Department of Civil and Environmental Engineering, University of Wisconsin-Madison, Madison, Wisconsin, USA
Corresponding author: S. P. Loheide II, Department of Civil and Environmental Engineering, University of Wisconsin-Madison, 1269C Engineering Hall, 1415 Engineering Drive, Madison, WI 53706, USA. (firstname.lastname@example.org)
 This research introduces and provides evidence of a novel mechanism of stream-aquifer exchange caused by dynamic ice formation on streams. At a stream site in southwestern Wisconsin, we document a significant fluctuation in stream depth and the potentiometric surface in the adjacent aquifer during periods of ice formation, suggesting that stream-aquifer interactions may be influenced by these transient events. Four years of winter stream data document that dynamic ice formation (1) causes an average increase in stream depth of 106%, (2) affects stream depth on 20% of days from December through February, and (3) substantially alters stream-aquifer interactions by reducing the hydraulic gradient toward the channel during ice formation events. Dynamic ice formation represents a potentially widespread mechanism for altering stream-aquifer interactions and associated biogeochemical transformations. The sensitivity of this process to air temperature indicates a need for further research on the impacts of climate warming.
 Traditionally, surface water and groundwater had been considered separately [Winter et al., 1998]; however, a growing body of research in recent decades has investigated surface water and groundwater as a complex and integrated system that plays a crucial role in how riparian ecosystems function [Kalbus et al., 2006]. Changes in water fluxes between a stream and its adjacent aquifer can lead to changes in stream and aquifer temperature, chemistry, and biota [Sophocleous, 2002], which have important management implications for these valuable ecosystems [Boulton and Hancock, 2006]. This temporal variability is commonly driven by changes in the hydraulic gradient between the stream and adjacent aquifer, which can be caused by flooding [Chen and Chen, 2003], snowmelt pulses [Loheide and Lundquist, 2009], and dam releases [Sawyer et al., 2009]. In this study, we add to this list another mechanism—dynamic ice formation—that readily occurs in cold climates during the winter and can quickly and dramatically increase stream stage and trigger an alteration in stream-aquifer interactions.
 Although the term dynamic ice formation is typically used to discuss the temporal and spatial variability of anchor ice dams in steep and turbulent streams [Stickler et al., 2010], this research documents dynamic ice formation in a low-gradient stream due to daily patterns of surface ice floes and static ice cover. Surface ice floes occur when suspended ice particles (frazil ice) flocculate and rise to the stream surface [Osterkamp, 1978; Daly, 1984]. The ice floe may solidify and form static surface ice cover (frazil ice jam) if the floe becomes impeded due to a channel bend, downstream ice cover, border ice, or other channel obstructions [Bergeron et al., 2011]. Prowse and Beltaos  explain that the additional shear stress created by the ice cover causes the mean water velocity in the stream to decrease. In order for the stream discharge to remain constant at lower mean velocity, the stream stage must rise, increasing the cross-sectional area of the stream. Flume experiments by Smith and Ettema  confirm that ice cover roughness increases flow depth and decreases mean velocity. For these reasons, ice floes and ice jams have been studied extensively due to their ability to cause flooding events [Tuthill et al., 1996; Beltaos, 1983; Healy and Hicks, 2007], disrupt rating curves [Pelletier, 1990; Sui et al., 2007], and alter channel erosion [Ettema, 2002]. However, their ability to modify stream-aquifer interactions has not been previously documented.
 This research explores a case where ice formation occurs on a dynamic basis throughout winter, causing large fluctuations in stream depth and potentially altering the interaction between surface water and groundwater. During periods of sustained cold weather, ice forms in the Upper East Branch Pecatonica River overnight and thaws during the day, causing a daily rise and fall of the stream stage. As stream stage rises during the ice formation, the hydraulic gradient from the aquifer toward the stream may be reduced, thus altering stream-aquifer interactions (Figure 1). The purpose of this paper is to (1) document the impact of dynamic ice formation on stream and aquifer water levels, (2) analyze the impact of this process on hydraulic gradients which drive surface water–groundwater interactions, and (3) determine the frequency of ice formation events that are likely to influence stream-groundwater hydraulic gradients.
