Instream wood increases riverbed temperature variability in a lowland sandy stream

The (re)introduction of wood into rivers is becoming increasingly popular in river restoration and natural flood management schemes. While instream wood is known to promote geomorphic and hydraulic diversity, the impact of wood in driving surface water‐streambed exchange and subsequent streambed temperatures remains under‐researched, particularly in lowland rivers. We make use of the occurrence of three naturally occurring wood structures in a small, lowland sandy stream to determine how the presence of wood alters the geomorphic, hydraulic and thermal properties of the streambed. Our results show that instream wood plays an important role in promoting localized geomorphic complexity and thermal variation in the streambed. Locations within and immediately downstream of wood structures displayed the highest temperature range and daily variation. Locations upstream of wood structures were characterized by weaker daily temperature variation, while areas without wood displayed relatively stable streambed temperatures, with little diurnal fluctuation. Our study indicates that at this lowland site, instream wood increased seasonal temperature extremes (increased summer and decreased winter temperatures) at shallow depths by enhancing infiltration of warmer (summer) and colder (winter) surface water. This reduction in thermal buffering is likely to have significant implications to streambed‐dwelling communities and highlights that the thermal impacts of wood reintroduction in lowland rivers should be considered prior to river restoration.


| INTRODUCTION
The promotion and restoration of natural process dynamics in rivers, in particular to mitigate anthropogenic impacts and climatic pressures, is an important component of river catchment science and management (Fryirs & Brierley, 2016;Wohl et al., 2005;Wohl, Lane, & Wilcox, 2015). Following years of active removal of instream wood from rivers as part of river and catchment management practices, there is an increasing body of evidence now suggesting that wood plays a vital role in creating and maintaining a range of ecosystem processes and functions Hester, Hammond, & Scott, 2016;Krause et al., 2014). Instream wood has been shown to improve ecosystem structure, including provision of instream habitat (Bocchiola, 2011;Harvey, Henshaw, Parker, & Sayer, 2018;Roni, Beechie, Pess, & Hanson, 2014), and biological diversity (Pilotto, Bertoncin, Harvey, Wharton, & Pusch, 2014;Thompson et al., 2018).
Wood has also been shown to enhance hyporheic exchange flows (HEF) by forcing surface water into the streambed at and immediately upstream of the wooden flow obstacle, and subsequent hyporheic flow and upwelling downstream of the wood (Hester & Doyle, 2008;Sawyer, Cardenas, & Buttles, 2011).
Temperature is therefore considered to be the "master variable" in determining hyporheic biogeochemical and ecohydrological processes (Webb, Hannah, Moore, Brown, & Nobilis, 2008). Heat transfer across the water-sediment interface and into the hyporheic zone is controlled by advection and conduction processes which pump surface water (advection) or diffuse heat (conduction) into the streambed (Arrigoni et al., 2008;Burkholder et al., 2008). In temperate environments, HEF usually results in cooler streambed temperatures in summer, and warmer hyporheic temperature in winter (Arrigoni et al., 2008;Caissie, 2006;Hannah, Malcolm, & Bradley, 2009;. This seasonal buffering provides important thermal microrefugia (Ashcroft, 2010) of particular importance to cold water species, including salmonids (Greer, Carlson, & Thompson, 2019;Malcolm et al., 2004).
To date, the majority of research on wood-induced HEF and its implications for thermal refugia in the hyporheic zone has focussed predominantly on small, upland gravel-bed rivers. These rivers are typically characterized by high hydraulic conductivities, which permit wood-induced HEF to penetrate the streambed to a greater degree than finer sediment or lower conductivity sediments typical of lowland streams (Hester et al., 2009;Krause et al., 2014;Magliozzi, Grabowski, Packman, & Krause, 2018). Meandering, finer sediment lowland rivers typically face a greater range of pressures than upland rivers. The flat fertile terrain (and high level of agricultural activity), proximity to large populations and past management mean that they are characterized as having increased point and diffuse pollution, altered flow, channelization and instream and riparian habitat degradation (Neal et al., 2012). Lowland rivers also typically have more extreme temperatures, owing to increased atmospheric input as a result of more stable discharges and decreased riparian shading (Poole & Berman, 2001b).
Given the differences in external pressures, environmental setting, and ecosystem integrity between upland and lowland rivers, it is not surprising to find an increasing body of evidence showing that wood restoration in lowland streams does not always confer the same benefits (outlined above) as observed in higher energy, upland streams (Daniels, 2006;Krause et al., 2014;Magliozzi et al., 2018). In order to address these opposing points, we made use of three naturally occurring wood structures to assess the wood's impact on instream structure and functioning within a sandy lowland river. More specifically, we undertook detailed geomorphic, hydraulic and thermal monitoring over the course of a year to assess how instream wood alters the streambed and temperature conditions which affect habitat diversity (ecosystem structure) and ecosystem functioning. We hypothesized that the instream wood in the investigated lowland stream would play an important role in promoting surface water downwelling into the streambed, resulting in increased thermal heterogeneity within the hyporheic zone. However, we also hypothesized that the finer sediments and more variable surface water temperatures found at our study site, typical of small lowland agricultural streams, would result in more extreme streambed temperatures, and therefore a reduction in thermal buffering within the hyporheic zone.

