Sediment transport and channel adjustments associated with dam removal: Field observations

Authors


Abstract

[1] This study documents changes in channel geometry, bed level profile, and bed grain size distribution and their relations with the sediment transport at the reach scale, following the removal of a low-head dam. After the removal, net sediment deposition occurred downstream of the dam, and net erosion occurred in the reservoir, but approximately less than 1% of the sediment stored in the reservoir was transported downstream. No bank erosion was evident either upstream or downstream of the dam. Bed deposition and scouring in the reservoir accounted for a decrease in the bed slope of 30%. The stations downstream of the dam had surface bed material sizes at least 40% finer than preremoval conditions. However, the sediment transport rates downstream of the dam were not significantly different from predam to postdam removal or from an upstream control. Overall, the removal of the dam had only minor effects on the channel adjustment downstream of the dam. A simple analysis linking transport to channel geometry explains this effect.

1. Introduction

[2] Approximately 75,000 dams over 1.8 m high have been registered in the United States [Graf, 1999; Heinz Center, 2002]. More than 80% of them will be beyond the designed life span (50 years) by 2020 [Evans et al., 2000]. Removal of the aging dams has gathered national attention during the past decade. During the past 20 years over 550 dams have been removed in the United States [Granata et al., 2007] and more may be removed in the future as dams age [Doyle et al., 2005]. Part of the interest and enthusiasm in dam removal is as a restoration tool [Grant, 2001]. Many dam removal studies have focused on ecological impacts of dam removal [Shuman, 1995; Scruton et al., 1998; Bednarek, 2001; Bushaw-Newton et al., 2002; Stanley et al., 2002; Pollard and Reed, 2004; Sethi et al., 2004; Ahearn and Dahlgren, 2005; Downward and Skinner, 2005; Gillette et al., 2005; Graf, 2005; Thomson et al., 2005; Tiemann et al., 2005a, 2005b]. One of the key issues surrounding restoration is sediment transport after dam removal and has been addressed in a series of papers being published in a conference proceeding on sediment transport after dam removal [Moglen, 2005]. Other studies have also been done to compare predam and postdam removal conditions to address mechanisms of channel recovery [Bushaw-Newton et al., 2002; Doyle et al., 2003]. Flushing of bed sediments from the reservoir after the removal is the main concern [Pizzuto, 2002] given the large uncertainty of how these sediments are transported. Hart et al. [2002] documented that 15 of 15 studies reported increased bed material transport after sediment was released from reservoirs, and Doyle et al. [2002] reported an additional four studies on removals or dam failures with high sediment loading. The major factors regulating sediment transport from a reservoir are discharge rate and duration, sediment size and type, and sediment supply [Wohl and Cenderelli, 2000; Doyle et al., 2003], however, the effects of transport on channel adjustment are varied [Bushaw-Newton et al., 2002]. For dams impoundments filled with sediment, the current theories on channel adjustment suggest that after dam removal there could be incision of dam fill and widening of the river channel via headcutting of dam fill [Doyle et al., 2003], fill dispersion [Lisle et al., 1997, 2001; Pizzuto, 2002], and coarsening of fill [Bushaw-Newton et al., 2002]. Flume studies of noncohesive sediments behind a model dam showed incision of fill but narrowing of the channel after the removal [Cantelli et al., 2004; Wong et al., 2004]. Reservoirs of many low-head dams (<5 m in height) are not filled with sediment [MacBroom, 2005], however, and transport associated with these dam removals may differ. For example, Granata et al. [2007] showed minor suspended sediment loading during the removal of a low-head dam which had sediment deposits farther upstream in the impoundment but not directly behind the dam, but which may be typical of most small dam removals.

[3] The general goal of this study was to resolve uncertainties in the removal of a low-head dam where sediments accumulated far upstream in the dam impoundment but not immediately behind the dam (hence no headcutting occurred at the dam). For 10 months following the removal of the St. John Dam, repeated observations documented the changes in channel geometry and bed material relative to sediment transport by comparing a control reach to reaches in the reservoir and downstream of the dam. Specifically, the study addresses: (1) the adjustment in channel geometry from predam to postdam removal; (2) the changes in transport rates; (3) the changes in the size of the bed material; and (4) the link between sediment transport and channel adjustment.

[4] This paper is organized into six sections. Section 2 is a description of the study site, the dam removal, and the locations of the sampling stations. Section 3 explains the sampling design, methods, and data analysis. The design included predam and postdam removal surveys to determine changes in (1) bed levels of cross sections, longitudinal profiles of the bed, and reach scale elevations using digital elevation models (DEMs), (2) sediment loading rates (estimated from sediment traps and bed volumes from differenced DEMs), and (3) the size distributions of the material on the surface and subsurface of the bed and in the sediment traps. Section 4 presents the results from preremoval to postremoval showing that the reservoir degraded after the removal yet only small volume of fine sand aggraded downstream. Estimated rates of transport were low and confirmed the morphological data. Section 5 is a discussion of the results and of past studies. The size distribution of surface and subsurface sediments from upstream to downstream of the dam indicate that only fine sediments were transported downstream. An analytical equation was derived and used to explain how channel width and elevation are related to sediment transport postdam removal for this and other studies. Finally, the conclusions in section 6, restate the major themes in the paper, namely, that (1) downstream sediments are finer after the removal, (2) the bed slope in the reservoir decreased as sand deposits in the reservoir were eroded and reservoir depressions were filled, and (3) the decrease in channel slope promoted low sediment transport rates downstream.

