detecting cumm.ative effects on headwater streams … · detecting cumm.ative effects on headwater...

15
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION AMERICAN WATER RESOURCES ASSOCIATION VOL. 34, NO.5 OCTOBER 1998 DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl Elizabeth S. Schnackenberg and Lee H. MacDonald2 ABSTRACT: This study evaluated the use of a modified pebble count procedure (Bevenger and King, 1995) to detect cumulative watershed effects on headwater streams in the Routt National For- est in north central Colorado. The 42 sample reaches were stratified by disturbance (reference or disturbed) and geologic terrane (granitic or mixed sedimentary-volcanic). Water surface slope was a significant control on the number of fine particles in the reference reaches in both terranes, and the data from the disturbed reaches were adjusted accordingly. The disturbed reaches in the granitic terrane generally had a higher percentage of fine particles, and the adjusted number of fine particles was significantly correlated with the number of road crossings. Disturbed reaches in the sedimenta- ry-volcanic (s-v) terrane generally did not have significantly more fine particles, nor were the adjusted numbers of fine particles sig- nificantly correlated with any management index. The lack of sig- nificant trends in the s-v streams is probably due to differences in weathering between the two rock types, and the location of the sample reach relative to sedimentary outcrops. Two other proce- dures were also used to assess cumulative watershed effects, with the Pfankuch channel stability rating yielding stronger and more consistent differences between the reference and the disturbed streams than the TaI"Lwell substrate ratio. We conclude that it may be difficult to define a standard reference condition, and that the number of road crossings is more strongly correlated with the num- ber of fine particles than equivalent clearcut area. (KEY TERMS: watershed management/wildland hydrology; cumu- lative watershed effects; sedimentation; stream channels; bed material particle size; forest management; roads.) activities over time and space. In many forested areas a primary concern is the cumulative effect of manage- ment activities on the designated beneficial uses of water, particularly coldwater fisheries (e.g., MacDon- aId et al., 1991). Numerous methods have been devel- oped to quantify the cumulative watershed effect of forest management activities, and these include the equivalent clearcut area (USFS, 1974), equivalent roaded area (Cobourn, 1989), and computer models such as HYSED (Silvey, 1981) and R1-WATSED (USFS, 1992). However, very few studies have directly tested these procedures with respect to water or sedi- ment yields, or related these indices to stream chan- nel condition, quality of aquatic habitat, or coldwater fish production (Reid, 1993). Cumulative watershed effects on the Routt Nation- al Forest (RNF) in north central Colorado result from multiple land use activities including timber harvest, unpaved roads, grazing, and recreation. Timber har- vest has been shown to increase annual water yields (Bosch and Hewlett, 1982; Stednick, 1996), the size of peak flows (Bates and Henry, 1928; Troendle and King, 1985), the rate of surface erosion (Everest et al., 1987), and the frequency of mass movements (Sidle et al., 1985; Swanson et al., 1987). Forest har- vest and the associated road network usually increase the amount of sediment delivered to the stream channel and basin-scale sediment yields (Everest et al., 1987; Reid, 1993). Literature reviews suggest that roads may be the primary source of sediment rather than the forest harvest per se (Everest et al., 1987; Reid, 1993). The deposition of fine sediment is of concern because this can adversely affect macroinver- tebrates and fish by filling pools and interstitial INTRODUCTION The National Environmental Policy Act requires federal land managers to assess the impact on the environment of any proposed management activity (Thatcher, 1990). Although single actions can create an adverse environmental impact, the degradation of water resources is usually a result of multiple IPaper No.97125 of the Journal of the American Water ResourcesAssociation. DiscU88ions are open until June 1,1999. 2Respectively, Hydrologist, Routt National Forest, 925 Weiss Drive, Steamboat Springs, Colorado 80487-9317; and Associate Professor, Department of Earth Resources, Colorado State University, Fort Collins, Colorado 80523 {E-Maii!MacDonald: [email protected]). 1163 JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

Upload: dinhnhi

Post on 15-Feb-2019

225 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATIONAMERICAN WATER RESOURCES ASSOCIATIONVOL. 34, NO.5 OCTOBER 1998

DETECTING CUMm.ATIVE EFFECTS ON HEADWATERSTREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl

Elizabeth S. Schnackenberg and Lee H. MacDonald2

ABSTRACT: This study evaluated the use of a modified pebblecount procedure (Bevenger and King, 1995) to detect cumulativewatershed effects on headwater streams in the Routt National For-est in north central Colorado. The 42 sample reaches were stratifiedby disturbance (reference or disturbed) and geologic terrane(granitic or mixed sedimentary-volcanic). Water surface slope was asignificant control on the number of fine particles in the referencereaches in both terranes, and the data from the disturbed reacheswere adjusted accordingly. The disturbed reaches in the graniticterrane generally had a higher percentage of fine particles, and theadjusted number of fine particles was significantly correlated withthe number of road crossings. Disturbed reaches in the sedimenta-ry-volcanic (s-v) terrane generally did not have significantly morefine particles, nor were the adjusted numbers of fine particles sig-nificantly correlated with any management index. The lack of sig-nificant trends in the s-v streams is probably due to differences inweathering between the two rock types, and the location of thesample reach relative to sedimentary outcrops. Two other proce-dures were also used to assess cumulative watershed effects, withthe Pfankuch channel stability rating yielding stronger and moreconsistent differences between the reference and the disturbedstreams than the TaI"Lwell substrate ratio. We conclude that it maybe difficult to define a standard reference condition, and that thenumber of road crossings is more strongly correlated with the num-ber of fine particles than equivalent clearcut area.(KEY TERMS: watershed management/wildland hydrology; cumu-lative watershed effects; sedimentation; stream channels; bedmaterial particle size; forest management; roads.)

activities over time and space. In many forested areasa primary concern is the cumulative effect of manage-ment activities on the designated beneficial uses ofwater, particularly coldwater fisheries (e.g., MacDon-aId et al., 1991). Numerous methods have been devel-oped to quantify the cumulative watershed effect offorest management activities, and these include theequivalent clearcut area (USFS, 1974), equivalentroaded area (Cobourn, 1989), and computer modelssuch as HYSED (Silvey, 1981) and R1-WATSED(USFS, 1992). However, very few studies have directlytested these procedures with respect to water or sedi-ment yields, or related these indices to stream chan-nel condition, quality of aquatic habitat, or coldwaterfish production (Reid, 1993).

Cumulative watershed effects on the Routt Nation-al Forest (RNF) in north central Colorado result frommultiple land use activities including timber harvest,unpaved roads, grazing, and recreation. Timber har-vest has been shown to increase annual water yields(Bosch and Hewlett, 1982; Stednick, 1996), the size ofpeak flows (Bates and Henry, 1928; Troendle andKing, 1985), the rate of surface erosion (Everest etal., 1987), and the frequency of mass movements(Sidle et al., 1985; Swanson et al., 1987). Forest har-vest and the associated road network usually increasethe amount of sediment delivered to the streamchannel and basin-scale sediment yields (Everest etal., 1987; Reid, 1993). Literature reviews suggest thatroads may be the primary source of sediment ratherthan the forest harvest per se (Everest et al., 1987;Reid, 1993). The deposition of fine sediment is ofconcern because this can adversely affect macroinver-tebrates and fish by filling pools and interstitial

INTRODUCTION

The National Environmental Policy Act requiresfederal land managers to assess the impact on theenvironment of any proposed management activity(Thatcher, 1990). Although single actions can createan adverse environmental impact, the degradationof water resources is usually a result of multiple

IPaper No.97125 of the Journal of the American Water Resources Association. DiscU88ions are open until June 1,1999.2Respectively, Hydrologist, Routt National Forest, 925 Weiss Drive, Steamboat Springs, Colorado 80487-9317; and Associate Professor,

Department of Earth Resources, Colorado State University, Fort Collins, Colorado 80523 {E-Maii!MacDonald: [email protected]).

