4.1. Applicability of Groundwater Flow Models in Hyporheic Studies
 Our use of MODFLOW in this study follows that of several previous studies [Kasahara and Wondzell, 2003; Storey et al., 2003; Cardenas et al., 2004; Lautz and Siegel, 2006] that used models in a “sensitivity analysis” to explore different factors influencing HEF. As is common in applying groundwater flow models to hyporheic studies, we did not have sufficient information to rigorously test our model predictions. This was especially true between the initial conditions and 4 years, for which we only had stream water elevations at the cross-sectional transects. Because we lacked water table elevations for these dates, calibration of our models was impossible. Therefore, we used “uncalibrated” models in this analysis.
4.1.1. Model Performance
 We assumed that sediment properties and basic boundary conditions did not change during the 16-year study period. Therefore, we measured K in 2003, extrapolated these to the model domain, and used those data for all model runs. We tested the fit of the model for the 2003 simulation to the piezometric head observed in the small piezometer networks installed in 2003 (Figure 1). The root-mean-square error (RMSE) of the residuals (simulated versus observed heads) was 0.02 m, whereas the total change in surface water elevations over the study reach is approximately 1.10 m (see Figure 2 for more information). However, as Wondzell et al.  have shown, close agreement between observed and simulated heads is to be expected when observation points are located close to specified head boundaries. Because stream water elevations can be very accurately surveyed, and because the water table elevations in near stream piezometers are tightly controlled by stream water elevations, the flow lines are likely quite accurate. However, the close agreement between the observed and simulated values should not be accepted as a rigorous test of the model's ability to simulate the magnitude of hyporheic fluxes nor hyporheic residence times.
 Independent measures (tracer test or other data) of flux through the model domain provide a much better test of a hyporheic simulation [Wondzell et al., 2009]. Therefore, we also tested model performance by comparing observed median travel times between piezometers located on the gravel bar (described in section 2.1) with simulated median travel times from MT3D [Zheng, 1990]. Simulated median residence times were within a factor of 2 of observed residence time for eight out of nine piezometers, indicating the model estimates of flow through the gravel bar were reasonable (Figure 2). We did not recover usable tracer breakthrough curves from the floodplain piezometers and were therefore unable to test the model simulations from the larger model domain.
4.1.2. Effects of Uncertainties and Model Simplifications
 We acknowledge that our use of uncalibrated models with uncertain, homogeneous hydraulic conductivity and uncertain boundary conditions increased the overall uncertainty of our simulations and may limit the interpretation of the absolute values of model predictions. However, the modeling uncertainty does not invalidate our efforts to examine the effect of large wood on HEF. Instead, we consider our efforts to be consistent with a sensitivity analysis approach. We kept boundary conditions and the spatial distribution of K unchanged, thus allowing us to isolate the changes in HEF resulting from changes in channel morphology following wood removal.
 The constant head boundaries at the side and the no-flow boundaries at the bottom of the model are uncertain and may have slightly truncated the end of the residence time cdf. It is possible that a very small fraction of the flow that exited the side model should have returned to the stream and been counted as HEF with a long residence time. Similarly, flow is forced to return to the surface by the bottom no-flow boundary. Together, these boundaries may have generated a smaller number of very long and very slow flow paths than in reality. We did a sensitivity analysis on the HEF to bottom boundary depth and found no significant changes after 20 m. We did not do a sensitivity analysis to the side boundaries, but these are a minimum of 20 m from the stream and involve a negligible amount of flow originating in the stream reach. These results are reported by LaNier . For a further analysis of the effects of boundary conditions on long-time hyporheic flow, see Wörman et al. , Cardenas [2007, 2008], and Cardenas et al. [2008b].
