We examined channel response following the removal of low-head dams on two low-gradient, fine- to coarse-grained rivers in southern Wisconsin. Following removal, channels eroded large quantities of fine sediment, resulting in deposition 3–5 km downstream. At one site (Baraboo River), upstream changes were rapid and included bed degradation, minimal bank erosion, and sediment deposition on channel margins and new floodplain. Sand was transported through the former impoundment and temporarily deposited downstream. At the second site (Koshkonong River), head-cut migration governed channel adjustments. A deep, narrow channel formed downstream of the head-cut, with negligible changes upstream of the head-cut. Fluvial changes were summarized in a conceptual channel evolution model that highlighted (1) similarities between adjustments associated with dam removal and other events that lower channel base-level, and (2) the role of reservoir sediment characteristics (particle size, cohesion) in controlling the rates and mechanisms of sediment movement and channel adjustment.
 Despite being considered and practiced for several decades [Miles, 1978; American Rivers et al., 1999], dam removal has recently emerged as a critical issue in the United States because of a convergence of science, management, and policy concerns [Grant, 2001]. Recent interest in dam removal has exposed a profound lack of fundamental information on the physical, chemical, or biological impacts of removing dams [Shuman, 1995]. Given that 85% of dams in the United States will be at the end of their operational design lives by 2020 [Evans et al., 2000], and that major policy decisions are considering dam removal as a viable river management alternative [Wik, 1995; Kareiva et al., 2000], the luxury of postponing the development of a scientific basis for approaching dam removal no longer exists.
 Previous studies of dam removal have focused on dam failures, as opposed to planned removal, and thus have not been able to carefully document preremoval and postremoval conditions [Evans et al., 2000]. Numerical modeling of sediment transport and channel adjustments provides some basis for understanding dam removal [e.g., Williams, 1977], but model predictions lack long-term validation [Rathburn and Wohl, 2001]. Important issues that have not yet been quantified include the rates and mechanisms of sediment removal from reservoirs; how watershed geomorphology and hydrology affect these rates and mechanisms; how far and quickly sediment will be transported downstream; and how downstream sedimentation will affect channel morphology and biotic communities [American Society of Civil Engineers(ASCE), 1997; Shuman, 1995].
 In cases where a reservoir is mostly filled with fine sediment, the geomorphic conditions of dam removal are analogous to those of an alluvial channel responding to base-level lowering, specifically, alluvial channel incision [Schumm et al., 1984]. As such, the forms and processes associated with dam removal should be analogous to those of incising channels, and existing models of channel incision may be applicable to some dam removal cases [Doyle et al., 2002]. Further, if this analogy is accurate, then methods of managing incising channels [e.g., Shields et al., 1999] would be applicable to some dam removal cases.
 In addition, geomorphic response to dam removal serves as an ideal experiment for studying channel response to disturbance. Dam removal constitutes a discrete lowering of local base-level for upstream reaches [Simon and Darby, 1997], as well as a step increase in sediment load to downstream reaches [Howard, 1982], and thus is analogous to classic geomorphic studies of disturbance and response [Gilbert, 1917]. Dam removal also permits the luxury of knowing when and where a disturbance will occur, allowing preremoval and postremoval disturbance monitoring.
 The purpose of this study was to examine the physical changes associated with low-head dam removal. Low-head dams comprise the vast majority of dams in the United States, are the only dams being systematically removed, and represent a first step in building our understanding of the environmental impacts of dam removal prior to undertaking removal of more substantial structures [Heinz Center, 2002]. Because of the paucity of detailed geomorphic studies of any dam removal, we were interested in documenting channel adjustments, classifying the forms and processes we observed, and quantifying the dominant processes of channel change. Further, in earlier work we developed a conceptual model for predicting geomorphic response to dam removal [Doyle et al., 2002] and examined the effect of these geomorphic changes on some ecosystem processes [Stanley and Doyle, 2002]. Thus a secondary purpose of this study was to test the accuracy of our earlier conceptual model, estimate the rate at which a channel will progress through the model, and examine its applicability for predicting geomorphic response to dam removal in settings comparable to those of our study.