1163 JAWRAJOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

Page 2: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Schnackenberg and MacDonald

series of conceptual and methodological issues, andrelatively little additional effort was needed to collectdata using two other channel assessment procedures.Hence this study expanded into a more comprehen-sive attempt to detect CWEs and their potential caus-es. The specific objectives were to: (1) determine iflithology had a significant effect on the number offine particles in undisturbed ("reference") reaches;(2) determine if there was a significant autocorrela-tion between sampled particles (i.e., to verify thateach particle was an independent sample); (3) evalu-ate consistency between observers; (4) determine ifone or more local controls (water surface slope,drainage area, elevation, valley slope, or amount oflarge woody debris) significantly affected the amountof fine sediment in the reference reaches; (5) evaluatethe relationship between the number of fine particlesin the disturbed ("study") reaches and three indices offorest management; (6) compare the results of themodified pebble count to two other commonly-usedprocedures to assess stream channel condition(Pfankuch channel stability rating and Tarzwell sub-strate ratio); and (7) evaluate the relationshipbetween each of the three channel assessment proce-dures and the management indices. The ultimate goalis to develop and validate one or more procedures todetect the adverse cumulative effects of managementactivities on stream channel characteristics and, byinference, aquatic ecosystems.

STUDY AREA

The Routt National Forest (RNF) encompasses4,450 km2 in north central Colorado (Figure 1), Eleva-tions range from 1,970 to 3,690 meters. The continen-tal divide runs along the Park and Gore Ranges andeffectively bisects the RNF. Annual precipitationranges from 40 cm in the lower elevations to morethan 150 cm in the higher elevations of the ParkRange. The annual runoff hydrograph is dominatedby the seasonal snowmelt peak in late Mayor June.

The RNF has two dominant geologic terranes:(1) crystalline rocks (primarily granitic) in the Parkand Gore Ranges; and (2) sedimentary-volcanic rocksin the Elkhead, Flattop, and Rabbit Ears mountainranges. Within the latter category, the volcanics typi-cally occur on the peaks and high ridgetops, and theseare underlain by sedimentary rocks such as sand-stones, siltstones, and shales.

spaces, decreasing intergravel dissolved oxygen con-centrations, and inhibiting fry emergence (MacDonaldet al., 1991; Meehan, 1991; Weaver and Fraley, 1993;Waters, 1995).

The bed material particle-size distribution isbelieved to be one of the first channel characteristicsto change in response to management activities (Diet-rich et al., 1989; Madsen 1994). Wolman pebble counts(Wolman, 1954) are the most common technique toassess changes in the surface particle-size distribu-tion and thus evaluate management effects on aquatichabitat.

There are several limitations to the use of Wolmanpebble counts to assess cumulative watershed effects(CWEs). First, measurements are usually made at achannel cross-section or within a single habitat typesuch as a rime or a pool. This does not provide anintegrated, reach-scale assessment, even though thereach is often the primary unit of concern to landmanagers. Second, the statistical procedures used forcomparing pebble counts typically focus on a singlevalue, such as the D50 or D84. Thus a 100-unit pebblecount is reduced to a single metric, and the resultingdecrease in sample size greatly weakens the statisti-cal power of the analysis. Finally, recent studies haveshown that observer bias and measurement accuracycan limit the accuracy and precision of pebble counts(Marcus et al., 1995; Wohl et al., 1996).

Bevenger and King (1995) addressed the first limi-tation by longitudinally sampling 150-300 particles atapproximately 1-m intervals in a several hundredmeter reach. The second limitation was resolved byfocussing on the total number of particles with anintermediate diameter less than eight millimeters.Each measured particle is therefore treated as onesample and this greatly increases the sample size.Comparisons over time or between streams can beanalyzed by chi-squared contingency tables (Potyondyand Hardy, 1994; Bevenger and King, 1995; Schnack-enberg, 1996).

Bevenger and King's study showed that reachesdisturbed by forest management had significantlymore fine « 8 mm) sediment than otherwise compa-rable reference reaches; the observed increase in finesediment was attributed to land use activities(Bevenger and King, 1995). A particle size of eightmillimeters was used because this is a break in phi(log 2) classes (Wolman, 1954) and is close to the 0.25-inch (6.35 mm) threshold considered detrimental tomany coldwater fish species (Chapman, 1988).

The primary goal of this study was to evaluate theuse of the Bevenger and King (henceforth referred toas B&K) pebble count procedure to assess cumulativewatershed effects in the Routt National Forest. How-ever, to accomplish this goal we had to address a

JAWRA 1164 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

Page 3: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Detecting Cumulative Effects on Headwater Streams in the Routt National Forest, Colorado

Areas in grey represent Routt National Forest boundaries. '1._- .,,1100 kn:i

Figure 1. Map of the Routt National Forest and Major Mountain Ranges in North central Colorado.The shaded area southeast of Kremmling is no longer being administered by the Routt National Forest.

METHODS

Forty-two reaches were sampled in the summer of1994. Reaches were stratified by land use and geolog-ic terrane (granitic or sedimentary-volcanic). We triedto focus on lower-gradient response reaches (Mont-gomery and Buffington, 1993, 1997), but the moun-tainous terrain meant that 20 of the 42 reaches had awater surface slope greater than two percent(WSLOPE in Table 1). The median bankfull widthwas only 4.5 m. If classified by Rosgen (1994), 14 per-cent of the reaches were type A, 31 percent type B, 38percent type C, and 17 percent type E.

In each 150-300 m reach 150 particles were sam-pled at approximately one-meter intervals whiletraversing diagonally upstream from bankfull tobankfull (Bevenger and King, 1995). Each particle

was selected by blindly placing a finger at the toe ofone's wader and picking up the first particle touched.The intermediate axis was measured with a ruler tothe nearest millimeter, and the percent of particlessmaller than eight millimeters was calculated foreach reach. To eliminate problems with observer bias,all of the data used in the primary analysis were col-lected by the same observer. However, we also evalu-ated observer variability by having two or morepeople independently sample the same reach. Forthese comparisons the sample size ranged from 50 to150 particles per observer, with a smaller sample sizeand correspondingly shorter reach being used in cold-er water.

Boxplots, summary statistics, and an F-test forequal variances were used to determine whetherbedrock lithology affected the number offine « 8 mm)

1165 JAWRAJOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

Page 4: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Schnackenberg and MacDonald

TABLE 1. Characteristics of Reference and Study Reaches in Granitic and Sedimentary-Volcanic Terranes.The five local control variables are elevation, drainage area (DAREA), water surface slope (WSLOPE), valley slope

(V SLOPE), and number of pieces of large woody debris per 300 m (LWD). Indices of management activity werethe equivalent clearcut area (ECA), number of road crossings per square kilometer (RDXING), and number ofroad segments within 60 m of a stream per square kilometer (RDTWO). Indices of stream channel conditionwere the number of particles less than eight millimeters (L8), Tarzwell substrate ratio (TSR), and Pfankuch

channel stability rating (PCSR). n/a indicates data were not available for that reach.

