INTRODUCTION
The variation of stream channel width with flow, basin area, habitat unit type, vegetation and soil type has been considered by many (e.g., Zimmerman et al., 1967; Leopold and Maddock, 1953; Murgatroyd and Ternan, 1983; Touysinhthiphonexay and Gardner, 1984). Studies of pools included spacing (Keller, 1978; Keller and Melhorn, 1978; Grant et al., 1990; Montgomery et al., 1995), variation of pooi area (Lanka et al., 1987; Kozel et al., 1989; Myers and Swanson, 1991), mechanisms (Robison and Beschta, 1990), formation, formative features and resiliency of the pooiriffle sequence (Tinkler, 1970; Gregory et al., 1994;
Myers and Swanson, 1994). Studies of the impact of land management on streams often compare changes in width or pooi area (PA) or spacing with time (Heede, 1981 and 1986; Ryan and Grant, 1991; Gregory et al., 1994; Myers and Swanson, 1996a, b), among streams receiving different treatments (Myers and Swanson, 1994, 1995), or along stream reach continuurns that experienced different treatments (Gunderson, 1968; Kondolf, 1993).
Variability at a site affects width or PA studies. Wolman (1955) and Knighton (1975) found that width measurements varied substantially along short reaches and caused scatter around downstream hydraulic geometry relationships. Hydraulic exponents may vary substantially within short reaches where the flow is constant (Knighton, 1975; Richards, 1976a; Phillips and Harlin, 1984). Myers and Swanson (1991) found that the high variance, or low precision, of measurements limited the value of stream measurements of PA.
Precision is the degree to which an estimate represents the appropriate population value (Thompson, 1992) and is expressed as a confidence limit or coefficient of variation. Although measurement precision affects the design of monitoring studies and comparisons among treatments (Thompson, 1992), and many studies cited above found that variability decreased the value of their results, we found no published studies of the precision of width or PA measurements. Our objective is to improve future studies and provide a better understanding of previous studies of the variation of stream width and pool area with location or time by considering the precision of transect-based width and PA estimates. We make recommendations for transect numbers and spacings that optimize measurement precision and survey time. Our study relates to the pool-riffle scale of Frissel et al. (1986). Because randomly located features such as boulders and coarse woody debris vary among watersheds (Robison and Beschta, 1990; Myers, 1996), some streams may be inherently more variable. This potentially affects survey strategies. Therefore, we also analyzed the variation of precision estimates with stream morphology to assist in the selection of survey methods based on the type of stream.
Background
Channel and Water Width
The ability to detect width changes with time, w = b(t)Q(t) depends on measurement precision which depends on in-stream structural features and the selection of reach lengths. Surveyors should select stochastically homogeneous reaches to minimize measurement variance. Stochastic homogeneity is the condition that randomly chosen measurements are drawn from the same probability distribution such that expected value and variance for any measurement y, E(y x) and Var(y I x), is constant for any location x on the reach. For example, a survey reach combining reaches with scour pools and reaches with step-pools would be inhomogeneous yielding descriptive statistics not representative of either reach.
Confidence limits are higher for correlated observations because of decreased effective n (where n is sample size). Although spatial autocorrelation of channel width is generally insignificant after two widths (Furbish, 1985; Robison and Beschta, 1990) some studies have found cyclic variations in width along a channel in relation to the spacing of pools and riffles (Harvey, 1975, Richards, 1976a, b). Our transects were spaced to measure the phenomenon of autocorrelation.
Water width, measured at baseflow, represents the low flow channel which may be a stable indicator of basin conditions (Richards, 1982). The baseflow channel coincides within a few centimeters with various vegetative indicators (Hupp and Osterkamp, 1985), has been useful in analyzing long-term changes due to land management (Dose and Roper, 1994; Myers and Swanson, 1996a) and is essential in definitions of pool area (BLM, 1978; USFS, 1985, 1992; Andrus et al., 1988; Hankin and Reeves, 1988; Myers and Swanson, 1991). Thus, many types of studies and surveys depend on the precision of water width which we analyzed herein.
Pool Area
Many studies showing the importance of pools as fish cover (Baltz et al., 1991; Heggenes et al. 1991) and indicators of habitat quality (Andrus et al. 1988; Myers and Swanson, 1995) estimated PA using systematically spaced transects along a reach. If regularly spaced pools coincide with the transect spacing (i.e., Richards 1976a, b), it is. possible to miss all pools or to sample every pool and no riffles. Thus, the estimate varies based on the correspondence of transects and pool location.
