Stream temperature under contrasting riparian forest

• We assess stream temperature and energy fluxes under 3 riparian vegetation types. • Stream temperature varies significantly between different vegetation types. • Net energy fluxes are greatest in open grassland and lowest in coniferous

Stream temperature under contrasting riparian forest cover: Understanding thermal dynamics and heat exchange processes Water temperature is one of the principal river habitat variables, exerting an influence on chemical, physical and biological processes within the river channel (e.g. Caissie, 2006). River temperature is therefore important to freshwater fish species, many of which inhabit relatively tight thermal niches (e.g. Jonsson and Jonsson, 2009). In addition to influencing levels of growth and survival, river temperatures can alter behaviour and drive mortality events, eventually leading to population decline (Breau et al., 2011;Dugdale et al., 2016;Martins et al., 2012). Even under conservative emissions scenarios (see IPCC, 2013), most global climate models indicate that global river temperatures will increase by~1.0-3.0°C (Ficke et al., 2007;Morrill et al., 2005;van Vliet et al., 2013;van Vliet et al., 2011) by 2070-2100. Although the magnitude of temperature increase will inevitably vary between locations, large temperature rises are expected in some northern European watersheds (e.g. van Vliet et al., 2011;Webb, 1996) where temperature-sensitive fish species are often resident. Within these locations, exposed upland streams such as those in Scotland will be particularly susceptible to warming due to their relatively low thermal inertia (in comparison to larger rivers) and their lack of substantial shading (e.g. Hrachowitz et al., 2010;Jackson et al., 2017). Indeed, recent research suggests that temperatures in Scottish upland rivers could increase by an average of 2.2°C by 2050 alone (Capell et al., 2013;Punzet et al., 2012), and the high degree of exposure of large tracts of Scottish rivers (forest cover in Scotland is only 17.1%; Forestry Commission Scotland, 2006) means that such a warming could have serious impacts on Scotland's fluvial ecosystems. Research is therefore needed to help understand and mitigate river temperature extremes in these exposed upland environments.
Over the past twenty years a large number of river temperature studies have provided valuable insights into the dominant processes driving river temperature regimes (e.g. Caissie, 2006;Hannah and Garner, 2015;Webb et al., 2008). At the largest scales, temperature is controlled by energy fluxes at the air-water and streambed interfaces (Caissie, 2006;Webb, 1996). Although advective and sensible heat fluxes at these interfaces exert an influence on river temperature (Webb and Zhang, 2004;Yearsley, 2009), radiative and evaporative fluxes at the water surface are generally considered the dominant heat exchange mechanisms (Caissie, 2006;Caissie, 2016;Maheu et al., 2014;Morin and Couillard, 1990). While the role of these heat fluxes in determining stream temperature is reasonably well understood, the magnitude and direction of energy transfer processes can be substantially altered by the properties of the landscape through which the river flows (e.g. Benyahya et al., 2012;Dugdale et al., 2015;Garner et al., 2014;Leach and Moore, 2010). Indeed, the degree to which a river responds to radiative and climatic forcing is heavily dependent on patterns of land use and topography within the basin, and is thus complex and multi-faceted Laizé and Hannah, 2010).
Despite the difficulty in elucidating many of the landscape properties controlling river temperature, some landscape -river temperature interactions have long been recognised. For example, experiments conducted during forest clearcutting since the early 1970s have demonstrated how the removal of riparian vegetation can result in greatly increased summer water temperatures under some circumstances (e.g. Brown and Krygier, 1970;Burton and Likens, 1973;Keith et al., 1998;Moore et al., 2005a;Rishel et al., 1982;Roth et al., 2010;Zwieniecki and Newton, 1999). Given the reasonable assumption that the addition of new riparian vegetation will have the opposite effect, river managers have started to pursue riparian tree planting as a strategy for moderating high temperature events in summer (e.g. Drainey, 2012;EA, 2011;Lawrence and Dandy, 2014;Lowe et al., 2012;Parrott and Holbrook, 2006;Withrow-Robinson et al., 2011). However, it is only recently that studies have started to identify the precise manner by which riparian tree cover causes an apparent cooling effect on river channels (see Garner et al., 2014). Furthermore, while previous research has documented how stream temperature (and corresponding energy fluxes) can differ between forested and exposed river reaches (e.g. Garner et al., 2015;Hannah et al., 2008;Leach and Moore, 2010;Webb and Zhang, 1997, little is known about how stream temperature dynamics respond to variability in riparian forest cover or species. Indeed, while the majority of previous studies (reviewed in Moore et al., 2005a) have concentrated on the impact of dense riparian conifer woodland on river temperature, riparian planting is increasingly focused around achieving multiple environmental benefits, and seminatural deciduous riparian woodland plantation is instead often preferred in the UK. There is therefore a pressing need to quantify how riparian energy fluxes respond to differing forest types in order to better understand the impact of different tree plantation strategies on stream temperature dynamics with a view to improved climate change mitigation.
This paper therefore presents a detailed comparison of stream temperatures and energy fluxes associated with three different riparian vegetation types in Loch Ard Forest, Scotland. First, we characterise spatial and temporal variability in stream water temperature regimes between open grassland, semi-natural deciduous and commerciallyplanted coniferous forested reaches of three streams. We then examine differences in energy fluxes under the three forest cover types. Finally, we assess how observed variations in energy fluxes are responsible for the observed patterns of stream temperature, with a view to understanding the physical processes driving stream temperature regimes under varying forest cover. We hypothesise that stream temperature regimes and energy fluxes will differ significantly between different forest treatments and that variability in these parameters will be largely driven by alterations in the amount of solar radiation received at each site as a result of contrasting riparian shading.