4.2. Changes in Hyporheic Exchange
 Previous work by Smith et al. [1993b] clearly showed that wood limited the development of gravel bars in this small, low-gradient stream. Once wood was removed, the channel began adjusting toward a more free-formed pool-riffle morphology. Our simulations suggest that these changes were accompanied by increased hyporheic exchange flow and increased residence times of stream water in the hyporheic zone. Because wood buried in the floodplain was cut flush with the stream banks and only the portion projecting into the active channel was removed, the degree of bar development observed since wood removal was not entirely “free form.” The residual wood defends stream banks against erosion, deflects flows, and tends to fix the preexisting configuration of alternating bars in place. If both the active channel and floodplain sediments were wood free, further bar development might occur, in which case a fully free-formed channel might have still greater sinuosity, more or larger point bars, and better developed pool-riffle sequences which could further increase the amount of hyporheic exchange.
4.2.1. Spatial Patterns of Downwelling and Upwelling
 The spatial patterns of simulated exchange fluxes are quite complex within our study reach. There is a tendency for a net increase in nonhyporheic exchange (upwelling of nonhyporheic water) toward the top of the reach and a net loss of nonhyporheic exchange (downwelling) toward the tail of the reach (Figure 5). Much of this spatial trend is likely an artifact of our methods. We defined HEF as water that originated in stream cells, flowed into the subsurface, and subsequently flowed back into a stream cell within the study reach. Much of the nonhyporheic water upwelling through the streambed near the head of the reach, or downwelling near the tail of the reach, represents “bypass flow” that would have originated in, or returned to the stream, if we had modeled a longer reach. Certainly, some of the water flowing into the stream cells would have originated from adjacent hillslopes. We cannot distinguish among the ultimate sources or sinks of this water in our modeling approach. Therefore we refer to this water as “nonhyporheic water,” and our estimation of HEF is therefore conservative.
 Although the spatial pattern of downwelling and upwelling is quite complex, it is clear that head gradients imposed by changes in the WSE profile along the reach are one of the dominant drivers of hyporheic exchange (Figure 5). Alternating zones of downwelling and upwelling occur in all simulations. Further, the transition between downwelling and upwelling usually occurs midway along each riffle. Under conditions immediately prior to wood removal, there was one wood-forced step controlled by a channel spanning log just upstream of cross section B. A wedge of sediment had accumulated above this log, with a steep riffle downstream of the log. Removal of the log initiated erosion and channel incision, and by 2 years post wood removal, a sharp knickpoint had migrated upstream to cross section A. The zones of downwelling and upwelling also migrated upstream, so that 2 years after wood removal, little exchange flux occurred at cross section B. Sediment was scoured from the channel near the head of the reach while deposition occurred downstream, smoothing the longitudinal profile and reducing hyporheic exchange. Continued adjustment of the channel over the next 14 years created a number of low-gradient zones separated by riffles, with substantial hyporheic exchange.
 Another potentially important driver of hyporheic exchange which we did not include in our models is heterogeneity in K. For an analysis of this effect, refer to Cardenas et al. , Salehin et al. , Ryan and Boufadel , and Marion et al. . Interestingly, the deep pool scoured into the streambed just downstream of cross section E had little effect on HEF. Two large logs were removed from the channel in this location, but these logs were oriented parallel to flow and thus had relatively little influence on sediment or the WSE profile. Lateral channel migration is limited in this location by the edge of a high terrace. Removal of the wood apparently allowed additional scour and enlargement of the pool, with relatively little effect on HEF.
 Calculations of gross exchange fluxes suggested that downwelling and upwelling of both hyporheic and nonhyporheic water occurred within the short stream segments between adjacent cross-sectional transects (Figure 5). The spatial patterns of exchange flow on the streambed show how such complex patterns in gross exchange can occur at such small scales (Figures 6a–6c). Although the simple pattern of alternating zones of downwelling and upwelling can be discerned, these zones are often quite elongated, especially around meander bends. For example, consider the zone from cross section C to the point bar downstream of cross section D under initial conditions (Figure 6a). Upwelling occurs on the left side of the channel (facing downstream), downstream of C where cross-meander flow paths rejoin the stream. The right side of the channel, however, is a zone of downwelling, feeding stream water into cross-meander flow paths through the next point bar. These flow paths rejoin the stream in a very elongate zone of upwelling, all along the downstream face of this point bar. Zones of downwelling and upwelling are not immediately contiguous but are separated by zones of relatively neutral exchange flux. Also, exchange fluxes of nonhyporheic water occur over large portions of the streambed.