2. Study Sites
 The Koshkonong River at the Rockdale Dam (Figure 1) drains approximately 360 km2 in the low-relief, glaciated region of south central Wisconsin. The channel contains a mixture of sediment, ranging from silt to coarse gravel. Upstream of the reservoir, the channel is primarily sand and silt-bedded, with sporadic fine and medium gravel in the thalweg. Immediately downstream of the dam the channel is gravel and cobble-bedded, and 1.2 km downstream of the dam, then gradually changes to bed conditions like those upstream of the reservoir. The dam was located on a distinct break in river long-profile slope (Figure 2), between an upstream low-gradient section (0.0007 slope) and a downstream steeper section (0.004 slope).
 The Baraboo River at the LaValle Dam (Figure 3) drains approximately 575 km2 at the edge of the Driftless area of southwestern Wisconsin, a region of unglaciated terrain characterized by relatively steep hill slopes and high relief [Trimble, 1983]. The channel in the study reach contains a relatively uniform mixture of fine sand and silt with no gravel. Channel slope through the study reach ranges from 0.0005 upstream of the dam to 0.0002 downstream of the dam (Figure 2). The primary land use in both basins is agriculture, and the hydrology of the region is dominated by thunderstorm frontal systems, resulting in relatively flashy hydrology, although seasonal flooding is common in spring (March–May) due to snowmelt and rain-on-snow events.
 The Rockdale Dam on the Koshkonong River was originally constructed of rock and timber in 1848 to power a grain mill and was modified to a fixed-crest stone and concrete dam in 1887. At the start of the study in August 2000, the Rockdale Dam was 3.3 m high and created an impoundment of approximately 42 ha. The impoundment was approximately 2–3 m deep during the 1960s (personal communication with local landowners, 2000). In August 2000, maximum depth had declined to 1.5 m and average depth was 0.6 m, indicating progressive infilling at a rate of approximately 0.04 m/yr. The pattern of sediment deposition was characterized by fine sediment (silt and clay) throughout most of the reservoir area and sand at the upstream end of the reservoir in the form of a prograding delta.
 The Rockdale Dam was breached to grade on 12 September 2000. The remainder of the structure (i.e., the lateral portion) was removed during late June 2001. No other attempts were made to stabilize the impoundment sediment via structure through August 2001.
 The LaValle Dam on the Baraboo River (Figure 3) was constructed in the mid-1800s to power a grain mill. The impoundment was greater than 2 m deep in the early 1900s, but sedimentation decreased the depth of the impoundment to less than 1 m by the 1970s (personal communication with local landowners, 2000). The impoundment was subsequently dewatered by opening gates in the dam every 5–10 years to flush some of the stored sediment and/or to make repairs on the dam, although the majority of the sediment remained in the impoundment.
 The LaValle Dam reservoir was dewatered by opening the gates on 11 July 2000, and the remaining structure was removed during February 2001. Riprap was placed at the former dam site to create a long, steep riffle that acted as a grade control. Like the Koshkonong River, no other attempts were made to stabilize the impoundment sediment through August 2001.
 Because breaching on the Koshkonong and dewatering on the Baraboo represented the dominant disturbances of the river systems, we will hereinafter refer to these events as “removals.” Both channels were allowed to adjust to the dam removals through the summer of 2001 without any anthropogenic modifications, after which some physical manipulations were begun (channel realignment, widening, and stabilization). This study encompasses only the natural channel adjustments that occurred during the first year following removal.
 A series of cross sections were surveyed from June 2000 through August 2001 at both sites within reaches extending approximately 4–5 km upstream and downstream of the dams (Figures 1 and 3). All cross sections are referred to by their distance relative to the dam, with upstream distances designated as negative and downstream distances designated as positive. Cross sections were spaced more densely close to the dams where we expected channel form changes to be most pronounced. Cross sections were located to include both pools and riffles to test for possible variability in response to dam removal. Each cross section was surveyed at least once before and after dam removal, with most being surveyed at least twice after removal. We used the channel surveys and standard modeling analysis [Hydrologic Engineering Center, 1995] to quantify the hydraulic conditions of the channels throughout the study.