WSLOPE V SLOPE(%) (%)

DAREA

(hectares)

Elevation(meters) La TSR PCSRLWD ECA RDXING RDTWOStream

Granitic Reference

0.12 1.3

5.9 6.8

1.5 3.7

1.4 4.1

0.69 5.0

0.59 2.4

1.5 4.5

1.1 3.3

1.8 1.8

3.7 4.2

3.6 4.8

2.7 3.8

0

0.

0

0

0

0

0

0

0

0

0.

0

0

0

0

0

0

0

0

0.11

0.04

0

0

0

691045383575261913211728

126

34

180

221

36

141

208

38

35

80

34

42

93

53

58

77

80

59

3

13

4

21

10

23

7

2

8

8

24

46

O

1

O

O

O

O

O

O

4

O

1.2

O

271521453006264227762836293326972533284225942679

21404380

533931

1500642506

26202760121031201400

Green Cr.1AJwer Harrison Cr.S. Fork Silver Cr.Middle Fork Mad Cr.Newcomb Cr.N. Fork Service Cr.Porcupine Cr.Service Cr.S. Fork Hog Park Cr.S. Fork Mad Cr.Upper Harrison Cr.w. Fork Encampment

6549475065

Granitic Study

2.8 4.2

0.5 0.97

1.1 3.7

4.8 9.0

1.7 2.2

6.0 6.7

0.76 2.5

2.6 2.8

1.8 2.4

2.2 3.0

0.60 1.2

6.7 11

n/a11

601268

78410

47

5.3

33

21

6.4

43

4.6

21

2.3

29

2.8

4.2

11

0.50.730.961.41.93.3°0.160.612.92.01.3

0.25

0

0

0

1.4

0

0

0

0.3

3.0

1.8

0

225437454054481136527844

27

30

55

45

25

30

38

81

69

10

7

27

878583919489905679

1079161

257629702945287928242970297027522848266726552230

103010601900363948159405

1660846533

1020611

Elkhom Cr.High Rock Cr.Rock Cr.Famham Cr.Gore Cr.French Cr.Red Dirt Cr.Damfino Cr.Line Cr.Upper Little Rock Cr.Lower Little Rock Cr.Burgess Cr.

0,

0

0

0

0

0

0

0

0

0.24

0

0

0

0

0

0

0

0

373243171735364439

272420583316172737

777791697695828989

252124612497278829943042249728482770

8081080328

108017201490809493457

Sedimentary.Volcanic Reference

1.9 2.3 3 4

3.2 5.7 18 0

4.6 6.7 18 0

3.9 3.8 8 0

4.8 5.4 29 0

0.44 2.0 8 0

1.8 3.1 33 0

1.3 3.4 14 0

2.5 10 n/a 0

S. Fork Little SnakeJohnson Cr.Lopez Cr.Lower Roaring ForkLower Trout Cr.Upper Trout Cr.Oliver Cr.Upper Roaring ForkHorse Cr.

Sedimentary-Volcanic Study2.8 3.9 11

0.82 2.0 2

3.8 3.9 9

1.4 2.3 24

2.7 3.7 2

1.5 2.3 5

0.68 2.0 4

2.6 3.5 15

2.9 4.1 10

2.5

0.17

0.8

0

0.590.341.3°0.420.350.352.3°

415833462835792431

29

24

26

32

24

57

18

32

71

n!a

123

90

108

105

82

129

90

99

17603030966

11401840221014601220460

14

0

1.7

1

9.2

36

19

9.6

0.7

276424092606263625582776271524852788

1.5

1.2

1.4

2.8

0.56

Corral CroFirst CroGrizzly CroKing Solomonwwer So Fork SlaterWillow CroSnyder Cro

Mill CroN. Fork West Prong

1166 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATIONJAWRA

.16

.15

.25

22

.24

Page 5: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Detecting Cumulative Effects on Headwater Streams in the Routt National Forest, Colorado

particJes in the reference reaches. Autocorrelationanalyses were run for 1 to 16 lags in each reach todetennine if each pebble was an independent sample.Samples from different observers were comparedusing chi-square contingency tables and Fisher'sexact two-tailed test (SAS, 1990).

Because this was an exploratory study rather thana monitoring project, five local controls were evaluat-ed for their effect on the particle-size distribution.The five local control variables were elevation,drainage area, valley slope, water surface slope, andnumber of pieces of large woody debris. Elevation,drainage area, and valley slope were determined fromU.S. Geological Survey 1:24,000 topographic maps.Water surface slope was surveyed in the field usingan engineers level and stadia rod. The number ofpieces of large woody debris was normalized to areach length of 300 meters. To be counted, apiece oflarge woody debris had to be longer than half thechannel width, thicker than five centimeters at thelarge end, and influence stream dynamics duringbankfull flow. Log jams were relatively rare and werecounted as a single piece of large woody debris.

Simple and stepwise regression (SAS, 1990) wereused to determine the effect of the five local controlson particle size in the reference reaches in each ter-rane. The regression model developed from the refer-ence reaches was then used to remove the effect ofsignificant local controls on the number of fine parti-cles in the study reaches. The resulting residualswere regressed against the three management indicesto evaluate the relative strength of the relationshipbetween the adjusted number of fine particles andeach management index (Helsel and Hirsch, 1992).We used this approach rather than pairwise compar-isons because the primary objective was to evaluatethe B&K methodology.

The two other indices used to evaluate streamchannel condition were the Tarzwell substrate ratio(TSR) and the Pfankuch channel stability rating(PCSR). The TSR is intended to be a dimensionlessindex of macroinvertebrate productivity (Tarzwell,1936, 1938), although the values were originallyderived from biomass data (J. Chanat, Colorado StateUniversity, personal communication, 1995). This pro-cedure ocularly classifies stream substrates accordingto the dominant particle size, presence of sand infill,and presence of aquatic plants (Ohlander, 1994). Eachsubstrate class is assigned a coefficient that quanti-fies the standing crop of macroinvertebrates relativeto a sand substrate. The TSR value for a reach isequal to the average of 100 samples taken at two-meter intervals along the thalweg (Ohlander, 1994).

The PCSR was developed in the northern RockyMountains (Pfankuch, 1978) and is widely used in thewestern U.S. (Myers and Swanson, 1992). The PCSR

quantitatively assesses stream channel condition byevaluating 15 different attributes of the upper banks,lower banks, and channel bottom. The range of possi-ble values for each attribute varies according to anarbitrary weighting system, and the sum of theattribute values is the PCSR. Higher point totals rep-resent less stable channels, and fish biologists haveinterpreted high PCSR values as less productivereaches.

Three management indices were used to character-ize the amount of roads and forest harvest upstreamof each sample reach. The first index was the equiva-lent clearcut area (ECA), and this was originallydeveloped to predict the effect of roads and differenttypes of timber harvest on annual and monthly wateryields (USFS, 1974). Coefficients are used to convertthe area of each timber harvest, road, or certain otherground-disturbing activities into an equivalent num-ber of clearcut acres or ECAs (e.g., two acres of selec-tion cut harvested with tractors might be consideredjust under one ECA). This procedure and the analo-gous equivalent roaded area method have been widelyused by the U.S. Forest Service to evaluate water-shed-scale disturbance (e.g., Cobourn, 1989; McGurkand Fong, 1995). Hydrologic recovery curves havebeen explicitly incorporated, and this study assumeda linear 80-year recovery curve for timber harvest(Troendle and King, 1985).