 By 2 years after wood removal, the WSE profile was smoothed, with large distinct breaks in the longitudinal gradient replaced by a much finer, small-scale pattern (Figure 5). This is also reflected by the increase in number (Table 1) and the decrease in size (Figure 6b) of downwelling and upwelling patches on the streambed. By 16 years post wood removal, with the alternating bar and pool-riffle morphology well developed, the patches had coalesced, forming numerous elongate zones of downwelling and upwelling.
 Model-based estimates of hyporheic exchange through the streambed, averaged over the study reach, ranged between 19 and 30 L m2/h. Estimates for individual segments of the streambed, however, can exceed 400 L m2/h, with even higher rates simulated from individual stream cells. Significant spatial variability in streambed flux has been observed in the field [Burkholder et al., 2008]. Aquatic organisms can be sensitive to these fluxes. For example, many cold water–dependent fishes seek cooler upwelling locations during summer low flows when the bulk stream temperatures exceed the species temperature preferences [Berman and Quinn, 1991; Ebersole et al., 2001]. Similarly, patches of upwelling or downwelling water may influence spawning site selection of cold water fishes [Baxter and Hauer, 2000; Geist et al., 2002].
 Kasahara and Wondzell  argued that expressing hyporheic upwelling rates relative to the streambed area may better reflect the importance of HEF to stream ecosystem processes than the ratio of exchange flow to stream discharge (QHEF:Qstream) because biological activity tends to be concentrated on benthic surfaces. The presence and importance of fine-scale environmental patchiness on streambeds generated by groundwater inflows and hyporheic return flows are increasingly recognized [Burkholder et al., 2008, Cardenas et al., 2008a; Poole et al., 2008]. Here, we have shown that there is substantial spatial and temporal variability in the amount of water downwelling or upwelling through these patches and variability in the size, shape, and location of these patches. Patches with high rates of HEF may be hot spots of biological activity [McClain et al., 2003] and may provide distinct microhabitats critical to the success of stream dwelling organisms. Our results also suggest that human land use activities that influence channel morphology may strongly affect downwelling and upwelling patches.
4.2.2. Interactions Between Wood, Channel Morphology, and HEF
 The general trends observed in this stream contrast with the role of large wood in higher-gradient mountain streams [Kasahara and Wondzell, 2003; Wondzell, 2006]. Kasahara and Wondzell  used MODFLOW and conducted a sensitivity analysis of the effect of wood-buttressed sediment wedges on HEF in a stream. HEF increased with the abundance of wood-forced steps in the WSE profile, and the larger the steps, the greater the increase in HEF. Wondzell  found similar results when examining naturally formed stream channels with the stream tracer approach. In the case of these two high-gradient streams (∼13%), free-formed channels would have either been scoured to bedrock [Montgomery et al., 1996] or would have formed some boulder-buttressed steps [Faustini and Jones, 2003]. Large logs significantly increased sediment storage in these channels [Nakamura and Swanson, 1993], and similar relationships between large wood and sediment storage and channel morphology have been reported for other high-gradient channels [Beschta, 1979; Mosley, 1981; Díez et al., 2000; Faustini and Jones, 2003]. Thus, all available lines of evidence suggest that the removal or loss of large wood changes channel morphology and reduces the extent of the hyporheic zone in steep mountain streams.
 The interactions between in-stream large wood, channel morphology, and HEF appear to be more complex in low-gradient streams. Several authors have shown that in-stream wood promotes HEF at the scale of individual bed forms [Mutz et al., 2007; Hester and Doyle, 2008], the channel unit scale [Kasahara and Hill, 2006], and whole reach scales [Lautz et al., 2006].