 At several cross sections upstream and downstream of each dam, we established sampling stations for water discharge measurement and suspended and bed load sediment collection. At each sampling station, a DH-48 sampler was used to collect a minimum of 10 depth-integrated suspended sediment samples, which were evenly spaced across the wetted channel following the protocol of Edwards and Glysson . Samples were composited and then filtered onto preweighed 1.0-μm filters, dried for >48 hrs at 60°C, and weighed to determine total suspended solids (TSS) concentration. Also at each sampling station, a composited bed load sample was collected by sampling at least 20 points across the wetted channel using a 76.2 × 76.2 mm Helley-Smith sampler [Edwards and Glysson, 1999]. Bulk bed material samples were collected at the sampling stations by collecting samples of the top 10–15 cm of the bed substrate from three points across the channel and compositing these three samples. Composited bed material samples greatly exceeded the minimum sample size criteria for the resident sediment [Ferguson and Paola, 1997], with the exception of station 380 on the Koshkonong River, where a pebble count (n > 100) was used due to the size of the material [Wolman, 1954]. Following floods, sediment deposits along the channel margin or on the floodplain were collected for grain size analysis. Automated water samplers (Isco samplers) were used intermittently at sampling stations to collect 1-L point TSS samples at a greater temporal resolution. Intakes of the ISCO samplers were positioned approximately 1 m from the toe of the channel bank and off of the channel bed so as not to collect bed sediment. Point samples for TSS determination were also collected at several sampling stations throughout the study period. Sediment cores were collected from the reservoirs using a 13-cm diameter core. Descriptions of core material were recorded in the field, noting the size of material, and color of sediment as well as the presence of organic debris. Subsamples of the cores were retained for laboratory particle size analysis. Reservoir stratigraphy and survey information were used to estimate the volume of sediment deposited since dam construction.
 Following dam removal, aerial photographs were taken on the Koshkonong River on 9 May 2001. We also visited several sites above and below the dam on at least a bimonthly basis, and visual estimates were made of the location of any prominent changes in channel morphology. In particular, the formation and location of head-cuts were recorded on maps.
4.1. Predam Conditions
 Prior to dam removal, both reservoirs were sediment sinks and had accumulated approximately 140,100 and 287,000 m3 of sediment in the Baraboo and Koshkonong River reservoirs, respectively (Table 1). Within the Koshkonong River reservoir the accumulated fine sediment was approximately 2 m in thickness from the dam to station −680, and approximately 1 m thick from station −680 to the prograding delta (approximately at −1500 m). In the downstream 1000 m of the reservoir the sediment was highly consolidated. There was a distinct difference between the fine surface sediment (36% sand, 45% silt, 19% clay, based on hydrometer analysis) and the underlying coarser sediment (gravel with d50 > 3 mm) (Table 2). This underlying gravel layer was interpreted as the predam channel substrate [Stofleth et al., 2001]. At all stations upstream of the reservoir the channel was covered with approximately 10–20 cm of fine sediment, which was comparable to the unconsolidated sediment on the surface of the lower reservoir.
Table 1. Sediment Stored and Exported From Reservoirs on the Baraboo River and Koshkonong River, Wisconsina
Surface Area of Reservoir, ha
Volume of Sediment Stored in Reservoir, m3
Volume of Sediment Eroded Between Removal and May 2001,b m3
Volume of Sediment Eroded From Reservoir Between May 2001 and August 2001,b m3
Reservoir on Koshkonong River extends from dam to section −2040. Reservoir on Baraboo River extends from dam to section −1030.
Number in parentheses is the value as a percentage of total stored sediment.
LaValle Dam was removed on 11 July 2000, and Rockdale Dam was removed on 12 September 2000.
By August 2000, 8200 m3 (5.9%) were eroded.
Does not include 109,000 m3 of coarse sediment in delta.
 On the Baraboo River the accumulated fine sediment varied in thickness from 1.4 to 1.7 m throughout the reservoir and was less consolidated compared with the Koshkonong River reservoir. This fine sediment (46% sand, 41% silt, 13% clay) was underlain by a fine sand layer (d50 = 0.21 mm) (Table 2) that was interpreted to be the predam channel substrate. At all stations upstream of the reservoir the channel was covered by approximately 20–40 cm of fine sediment, which was comparable to the fine sediment on the surface of the reservoir.
 Ratios of total suspended solid (TSS) concentration at a station to TSS concentration at the most upstream sampling station were consistently less than or near unity, indicating sediment retention in both reservoirs prior to removal (Figure 4). TSS values prior to dam removal were high in late May and early June, 2000, during out-of-bank flooding on both rivers. At the Koshkonong River, TSS concentration decreased downstream in comparison with sediment entering the study reaches (Figure 4). The limited data on the Baraboo River do not allow the same determination. Flooding prior to removal caused fine-sediment deposition along the floodplains at both sites; 3–5 cm of deposition were measured on the floodplain upstream of the impoundments at stations −4180 and −2860 on the Koshkonong River and at stations −5390 and −2330 on the Baraboo River. No fine sediment deposition was noted downstream of either dam after the flood. Other rivers in the region experienced approximately 10-year floods during this time period (simple recurrence interval analysis [Leopold et al., 1964] of annual peak flows of nearby stream gages with at least 50 years of data), and we expect that our study sites experienced floods of comparable recurrence interval. Suspended sediment samples were collected at both sites during high flows, although not at the peak stage of the flood event (Table 3). In addition to fine-sediment deposition, patches of sand (∼1 m2 × 10 cm deep, Table 2) were deposited on the channel-floodplain boundary at station −2860 and other meander bends in this area on the Koshkonong River. These patches indicated that some fine sand was transported during the flood, although the volume was relatively small. Only fine sediment was deposited on the Baraboo River channel margins and floodplain.
