown to what degree these depressions result from streambed “fining” (sedimentation), as opposed to increases in bed shear stress (e.g., see discussions by Millar and Rennie, 2001; Wilcock, 2001). In this section, we build on weight-of-evidence from previous sections supporting the assertion that negative LRBS* associations with human activities we observed are probably causal, regardless of whether the effect is on bed particle size or shear stress. Further, we then argue that streambed fining is a larger and more pervasive response than is increased shear stress in wadeable streams draining sedimentary lithology in our study region, while both mechanisms are evident in streams draining volcanic basins that are generally steeper and more resistant to weathering and surface erosion but more prone to upslope mass-failures.

Mode of Response: Fining vs. Change in Competence.  We used correlations between LRBS* and its log-transformed numerator (logD_gm) and denominator (logD*_cbf) to discern the relative contribution of each term to the variation in LRBS* over the region and its subclasses (Table 5), knowing well that the combination of D_gm and D*_cbf necessarily explains all the variation in LRBS*. In the region as a whole, variation in LRBS* was much more strongly correlated with logD_gm (Pearson r = +0.91, p < 0.0001), than with logD*_cbf (r = +0.09, p > 0.1) or with logD_cbf (r = −0.07, p > 0.1), the unadjusted critical diameter (log-transformed). When streams were grouped by lithology, LRBS* variation in the sedimentary group was again correlated much more strongly with logD_gm (r = +0.93, p < 0.0001) than with logD*_cbf or logD_cbf (r = +0.05 and −0.13, both with p > 0.1). Large streams in volcanic lithology followed the same pattern as that for large and small sedimentary streams. In small streams of the resistant lithology group, all but one of which were in the steeper slope category in our sample, correlations between LRBS* and logD_gm were similar in magnitude to those with logD*_cbf or logD_cbf. These results suggest that in the region as a whole, and in the sedimentary lithology subgroup in particular, LRBS* depressions resulted from bed textural fining (e.g., by the mechanisms described by Buffington and Montgomery, 1999b, 2001) rather than variations in transport competence, i.e., shear stress. In small streams within the more resistant volcanic lithology subgroup, the evidence suggests that variations in both bed textural fining and transport competence (e.g., see Millar, 2000; Millar and Rennie, 2001) account for the smaller amount of regional variation we see in LRBS*, which ranged less than three orders of magnitude (−1.9 to +0.80) in streams of volcanic lithology compared with more than five orders of magnitude (−4.2 to +0.98) in the sedimentary group.

Streams Draining Sedimentary Basins.  In all subgroups of streams draining sedimentary lithology, as in the region as a whole, LRBS* and logD_gm were positively correlated with WRCOND (see Equation 2), a measure of basin + riparian condition inversely related to human disturbance (Table 6). Pearson r values with p < 0.1 varied from +0.46 to +0.54 for LRBS* and from +0.38 to +0.53 for logD_gm in various groupings of sedimentary streams by basin size and channel slope, with no significant negative correlations (see Table 6 for significance levels). By contrast, for sedimentary basins, only in larger streams (basin areas ≥15 km²) with channel gradients ≤1.0% were logD*_cbf or logD_cbf significantly correlated with WRCOND (r = −0.38 and −0.37, respectively, and p < 0.1), and the strength of those correlations did not exceed those between logD_gm and WRCOND. These results lend support to the hypothesis that LRBS* was lowered by human activities in sedimentary streams of the region, and this is due more to streambed fining from increases in sediment supply than to increases in transport competence.

Streams Draining Small Volcanic Basins. LRBS* in small streams draining volcanic basins was even more strongly correlated with basin + riparian condition (r = +0.80, p < 0.001) than observed for sedimentary streams (Table 6). LRBS* variation was driven nearly equally by variation in D_gm and D_cbf or D*_cbf (Table 5). However, the correlation of logD_gm with WRCOND (r = +0.39, p > 0.1) was somewhat weaker than the moderate negative correlations of logD_cbf and logD*_cbf with WRCOND (r = −0.42, p < 0.1 and −0.62, p < 0.01). In addition to bed fining that would result from an augmentation of sediment supply, this pattern suggests possible increases in competence that could have resulted from anthropogenic augmentation of peak flows and reduction of channel roughness (see bottom of Table 6). Correlations between logD_cbf (unadjusted for hydraulic roughness) and disturbance would suggest that disturbance-related increases in bed shear stress may be the result of changes in discharge or channel morphology (influencing bankfull depth); whereas correlations of logD*_cbf, but not logD_cbf, with disturbance would suggest that disturbance may be affecting channel complexity (e.g., woody debris). Although tentative, these findings suggest that human activities in small basins draining resistant lithology may have led not only to increases in sediment supply, but also to hydrologic or morphologic changes that increase bed shear stress, thereby decreasing LRBS*. Millar and Rennie (2001) argue that this pattern of channel response to riparian disturbance, as was observed by Millar (2000) in coastal streams of British Columbia, would alter shear stress available for sediment transport.

Streams Draining Large Volcanic Basins.  The pattern for large streams in volcanic lithology is mixed and interpretation is hampered by small sample sizes in this class. Overall, our data showed that LRBS* variation in those streams was primarily driven by variation in D_gm (Table 5), just as observed for nearly all the other classes of streams in the region. In contrast with other stream classes, however, these large, volcanic streams showed only weak (r < 0.35) uncertain (p >0.10) correlations between LRBS*– or log-transformed D_gm, D*_cbf, and D_cbf– and the basin + riparian condition index, WRCOND (Table 6). An uncertain exception (not shown in Table 6) was that LRBS* was strongly negatively correlated with basin road density (r = −0.85, p = 0.1) in the small subset of six large volcanic streams with channel gradients ≤1.0%. Although the sample sizes are rather small for the entire class of large volcanic streams (n = 14), weak correlation between disturbance and LRBS* might also reflect comparatively long lags in channel response between disturbance (or recovery) and channel response in these large streams. Harding et al. (1998), for example, found basin land use 40 years previous to be a better predictor of present-day stream biodiversity than contemporary land use. The low correlations between LRBS* and indicators of current basin and riparian land use in large volcanic streams of our study might also reflect greater resistance of large streams to perturbation. Furthermore, because quantifying the proximity of disturbances may be more important in larger basins than in smaller ones, our measures of riparian and near-stream disturbance may not be as good surrogates for total basin + riparian disturbance in large streams as they are for small streams.