 Mutz et al.  examined a low-gradient, sand bed channel in a flume and demonstrated that the addition of large wood significantly increased bed form irregularities, leading to increases in hyporheic exchange and the depth to which stream water penetrated into the streambed sediment. This flume study, however, used a straight, plane bed channel entirely lacking bed forms as the control condition. They did not contrast channels with free-formed bed forms lacking wood with those formed in the presence of wood. Further, the flume walls tightly bounded the stream, preventing lateral adjustments in the channel planform under either wood-loaded or wood-free conditions. As our simulations of the Bambi Creek study site showed, loss of wood resulted in reduced hyporheic exchange 2 years after wood removal because scour and deposition smoothed the WSE profile. To that extent, our results from the first few years after wood removal are consistent with those of Mutz et al.  in that HEF was positively correlated to bed form irregularity. However, our longer-term results contrast sharply with these previous studies, as wood removal eventually increased HEF. It took 4 years of channel adjustments in Bambi Creek before our simulations showed increased hyporheic exchange.
 Both riffles [Storey et al., 2003; Kasahara and Hill, 2006] and meander bends [Boano et al., 2006] have been shown to be important locations of hyporheic exchange in low-gradient streams. Lautz and Siegel  examined the relative importance of these features in a low-gradient, meandering stream, Red Canyon Creek. Using a groundwater flow modeling analysis, they demonstrated that exchange fluxes driven by head gradients imposed by beaver dams were much larger than exchange fluxes through meander bends. Of course, these hyporheic exchange flows would be lost if the obstructions were removed from the channel. A critical question, then, is how does that channel adjust to the loss of wood or other obstructions over the long term?
 Small stream studies have shown that active channels in forested reaches are wider than equivalent channels in meadows or pasture because of the combined effects of decreased bank cohesion where forest canopy shades out stream bank vegetation and bank scour caused by large wood [Trimble, 1997; Davies-Colley, 1997; Sweeney et al., 2004; Allmendinger et al., 2005]. In contrasting Bambi Creek (this study) to Red Canyon Creek [Lautz and Siegel, 2006], we suggest that the relatively wide channel at Bambi Creek allowed sufficient room for the development of alternating bars within the active channel following wood removal. At low flow, the wetted channel meandered around these unvegetated gravel bars. The development of the associated pool-riffle morphology resulted in a stepped longitudinal profile (Figure 3) that enhanced HEF. In contrast, cohesive stream banks resulting from dense herbaceous vegetation would promote a narrow active channel and limit gravel bar development within the channel of Red Canyon Creek. Because of relatively fine textured floodplain sediment at both Bambi Creek. and Red Canyon Creek, hyporheic exchange is dominated by short exchange flow paths in the near stream zone. Thus, it seems reasonable that development of gravel bars within the active channel would lead to increased hyporheic exchange at Bambi Creek, whereas the narrower channel at Red Canyon Creek would limit bar development such that steps in the energy profile of the stream created by beaver dams would dominate HEF.
 Overall, the Bambi Creek wood removal experiment [Smith et al., 1993a, 1993b] (also this study) is different than previous studies on higher-gradient streams in that wood removal increased both sediment storage and HEF. Studies on lower-gradient streams or flumes mostly report in situ conditions or those immediately following changes in wood loading (removal or addition experiments), thereby highlighting the direct effects of wood on sediment storage, channel morphology, and hyporheic exchange. To resolve the apparent contradictions between these studies, we must consider more than just the direct effects of wood removal. We must consider the long-term channel adjustments set in motion through changes in wood. Clearly, not all channels have equal potential to adjust their planform in response to changes in wood loading. Some low-gradient natural channels are tightly constrained and others are incised, often from human land use impacts. We expect these would respond more like the flume study of Mutz et al. . Channel adjustments may be very slow in meadow streams where grassy vegetation maintains narrow active channels so that woody obstructions, if present, are likely to be the most important contributors to HEF. Finally, channels that are sediment supply limited are likely to incise following wood removal resulting in reduced HEF. The Bambi Creek study is relatively unique. It is a 16-year-long study of a small, low-gradient stream in a near-pristine environment. The channel is not incised, and the long record of study provided time for secondary effects of wood removal to become evident; that is, the major planform adjustments resulting from sediment redistribution into a more fully developed alternating bar channel with well developed pool-riffle morphology. Further study in other low-gradient streams will be needed to know if the pattern observed at Bambi Creek is repeated elsewhere.