Table 3. Water and Sediment Discharge Characteristics of the Baraboo River and Koshkonong River
Bank-full Discharge, m3/s
Peak Discharge Sampled for TSS, m3/s
Peak Preremoval TSS, g/L
Peak Postremoval TSS, g/L
Ratio of Peak Postremoval to Peak Preremoval Sediment Concentration
This discharge was not the peak discharge of the flood event.
Collected on 1 June 1 2000 at station −5390.
Collected during dam removal on 11July 2000 at station 160.
Collected 2 June 2000 at station −2860 and 11 September 2000 at station −4180.
Collected during dam removal on 12 September 2001 at station 380.
 Removal of both dams resulted in (1) the export of fine sediment from the reservoir as the reservoir dewatered (Table 3, Figure 5), and (2) the conversion of the reservoir from a sediment sink to a sediment source (Figure 4). Downstream sediment concentrations immediately following removal exceeded the greatest concentrations recorded during the preremoval June flooding by factors of 10 and 4 for the Koshkonong and Baraboo, respectively (Table 3). Subsequent storm events remobilized fine sediment, but not to the degree observed immediately following removal (Figure 4).
 Postremoval TSS concentrations were nearly always greater downstream of the dam than upstream, indicating a net export of fine sediment from the reservoirs (Figure 4). TSS concentrations in the Koshkonong River were generally greatest immediately below the dam and then decreased with distance downstream (Figure 4), suggesting net deposition of fine sediment along the channel.
4.3. Upstream Channel Morphology Changes
4.3.1. Baraboo River
 In the Baraboo River the transport of unconsolidated sediment out of the preexisting channel controlled sediment export from the reservoir as well as initial postremoval channel development (Figure 6). Export of sediment increased depth and cross-sectional area within and upstream of the reservoir (Figure 7). Following this initial flushing, the channel incised vertically throughout the length of the reservoir within a period of weeks (Figure 7). Incision of the channel bed occurred at all upstream sites, exposing the underlying sand (Table 2). Bed degradation, coupled with a lowering of the water surface, caused deep-seated rotational slumping of channel banks near cross sections closer to the dam (Figure 3). However, such mass-wasting was rare, and most of the slumped material remained in place and became partially covered by subsequent sediment deposition, resulting in little or no change of channel width throughout the study reach.
 Erosion at the upstream end of the reservoir exposed coarse sediment (sand) on the bed, which was then available for transport downstream. Sand eroded from upstream was deposited within the reservoir along the channel margins and on the previous reservoir sediment surface (Figures 3 and 6). Thus much of the channel underwent a sequence of bed degradation followed by deposition on the channel margins (e.g., section −1020) over a period of 13 months following removal.
4.3.2. Koshkonong River
 Initial fine sediment export from the Koshkonong Reservoir was substantial (Figure 5) but had little effect on channel formation, essentially lowering the reservoir sediment surface but not forming a channel. Instead, channel development was initiated by a head-cut, which formed within a day of dam removal and began migrating upstream at approximately 10 m per hour over the first 24 hours following removal (Figure 1). However, this rate of migration decreased dramatically during the ensuing week, and over the course of 11 months, the head-cut migrated approximately 40 m per month (Figure 8). The reservoir sediment surface upstream of the head-cut remained largely undisturbed after initial fine sediment removal, while the channel downstream of the head-cut changed substantially (Figures 9 and 10).