| Site description
Experimental investigations within this study focused on the Hammer Stream, a third order tributary of the lowland Western Rother river, West Sussex, comprising 25 km 2 of agricultural and forested land (Foster et al., 2019). The underlying Greensands and Mudstone geology result in the dominance of a fine sandy bed (D 50 = 0.28 mm), with peat and clay lenses throughout the river bed and floodplain to a depth of approximately 1-2 m (Shelley et al., 2017). These impermeable layers were found to effectively isolate the study reach from the underlying aquifer and locally supressed groundwater upwelling (Dara, 2017), resulting in hyporheic flow being dominated by bedform and river bank geometry forced surface water downwelling.  comprised numerous 1 m length wood pieces racked together in the centre of the channel, creating a large sandbar which at base flow, was often exposed.

| Data collection/instrumentation
Areas around the three wood structures were distinguished into upstream, at/within and downstream of wood categories to represent their spatial relationship to the wood. The 20 m section with no wood or other large structural elements was termed "no wood"; these delineations were used for subsequent categorical analyses as outlined below. Monitoring was predominantly focused around the wood, given the complexity of the structures and objectives of the study. MA, USA) were deployed flush to the outside of the pipe. The temperature loggers were connected to Hobo U12 outdoor loggers (Onset Computer Corporation, Bourne, MA, USA) which were installed on the river banks. A total of 10 temperature lances were placed in the streambed following the thalweg near the centre of the channel to monitor longitudinal profiles of streambed temperatures ( Figure 1f). The sensors were calibrated prior to deployment by placing them into a bucket held at 20 C for 2 hr to determine correction values for each lance and every sampling depth which were subsequently applied in the data analysis.

| Streambed temperature
Lances were pushed manually into the streambed sediments, with the tops being flush with the bed surface. The exact location of the lances were determined using a total station as detailed below.
A separate surface water level, temperature and conductivity logger (Levelogger 3001 LTC Junior; 0.2 cm/±0.1 C/20 μS/cm accuracy, Solinst, Ontario, Canada) was installed at the bottom end of the study reach, protected from incident solar energy using a plastic pipe as a stilling well to house the logger. An air temperature and atmospheric pressure logger (Barrologger Edge, Solinst, Ontario, Canada) was placed on the river bank, adjacent to the surface water logger and out of direct sunlight, for subsequent atmospheric correction of the surface water logger. Surface water and atmospheric temperature data were collected every 15 min for a total of 418 days (December 19, 2014 -February 10, 2016). change in streambed elevation. These data were interpolated using IDW and differences in streambed topography between DEMs were calculated to produce seasonal maps of changes in topography. A minimum level of detection (LoD) was calculated as outlined by Smith and Vericat (2015) to distinguish between real topographic change and errors and uncertainties between DEMs. Data below the minimum LoD were omitted from the bedform change maps (Williams, 2012).

| Streambed geomorphology
Channel centreline transects at temperature lance locations were extracted from the bed elevation change maps to determine topographical change of the stream bed at each sampling season.