2. Site Description

[5] The Sandusky River Watershed is a coastal watershed that drains 3637 km2 in the predominately (80%) agricultural region of north central Ohio. The St. Johns Dam was a 46 m long, 2.2 m high concrete structure located on the Sandusky River at river kilometer (RKm) 80 (Figure 1). The dam was constructed during the early 1900s for power and water supply. The drainage area at St. Johns Dam is 1974 km2, the average river width downstream of the dam is 20 m, and the slope of downstream channel is 0.06%. The reservoir impounded by the dam extended 13 km with an average width of 30 m and a channel slope of 0.01%. Total reservoir area and storage was approximately 0.38 × 106 m2, and 0.56 × 106 m3, respectively. Sediment did not fill the impoundment from the dam to 1 km upstream of the structure. Rather, sediments in the impoundment accumulated 1 to 12 km upstream of the dam and were mostly composed of sand and gravel. The average thickness of the impounded sediments was 1 m, determined using a staff to probe to the underlying bedrock in 10 locations throughout the reservoir. The volume of sediment in the impoundment was estimated as 0.2 × 106 m3. Throughout the study region the underlying material was limestone bedrock, however it was only exposed at the dam site, 50 m upstream and 50 m downstream of the dam. The St. Johns Dam was partially breached in March 2003 and removed completely in November 2003 [Granata et al., 2007]. Prior to the breach and removal, a series of seven sampling stations were established, three upstream of the dam (UP3, UP2 and UP1) and four downstream (DN1, DN2, DN3, and DN4) of the dam (Figure 1). Stations UP3, UP2 and UP1 were located at distances of −12.5 km, −2.3 km, and −50 m upstream of the dam and stations DN1, DN2, DN3 and DN4 were located at distances of 75 m, 200 m, 600 m, and 3.6 km downstream of the dam (Table 1). The criteria for choosing the stations were (1) easy access, (2) no log jams within 300 m (i.e., >10 times river width) upstream and downstream, (3) smooth cross-channel slope change, and (4) no major tributaries flowing into the river section between the farthest upstream station (UP3) and the farthest downstream station (DN4). The upstream most station (UP3) was beyond the backwater influence of the dam and therefore acted as a control to determine the natural variability of the river. Only station DN3 was located in a meandering reach; the other stations were in relatively straight runs. Table 1 lists the variables sampled and sampling dates for each station.

Figure 1.

Diagram of the study site with the location of the stations upstream and downstream of the dam. The insets show the location of the dam within the state of Ohio and the Sandusky River watershed.

Table 1. Summary of Observations at Each Station With the Dates of Sampling Noteda
Station NameSampling LocationbCross Sections, DEMsChannel Bed ProfileSurface Material SizeDates of Pit Trap LoadTrap Material Size
  • a

    Italicized dates signify predam removal, while others are postdam removal.

  • b

    Negative distances signify the distance upstream of the dam, while positive values are distances downstream of the dam.

UP 3 (control)upstream unimpounded −12.5 kmnonenoneAug 2003, Aug 20041 Oct to 11 Nov 2003, 1 Jun to 5 Jul 2004, 6 Jul to 13 Aug 2004, 14 Aug to 5 Sep 2004yes
UP 2upstream, impounded −2.3 kmNov 2003, Aug 2004Oct 1998, Nov 2003, Jul 2004Jun 2005nonenone
UP 1upstream, impounded −50 mNov 2003, Aug 200410/199811/2003 07/2004nonenonenone
DN 1downstream, 75 mNov 2003, Aug 2004noneAug 2003, Aug 20041 Oct to 11 Nov 2003, 1 Jun to 5 Jul 2004, 6 Jul to 13 Aug 2004, 14 Aug to 5 Sep 2004yes
DN 2downstream 200 mnonenoneAug 2003, Aug 20041 Oct to 11 Nov 2003, 1 Jun to 5 Jul 2004, 6 Jul to 13 Aug 2004, 14 Aug to 5 Sep 2004yes
DN 3downstream, 600 mNov 2003, Aug 2004noneAug 2003, Aug 20041 Oct to 11 Nov 2003, 1 Jun to 5 Jul 2004, 6 Jul to 13 Aug 2004, 14 Aug to 5 Sep 2004yes
DN 4downstream, 3.6 kmJun 2003, Jul 2004noneAug 2003, Aug 20041 Oct to 11 Nov 2003, 1 Jun to 5 Jul 2004, 6 Jul to 13 Aug 2004, 14 Aug to 5 Sep 2004yes