 Upstream of the head-cut, boundary shear stresses calculated for a flow of 2.7 m3/s (the mean annual flow and the flow at which active head-cut migration was observed) were insufficient to erode either the fine, cohesive sediment (τcrit = 8 N/m2; Chow ) or the coarser underlying sediment (τcrit = 9 N/m2; Shields relation with D50 = 12 mm and τ* = 0.047 [Buffington and Montgomery, 1997]) (Figure 11). Further, boundary shear stresses during bank-full discharge never exceeded 7 N/m2 and thus were also insufficient to erode the reservoir sediment. In contrast, boundary shear stresses at and below the head-cut were sufficient for erosion of the fine and coarse sediment at the mean annual flow (Figure 11).
 Indeed, these high boundary shear stresses downstream of the head-cut induced substantial incision into the reservoir sediment. Initial incision was primarily through fine sediment, but coarser sediment was exposed after approximately 2 m of incision in the May 2001 survey. This coarser sediment was comparable to that of the bed material downstream of the dam and upstream of the reservoir (Table 2). As the head-cut progressed upstream, coarse material eroded from the region near the head-cut was deposited at the downstream end of the reservoir, resulting in aggradation of the channel bed between May and August, 2001 (e.g., section −40 on Figure 9).
 Incision near the former dam site led to mass-wasting of the banks and channel widening (Figure 1). However, widening was limited by the combination of highly cohesive and consolidated bank material, groundwater lowering, and later aggradation of the channel bed. Incision at station −400 formed a narrow channel with steep, almost vertical banks, although the depth of incision was not sufficient to induce channel widening.
 Like the Baraboo River, some sand was deposited on the margins of the newly forming channel and on the reservoir sediment surface. However, the deposits on the Koshkonong Reservoir sediment surface were never greater than 2–3 cm in depth.
4.4. Downstream Channel Morphology Changes
4.4.1. Baraboo River
 Initial changes in downstream channel morphology immediately following dam removal in the Baraboo River were limited to deposition of fine sediment along channel margins and in backwater regions (e.g., inside of meander bends, near debris jams). However, as sand was eroded from the upper part of the reservoir and moved through the reservoir into downstream reaches, changes in channel morphology became more evident (Figures 6 and 7). A large slug of sand was deposited immediately downstream of the dam and mostly upstream of cross section 160, between 1 and 10 months following dam removal (Figures 3, 6, and 7), decreasing channel depth by ∼20%. The sand was deposited only temporarily, and depth and cross-sectional area returned to near preremoval magnitude within 3 months. In contrast, deposition increased the size of the point bar at section 620, and while some of this deposition was only temporary, sand deposited on the upper point bar remained (Figure 6). Deposition was most evident at station 5300, where sand gradually accumulated on a point bar, redirecting the channel thalweg closer to the outside bank (Figures 6 and 7).
4.4.2. Koshkonong River
 There was very little in-channel deposition downstream of the Rockdale dam, and all of this deposition was of fine sediment (Figure 1). However, vegetation colonized some sediment deposits, which were not eroded by subsequent flows (Figure 1). In particular, persistent deposition on the right bank at stations 1720 and 2600, coupled with rapid vegetation establishment, narrowed the channel, which in turn caused deepening of the thalweg (Figures 1 and 9).
5.1. Upstream Channel Formation and Evolution
 While the geomorphic forms and processes associated explicitly with dam removal have been unstudied until recently, there is a wealth of information on how channel form responds to base-level lowering, the conditions created by dam removal. Such channel adjustments have been well documented and synthesized in channel evolution models [e.g., Schumm et al., 1984; Simon, 1989; Simon and Hupp, 1992], and we hypothesized in earlier work that channel changes following dam removal would be consistent with previous channel evolution models [Doyle et al., 2002]. Indeed, the forms and processes we observed at both field sites were largely consistent with those developed for incising channels [e.g., Simon, 1989] and those we hypothesized earlier (Figure 12). This conceptual model is likely to be applicable to other dam removals in fine-grained, alluvial channel systems with reservoirs that have accumulated a significant amount of fine sediment. We have changed some of the stages previously used in other evolution models [Schumm et al., 1984; Simon and Hupp, 1992] to fit the dam removal context and our observations. Further, some stages were included due to their potential ecological significance [Stanley and Doyle, 2002] rather than because of significant geomorphic change.
 In our model, stage A represents existing conditions, wherein the dam is in place creating backwater conditions upstream, (i.e., high water elevation but low velocity) and the reservoir has accumulated a large amount of fine sediment. Stage B represents conditions immediately following removal, when the water surface elevation decreases dramatically but the reservoir sediment surface remains undisturbed. Channel flow during this stage is wide and shallow and has relatively low velocity. Degradation characterizes stage C, wherein the channel bed incises, concentrating flow into a narrow, deep channel with steep banks and high flow velocity. A head-cut will often be the upstream terminus of channel adjustments. Large amounts of fine sediment are exported from the reservoir during this stage, and if degradation progresses to underlying coarse sediment, then coarse sediment may also become mobilized and transported downstream.