| Data analysis
Temperature data from the lances were visually inspected for erroneous data and errors prior to any subsequent analysis. 30 cm depth data at TL4 were excluded from all temperature analysis due to a faulty temperature probe, and data from 20 cm depth at TL10 were only used in the analysis from October 20, 2015 onwards due to an early problem with the probe. Data for TL5 had a later start date from April 17, 2015 onwards due to logger malfunction. All statistical analyses were performed in the R software package (v 3.4.1; R Core Team, 2015).
Due to the large amount of both spatial (4 depths at 10 locations) and temporal data (12 months 10) and "no wood influence" (TLs 4, 5 and 6). Grouping of TLs in this manner was validated using Kruskall-Wallis sum rank tests to ascertain any inherent variation between TLs which may preclude such grouping.
Differences in temperature between surface water and the streambed at the four location categories were assessed using effect size, calculated as Cohen's D (using the "effsize" R package) which is the difference in mean average temperature divided by the average of their pooled standard deviation (Cohen, 1988). Effect size, rather than traditional probability-based statistics (e.g., ANOVA) were used due to the large sample size (n > 25,000 points in all cases) which conflates Type I statistical errors (false positives) and thus prevents accurate probability-based statistical testing (Coe, 2002). Differences between streambed and surface water temperatures were considered to be significantly different when Cohen's D was >0.5 (medium effect;Cohen, 1988). Upper and lower confidence intervals at 95% confidence were calculated to provide an equivalent estimate of certainty. Coefficients of variation (ratio of standard deviation to the mean, expressed as a percentage) were calculated to express the dispersion of data around the mean which represents the stability and variation in temperature. Seasonal temperature duration curves at each depth were calculated to illustrate the thermal characteristics of the streambed (Hannah et al., 2009).
Diel (24 hr) temperature dynamics at each temperature lance and depth were analysed relative to their respective location relative to wood, to determine the average daily temperature ranges and dynamics experienced during each season. Average diurnal trends in streambed and surface water temperature were calculated for each hourly timestep at each of the four seasons to determine the magnitude of diurnal temperature fluctuation. The hottest (July 1, 2015; air temperature 21.6 C, surface water temperature 17.7 C) and coldest (January 20, 2016; −3.4 C air, 1.1 C surface water temperature) recorded air and surface water temperature days (both of which occurred during prolonged extreme temperature periods; Figure 1) were analysed using a 6-hr time step to illustrate how extreme surface water temperatures influence streambed temperature in relation to the relative position of in-stream wood features. Stream bed topography data from July and October 2015 were used to represent the location of temperature probes in the streambed at the time of analysis. Streambed temperature was then interpolated for each 6-hr timestep using IDW between points (using default parameters) to illustrate diel temperature change. The shapefiles were then plotted in R using "ggplot2", "akima" and "reshape" packages.

| Summer
The average summer surface water temperature was 14.1 C (air temperature 15.1 C). During this time, locations downstream of wood had the largest effect size (difference) to surface water temperatures to a depth of 20 cm, as did locations upstream and within wood at 5 cm depth. The temperature duration curves reveal a high variability in streambed temperature, particularly at locations downstream of wood which were characterized as being warmer than other locations at this time (to 30 cm depth) for a large majority of the time (approximately 80 % of the time to 30 cm depth at TL10), in contrast to sites with no wood which are much more stable as shown by the flat temperature duration curve. Diurnal patterns were strongest at depth in summer, with TLs downstream of wood showing a strong diurnal trend to 20 cm, and locations within wood following similar patterns to surface water ( Figure 5).

| Autumn
Surface water temperature averaged 10.9 C (11.0 C air temperature) during autumn. TLs upstream and downstream of wood and in areas with no wood were significantly different from surface water temperatures. TLs located in areas without wood in particular exhibited the highest Cohen's D values (1 and 1.03 at 10 and 30 cm depth, respectively). The temperature duration curves suggest that areas without wood had higher and more stable temperatures than any other locations, down to 30 cm depth. Lances within and downstream of wood structures, on the other hand, display more variable and colder temperatures for a large percentage (50 %) of the time. Figure 5 reveals a weak surface water diurnal fluctuation, however TLs within and downstream of wood show some evidence of a diurnal response at 5 cm depth which is more pronounced than that of surface water. The three TLs located in areas without wood influence (the control reach) all display warmer average streambed temperatures at 10 and 30 cm depths.

| Winter
Trends in streambed temperature during winter were similar to those found in autumn. Average surface water temperature was 7.7 C (air temperature 7.9 C). TLs located upstream, downstream and at areas without wood influence all had significantly different average streambed temperatures at all depths in comparison to surface water, with the exception of the 5 cm depth downstream of wood. Temperature duration curves reveal stable streambed temperatures in comparison to other seasons; TL6 once again had warmer, more stable streambed temperature at 10 cm. All locations and depths displayed warmer average diurnal temperatures than surface water, particularly at locations upstream, downstream and without wood influence.