The second management index was the number ofstream crossings by roads per square kilometer ofwatershed area (RDXING), while the third manage-ment index was the number of 400-m long road seg-ments within 60 m of a stream channel per squarekilometer of watershed area (RDTWO). Both of theseindices were selected because roads can generatemost of the sediment associated with forest manage-ment activities (e.g., Hafley, 1975; Megahan, 1984;Eaglin and Hubert, 1993; Waters, 1995). RDXINGand RDTWO were also relatively easy to obtain froma geographic information system that superimposedthe road network on the stream network as delineat-ed on USGS 1:24,000 topographic maps. The trans-portation layer included open and closed roads, butdid not include temporary roads or skid trails used inlogging operations.

For this study we defined reference reaches as hav-ing less than two percent ECA, and RDXING andRDTWO values of less than 0.25 per square kilome-ter. In a few cases sample reaches were located imme-diately downstream of roads and harvest units, andthese reaches were classified as study reaches eventhough the basin-scale disturbance indices were rela-tively low (Table 1). Conversely, two of the referencereaches had unexpectedly high management indices,but in each case the disturbance was primarily on theridges in the uppermost parts of the basin.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1167 JAWRA

Page 6: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Schnackenberg and MacDonald

T-tests were used to compare the mean TSR andPCSR values between the study and reference reach-es. Differences in the variance between study and ref-erence reaches were assessed by F-tests; if thevariance was significantly different, an unequal vari-ance t-test was used to evaluate the differencebetween means. Stepwise regression was used toassess whether the TSR and PCSR values could bepredicted from the three management indices.

(r > 0.80) but were still highly significant. Hence theresults are quite robust with regard to the chosenbreakpoint in particle size, and subsequent analysesused just the L8 values.

TABLE 2. Correlation Coefficients (r) and p-Values for DifferentSize ClssBes in the Granitic Reaches (n = 24) in the Upper

Half of the Table and Sedimentary- Volcanic Reaches (n = 18)in the Lower Half of the Table. L2, L4, La, and L10 refer to

the number of particles with an intermediate diametersmaller than 2, 4, 8, and 10 mm, respectively.

RESULTSL2 L4 L8 LlO

Overall Analysis 0.970.0001

0.840.0001

0.93

0.0001

L2 => r

=> p-value

The granitic reference reaches generally had fewersmall particles and significantly (p = 0.04) more vari-ability than the sedimentary-volcanic reference reach-es (Figure 2). On this basis the two terranes wereseparated for all subsequent analyses.

0.940.0001

0.910.0001

0.970.0001

lA ~r

~ p-value

0.850.0001

0.960.0001

0.970.0001

L8=>r

=> p-value

0.800.0001

0.930.0001

0.99

0.0001

LlO ~ r~ p-value

80 I -

.

,

EEa)v00Q)ut(\Ic.

"6

0;.0E:J

z

60

40

20

oL-Granitic

.

Sedimentary-

volcanic

Figure 2. Boxplots Comparing 1..8 Values for the Granitic andSedimentary-Volcanic Reference Reaches. The middle line

represents the median, while the top and bottom of thebox represent the 25th and 75th percentiles, respectively.

Whiskers show the 5th and 95th percentiles,and the solid circles indicate outliers

The autocorrelation analysis indicated that therewas a significant autocorrelation (p < 0.05) in at leastone sample reach for 13 of the 16 lags tested (Table 3).Seven of the 42 sample reaches had a significant lag-one autocorrelation (highest R2 = 0.22), and five of thereaches were significantly correlated at lag 10 (high-est R2 = 0.22). These results suggest that samplestaken one meter apart may not be completely inde-pendent, and this would violate one of the keyassumptions underlying the use of chi-squared contin-gency tables. On the other hand, the 36 significantautocorrelations is only slightly greater than the 34autocorrelations that would be expected by chance(i.e., 16 lags times 42 reaches with a level of signifi-cance of 0.05). The mean R2 for the statistically signif-icant autocorrelations ranged from 0.08 for lags oneand ten to 0.04 for most of the other lags (Table 3).These low values, plus the fact that 95 percent of thelags were not significant, led us to conclude that auto-correlation can be a potential problem, but did notsignificantly affect the results of our study.

Twenty-two pairwise comparisons between elevendifferent pairs of observers found no significant differ-ences (p < 0.10) between observers (Table 4; Figure 3).While this suggests a lack of significant observer bias,the mean difference of 4.2% and the standard devia-tion of 2.9 percent suggests that observer variabilitycould cause up to a 10 percent shift in the number offine particles.

Correlation tables indicated that the number ofparticles less than B mm in diameter (LB) was veryhighly correlated (r > 0.9; p = 0.0001) with the num-ber of particles less than 4 and 10 mm, respectively(Table 2). Correlations between LB and the number ofparticles less than 2 mm were not quite as strong

1168JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

Page 7: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Detecting Cumulative Effects on Headwater Streams in the Routt National Forest, Colorado

TABLE 3. Number, Mean, and Maximum of Statistically Significant (p < 0.05) Autocorrelations for One to Sixteen LagsNegative correlations were converted to absolute values to calculate the mean autocorrelation for each lag.

16 Total7 8 9 10 u 12 13 14 15Lag 1 2 3 4 5 6

2 2 0 2 2 2 362 1 0 0 2 2 5 4Number of Significant

Correlations

7 3

0.23 0.20 0.24 0.21 0.20 0.240.28 0.23 0.20 0.24 0.23 0.22 0.19 0.28Mean Con'elation

0.23 0.24 0.20 0.47 0.32 0.23 0.30 0.24 0.20 0.47Maximum Correlation 0.47 0.31 0.21 0.24

TABLE 4. Comparison of L8 Values Between Observers.

Numberof

Particles<8mm

Numberof

Particles<8mm

Number ofParticlesCounted

(per observer)

PercentDifferenceBetween

Observers P.ValueObserver I1 Observer .2Stream

0.25

0.62

0.63

0.68

0.21

1.00

0.13

0.26

1.00

0.36

0.29

1.00

0.72

1.00

0.18

0.34

1.00

1.00

0.88

0.21

0.50

0.80

21

21

29

15

53

81

32

42

42

28

28

36

96

11

47

60

31

31

30

39

39

49

T. FrattJ. AlmyJ. Almys. Cowmans. Cowmans. CowmanS. CowmanK. FosterS. CowmanS. CowmanL. MacDonalds. CowmanS. Coulsons. Cowmans. Cowmans. CoulsonS. CowmanC. ClapsaddleS. Cowmans. CoulsonL. Beltons. Coulson

29

25

25

12

42

80

21

52

43

35

36

35

100

11

59

51

32

30

32

50

33

46

8.0

4.0

4.0

3.0

7.3

0.7

7.3

6.7

0.7

7.0

8.0

1.0

2.7

0.0

8.0

6.0

1.0

1.0

2.0

7.3

4.0

2.0

100

100

100

100

150

150

150

150

148

100

100

100

150

50

150

150

100

100

100

150

150

150

K. WolfT

K. WolfT

T. Fratt

L. Schnackenberg

L. Schnackenberg

L. Schnackenberg

L.SchnackenbergL. Schnackenberg

L. Schnackenberg

L. Schnackenberg

L.SchnackenbergL. MacDonald

L. Schnackenberg

L.SchnackenbergL. Schnackenberg

L. Schnackenberg

L. Schnackenberg

L. Schnackenberg

C. Clapsaddle

L. Schnackenberg

L. Schnackenberg

L.Schnackenberg

Unrompahgre Cr.Unrompahgre Cr.Unrompahgre Cr.Upper South Fork Slater Cr.Upper Little Rock Cr.Lower Little Rock Cr.Mill Cr.Horse Cr.Gore Cr.High Rock Cr.High Rock Cr.High Rock Cr.Mule Cr.North Fork West Prong Cr.King Solomon Cr.First Cr.North Fork Elkhead Cr.North Fork Elkhead Cr.North Fork Elkhead Cr.Upper Trout Cr.Middle Fork Mad Cr.Burgess Cr.