 If incision continues beyond the critical bank height [Thorne, 1999] of the reservoir sediment, then channel widening begins via mass-wasting of banks and marks the beginning of stage D. Large amounts of fine sediment are exported due to bank erosion, and there is continued transport of both fine and coarse sediment from the channel bed. Channel depths in excess of the critical bank height cause widening to continue during stage E, although sediment derived from upstream fluvial erosion begins to be deposited as the local energy slope is reduced via vertical and lateral channel adjustments. The sediment deposited is the coarsest fraction of sediment derived from upstream, as finer sediment is transported through and out of the reach. Significant reduction of bank heights by channel bed aggradation, establishment of vegetation, and reduction in groundwater elevation within the reservoir cause bank erosion to decrease, and equilibrium conditions develop, stage F.
 The forms and processes described in the six stages may not all appear at every dam removal site. For instance, if sediment is unconsolidated and easily eroded, then stage C may be absent because any degradation of the channel bed will induce widening as well, causing a direct transition from stage B to stage D. Also, the aggradation described in stages E and F may be temporary if upstream sediment moves through the reservoir as a wave [Lisle et al., 2001].
 Although simply describing the forms and processes likely to occur following dam removal is useful, it is also important to estimate how local environmental conditions determine the rate of progression through the stages and the magnitude of changes that can be expected. The dominant control on channel form evolution at our sites was the character (i.e., size, cohesiveness, and consolidation) of sediment stored in the reservoirs, particularly if sediment character led to head-cut formation and migration rather than fluvial erosion of reservoir sediment. Channel evolution on the Baraboo River was rapid, as degradation along the upstream reach occurred immediately following removal. In terms of the conceptual model, the transition from stage B to stage C on the Baraboo was immediate. Even though reservoir sediment was fine, sediment within the predam channel did not become consolidated because of regular reservoir dewaterings and was easily eroded. Further, channel morphology immediately following dam removal appeared to be very close to the predam channel, and thus there were few bank adjustments in the months following dam removal (e.g., sections −460 and −1020). In subsequent months (the winter and spring of 2001), high flows mobilized sand in the upper reaches of the reservoir, and while some of this sediment was deposited on channel margins and the newly forming floodplain, most was transported downstream of the dam. In terms of the conceptual model the transition from stage A to stage D on the Baraboo River occurred within a period of days to weeks following dam removal. It currently appears that aggradation will not occur (i.e., stage E is absent), so the current channel morphology may represent the long-term equilibrium channel morphology.
 The consolidation of fine sediment in the reservoir on the Koshkonong River caused drastically different rates of channel evolution compared with the Baraboo River. Because boundary shear stresses upstream of the head-cut were insufficient to mobilize the reservoir sediment, channel evolution rates were governed by the rate of head-cut migration, which in our case averaged 40 m per month. Slow migration of the head-cut on the Koshkonong River resulted in a period of months between stages B and C at sections in the lower reservoir, and after a year, most of the reservoir remained in stage B because the head-cut had yet to reach the upper reservoir.
 Slow head-cut migration rate on the Koshkonong River also limited progression through latter stages of the model, as only the most downstream station in the reservoir progressed through stage E of the model. The transition from stage D to E requires reduction in the energy slope via degradation upstream, aggradation downstream, or both. If coarse material is available for transport, then aggradation in downstream reaches can greatly enhance channel recovery [Simon, 1992] by reducing the local energy grade and reducing bank heights. In a reservoir, coarse material is primarily stored in the upstream end of the reservoir. During the time that the head-cut remains downstream of the delta, this coarse sediment will not be available for transport, leaving sediment in deeper strata as the only possible source of material for aggradation. Such conditions, then, would necessitate greater reservoir degradation than if coarse material were transported from the delta to downstream aggrading reaches. In all, if channel evolution is controlled by head-cut migration because of reservoir sediment conditions, as was the case on the Koshkonong River, then the transition from stage D to stage E will take longer and require a greater degree of bed degradation.