2 Site Description: Upper East Branch Pecatonica River
 The study site is located along the Upper East Branch Pecatonica River on The Nature Conservancy's Barneveld Prairie in Iowa County, Wisconsin (Figure 2a). A typical cross section of the floodplain consists of approximately 1 m of silt alluvium overlying approximately 2 m of Late Wisconsin gravel underlain by St. Peter sandstone [Knox, 1972; Booth and Loheide, 2010]. Due to the differences in hydraulic conductivity between the coarse gravel layer and the silt alluvium, the silt layer functions as a semiconfining unit and is common in the floodplains of the region [Hunt et al., 1999].
 The stream at the research site has a base flow of approximately 0.12 m3/s, average depth of 16 cm, channel bed slope of 0.0016, and contributing watershed of approximately 13 km2. The stream is largely fed by contributions from the gravel aquifer, though a layer of silty-clay deposits along the streambed reduces the connectivity between the stream and the aquifer. Together, the silty-clay deposits in the streambed and the silt-confining unit cause the saturated gravel layer to function as a semiconfined aquifer that is relatively well connected to the stream. Piezometers that penetrate into the gravel show a potentiometric surface that is above the water table measured in the shallow silt alluvium.
3 Methods: Data Acquisition
 Three piezometers were installed perpendicular to the stream during the winter of 2010–2011 at the location of an existing stream stage and temperature sensor, which were installed in November 2007. The wells were placed at 2, 4, and 10 m from the stream bank and are referred to as Z1, Z2, and Z3, respectively (Figure 2a). Due to the meandering nature of the stream and the possibility of heterogeneity in the stream stage fluctuations, an additional stream stage sensor was placed upstream at a bend closest to well Z3. Pressure transducers (HOBO water level loggers) were placed in each well and stream stage location to log absolute pressure data every 5 min. A weather station on site provided air temperature and barometric pressure data for analysis. The barometric pressure data were used to correct the absolute pressure recorded by the pressure transducers so that the height of water above the sensors could be monitored. Last, a time-lapse camera with flash for nighttime photography was used to provide visual confirmation of the stream ice processes.
 Simultaneous measurements of stream stage and temperature reveal that when the water temperature approaches 0°C and ice formation is initiated, the stream stage rises dramatically and then later recedes as the ice melts and water temperature increases (Figure 2b). To further verify the ice formation process and its impact, time-lapse images show the stream at base flow conditions just prior to ice formation, at peak stage with ice cover, and then after ice melt when the water has receded, leaving behind hanging border ice (Figure 2b). The prevalence of these dynamic ice cover events over the winter season is illustrated for 2008–2009 when 27 distinct ice formation events occurred (Figure 3).
 More intensive monitoring during the winter of 2010–2011 documented the effect of the stream stage increase during freeze events on the adjacent aquifer. Six distinct freeze events were observed from 20 January to 13 February, including a multiday event from 2 February to 4 February when stream temperature remained near zero (Figure 4). This third event led to water level increases of 0.42, 0.30, 0.27, and 0.17m at the stream, well Z1, Z2, and Z3, respectively (Figure 4b). These data show that ice formation causes a dramatic increase in stream depth, a 248% increase during the third event, and a related increase in the potentiometric surface within the adjacent gravel aquifer.
 The hydraulic gradient in this aquifer is also modified by these freeze events (Figure 4c). The hydraulic gradient between the stream and Z1 is strongly reduced following freeze events with reductions ranging from 39 to 60% for the events shown. This indicates a correspondingly large reduction in groundwater discharge to the stream.
 Last, stream stage records from 2007 to 2011 show that, on average, the stream depth increased by 106% during ice formation with the largest event causing a 248% increase. The frequency of ice formation impacting stream stage is also significant. During the 4 year record, ice formation caused an average of 19.25 stream stage responses from December through February or about one event every 5 days. The intensity and frequency of these events suggest that dynamic ice formation may provide a significant control on stream-aquifer interactions during winter months in cold regions.