Reaches in Granitic Terra ne

Stepwise regression showed that water surfaceslope was the only significant local control on LB inthe granitic reference reaches. The significance of thecorrelation was strengthened by using the naturallogof the water surface slope (LNWSLOPE) (Table 5; Fig-ure 4). The resulting regression model was:

Predicted L8 = 38.6 -16.3 * LNWSLOPE (1)

u sing the reference reach data, ninety-five percentconfidence bands were calculated for both individualpoints and expected means. The L8 values for thestudy reaches were plotted on this graph (Figure 5),and all but one of the study reaches plot on or abovethe regression line. Six of the study reaches fell out-side the mean value confidence bands developed fromthe reference reach data, while four values plotted onor above the 95 percent confidence bands for individu-al points (Figure 5).

The effect of water surface slope on the L8 valuesin the study reaches was removed by applyingthe regression model developed from the referenceThis had a R2 of 0.64 and a standard error of 3.8 per-

cent.

JAWRA1169JOURNAl Of THE AMERICAN WATER RESOURCES ASSOCIATION

Page 8: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Schnackenberg and MacDonald

-71

7/

/

T..

./

..

..~

80

60

40

20

0 ,.

~. .I~-~quar~=~:~~ ;..

.

.~ ...

~

10080200 40 60

Observer 1: number of particles <8 mm

Figure 3. Comparison of L8 Values as Determined by TwoObservers in the Same Reach With a 1:1 Reference Line.

TABLE 5. Correlation Matrix for L8 and Local Controls for Granitic Reference Reaches in the Upper Halfof the Table, and Sedimentary- Volcanic Reference Reaches in the Lower Half of the Table.

The top number in each box is the correlation coefficient, and the lower number is the p-value.Variable names are the same as in Table 1; LNWSLOPE is the naturallog of the water surface slope.

WSLOPE LNWSLOPE V SLOPE LWD1.8 ELEV DAREA

-0.570.05

0.00 10.999

0.48

0.12

-0.540.07

-0.68

0.02

-0.800.002

L8

-0.380.22

-0.170.59

-0.890.000

-0.650.02

-0.39

0.21

-0.360.34

0.600.04

0.300.35

0.320.31

-0.070.83

DAREA .0.760.02

0.550.12

0.860.0004

0.710.009

0.230.47

WSLOPE -0.50

0.17

-0.160.67

0.0690.86

N/A N/A 0.690.013

0.310.32

LNWSLOPE N/A N/A

N/A 0.140.67

V SLOPE 0.090.81

-0.11

0.78

-0.370.33

0.490.18

0.320.44

N/A 0.420.30

00.130.76

-0.110.79

0.110.79

reaches. Stepwise model selection indicated thatRDXING was the only management index significant-ly correlated with the adjusted (i.e., residual) LB val-ues (R2 = 0.61) (Table 6; Figure 6). This relationshipsuggests that road crossings have a greater effect on

the bed material particle size than the area disturbedby timber harvest.

With regard to the other two channel assessmentprocedures, the Tarzwell substrate ratios were signifi-cantly higher and more variable in the reference

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION1170JAWRA

00v{/)Q)"'(3:e:ro0.

'0'-Q)

.0E:Jc:

N'-Q)~Q){/).0O

Page 9: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Detecting Cumulative Effects on Headwater Streams in the Routt National Forest, Colorado

EE

00c:ro

.c-(/)(/)Q)

(/)Q)

U

troCoo

"-0L-Q)

-0

E~

Z

.

.

I R-square=O.~.J

-3

Figure 4. Plot of La Versus the Natural wg of the Percent Water Surface Slope (LNWSLOPE)for the Granitic Reference Reaches. The standard error of the regression is 13.1.

120

100

80

60

40

20

0

-20

-40

0-

---0.

.... ~

::!"',, .

*'-"",-."".=-

--.::: I::G..

~-

(/)Q):J(0>

co-.J-0Q)C:Q)(/)

.0o

.I::.u(0Q)L-

>--0:J

U5

-3 -2 -1 O

Naturallog of wslope

1 2

Figure 5. Plot of the La Values for the Granitic Study Reaches Versus the Natural Log of Water Surface SlopeThe 95 percent confidence bands were developed from the granitic reference reaches.

reaches than the study reaches (p < 0.02 andp < 0.001, respectively). TSR values were significantlycorrelated (p = 0.04) with both the number of roadcrossings and the number of road segments adjacentto the channel, but neither management index

explained more than 37 percent of the variability inTSR (Table 6).

Results from the Pfankuch stability ratings weregenerally similar, in that the PCSR values in the ref-erence reaches were significantly lower (i.e., better

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1171 JAWRA

Page 10: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Schnackenberg and MacDonald

significant local controls on L8 in the reference reach-es in the s-v terrane (Table 5). However, the L8 valueswere only adjusted for WSLOPE because the refer-ence reaches had significantly smaller drainage areasthan the study reaches (p < 0.03). The resultingregression model was:

stream channel condition) than in the study reaches(p = 0.001). In contrast to the TSR, there was no sig-nificant difference in the variance between the refer-ence and study reaches. The PCSR values were alsosignificantly correlated with both RDTWO and RDX-ING, but again these management indices did notexplain more than 37 percent of the variability in thePCSR values (Table 6). Neither the TSR nor thePCSR values were significantly correlated with equiv-alent clearcut area (Table 6).

Predicted L8 = 42.4 -3.36 * WSLOPE (2)

TABLE 6. Correlation Coefficients (r) and p- Values for theGranitic Study Reaches Between the Three AssessmentProcedures [L8 residuals (L8RES), TSR, and Pfankuchchannel stability rating) and the Three Management

Indices (ECA, RDXING, and RDTWO).

PCSRLaRES TSR

.0.380.23

0.100.74

0.100.75

ECA r

p-value

-0.60

0.04

0.530.08

0.780.003

RDXING r

p-value

The relationship between WSLOPE and LS was muchweaker than for the granitic reference reaches (R2 =

0.25}.In contrast to the granitic study reaches, there was

less of a tendency for the study reaches in the s-v ter-rane to plot above the confidence bands developedfrom the reference reaches (Figure 7}. Only three ofthe nine study reaches plotted on or above the 95 per-cent confidence band for mean values, while only onereach was above the 95 percent confidence band forindividual values. One reach (Mill Creek} plottedbelow the 95 percent confidence band for mean val-ues. Stepwise model selection indicated that none ofthe management indices was significantly correlatedwith the adjusted L8 values for the study reaches.

With regard to the other two channel assessmenttechniques, the mean PCSR in the study reacheswas significantly higher (p = 0.001} than the meanvalue in the reference reaches. The variance in thestudy reaches was also greater than in the referencereaches (p = 0.09}. However, none of the managementindices was significantly correlated with thePCSR values from the study reaches. There was no

.0.612

0.04

0.610.03

RDTWO 0.240.45

r

p-value

Reaches in the Sedimentary- Volcanic Terrane

Stepwise model selection chose both drainagearea (DAREA) and water surface slope (WSLOPE) as

50

40

30"~ 20-0in~ 10ro-J

0

-10

-203.52.5 320 0.5 1 1.5

RDXING

Figure 6. Plot of the Adjusted La Values for the Granitic Study Reaches Against theNumber of Road Crossings Per Square Kilometer of Watershed Area (RDXING).