3. Methods

3.1. Cross Sections and Topographic Channel Surveys

[6] The bed elevation was surveyed at all seven stations before and after the dam removal. At each station, 2 to 6 cross sections were surveyed, spaced 2 m to 3 m apart in the downstream direction. When water level permitted, stations DN1, DN2, DN3, DN4, and UP3 were surveyed with a Trimble® 5700 GPS receiver (rover) sampled at 1 Hz. By slowly traversing the cross section perpendicular to the bank, the GPS produced a horizontal spacing between points of <5 cm. The stream banks (boundaries) and floodplains were surveyed using a SOKKIA Set5E total station. At higher water levels (i.e., in the reservoir at UP1 and UP2), the GPS was integrated with an acoustic Doppler profiler (ADP) deployed from a canoe, where the GPS measured the water surface elevation and the ADP measured the depth from the submerged ADP to the channel bed. Total water depth was determined by adding the depth from the ADP to the bed, to the water depth above the ADP, measured using a pressure sensor in the ADP housing. GPS data were averaged over a 5 s window to match the sampling frequency of the ADP (0.2 Hz). The bed elevation of the channel was computed by subtracting the total water depth from the averaged water surface elevation. Water depth was verified using a calibrated staff. A description of the principles of the ADP can be found in Rennie et al. [2002] and the SonTek user manual (SonTek, San Diego, California). GPS rover data were referenced to the National Geodetic Survey base station at Tiffin, Ohio (7 km baseline distance) using Trimble Geometrics Office® in the postkinematic mode giving a vertical accuracy of 1 cm.

[7] The ADP-GPS system was also used to measure longitudinal profiles of the thalweg elevation in the reservoir. This was done by repeatedly crossing the thalweg while navigating downstream and mapping bed elevation relative to the downstream distance. Unfortunately, elevation (GPS) data from the preremoval profile were corrupted and unusable. Thus the preremoval profile only gave depth as a function of longitudinal distance. Further, the survey could not be repeated prior to the removal. Instead, a preremoval profile was used from a total station survey done in 1998 by the Geologic Survey, Ohio Department of Natural Resources (ODNR). Bed elevations taken at select stations in 2003 verified the 1998 profile was similar to the 2003 profile. This was done by comparing the elevations and depths in the 1998 profile to elevations from cross sections taken prior to the removal and to the longitudinal depth distribution from the preremoval ADP survey.

[8] DEMs were produced at DN1, DN3, and UP1 stations. At each station, DEMs with a 0.25 m grid size were generated by Kriging interpolation in ArcGIS. All the surveyed points within the grid were averaged, and the mean represented the elevation value of that grid location. The averaged survey point density was greater than 2 points/m2, except for the UP1 preremoval survey which had 0.1 points/m2 due to the ADP sampling frequency of 0.2 Hz. The resolution of the DEMs were much higher in the cross-stream direction than in the downstream direction, which was acceptable since it was assumed that channel adjustments were more pronounced in the cross stream direction than in the longitudinal direction. The areas surrounding the DEMs were blanked by setting the elevation to zero, such that the area of interest was the same area of the river bed for each pair of DEMs. In this way, potential edge problems were eliminated [Lane et al., 2003].

[9] Sediment deposition or erosion was estimated from DEMs by differencing predam and postdam removal images and applying equation (1),

equation image
equation image

where ie is the estimated sediment storage (expressed in kg/m2), ρs is the sediment density (ρs = 2650 kg/m3), ɛ is the sediment porosity, which is estimated from equation (2) [Carling and Reader, 1982], V is the volume of deposition (m3), A is the area of deposition or erosion (m2) and D50 is the median particle size of the bed surface material. DEM sediment storage (ie) was converted to a storage rate, qe, as ieA/T, where T is the duration of the study in seconds.

3.2. Size Distribution of Bed Surface and Subsurface Material and Bed Load

[10] Bed surface material size at UP3, DN1, DN2, DN3, and DN4 were estimated using the pebble count method [Wolman, 1954] by randomly walking the cross section. The surface layer was defined by the maximum substrate size. Measurements were repeated by the same operator at the same locations at least twice before and after the dam removal to avoid bias [Wohl et al., 1996]. Subsurface materials were also collected. Enough substrates samples were collected so that the largest particle in the sample weighed no more than 5% of the total sample weight [Church et al., 1987]. Substrate maps of the river bed (<3 m horizontal resolution) were provided by the Geologic Survey of the Ohio Department of Natural Resources.

[11] Pit traps were deployed at some cross sections to simultaneously sample transport of bed material over the course of the study [Lisle, 1995; Wilcock et al., 1996]. Traps were made from PVC pipes, 0.2 m in diameter and 0.3 m deep. Pit traps have a higher trapping efficiency than the Helley-Smith sampler for materials larger than 0.8 mm [Wilcock, 2001; Sterling and Church, 2002]. The pit traps were emptied when less than 50% full, which improved the trapping efficiency because of their low width:depth aspect ratio. Sampled in this manner, pit traps do not clog or bias sampling, as do Helley-Smith samplers [Vericat et al., 2006]. Three to six pit traps were installed across the channel at three stations DN1, DN2, and UP3. However, some of them were inaccessible during high flows at DN1, and DN2, therefore only the traps at the shallow positions at each station were analyzed. The peak discharge during each rain event (see discharge description later in this section) was taken as the discharge for bed load transport [Wilcock, 2001]. All the pit traps were under water when discharge was greater than 3.4 m3/s (measured with a FlowTracker velocimeter). The particles captured in the traps were dried at a temperature under 60°C and sieved using US standard sieves. The transport rate of the material was based on sediment accumulated in the pit traps and was computed as qb (g/s) = AcΣMb/NAtT, where ΣMb is the total mass of the particles collected in the pit traps (in g), T is the duration of sampling period (in s), N is the number of traps at the station, and Ac/At is a scaling factor of the ratio of channel area, Ac to trap area, At, where At = πD2/4 and Ac = WD or the channel width multiplied by the trap diameter (D).