Study area
Loch Ard is a forest park situated between Loch Lomond and Aberfoyle (districts of Argyll & Bute and Stirling) on the west coast of Scotland (Fig. 1). The park is home to a long-term monitoring programme by Marine Scotland Science (started 1976) and has hosted a variety of studies on hydrology, hydrochemistry, acidification, freshwater fish and invertebrates (e.g. Harriman et al., 1995;Harriman et al., 2003;Malcolm et al., 2014;McCartney et al., 2003;Tetzlaff et al., 2007). The park contains a range of different riparian vegetation types, ranging from open grassland to both semi-natural deciduous and commercially-planted conifer woodland (Forestry Commission Scotland, 2015). Geology within the area is dominated by lowpermeability metamorphic bedrock covered by slow-draining peaty gleys and podsols (Tetzlaff et al., 2007). The area is subject to high levels of precipitation, receiving an average of 1980 mm per year (Tetzlaff et al., 2007). Winters are relatively mild for the latitude (mean January air temperature of 2.8°C) although mean summer air temperatures also remain low (mean of 13.7°C; Tetzlaff et al., 2010). Although snowfall does occur in winter, it is usually a relatively minor component of the annual hydrograph .
Within the forest park, three sites comprising stream reaches with different forest cover types were selected to contrast variability water temperature and energy fluxes. One grassland site was clear of any significant forest cover and was thus used as a control; the remaining two were chosen to characterise two riparian forest types typically found in Scotland: semi-natural broadleaf and commercial non-indigenous conifer. The control site (termed OS (open site) hereafter; plotted in red) was characterised by open (exposed) riparian grassland with no significant canopy cover. Clear felling within the 10-year period prior to this study ensured that no vegetation taller than~1 m existed within the riparian zone, where herbaceous plants/shrubs dominate. The OS sub-basin covers approximately 3.1 km 2 with an average altitude of 253 m. The semi-natural forest site (termed SNS hereafter; plotted in green) was characterised by mixed native deciduous species (birch, rowan and alder; approximate heights 10-20 m) growing directly on the stream banks. The SNS sub-basin covers approximately 1.3 km 2 with a mean altitude of 243 m. The commercial conifer plantation site (termed CS hereafter; plotted in blue) comprised Sitka spruce monoculture originally planted in 1978 (approximate heights 40-50 m). Mean altitude for the CS sub-basin was 367 m, with a basin area of 1.17 km 2 . The active channel width for all study reaches varied between 1.3 m and 1.5 m. The SNS and CS study reaches were oriented in the same approximate SW -NE direction. Although OS was oriented NW-SE, it is unlikely that this orientation will substantially alter energy fluxes in relation to SNS and CS because of the lack of riparian vegetation. Mean discharge at OS and CS (recorded from stage-discharge curves established between pressure transducers installed at the sites and a nearby gauging station) was 0.09 and 0.08 m 3 s −1 respectively. Although no pressure transducer was available for the SNS sub-basin, its similar size and location near to CS means that its hydrometric regime (i.e. average discharge) is very similar.