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION1172JAWRA

Page 11: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Detecting Cumulative Effects on Headwater Streams in the Routt National Forest, Colorado

80

I/)0>:J

ro>

00-.J-00>~0>I/).0O..I::Um0>L-

>--0:J

ci5

0

50 1 2 3 4Water slope

.Study reach observed LB values

o Individual point confidence bands

.Mean value confidence bands

Figure 7. Plot of the La Values for the Sedimentary- Volcanic Study Reaches Versus Percent Water Surface SlopeThe 95 percent confidence bands were developed from the sedimentary-volcanic reference reaches.

significant difference in the mean TSR values fromthe study and reference reaches, and none of the man-agement indices was significantly correlated with theTSR values from the study reaches.

DISCUSSION

large-scale data collection program using theBevenger and King (1995) approach. Given the physi-cal processes that influence the particle-size distribu-tion within stream channels, one would expect someautocorrelation between closely-spaced samples. Thisauto correlation is a concern because it violates theassumption of independence between successive sam-ples (pebbles). On the RNF this potential problemmight be addressed by extending the sampling inter-val to two meters, as the streams were relativelysmall and dominated by smaller particles (medianD50 was 41 mm). On the other hand, extending thesampling interval while maintaining the sample sizewill require longer stream reaches with no change ingradient or other factors. This may be a problem intopographically diverse areas or streams with sub-stantial beaver populations.

The sudden increase in significant autocorrelationsat lags 7-12 was more surprising, and probably stemsfrom the zigzag sampling path up the stream channel.Depending on the size of the stream, the stream type,and the angle taken by the observer, one could be sys-tematically sampling sequential geomorphic unitssuch as pool-riffie sequences or the thalweg and chan-nel edges.

The known differences in particle-size distributionsfor different channel locations has led some peopleto recommend stratified sampling. Samples from

Although none of the 22 pairwise comparisonsbetween observers were significantly different, thepercent difference in the number of fine particlesranged as high as eight percent (Table 4). Our experi-ence was that careful training and close adherence toa standardized sampling methodology are essential toreduce observer variability. If pebble count data are tobe collected by more than one observer, the variabilitybetween observers must be quantified before makingany statement regarding differences over time orspace. The importance of training in decreasingobserver variability has also noted by Bevenger andKing (1995), Marcus et al. (1995), and Wohl et al.(1996). This problem of observer variability may beparticularly acute when the data are collected by dif-ferent employees with widely varying levels of knowl-edge and experience (Wohl et al., 1996).

Similarly, the potential autocorrelation betweensamples should be evaluated before embarking on a

1173 JAWRAJOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

60

40

20

Page 12: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Schnackenberg and MacDonald

different geomorphic units are then weighted by theproportion of the channel that falls in each of the dif-ferent units (Kondolf, 1997). This approach may workwell in those stream types with distinct geomorphicunits and low enough flows that each unit can be ade-quately sampled, but it implicitly assumes that onecan consistently identify distinct habitat units andaccurately estimate their respective areas in thereach under study. Both of these assumptions arequestionable (e.g., Pleus, 1995; Myers and Swanson,1997). In plane-bed or other streams where distincthabitat units are not easily identifiable, or in smallerstreams where one traverses a representative sampleof the different habitat units, the B&K method isprobably better suited to provide an estimate of thereach-scale particle-size distribution. In these situa-tions the auto correlation issue is less of a problemthan might be suggested by a simple autocorrelationanalysis.

Another issue that typically complicates the detec-tion of CWEs is the inherent variability in relativelyundisturbed stream channels. Water surface slopeaccounted for almost two-thirds of the variability inL8 values in the granitic reference reaches. The selec-tion of water slope as a significant local control is con-sistent with our understanding of the energyavailable for entrainment and sediment transport(Knighton, 1984). The importance of water surfaceslope in determining channel characteristics -partic-ularly bed material particle size -was also document-ed in a study of nearly 100 reaches in northwesternMontana (Madsen, 1994).

A stepped classification of stream type (Mont-gomery and Buffington, 1993; Rosgen, 1994) will helpaccount for the effect of gradient, but this stratifica-tion is not the statistically most effective means toremove the effect of continuous variables such aswater surface slope. Failure to account for gradient orother local controls will make it extremely difficult todefine standard reference conditions. On the otherhand, it is not practical, and probably impossible, tosample enough reaches to quantify the effect of all theimportant local controls. Thus the combination ofmeasured and unmeasured local controls will alwayscreate a range of variability within reference reaches.

This study indicates that it may also be necessaryto stratify streams by bedrock lithology to reduce thevariation within reference reaches. A comparison ofthe granitic and s-v reference reaches showed that theTSR was significantly higher (p = 0.01) and the PCSRwas significantly lower (p = 0.001) in the granitic ref-erence reaches than in the s-v reference reaches. Thevariability in L8 and TSR was also significantlydifferent between the two sets of reference reaches.Although the granitic reference reaches did have

larger (p = 0.03) and more variable (p = 0.01) drainageareas than the s-v reference reaches, this does notreadily explain the observed differences between thetwo sets of reference reaches. Thus geology can have asignificant effect on channel characteristics in undis-turbed reaches, and this effect will vary with thechannel assessment procedure being used.

Geology can also affect the channel response toanthropogenic disturbance. In the granitic terranethere were much stronger and more consistent rela-tionships between the stream channel characteristicsand the management indices. We believe that the lackof correlation in the s-v terrane between the threechannel assessment procedures (L8, TSR, and PCSR)and the three management indices, as well as thepoorer correlation between water slope and L8, islargely due to the location of the sample reaches rela-tive to volcanic outcrops. Field observations suggestthat reaches closer to volcanic outcrops had more vol-canic particles and lower amounts of fine sediment.Conversely, reaches located in or immediately down-stream of sedimentary outcrops had a finer substrate.For the sedimentary-volcanic reaches this geologiccontrol may have a greater effect on bed material par-ticle size than moderate levels of forest managementactivities. Failure to consider these differences inlithology and weathering could easily bias the resultsof comparisons between managed and unmanagedbasins (Rinne and Neary, 1996).

The observed difference in drainage area betweenthe reference and study reaches in the s-v terrane isanother complication in the identification of CWEs.One would normally expect a fining of the bed materi-al with increasing drainage area, and hence the studyreaches in the s-v terrane should have plotted evenhigher above the regression line in Figure 7. From apractical point of view, it is difficult to obtain com-pletely comparable pairs or sets of reference andstudy reaches, and this will always hinder the identi-fication of CWEs.

In terms of predicting and minimizing CWEs,equivalent clearcut area was not significantly corre-lated with any of the channel assessment procedures.On the other hand, the strong relationship betweenthe number of road crossings and the three channelassessment procedures is consistent with recent stud-ies that emphasize roads as the primary cause ofchanges in runoff and sediment production in forestedareas (Eaglin and Hubert, 1993; Madsen, 1994; Jonesand Grant, 1996). The much poorer correlation in thegranitic terrane between the adjusted L8 values andRDTWO may be due to the large variability in theeffects of nearby road segments on stream channels.Road crossings provide a direct input of sedimentfrom unpaved roads into the stream, while runoff

JAWRA 1174 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

Page 13: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Detecting Cumulative Effects on Headwater Streams in the Routt National Forest, Colorado

management impacts could strengthen the linkagebetween land-use activities and substrate changes,but it is unrealistic to expect that a single index canaccurately quantify changes in both runoff and sedi-ment production.