 In addition to vertical channel adjustments, lateral adjustments are also highly sensitive to the reservoir sediment conditions and the rate of migration of head-cuts. The degree of degradation before the onset of widening, i.e., the transition from stage C to D, is a function of bank sediment characteristics and sediment saturation conditions [Simon et al., 1999]. Simon and Darby  showed that channels with sand banks respond to disturbance primarily through widening with limited incision, while channels with clay banks respond primarily through incision and very little widening. Bank saturation is also important in that saturated banks are much less stable than unsaturated banks [Thorne, 1999; Simon et al., 1999]. Thus the magnitude of incision before bank failure begins will be less for rapidly migrating head-cuts because bed degradation will occur when reservoir sediment is saturated and more vulnerable to mass-wasting [Simon et al., 1999]. Slower head-cuts will cause bed degradation when sediment has been dewatered and thus will result in a deeper and narrower channel.
5.2. Downstream Channel Morphology
 Changes in channel morphology are driven by transport and storage of the coarser material in a system [Leopold, 1992]. In our study sites, coarser material was in the sand and gravel range and stored in the deltas or buried strata of the reservoirs. The head-cut on the Koshkonong River restricted access to coarse sediment, and there were correspondingly few changes to downstream channel morphology during our study. Presumably, if the head-cut on the Koshkonong River continues to migrate approximately 1.3 m/d, then the coarse delta sediment will not be accessed until the winter of 2004. Until then, channel changes downstream of the dam will not occur, or will only be small, unless sufficient quantities of sediment are eroded from the underlying strata of the reservoir. In contrast, material eroded from the delta on the Baraboo River was rapidly transported through the reservoir and into downstream reaches, inducing clear changes in channel morphology. Thus the variation of upstream erosion processes at our sites, as controlled by reservoir sediment characteristics, exerted a strong control on the rate and magnitude of downstream channel adjustments.
 In all dam removal cases, there will be a time lag between when a dam is removed and when downstream changes begin to occur. Other studies in disturbed channels have shown similar lags between the time of a channel disturbance and the time of downstream channel response and have attributed this time lag to the rate of upstream channel adjustments [Ritter et al., 1999]. At dam removal sites this time lag will be a function of the rate of reservoir sediment erosion, the thickness of the fine sediment layer, and the length of the reservoir. In cases like the Baraboo River where reservoir sediment is easily transported, erosion may occur throughout the reservoir immediately following dam removal, and so the time between removal and downstream changes will be relatively short, i.e., of the order of weeks or months [Wohl and Cenderelli, 2000]. In contrast, if reservoir sediment is consolidated, or so coarse that it is mobilized only by infrequent flows, then the time between dam removal and downstream geomorphic changes may be of the order of years.
 Once sediment is delivered to downstream reaches, its effect on channel morphology can vary greatly in terms of magnitude and duration [Doyle et al., 2002]. Several studies of sediment from dam removals or reservoir sediment releases have shown sediment routed through downstream reaches with little or no long-term impact on channel morphology [Simons and Simons, 1991; Wohl and Cenderelli, 2000; Stanley et al., 2002], and this is what appears to be happening at the Baraboo River. However, geomorphic analogies of dam removal (e.g., landslides) suggest that the infusion of large quantities of sediment to a channel can induce substantial geomorphic adjustments [Madej and Ozaki, 1996] and significant ecological changes [Stanley et al., 2002]. How such sediment inputs are routed through channels is poorly understood, as is their potential impact on channel morphology [Lisle et al., 2001].
5.3. Implications for Dam Removal
 The match between observed channel adjustments following dam removal and available channel evolution models suggests that current understanding of incised channel processes can be adapted and applied to managing reservoir sediment following dam removals in cases where river and reservoir characteristics are comparable to those at our study sites. For example, bank stability algorithms developed from studies of incising channels [Simon, 1989; Simon et al., 1999] were applied to the Koshkonong River and accurately predicted critical heights for the channel developing in the former Koshkonong Reservoir [Doyle, 2002]. Our observations of channel evolution at the Koshkonong River site, along with these modeling results, suggest that staged drawdown of a reservoir [e.g., Harbor, 1993] and establishing vegetation following dam removal can reduce the quantity of sediment eroded from a reservoir. The long-term accuracy of these predictions and their applicability to other river and reservoir characteristics remain significant unknowns.