Field data collection
A range of measurements necessary to characterise the stream temperature regime and energy fluxes at each site were collected between 1st January and 31st December 2010. Automated weather stations (AWS) installed above the stream at each site were used to collect data following methods described by Hannah et al. (2004). Stream temperature (T w ) measurements were obtained from thermistors anchored in the water column and protected from the effects of sunlight by PVC tubing. Direct measures of incoming (K s ↓) and reflected shortwave radiation (K s ↑) and net radiation (Q*) were used to characterise radiative fluxes at the water surface, while air temperature (T a ), relative humidity (RH), wind speed (WS) were recorded in order to compute latent (Q e ) and sensible (Q h ) heat exchanges (see Section 2.3). Bed heat flux (Q bhf ) and bed shortwave radiation (K b ↓) were measured using instrumentation installed within the channel bed, while stream bed temperatures (T b ; measured using thermistors buried at 0.05 m, 0.20 m and 0.40 m) were used to compute new streambed longwave flux (L b *; see Section 2.3). A complete list of variables (and instruments used to collect the data) is given in Table 1. All instruments were cross-calibrated prior to installation and found to be in good agreement. Instruments were polled at 10 s intervals and mean values (maximum for relative humidity) logged at 15 min intervals. The automated weather stations (AWS) were serviced and downloaded at approximately 6-week intervals.
All data for CS between 14th April and 3rd June were lost due to battery failure. Bed radiation (K b ↓) data were rejected on several occasions where high flows following precipitation events had dislodged the pyrometer. Bed heat flux (Q bhf ) values at CS between 15th July and 31st December were found to be erroneous and were removed from the dataset. Streambed temperature (T b ) data at 0.40 m at OS were discarded due to a sensor fault that persisted over the duration of the study; a similar problem led to the rejection of bed temperature data at 0.05 m from SNS between 1st January and 30th May. Instruments were periodically obstructed following high snowfall/precipitation events; these periods were identified in the measurement time-series and removed from subsequent analysis.

Stream energy balance
Webb and Zhang (1997) express the energy budget of a stream not receiving significant inflow as: where Q n is the total energy flux, Q* is the net radiative flux (all wavelengths), Q h is the sensible heat flux, Q e is the latent heat flux, Q b is the energy flux from bed conduction, Q f is the energy flux resulting from friction with the bed and banks and Q a is the advective heat flux resulting from runoff and/or groundwater inputs. Fluid friction in this study was considered to be negligible and was omitted from the energy budget calculation. Similarly, in the absence of data regarding advective groundwater/runoff inputs, Q b and Q a were substituted with Q bhf which aggregates conductive, convective, radiative and advective processes at the stream bed . The stream energy budget at each of the three sites was thus based on the following modified equation: The following flux terms needed to complete the energy budget at each site were measured or estimated from data observed at the three AWS, according to the methods detailed in Hannah et al. (2004Hannah et al. ( , 2008: 1. Net short-wave (K s *) and net long-wave (L s *) radiation at the stream-air interface. 2. Streambed net short wave (K b *), net long wave (L b *), and net allwave streambed radiation (Q b *). 3. Latent heat flux (Q e ), using a Dalton-style equation to derive heat lost by evaporation or gain by condensation (Evans et al., 1998;Webb and Zhang, 1997). 4. Sensible heat flux (Q h ), as a product of Q e and the Bowen ratio (Bowen, 1926).
All energy fluxes are subsequently reported in MJ m 2 d −1 and are positive (negative) when adding (removing) heat to (from) the water column.