[12] Velocity profiles were measured just upstream of the pit trap via a handheld Sontek® FlowTracker acoustic Doppler velocimeter. Five to six velocities were measured at water depths from 13 cm to 50 cm from the surface. Relationships of u(z) vs. ln(z) were plotted for each profile, and a least squares fit applied based on equation (3) [Wilcock, 1996].

equation image

[13] Depth-averaged velocities (U) were calculated from the fitted lines and then averaged within each station. The averaged U was considered to be the mean velocity for the averaged water depth.

equation image

u* was calculated from equation (4), where u* is shear velocity (u* = (τ0/ρ)equation image); h is the water depth; κ is the von Karman constant (κ = 0.4); z0 is bed roughness elevation when u = 0, τ0 is bed shear stress, and ρ is the density of water. Because, z0 approximately equals 0.1 D84 [Whiting and Dietrich, 1990], equation (4) was written as

equation image

where D84 is the size for which 84% (by weight) of the particle distribution is finer. Finally, the Shields parameter, τ*, was calculated from equation (6) [Buffington and Montgomery, 1997],

equation image

where g is the gravitational acceleration. It was assumed that the incipient motion occurred for τ* > τc*.

[14] Discharge near station UP3 was calculated by the minimum width:depth ratio method using historic USGS gauge station records. Discharges at stations DN1 and DN2 were calculated using Manning Equation with channel cross sections [Williams, 1978] and a Manning's coefficient of n = 0.03.

4. Results

4.1. Cross Sections, Bed Profiles, and DEMs

4.1.1. Cross Sections

[15] Ten months after the removal of the St. Johns Dam, channel cross sections showed the river bed was slightly aggraded at two downstream stations and degraded in the reservoir but had no appreciable change in channel width (Figure 2). At the downstream most station, DN4, there was no substantial change in bed level from the predam to postdam removal period (Figure 2a). However, closer to the dam site, stations DN2 and DN3 showed the bed aggraded (Figures 2b and 2c). The upstream and downstream sections near the dam site, DN1 and UP1 degraded slightly (Figures 2d and 2e), while within the reservoir at, UP2, the channel degraded by nearly 0.5 m (Figure 2f).

Figure 2.

Cross-sectional channel elevation for predam and postdam removal conditions at stations (a) DN4, (b) DN3, (c) DN2, (d) DN1, (e) UP1, and (f) UP2. The predam removal is the dashed line, and the postdam removal is the solid line.

4.1.2. Channel Bed Profiles

[16] A comparison of bed level profiles from the dam site to −5 km (i.e., upstream) showed both erosion and deposition at different locations in the reservoir after the dam was removed. The bed was eroded 20 cm to 40 cm at distances of −3 km to −4 km upstream of the dam (section A, Figure 3) and by 25 cm to 40 cm at −1 km to −2.5 km upstream (section C, Figure 3). In contrast, the river bed degraded by at least 30 cm at −2.5 km to −3 km upstream (section B, Figure 3). Within, −500 m upstream of the dam (section D, Figure 3), bed level had not changed substantially. Substrate maps (not shown) indicated that sands dominated sections A, B, and D, while section C had mostly gravel. Overall, the average bed slope in the thalweg of the reservoir (from −1000 m to −4500 m upstream of the dam) decreased from 10−4 m/m to 0.07 × 10−3 m/m, a 30% decrease, excluding the pools between −2300 m and −3100 m.

Figure 3.

Profiles of the bed elevation in the thalweg of the reservoir before and after dam removal. The moving average is a five-point mean of the bed level after the removal. Sections A and C signify degraded regions, and section B indicates an aggraded region.

4.1.3. Digital Elevation Models

[17] The digital elevation models (DEMs) at three stations showed that bed level changes in the 3-D topographic surfaces were highly variable upstream to downstream and from bank to bank. Stations DN3 and DN1 showed slightly aggraded reaches with volumes changes of 9 m3 and 20 m3, respectively (Table 2 and Figures 4a and 4b). (Details of volume calculations and errors are discussed in Appendix A.) A bar formed on the left bank of DN1, 25 cm above the old bed, while downstream the channel degraded (Figure 4b). From the dam site to 50 m upstream (UP1), there was one section where the bed aggraded from bank to bank forming a riffle, and other areas near the bank that degraded (Figure 4c). When volume changes were averaged over the area of the DEM, the average nominal elevations changes were less than 5 cm.

Figure 4.

Digital elevation maps (DEMs) of the change in the elevation of the channel bed at stations (a) downstream 600 m (DN3), (b) downstream 50 m (DN 1), and (c) upstream −50 m (UP1). DEMs are oriented with downstream to the top of the image. The right bank of the channel is oriented to the right of the image (indicated by the letter R), and the arrows indicate the flow direction.

Table 2. Summary of DEM Calculations
StationSection Length, mChannel Width, mElevation Change, mmVolume Change, m3Storage Rate qe, g/s
UP1, 50 m upstream6043−40−94 ± 0.7N/A
DN1, 75 m downstream2123248.9 ± 0.30.07
DN3, 600 m downstream20293520 ± 0.50.15

[18] For DN1, the sediment transport calculated from the volume change in the DEM was i.e. = 3.5 × 103 g/m2 (based on equations 1 and 2 where D50 = 32 mm, V = 8.9 m3, A = 483 m2) giving a net sediment storage rate of qe = 7.2 × 10−2 g/s over the study period. The estimated sediment transport at DN3 was 6.2 × 103 g/m2 (D50 = 40 mm, V = 20 m3, A = 580 m2), giving a storage rate of qe = 0.15 g/s.