STUDY AREA
We used 47 stream segments in six mountain ranges of central and northwestern Nevada (Figure 1). All sites are within the basin and range geologic province (Stewart and Carlson, 1978). Sagebrush steppe to pinyon-juniper woodlands dominated the upland vegetation; Riparian vegetation included grasses, sedges, shrubs, and trees. Based on the distribution of stream types found by Myers and Swanson (1991), we chose sites attempting to represent small streams in Nevada rangelands. There were three primary stream types with a variety of subtypes based on substrate (Rosgen, 1994). Table 1 presents site characteristics. Table 2 describes Rosgen stream types used herein to describe the stream crosssection.
METHODS
At baseflow, we surveyed transects, cross-sections perpendicular to the stream centerline, at spacings of 1 to 1.5 active channel widths. The spacing was chosen based on the lack of autocorrelation found by other authors on similar small streams (e.g., Furbish, 1985; Robison and Beschta, 1990). Baseflow exists when streamflow consists almost entirely of groundwater discharge (Mosley and McKerchar, 1993). We assumed that baseflow occurred when spring runoff had ceased and flow rates had become essentially constant. A sampling unit is a reach sampled with 25 transects.
The active channel is the point where an area width curve would substantially change slope (Williams, 1978) as estimated in the field. Vegetation, scour marks, and bar height aided delineation (Leopold, 1994). At each transect, we measured water width (WW), channel width (CHW), fraction of the water width that is a pool (PF), maximum channel depth (CDEP) and interbank width at twice CDEP,, water depth at the quarter points (d1j4, d112, d314), both banks and the maximum depth (d1, dr, and dmax), substrate fractions [silt/clay, < 0.062 mm; sand, 0.062< sand < 2 mm; gravel, 2 mm < gravel < 64 mm; cobble, 64 mm < cobble < 256 mm; and boulders, > 256 mm, CWD (woody debris with diameter greater than 25 mm)]. We combined CWD and BLD into CWDBLD because of many zero values in each category and similar pool-forming function (Myers and Swanson, 1994).
Water width depth ratio is the quotient of WW and d. Channel width/depth ratio (CHWD) is the quotient of CHWand (CDEP - dmax + d). Pool area, PA, is the fraction of stream surface area classified as a pool:
where n is the number of transects. Pools are distinct habitat units with hydraulic gradient less than the, stream average and subcritical flow conditions except for an entry jet which may affect up to 15 percent of the surface (Grant et al., 1990). Table 3 presents the definition of variables calculated from transect and plan measurements and their simple correlation. We also measured the length and water width at each end and in the middle of each habitat unit according to Hankin and Reeves (1988). All length and width measurements were standardized by dividing by the average water width. Thus, the expected value of water width is always 1. Repeatability of the identification of basic units (pools and nonpools) is high (Roper and Scarnecchia, 1995). However, for many units, end points are nebulous (Montgomery et al., 1995). We identified ends of units as the points where their width expressed as a fraction of water width reached 0.5. Pocket pools, Edgewater and other small units were not identified as separate units. Any unit not a pool is considered a nonpool to avoid differences of rapids, riffles, cascades and other nonpool units.
CONCLUSION
Based on this data set, we recommend ten transects per reach at a three channel width spacing for future survey and research studies. This implies a sampling scale of about 30 channel widths per reach which corresponds to a range of three to 20 pool-riffle sequences (dependent on pool spacing). Additional transects within this length will not improve the estimate because spacing closer than three widths will cause autocorrelation between transects. Channel lengths of 30 widths are generally stochastically homogeneous. Additional length only combines adjacent, inhomogeneous reaches leading to decreased measurement precision. Transect locations should not correspond to specific locations within the pool-riffle sequence because of potential cyclic tendencies from pool to pool (Knighton, 1975) on reaches with regularly spaced pools (Myers, 1996).
Stream sizes and the pool formative features of the streams studied herein affect the measurement precision. Pools formed by structural features on small streams tend to be randomly located (Myers, 1996). Regular spacing of transects randomly samples features on these streams which is essential for accurate estimates of pool area. Different recommendations may pertain to larger streams with well-developed unforced pool-riffle sequences with regularly spaced pools. For example, transect spacing that corresponds to the pool spacing may result in repeated sampling of just pools or riffles leading to inaccurate pool area estimates. Spacing transects systematically to alternately sample riffles and pools will not accurately estimate pool area if pool and riffle lengths are different.
In light of our recommendations, consideration of agency monitoring methods is instructive. BLM (1978) prescribes four transects at 30.5 m spacings and USFS (1985) prescribes five transects at 15.2 m spacings without regard to stream size. On a 2 m wide stream, the 122 m reach length and 31 m spacing of the BLM method is two and five times as long as we recommend. The BLM method samples a proper length only on streams wider than 4 m and has proper spacing on 10 m wide streams. The 61 m length of the USFS method is about correct for a 2 m stream, but the 15.2 m spacing is 2.5 times too long. Precision of each method would improve by doubling transects within the same reach.