| Seasonal temperature extremes
More detailed analysis of the diel temperature fluctuations during the hottest and coldest days ( Figure 6) shows that on the hottest time, the areas without wood maintained higher temperatures (day average of 3.9, 6.2, 5.8 and 6.8 C at 5, 10, 20 and 30 cm depth, respectively, in comparison to 3.2 C day average for surface water), while areas around instream wood and LW1 & 3 in particular displayed a decrease in temperature from 1200-2400 hr. The range of temperatures at this time were less variable, with the exception of the most upstream area between 0600 and 1200 hr.
A summary of topographic and thermal characteristics of the streambed in relation to wood position is provided in Table 2 (Krause et al., 2014), with significantly fewer studies aimed at small agricultural sand-bed streams, which differ significantly in terms of character (gradient, sediment size and stream power) and management (Neal et al., 2012). This study aimed to provide a detailed assessment of the role of instream wood in altering streambed topography and temperature in a lowland sandy stream. Our results show that instream wood promotes localized geomorphic complexity, which increased HEF and thermal variation within the streambed; however, the timing and location of streambed temperature variability observed was markedly different than that previously reported in upland sites.
Monitoring of the seasonal dynamics of streambed topography revealed that the sandy streambed in the research area was highly mobile, with the presence of instream wood being particularly important in creating topographic variability by inducing forced scour and deposition. The wood was found to play an integral role in creating mid-channel bars and areas of fine sediment deposition which may otherwise be rapidly washed out (Smock, Metzler, & Gladden, 1989).
Sediment accumulation behind the downstream wood structure (LW3) in winter and spring was gradually eroded again in summer  from surface water to the streambed rather than wood-induced HEF, resulting in the more stable temperatures observed.
Areas downstream of wood were variable in their response to daily and seasonal temperature fluctuations. TLs 9 & 10, which were located in a mid-channel sandbar created by the wood had large temperature variation, and evidence of warmer average daily temperature than surface water to a depth of 30 cm in summer, resulting in steeper temperature duration curves and average daily temperatures as shown in the diurnal plots. The build-up of exposed sediment downstream of wood at TLs 9 & 10 and exposure of the sediments above the water level allows these areas to receive increased atmospheric heat input, resulting in this sustained increase in summer temperature due to heat conduction, as can be seen during the hottest day. Similar results have been reported by Arrigoni et al. (2008) in exposed gravels and cobbles of a gravelbed river and a sandy desert stream by Valett, Fisher, and Stanley (1990) (Baxter & Hauer, 2000;Ebersole, Liss, & Frissell, 2003) and microbes . In comparison, our results from a small lowland sand bed stream show that rather than providing thermal buffering to the streambed (colder temperatures in summer, warmer temperatures in winter), the wood increased the seasonal temperature extremes (increased summer and decreased winter temperatures) at shallow depths by enhancing infiltration of warmer (summer) and colder (winter) surface water. This relative difference in seasonal surface water temperature is due to the slower flow and more open nature of the vegetation canopy at lowland river sites, which result in more extreme surface water temperatures than those experienced at upland sites (Garner, Malcolm, Sadler, & Hannah, 2017).
This loss of temperature buffering and hyporheic refugia has the potential to have either negative or positive effects on streambed dwelling biota. Aquatic insects and cold-water fish species generally have an optimal thermal regime which determines the success, rate and duration of larval growth and development and adult survival (Burkholder et al., 2008;Malcolm et al., 2004;Vannote & Sweeney, 1980). In these cases, the promotion of more extreme streambed temperatures and loss of thermal refugia is likely to negatively impact these species. Previous research by Magliozzi, Usseglio-Polatera, Meyer, and Grabowski (2019) on the Hammer Stream reported that locations associated with wood structures typically contained hyporheic invertebrates associated with temporal instability and flow disturbance in comparison to control sites without wood, suggesting that the dynamic (geomorphic, hydraulic and thermal) conditions present near the wood altered the benthic community.
Other streambed biota may not be as adversely impacted by seasonal temperature extremes. For example, work by Silva and Williams (2005) has shown that hyporheic microbial richness was positively correlated with temperature and vertical hydraulic gradient, whereby alteration of streambed temperatures due to HEF manipulation led to an increase in the growth of nitrifying bacteria which were normally less dominant in the streambed. Streambed microbial activity was also found to respond to diel temperature fluctuations, with a peak in activity when sediment temperatures were the highest (Claret & Boulton, 2003). These results, therefore, suggest that the observed variability in streambed temperatures around wood structures and in particular downstream of wood are likely to influence microbial communities and their subsequent activity which benefit from the improved spatio-temporal temperature diversity provided by the presence of wood. Indeed, our previous research within the Hammer Stream has shown that reach scale ecosystem respiration at baseflow conditions were highest in reaches with higher wood loading (Blaen et al., 2018), although nitrate reduction adjacent to instream wood (<1 m away) was only significantly higher at high flow events (Shelley et al., 2017). Evidence by Zheng, Cardenas, and Wang (2016) further revealed that higher streambed temperatures increased the nitrate removal efficiency of the streambed, resulting in the hyporheic zone becoming a nitrate sink rather than a source (particularly where nitrate supply is high). These results suggests that further work on the potential of wood restoration programs in attenuating high surface water nitrate levels, such as those found in lowland, agricultural areas remains. However, future reintroduction and promotion of wood in rivers should consider the local conditions and range of pressures which differently affect upland and lowland rivers to ensure that restoration efforts are designed to maximize the desired management objectives.