4.2. Bed Material Size and Transport

4.2.1. Bed and Bed Load Size Distributions

[19] After the dam removal, the size distribution of surface bed material decreased at downstream stations DN1, DN2, and DN3 (Figures 5b, 5c, and 5d). However, the size distributions for the predam and postdam removal for stations DN4 and UP3 had little change (Figures 5e and 5a). Prior to the removal, most of the stations had well sorted sediments except for a riffle on the right bank of UP3 and the bed at DN1. After the removal, all stations had well sorted sediments (Table 3). Despite the evenness in sediment sorting at the stations, for all the downstream stations, except the last one (DN4), the size of the finest fraction decreased (between 34 and 58%) as did the size of the coarsest fraction (between 33 and 55%). At these stations, the median size after the removal decreased between 43%, at DN2 and DN3, up to 60%, at DN1 (Table 3). In contrast, the control (UP3) was <9.1% finer and the station farthest downstream (DN4) was only 8.5% finer.

Figure 5.

Size distributions of surface bed material for predam and postdam removal from upstream to downstream stations. (a) UP3, (b) DN1, (c) DN2, (d) DN3, and (e) DN4.

Table 3. Sorting of Surface Particles Before and After Dam Removal
 Preremoval Size/Postremoval SizeδaPercent δ Decrease
D16, mmPercent D16 DecreaseD84, mmPercent D84 Decrease
  • a

    Here δ = equation image is standard error (bed materials are considered to be poorly sorted if δ > 2).

  • b

    Substrates were collected on the left (LB) and right bank (RB).

Upstream −12.5 km (UP1)b46.6/37.818106/9691.51/1.59−5
Upstream −12.5 km (UP1, RB)b0.28/0.34−211.4/1282.24/1.7123
Downstream 75 m (DN1)50/2158100/45552.24/1.4634
Downstream 200 m (DN2)49/235390/60331.36/1.62−19
Downstream 600 m (DN3)42/27.63482/46431.40/1.297
Downstream 3.6 km (DN4)38/42−1092/9201.56/1.485

4.2.2. Total Material Transport

[20] Generally, the median size of the materials transported into the pit traps was smaller after the removal at the two downstream stations, DN1 and DN2 (Table 4). At UP3, however, there was no difference in the median size of the materials trapped after the removal (Table 4). For all stations except the UP3 (right bank), the median size of the subsurface material was less than that of the trapped material (Table 4). The size distribution of the trapped material at UP3 remained unchanged over time, except for a coarsening of material in the July sample and a fining in the August sample (Figure 6a). This was not true of the materials trapped at DN1 and DN2 where size distributions shifted to both finer and coarser sediments after the removal (Figures 6b and 6c). Despite this temporal variability, there was a trend in fining of material for the <20% fraction at DN1 over time.

Figure 6.

Size distributions of bed material in pit traps at stations (a) UP3 (upstream −12.5 km), (b) DN1 (downstream 75 m), and (c) DN2 (downstream 200 m).

Table 4. Summary of D50 Values of Trapped Materials for Predam and Postdam Removal Compared to Bed Materials
StationTrap Material D50, mmSurface Material D50, mmSubsurface Material D50, mm
PreremovalPostremovalPreremovalPostremoval
  • a

    Substrates were collected on the left (LB) and right bank (RB). ND indicates not determined.

UP3-LBa0.50.50.550.510.9
UP3-RBaNDND68640.9
DN10.80.3803212
DN20.80.4714010

[21] Transport rates at the upstream control and stations downstream of the dam were generally low (≤0.1 g/s) and there were no clear trends in transport rate as a function of discharge or time (Figures 7a and 7b). Prior to the dam removal, transport rates were lower at the downstream stations than at the station upstream, which coincided with larger size bed materials downstream (Figure 7b). After the removal, transport rates increased at DN1 and DN2, then decreased to rates less than those of the control station (UP3). For this sampling period, maximum transport values occurred at DN2 and UP3 while minimum values occurred at DN1 (Figure 7b). The D50 of the sediment transported increased with discharge at downstream stations, but remained relatively constant with discharge at the upstream station (Figure 7c). Prior to the removal, the downstream station trapped coarser material than the control (Figure 7d). The D50 of trapped material at the upstream station remained relatively constant over time compared to downstream stations, which tended to decrease in size over time with some fluctuations (Figure 7d). Despite these fluctuations in particle sizes, the D50 values of the trapped material at the downstream stations attained the size of the material of the upstream station some 10 months after the removal.

Figure 7.

Sediment transport rates (g/s) as a function of (a) discharge (m3/s) and (b) date. D50 for trapped material (mm) as a function of (c) discharge (m3/s) and (d) date. The shaded symbols in Figures 7a and 7c are for the preremoval sampling period. The shaded area at 11/18 in Figures 7b and 7d indicates sampling period before the dam removal.