Regardless of the difficulties in quantifying man-agement impacts, the B&K procedure can be used todetect differences in the amount of fine sediment at areach scale. The inevitable presence of other sourcesof variability mean that this technique, like any otherchannel assessment procedure, is far from perfect,and this uncertainty must be recognized by land man-agers. The reach-scale approach of the B&K proce-dure is a useful scale for resource managers, and thefocus on bed material particle size does allow a directlink to coldwater fisheries.

With respect to the other two channel assessmentprocedures, the PCSR indicated significantly poorerchannel condition in the study reaches relative to thereference reaches in both terranes. In contrast, theTSR did not show a significant difference between thereference and study reaches in the s-v terrane, and inthe granitic terrane the differences between the refer-ence and study reaches were weaker and less consis-tent than for the PCSR. These results provide somesupport for the use of the PCSR relative to the TSR.

One limitation to the use of both TSR and PCSR isthat these have not been as directly linked to coldwa-ter fisheries as the composition of the bed material. Amore explicit evaluation of some of the specificattributes used in the PCSR procedure might providemore insight into stream channel response, and hencemore guidance for managers (Myers and Swanson,1992; MacDonald et at., 1997). Tracking individualattributes might also indicate which channel charac-teristics are more sensitive or responsive to particularmanagement activities, and provide a more directlinkage to specific designated beneficial uses.

It should also be noted that because this study wasconducted in one field season, we were not able toinvestigate the annual variability in LB or the othertwo channel assessment procedures. Knowledge of theannual variability is essential if the pebble count orother procedures are to be used to monitor changes ina reach over time. We purposefully sampled a largernumber of streams in one field season to evaluate therelationship between stream channel condition andmanagement activities, and this design also allowedus to address the variation between streams. Assess-ing change over time could avoid the issues of localcontrols and consistency between reaches, but thisthen requires either that sampling take place priorto any management activity, or that one need onlyto identify trends rather than condition on an abso-lute scale. In most cases monitoring begins aftermanagement activities have been initiated, and a

from road segments near the stream will usuallyinfiltrate downslope (Campbell and Stednick, 1983;Megahan, 1984).

The effect of timber harvest on a particular channelwill vary with the proximity to the stream, sideslopesteepness, and the amount of overland or channelizedflow. The lack of any significant correlations betweenECA and stream channel characteristics suggeststhat ECA is a poor surrogate for the changes in runoffand erosion that might then affect stream channelcharacteristics. A study in northern Idaho alsoshowed a poor correlation between ECA values andthe observed changes in flow due to forest harvest androad building (King, 1989).

One limitation inherent in all three managementindices is the problem of defining the amount ofrecov-ery over time. The ECA index explicitly includeshydrologic recovery, but the modeled reduction inECA values over time presumes successful regenera-tion. In reality, recovery rates will vary considerablywith site conditions. The effects of roads will also varyover time according to the rate of regrowth on cut andfill slopes, the amount of traffic, and the amount andtype of maintenance activities (Megahan, 1974; Reidand Dunne, 1984). Substantial revegetation on closedroads with adequate drainage may largely eliminatesoil erosion and the delivery of sediment to the streamchannels. On the other hand, a closed road with sur-face rilling and inadequate drainage may deliver asmuch sediment to the stream channel as a well-designed new road with frequent drainage onto vege-tated hill slopes.

Another problem in quantifying managementindices is the primary emphasis on timber harvestand roads. Other management activities -such asgrazing, campgrounds, water diversions, and skiresorts -affect runoff, erosion, and stream channelcondition, but these are rarely considered in mostCWE assessments. Much of the RNF was intensivelygrazed at the turn of the century, and although graz-ing rates have been reduced, the streams may still berecovering. Quantification of grazing effects is diffi-cult because of the confounding effects of wildlife anddifferences in the season, duration, and intensity ofgrazing within a basin.

Similarly, the Steamboat Springs ski area mayaffect water and sediment yields through forest clear-ing, snowmaking, soil compaction, and road and trailconstruction, and it is difficult to characterize theseeffects through a simple index such as ECA or num-ber of road crossings. The impact of the ski area issuggested by the fact that Burgess Creek, whichdrains most of the Steamboat Springs ski area, wasone of the two granitic study reaches with an adjustedL8 above the 95 percent confidence interval for indi-vidual reaches. Abetter quantification of these other

1175 JAWRAJOURNAl OF THE AMERICAN WATER RESOURCES ASSOCIATION

Page 14: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Schnackenberg and MacDonald

LITERATURE CITEDcomparative approach means that consistencybetween reaches must be addressed. The other pointis that land management decisions can rarely be puton hold until a trend is clearly established, and thisalso forces managers into a comparative approach asfollowed here.

CONCLUSIONS

The modified pebble count technique (Bevengerand King, 1995) can, in some cases, identify signifi-cant differences in the amount of fine sediment on areach scale in the Routt National Forest. The reliabil-ity and sensitivity of this procedure will be increasedby an explicit evaluation of observer variability, theeffect of local controls (such as gradient and geologicterrane), and the interannual variability. Because allthe different sources of variation in bed material par-ticle size cannot be accounted for, the range of condi-tions in the reference reaches must be defined beforedetermining whether one or more study reaches havebeen adversely affected by management activities.Hence our ability to detect change can be improved byremoving some of the variability by stratification orstatistical analyses.

While the modified pebble count procedure pro-vides quantitative information regarding changes inthe bed material size, supplemental monitoring tech-niques are recommended to address other streamcharacteristics and improve reliability. The Pfankuchchannel stability rating addresses the upper andlower banks as well as the stability of the substrate,and this appears to provide more consistent resultsthan the Tarzwell substrate ratio.

ACKNOWLEDGMENTS

The Routt National Forest provided both logistical and financialsupport for this study, and without their support this study wouldnot have been possible. Rudy King, John Potyondy, Chuck Troendle,and Ellen Wohl all contributed their time and technical expertise.Rob Sampson graciously conducted additional analyses that deep-ened our understanding of the data, while Mark Taylor and RodChimner improved the final figures. Publication costs were sup-ported by the U.S. Environmental Protection Agency under Con-tract X825780-01-0 and the U.S. Forest Service, and we aregrateful for their support. We also appreciated the relatively rapidreview and the comments of the three anonymous reviewers.

Bates, C. G. and A. J. Henry, 1928. Forest and Streamflow Experi-ment at Wagon Wheel Gap, Colorado. Monthly Weather ReviewSupp. No.3, USDA Weather Bureau, Washington, D.C., 79 pp.

Bevenger, G. S. and R. M. King, 1995. A Pebble Count Procedurefor Assessing Watershed Cumulative Effects. USDA Forest Ser-vice Rocky Mountain Range and Experiment Station ResearchPaper RM-RP-319, Fort Collins, Colorado, 17 pp.

Bosch, J. M. and J. D. Hewlett, 1982. A Review of CatchmentExperiments to Determine the Effect of Vegetation Changes onWater Yield and Evapotranspiration. Journal of Hydrology 55:3-23.

Campbell, D. H. and J. D. Stednick, 1983. Transport of Road-derived Sediment as a Function of Slope Characteristics andTime. Report from the Department of Earth Resources, ColoradoState University, to the Rocky Mountain Forest and RangeExperiment Station, Fort Collins, Colorado, 46 pp.