 Comparisons with incised channels also suggest that dynamic simulation models currently developed for incising channels [Langendoen, 2000] may be more appropriate for dam removal impact analysis than models that only account for bed elevation adjustments (i.e., one-dimensional versus two-dimensional), as width adjustments may comprise a significant mode of adjustment [Simon, 1992; Simon and Darby, 1997]. In addition, sediment stabilization techniques currently applied to incising channels may be used for stabilizing channels that develop in reservoirs following dam removal. For example, engineering projects on incising channels have often focused on accelerating the natural processes of channel evolution and recovery rather than attempting to limit further erosion [Shields et al., 1999], that is, actively promoting channel development from stage C to stage F (Figure 12) rather than attempting to structurally stabilize the channel at stage C.
 Using current understanding of geomorphic processes that are analogous to those conditions surrounding dam removal will promote more effective management of upcoming dam removals. However, channel response to dam removal may vary considerably as a function of channel characteristics, particularly the character of channel and reservoir sediment [Pizzuto, 2002]. Thus lessons learned from our study, and particularly the analogy of the channel evolution model, may not be applicable to sites with vastly different characteristics. These analogies must eventually be replaced by models tested and calibrated from a sufficient number of empirical studies of dam removals, conducted in various physiographic regions and from various sizes of removed dams.
6. Summary and Conclusions
 Removal of two low-head dams in Wisconsin instigated a suite of upstream and downstream geomorphic adjustments. Upstream of the removed dams, channels developed in the reservoir sediment through bed degradation, channel widening, and aggradation. These processes were analogous to those described in conceptual channel evolution models, and thus a modified channel evolution model was suggested for describing channel adjustments upstream of a dam removal. Upstream channel development and evolution were strongly controlled by the character of the reservoir sediment and whether or not the reservoir was regularly dewatered. A reservoir which was dewatered regularly having relatively little consolidated or coarse sediment (Baraboo River) progressed rapidly through the evolution sequence, with erosion occurring throughout the reservoir immediately following dam removal. The second reservoir (Koshkonong River), which had not been dewatered, had consolidated fine sediment and progressed much more slowly due to the limited migration of a head-cut which controlled subsequent channel development.
 At both sites a large amount of fine sediment was exported from the reservoirs immediately following dam removal, but subsequent erosion of reservoir sediment, and thus subsequent downstream sedimentation, was strongly controlled by the rate and magnitude of channel development and evolution within the reservoir. At the site where erosion occurred along the entire length of the reservoir (Baraboo River), sand was transported through the reservoir and into downstream reaches. Downstream aggradation, however, was temporary. At the other site (Koshkonong River), there was little downstream sedimentation through time due to limited reservoir sediment erosion. These contrasting results highlight the potential role of head-cut migration on controlling rates of both upstream erosion and corresponding downstream sedimentation.
 Current understanding of the geomorphic changes associated with dam removal is surprisingly limited. Studies like the one described here and others [Egan and Pizzuto, 2000] are desperately needed in light of the rapidly increasing number of dams being considered for removal. In the absence of empirical studies, geomorphic analogies of dam removal offer some basis for qualitative predictions, study design, quantitative models, and reservoir sediment management schemes.
 While our study is one of the first to quantify geomorphic changes caused by dam removal, the temporal and spatial scale of our project was limited and the size of the dams studied was quite modest. Small dams are being removed across the United States, and these projects should be viewed as opportunities to learn about the processes and impacts surrounding dam removal, particularly from a variety of regions and from a variety of dam types and sizes. Large-scale dam removal programs should move slowly given the current status of the science. As funds become available for dam removal, and as our scientific understanding of the issue improves through detailed studies of a small number of initial removal projects, we will be well placed to develop the strategies needed to prioritize and implement dam removals that balance long-term environmental, economic, and societal goals.
 Funding for this research was provided by the Purdue University Showalter Fund, the Bradley Fund for the Environment, a Horton Grant from the American Geophysical Union, and a Fahnestock Award by the Geological Society of America to M. Doyle and NSF grant DEB-0108619 to E. Stanley. John Stofleth, Ben Lubbers, Andy Selle, and Mike Barry assisted with field work and laboratory analysis. Contributions from Tom Hooyer and the Wisconsin Geological and Natural History Survey, as well as from Sue Josheff, Bernie Michaud, and other personnel at the Wisconsin Department of Natural Resources, are greatly appreciated, especially their effort in coordinating dam removals with our research. Residents of Rockdale and LaValle are thanked for their cooperation, particularly the Smithbacks, Kenseths, Raybucks, and the Roths for allowing us access to their property to collect data. We thank Darryl Granger and two anonymous reviewers for suggestions on improving the paper.