[22] The Shields number increased with discharge (Figure 8) and, when extrapolated to peak discharges, revealed that sand transport was initiated at 50 m3/s and sand mixed with gravel at >100 m3/s. For the 10 month period after the dam removal (November 2003 to September 2004), 80% of the discharge events resulted in critical shear levels too low to contribute to transport of bed material based on the Shields number (Table 5a). The remaining 20% of the time, only 12 % of the events were capable of transporting only sand and another 8% of transporting sand and gravel combined. The most extreme event in 2004 had a peak discharge of 270 m3/s with a 1.4 year return period. The average discharge of 17 m3/s (S.D. of 35 m3/s) from November 2003 to September 2004, the postremoval period, was lower than the 77 year mean of 34 m3/s (S.D. of 56 m3/s), but was within one standard deviation (p = 0.68) of the historical mean, indicating that flows were slightly skewed toward lower values (Table 5b).

Figure 8.

Shields number as a function of discharge and steam power per unit width of channel (W). The arrows at discharges of 50 m3/s and 100 m3/s signify incipient motion of sand and gravel, respectively.

Table 5a. Discharge and Transport for the Postdam Removal Period Near the Upstream Section of the Reservoir, UP3
Discharge, Q (m3/s)<5050–100>100
Shields number, τ*<0.03>0.03>0.047
Bed material transportno transportsandsand and gravel
Duration, days2352533
Percent of time80911
Table 5b. Statistics of Discharge Events for the Dam Removal Period (From November 2003 to September 2004) and for a 77-year Record (1921–1999)
 Mean ±SD Q, m3/sMedian Q, m3/sMaximum Q, m3/s
200417 ± 354270 (1.4 y)
1923–197934 ± 5617769 (77 y)

5. Discussion

5.1. Postdam Removal Changes in the Sandusky River

[23] Ten months after the low-head dam removal on Sandusky River, downstream cross sections within 600 m of the dam became aggraded with at least 40% finer bed material. Still the changes in volume of sediment and bed level downstream were minor compared to the changes in the reservoir. Channel adjustments were restricted to the river bed and accounted for a moderate decrease in bed slope. The reduced channel slope over the length of the reservoir indicated that depressions in the reservoir appeared to act as sinks for much of the sand in the reservoir. Thus, within the reservoir, sand beds were degraded while gravel beds were aggraded with materials transported from further upstream in the reservoir. Though the reservoir was a source of sediment, increased storage and the decrease in slope probably restricted the total amount of sediment moving from the reservoir to downstream reaches. Transport of material out of the reservoir was limited to a distance of no more than 3.6 km downstream of the dam, which was confirmed by both the stable cross section and unchanged surface size distributions at DN3. Digital elevation models (DEMs) also showed low transport and downstream deposition and provided more detailed information on erosion and deposition patterns over spatial scales of O(10–100 m), giving a better estimate of substantial bed level changes [Brasington et al., 2000]. However, small changes in the bed level (<5 cm) might not be detected due to the heterogeneity in river bathometry and substrate size.

[24] On the basis of discharge and shear measurements, transport in 2004 was low but within the historical mean. Despite this, transport based on cross sections, DEMs and sediment trap data showed only small amounts of sand were transported from the reservoir to downstream sections. Observations that the size of the sand at the upstream control station was the same as that in the reservoir, the subsurface material at the downstream stations were less than 30% sand (D < 2 mm), and the particles collected in the traps were 65% sands, support the notion that sediment transport during the study period was mainly fine sand. Since the transport of this material had a temporal variation independent of discharge, it was likely a result of the disturbed balance of sediment supply and transport capacity. A conservative estimate of transport of sediment from the dam to 3 km downstream of the dam is 1024 m3, which constitutes about 1% of the reservoir sediments (100 × 1024 m3/0.2 × 106 m3).

[25] Bed material load is known to have temporal variability and some researchers believed that the temporal variation in bed load is independent of variations in discharge [Gomez et al., 1989; McLean et al., 1999] and depends on the shear stress history [Monteith and Pender, 2005]. Whiting and King [2003] suggested that increasing the supply of the finer material is not likely to lead to channel aggradation because the stream has some capacity to transport more material of these sizes. However, aggradation could coincide with the increased supply of fine materials [Lane et al., 1995] and lead to bar formation downstream [Bushaw-Newton et al., 2002]. Indeed, more recent studies have shown that increased sand supplies caused channel widening and decreased bed slopes [Bartley and Rutherfurd, 2005; Curran and Wilcock, 2005; Gaeuman et al., 2005]. However, laboratory investigation have found when sand is released as a pulse from behind a dam, the channel may rapidly incise and narrow, followed by a slow widening process [Cantelli et al., 2004; Wong et al., 2004]. Therefore the initial conditions of the sediment distribution behind the dam [Cui et al., 2006] and channel stability [Bartley and Rutherfurd, 2005] are extremely important to subsequent channel adjustments after the release of a sediment pulse.