Chapman, D. W., 1988. Critical Review of Variables Used to DefineEffects of Fines in Redds of Large Salmonids. Transactions ofthe American Fisheries Society 117:1-21.

Cobourn, J. 1989. An Application of Cumulative Watershed Effects(CWE) Analysis on the Eldorado National Forest in California.In: Proceedings of the Symposium on Headwaters Hydrology,W. Woessner and D. F. Potts (Editors). American WaterResources Association, Bethesda, Maryland, pp. 449-460.

Dietrich, W. E., J. W. Kirchner, H. Ikeda, and F. lseya, 1989. Sedi-ment Supply and the Development of the Coarse Surface Layerin Gravel-bedded Streams. Nature 340:215-217.

Eaglin, G. S. and W. A. Hubert, 1993. Effects of Logging and Roadson Substrate and Trout in Streams of the Medicine Bow Nation-al Forest, Wyoming. North American Fisheries Management13:844-846.

Everest, F. H., R. L. Beschta, J. C. Scrivener, K. V. Koski, J. R.Sedell, and C. J. Cederholm, 1987. Fine Sediment and SalmonidProduction: A Paradox. In: Streamside Management: Forestryand Fishery Interactions, E. 0. Salo and T. W. Cundy (Editors).Contrib. No.57 , Inst. Forest Resources, Univ. Washington, Seat-tle' Washington, pp. 98-142.

Hafley, W. L., 1975. Rural Roads as a Source of Sediment Pollution-A Case Study. Watershed Management, Irrigation andDrainage Division, ASAE, New York, New York, pp. 393-405.

Helsel, D. R. and R. M. Hirsch, 1992. Statistical Methods in WaterResources. Studies in Environmental Science 49. Elsevier Sci-ence Publishers, Amsterdam, The Netherlands, 522 pp.

Jones, J. A. and G. E. Grant, 1996. Peak Flow Responses toClearcutting and Roads in Small and Large Basins, WesternCascades, Oregon. Water Resources Research 32(4):959-974.

King, J. G., 1989. Streamflow Responses to Road Building and Har-vesting: a Comparison with the Equivalent Clearcut Procedure.USDA Forest Service Research Paper INT-401, Ogden, Utah, 13

pp.Kondolf, G. M., 1997. Application of the Pebble Count: Notes on

Purpose, Methods, and Variants. Journal of the American WaterResources Association 33( 1):79-87.

Knighton, D. 1984. Fluvial Forms and Proc,esses. Edward Arnold,London, England, 218 pp.

MacDonald, L. H., A. W. Smart, and R. C. Wissmar, 1991. Monitor-ing Guidelines to Evaluate the Effects of Forestry Activities onStreams in the Pacific Northwest and Alaska. Paper No. 910/9-91-001, Region 10, U.S. Environmental Protection Agency, Seat-tle' Washington, 166 pp.

MacDonald, L. H., E. E. Wohl, and S. Madsen, 1997. Validation ofWater Yield Thresholds on the Kootenai National Forest. Finalreport submitted to the Kootenai National Forest, Libby, Mon-tana, 197 pp.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATK)N1176JAWRA

Page 15: DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS … · DETECTING CUMm.ATIVE EFFECTS ON HEADWATER STREAMS IN THE ROUTf NATIONAL FOREST, COLORADOl ... nel condition, quality of aquatic

Detecting Cumulative Effects on Headwater Streams in the Routt National Forest, Colorado

Madsen, S., 1994. Channel Response Associated with PredictedWater and Sediment Yield Increases in Northwest Montana.M.S. Thesis, Dept. of Earth Resources, Colorado StateUniversity, Fort Collins, Colorsdo, 230 pp.

Marcus, W. A., S. Ladd, J. Stock, and J. D. Stoughton, 1995. PebbleCounts and the Role of User-Dependent Bias in DocumentingSediment-Size Distributions. Water Resources Research31(10):2625-2631.

McGurk, B. J. and D. R. Fong, 1995. Equivalent Roaded Area as aMeasure of Cumulative Effect of Logging. Environmental Man-agement 19(4):609-621.

Meehan, W. R. (Editor), 1991. Influence of Forest and RangelandManagement on Salmonid Fishes and Their Habitat. Amer.Fish. Soc. Spec. Publ. 19, Bethesda, Maryland, 751 pp.

Megahan, W. F., 1974. Erosion Over Time on Severely DisturbedGranitic Soils: A Model. USDA Forest Service Research PaperINT-l56, Ogden, Utah, 14 pp.

Megahan, W. F., 1984. Road Effects and Impacts -Watershed. In:Proceedings, Forest Transportation Symposium, IntermountainForest and Range Experiment Station, Boise, Idaho, pp. 57 -97.

Megahan, W. F., J. P. Potyondy, and K. A. Seyedbagheri, 1992. BestManagement Practices and Cumulative Effects from Sedimenta-tion in the South Fork Salmon River: An Idaho Case Study. In:Watershed Management, R.A. Naiman (Editor). Springer-Ver-lag, New York, New York, pp. 401-414.

Montgomery, D. R. and J. M. Buffington, 1993. Channel Classifica-tion, Prediction of Channel Response and Assessment of Chan-nel Condition. Report TFW-SH10-93-002, WashingtonDepartment of Natural Resources, Olympia, Washington, 84 pp.

Montgomery, D. R. and J. M. Buffington, 1997. Channel-ReachMorphology in Mountain Drainage Basins. GSA Bulletin 109(5):596-611.

Myers, T. J. and S. Swanson, 1992. Variations in Stream Stabilitywith Stream Type and Livestock Bank Damage in NorthernNevada. Water Resources Bulletin 28(4):743-754.

Myers, T. J. and S. Swanson 1997. Precision of Channel Width andPool Area Measurements. Water Resources Bulletin 33(3):647-659.

Ohlander, C. A., 1994. Clean Water and T-Walk Training. RockyMountain Region, USDA Forest Service, Lakewood, Colorado.

Prankuch, D. J., 1978. Stream Reach Inventory and Channel sta-bility Evaluation. USDA Forest Service Northern Region, Mis-soula, Montana, 26 pp.

Pleus, A. E., 1995. Variability Associated With Salmon HabitatIdentification and Water Surface Area Measurements. M.S. The-sis, Evergreen State College, 130 pp., plus app.

Potyondy, J. P. and T. Hardy, 1994. Use of Pebble Counts to Evalu-ate Fine Sediment Increase in Stream Channels. WaterResources Bulletin 30(3):509-520.

Reid, L. M., 1993. Research and Cumulative Watershed Effects.USDA Forest Service General Tech. Report PSW-GTR-141,Albany California, 118 pp.

Reid, L. M. and T. Dunne, 1984. Sediment Production from ForestRoad Surfaces. Water Resources Research 20(11):1753-1761.

Rinne, J. N. and D. G. Neary, 1996. Fine Sediment and Fish: Indi-cators of Natural and Man.induced Landscape Influences. Pro-ceedings of the Sixth Federal Interagency SedimentationConference, Interagency Advisory Committee on Water Data,Washington, D.C., pp. IX-1 to IX-8.

Rosgen, D. L., 1994. A Classification of Natural Rivers. Catena22(3):169-199.

SAS, 1990. Statistical Analysis Software. SAS Institute, Inc., Cary,North Carolina.

Schnackenberg, E. S., 1996. Monitoring Cumulative EffectsThrough Modified Pebble Counts. M.S. Thesis, Department ofEarth Resources, Colorado State University, Fort Collins, Col-orado, 156 pp.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 1177 JAWRA