5 Discussion and Conclusion
 This work has demonstrated that dynamic ice formation on streams creates a rise in stream stage that reduces the hydraulic gradient toward the stream. The reduction in hydraulic gradient results in a longer residence time of groundwater in the riparian aquifer and reduces the groundwater discharge to the stream. The reduction of groundwater discharge, if ubiquitous on channels throughout the watershed, could substantially reduce streamflow during these events. The increase in residence time in the subsurface of the riparian zone may play the important role of enhancing biogeochemical processes in sediments and influencing nutrient cycles. For example, Gu et al.  showed that stream stage fluctuations create “riparian biogeochemical hot moments” and are a significant cause of denitrification in a North Carolina stream even when the flooding-induced fluctuations occurred on average 3 to 4 times per year. The high frequency of dynamic ice formation events that we document in this study suggests that winter may play a strong role in biogeochemical processing in stream systems that experience cold winters.
 In cases where weaker hydraulic gradients exist, the direction of flow could be reversed causing hyporheic exchange, which is the mixing of surface water and groundwater that occurs in the sediments surrounding a stream. Much research has focused on the importance of hyporheic exchange due to its fundamental role in biogeochemical processes that control water quality, nutrient cycling, and primary productivity within stream ecosystems [Boulton et al., 1998; Brunke and Gonser, 1997; Findlay, 1995; Krause et al., 2010; Stanford and Ward, 1988]. Often, hyporheic exchange can be described by the “gill model” [Sawyer et al., 2009], where spatial gradients in hydraulic head at the streambed create flow paths in which stream water enters the subsurface and later returns to the steam at a down-gradient location. This type of exchange can be due to streambed topography [Harvey and Bencala, 1993], large boulders or woody debris [Hester and Doyle, 2008], and bed forms and channel bends [Cardenas et al., 2004]. Hyporheic exchange of this type under ice covered channels has been simulated by Cardenas and Gooseff . In cases where hyporheic exchange occurs via the “gill model” in gaining streams, the reduction in groundwater gradients toward the channel associated with dynamic ice formation would increase the spatial extent of the exchange zones [Cardenas and Wilson, 2006]. Another class of hyporheic exchange, described by the “lung model” [Sawyer et al., 2009], involves water moving in and out of the aquifer due to fluctuating water levels and can be caused by daily snowmelt cycles [Loheide and Lundquist, 2009], flood events and bank storage [Chen and Chen, 2003], and dam operations [Sawyer et al., 2009; Francis et al., 2010]. Although not observed in this study, this research suggests the possibility of a new mechanism of hyporheic exchange within the lung model, where dynamic ice formation causes fluctuating water levels and reversals of hydraulic gradients. At this particular site, the large and rapid pressure response observed at the wells is a consequence of the relatively low storage coefficient associated with this semiconfined aquifer. Since the observed response is largely a pressure response controlled by elastic storage, the volume of water associated with this change in water level is also relatively small. This is not necessarily a generalizable result, and more water exchange could occur in cases where unconfined alluvial aquifers are well connected to streams.
 Since ice formation is driven, in part, by air temperature, a warming climate may change the frequency and reduce the duration of freeze events. Our 4 year record shows strong relationships between the frequency of events and the mean air temperature. The 2008–2009 winter season was the coldest and recorded the most freezing events, 27, with an average air temperature of −8.9°C. The mildest winter was 2011–2012, recording only six events with an average air temperature of −3.8°C. We predict that with a warming climate, streams that formerly had ice forming dynamically throughout winter will have less frequent events. In addition, streams that tended to form static ice cover in the past may transition to having a more dynamic ice regime.
 In conclusion, 4 years of winter stream data document that dynamic ice formation causes a significant rise in stream depth, with an average increase of 106%, and occurs with significant frequency, one out of every 5 days from December through February. This dynamic ice formation reduces the hydraulic gradient between the aquifer and the stream and therefore reduces the discharge of groundwater to the stream during ice formation events. Thus, dynamic ice formation is a formerly unrecognized driver of changes to stream-aquifer interactions in cold weather environments with possible implications for biogeochemical processes, hyporheic exchange, and riparian system response to climate change.
 Funding for this research was provided by the University of Wisconsin-Madison Hilldale Undergraduate/Faculty Research Fellowship and the National Science Foundation under Grant CBET-0954499. Any opinions, findings, and conclusions are those of the author and do not necessarily reflect the views of the funding agencies. We would like to thank The Nature Conservancy for allowing us to install equipment on their property in order to conduct this research. This manuscript benefitted greatly from helpful comments provided by two anonymous reviewers.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.