[26] After a pulse of sediment, aggradation may be transient as finer materials deposited within the downstream reaches could be entrained and transported further downstream over time [Lane et al., 1995; Wohl and Cenderelli, 2000; Cui and Parker, 2005]. In their study, Wohl and Cenderelli [2000] argued that loading rate and discharge showed no relationship because fine sediments were depleted in downstream pools after a pulsed reservoir release. They also attributed the trend of increasing sediment size with discharge to depletion of finer sediment. The fact that loading rate did not vary with discharge downstream of the dam, but remained nearly constant at the upstream station, could also indicate that fine material was depleted downstream. This would also explain why downstream transport rates approached upstream rates then subsequently diverged to lower values. Before the removal of the St. Johns Dam, the river bed below the dam had been scoured, leaving coarse gravels, probably a result of a limitation in sediment supply at DN1 to DN3 [Leopold et al., 1964]. In contrast, after the removal sand had been deposited downstream causing surface particles to become finer and the gravel bar downstream of the dam substantially aggraded as a result of deposition of fine bed materials from the reservoir. Because of the low sediment transport rates, limited supply of material, and limited extent of deposition (less than 3 km downstream), the removal did not adversely affect habitat in the downstream areas. The variations in the distribution of fine sand from the reservoir is consistent with the idea that the temporal variation in sand transport was caused by migration of bed forms and release of sand stored in pools and other depressions [Lisle et al., 2000; Wohl and Cenderelli, 2000]. Part of the reason for the low transport out of the reservoir was a result of the large depressions in the impoundment which effectively trapped material. Other reasons might be that the impoundment was not filled with sediment immediately behind the dam and the low frequency of high discharges in the months after the removal.

[27] Bed evolution in the reservoir did not show an obvious upstream migration of a headcut. In fact, headcuts in dammed sediments may not be that common [Moglen, 2005]. Headcutting has been noted in other dam removal cases [Pizzuto, 2002; Doyle et al., 2003]. Nearly one year after two dam removals, Doyle et al. [2003] found (1) channel degradation and widening followed by aggradation in the reservoir, (2) temporary or little aggradation in downstream sections, and (3) downstream transport that was greater for the reservoir with fine sediments (silt and sand) than for the reservoir with consolidated sediments (sand and gravel). In the Manatawny Creek study of Bushaw-Newton et al. [2002], no headcut was evident and while erosion and deposition occurred in different areas, there was channel incision of about 0.50 m at 3 cross-sections upstream of the dam. Bed material in the reservoir was composed of gravel (cobble-sized sediments) with patches of sandy pebbly sediments. Silt and clay deposits were limited to the margins of the reservoir. Although it is likely that transport rates in the Manatawny Creek study were low (J. Pizzuto, personal communication, 2005), sediment transport rates were not measured so there is no basis for comparison to our study, Still, observations from other dam removals for similar types of low-head dams are in stark contrast to those of this study where sediment transport and bed incision near and below the dam are low.

5.2. Linking Channel Adjustment to Sediment Transport

[28] The two most notable differences of the St. Johns Dam study, compared similar dam removals, are the redistribution of reservoir sediments, which reduced the overall bed slope, and the minor amounts of sediments transported downstream of the dam. To reconcile the St. Johns Dam observations with other studies, the following analysis of channel adjustment is proposed.

[29] The slope in the reservoir has the form S0 = −dz/dx where dz is the change in bed elevation from upstream to downstream sites, and dx is the longitudinal distance between the sites. If the bed elevations at the cross sections are denoted as zU and zD, then the channel slope is defined by the difference formula: S0 = −(zU0zD0)/Δx, where the 0 subscript indicates at time, t = 0, which is defined as the time when dewatering of the reservoir is complete (2–5 hours for most low-head dams [Granata et al., 2007]). At this time the water surface slope is the same as the bed slope which should reflect the preremoval channel slope unless extensive adjustment occurs for the entire impoundment. After a dam removal, the bed elevation changes can be represented as ST = −(zUTzDT)/Δx at sometime, t = T. On the basis of these equations, three responses are possible for the adjustment of the reservoir slope. The slope can increase, decrease, or remain the same, which can be expressed in the ratio of ST/S0. For example, the slope will increase (ST/S0 > 1) if the upstream bed is aggraded, the downstream bed is degraded, or both. The slope will decrease (ST/S0 < 1) if the upstream bed is degraded, the downstream bed is aggraded, or both. Lastly, the slope will remain the same (ST/S0 = 1) if both sites are aggraded or degraded at the same rate or there is no change in slope over time.

[30] Bed slope is related to sediment transport as, qs = (ρQS/W)3/2H−2/3D−1/2, where Q is discharge, S is bed slope, W is the channel width, H is water level, and D is sediment size [Bagnold, 1986]. The ratio qST/qSO can be used to assess changes in sediment transport following dam removal. Expanding this ratio gives

equation image

If water density, discharge, water level and sediment size remain the same after the removal, the equation reduces to

equation image

On the basis of the Manning equation, H is related to S as, equation image for a wide, rectangular channel. Evaluating the ratio of postdam to predam removal for the same flow conditions gives a ratio of unity (i.e., QT/Q0 = 1). The same is true of n and W if it is assumed they do not change, as is the case of the St. John Reservoir. Making these substitutions and rearranging the Manning ratio gives

equation image

Substituting this expression into the sediment discharge ratio produces an a relationship based only on W and S,

equation image
equation image

[31] Using this equation, the change in bed slope in the reservoir can be related to the change in sediment transport rate. For example if slope were to double, sediment transport would increase 3.2 times. If slope were to decrease by half, transport would be reduced 3.2 times. A similar approach was used by Doyle and Harbor [2003] to analyze timescales of degradation, although it is different in that the Einstein equation, not the Bagnold equation, was used and focused on knickpoint evolution. Still, their results show that the reservoir slope will increase toward some equilibrium, as the headcut moves upstream. Table 6 compares changes in the reservoir slope for preremoval and postremoval for the St. Johns Dam and three other reservoirs where both slope and transport rates could be estimated [Williams, 1977; Doyle et al., 2003]. These other studies found increased sediment transport following dam removal, which are in contrast to the low transport rates for the St. Johns Dam removal. The apparent difference can be explained by the increase in bed slope in the reservoirs for these three dam removal studies, which is consistent with a higher transport rate downstream. In contrast, the channel slope decreased in the St. Johns reservoir, which would explain the lower transport and deposition downstream of the dam (Table 6). This hypothesis fits with the idea that sand bed rivers adjust to perturbations through bed level changes such that excess sediment supply in upstream reaches (for this study in the reservoir) leads to decreased channel slope in downstream reaches [Gaeuman et al., 2005], depending on the initial conditions in the channel [Bartley and Rutherfurd, 2005] and provided transport of sediment is not limited by the supply of sediment [Wohl and Cenderelli, 2000; Curran and Wilcock, 2005].

Table 6. Comparison of Bed Slope and Sediment Transport for Dam Removal Studies
 S0, m/mST, m/mST/S0qST/qS0Reference
St. John Dam Reservoir0.1 × 10−30.07 × 10−30.70.5this study
Rockdale Dam Reservoir0.7 × 10−31 × 10−31.41.8Doyle et al. [2003]
La Valle Dam Reservoir0.7 × 10−32 × 10−32.85.7Doyle et al. [2003]
Washington Power Dam Reservoir0.6 × 10−31 × 10−31.72.5Williams [1977]

6. Conclusions

[32] The ongoing study of the St. Johns Dam removal represents one of the most complete data sets of a low-head dam removal in the United States. Available data from other dam removals suggest that post removal bed slope in the reservoirs increased and sediment transport downstream was greater. In contrast for the St. Johns Dam removal, the bed slope in the reservoir decreased which would explain the lower sediment transport and deposition downstream. Although finer sediment was transported downstream after the dam removal, the most dynamic region was upstream of the dam in the former reservoir where sand filled pools and depressions causing the reduction in channel slope. A modified form of the Bagnold equation was used to show that the decreased channel slope in the reservoir was consistent with the low export of sediment downstream. We suggest that the distribution of sediment and the channel slope upstream of the dam are keys to predicting sediment load downstream following dam removal.

Appendix A:: DEM Corrections and Errors

[33] It was assumed that the survey errors were independent and random, so a Gaussian filter (mean = 0, σ = 4; equation (A1)) was used to eliminate the local white noise.

equation image

σ = 4 was empirically selected such that every point was weighted by 34 adjacent points based on the Gaussian curve. After the Gaussian filter was applied, the locations with a smaller change in bed level were “blurred”, while the positions with larger change became more obvious and the spatial distribution was more concentrated in some specific locations instead of scattering all over the surface. However, the detailed spatial information was lost after Gaussian filtering, while the volume change was not influenced substantially.

[34] The error sources of DEM differencing are from (1) survey methods, mainly the precision of the GPS survey points and (2) the techniques used to generate DEMs and DEM comparison. The former error is the major source of error. Studies have shown that the morphological approach estimating the change of substrates volume by differencing the DEMs involves error propagation [e.g., Brasington et al., 2000]. The survey errors could be from survey pole tilt error and different particle sizes of substrates. Knowledge of the uncertainty of each data point is required for the error propagation into the DEMs of differences. The common way of knowing the point uncertainty is by a repeating measurements at the same point. However, it is nearly impossible to obtain uncertainties for a reach scale area. The alternative approach is a singular measurement of many closely spaced points across the surface. The precision of each data point is the standard deviation of all the points within the surface. This alternative approach was used in this study to compute the error associated with each survey point. All the GPS points were spatially divided into segments, where the distance between the last point of one segment and the first point of the adjacent segment was greater than 1 m. Then the points within individual segments were divided into 1 m length subsegments. It was assumed that all the points within each subsegment have the same precision. The uncertainty of the subsegment, σi, was calculated by equation (A2). Then the uncertainty of all the GPS points, σ, is the arithmetic mean of the segment uncertainties (equation (A3)),

equation image
equation image

where n is the number of points within the subsegment, (xjequation image) is the uncertainty of each individual GPS point, and m is the number of the subsegments within the entire surface. Finally, the uncertainty associated with the computation of volume change was computed by equation (A4)

equation image

where σ1 and σ2 is the uncertainty of each cell of the DEM, respectively, N is the number of cells within the surface, d is the cell size. The volume change is 8.9 ± 0.3 m3 at the downstream 75 m station, 20 ± 0.5 m3 at downstream 600 m, and −94 ± 0.7 m3 at upstream 50 m station, respectively (Table 2).

Acknowledgments

[35] The authors thank Bob Vargo and Bob Gable of the Scenic Rivers program (Ohio Department of Natural Resources-ODNR) for their logistical support during this project, Connie Livchak and Ryan Murphy (Geological Survey, ODNR) for their data on surface substrate distributions, and Michael Eberle (U.S. Geological Survey) for providing historical water data. We also thank members of our group, Jason Peterson, Matthew Nechvatal, Daniel Gillenwater, Ryan Wensink, Dale White, Ralph Greco, and Christopher Tomsic, for their help in the field. Finally, the authors are indebted to Martin Doyle, Jim Pizzuto, Laura Wildman, and Ellen Wohl for their constructive comments on earlier manuscripts. This research was supported by grant 671 from the Great Lakes Protection Fund.

Ancillary

Advertisement