Data analysis
Patterns in stream temperature and energy fluxes are examined both in terms of annual and seasonal (winter and summer) timescales. Although plots/tables show data for the entire year, sporadic data gaps during spring and autumn made analysis of patterns during these seasons difficult. We therefore focus on the winter and summer months because a) data availability was greatest during these time periods and b) winter and summer are of high importance from a biological perspective (e.g. Dugdale et al., 2016;Wirth et al., 2012). Furthermore, contrasts between sites were maximised during the winter and summer seasons, making inter-site differences more readily apparent. For the purposes of this study, winter was defined as December, January and February and summer was defined as June -August. Where data was missing at one or more sites, descriptive statistics and data analyses were only calculated using concomitant data from all three sites, meaning that like-forlike comparisons between sites are valid.
Preliminary analyses (Durbin-Watson tests, examination of autocorrelation/partial autocorrelation plots; see Dickson et al., 2012) revealed serial autocorrelation of all timeseries. Significance of inter-site differences was therefore quantified using a mixed-effects analysis-of-variance (ANOVA) incorporating a first-order autocorrelation (AR1) process; this ensured that the reporting of significant differences between sites was not influenced by the autocorrelation of residuals. The strength of association between pairs of variables (e.g. air and water temperature) was assessed using the coefficient of determination of linear regression (R 2 ). In order to account for the effect of serial autocorrelation on the linear regression, autoregressive integrated moving average (ARIMA) models (e.g. Benyahya et al., 2007;Gurnell et al., 1992) were fitted to each variable; linear regressions were subsequently established between the ARIMA model residuals, effectively removing the influence of autocorrelation.

Stream temperature
Stream temperature was significantly different across the three sites (mixed-effects ANOVA p b 0.01). Averaged across the study period, mean daily stream temperature was warmest at the semi-natural forest site (SNS; 7.6°C) and coolest in the commercial conifer plantation (CS; 6.5°C). However, despite SNS yielding higher average temperatures, OS (control open grassland site) attained both the highest summer (June-August) and coolest winter (December-February) daily mean temperature (13.8°C and 0.6°C respectively). Stream temperature at OS was also the most variable, as demonstrated by the relatively high standard deviation (5.7°C) compared to SNS or CS (Table 2). Inter-site differences in stream temperature also varied between seasons (Fig. 2a). While in the summer, daily mean temperatures are consistently ordered OS N SNS N CS (13.8, 13.3 and 12.8°C respectively), a transition occurs in the cooler months, meaning that in winter, SNS is considerably warmer (3.0°C) than OS or CS (0.6 and 0.8°C respectively). This temperature difference is presumably due (in part) to the effects of winter ice formation at OS and CS which was not replicated at SNS. Diurnal stream temperature dynamics follow a similar but modified annual cycle (Fig. 2bd). In summer, OS consistently records the greatest mean daily temperature range (4.1°C), followed by CS (3.4°C) and SNS (3.1°C). However, during winter, the mean daily temperature range at SNS (1.0°C) is higher than either OS or CS (0.6 and 0.7°C), presumably a function of the substantially higher winter stream temperatures at this site. Stream temperature was moderately well correlated to air temperature (T a ; Supplementary material; Fig. S1a, Table S1) at all sites (R 2 = 0.61, 0.49 and 0.40 respectively for CS, SNS and OS respectively). However, the decreasing correlation strength from CS to OS indicates that air temperature explains increasingly less variability in stream temperature as tree cover decreases.

Energy fluxes at the air-water interface
Seasonal variability in solar shortwave and longwave radiation means that net radiation is characterised by a strong annual cycle, whereby Q* represents a heat sink (due to outgoing longwave radiation) in winter and a substantial heat source (incident shortwave) during warmer months. In terms of net solar shortwave radiation (K s *), minimum and maximum values track the winter and summer solstices (Fig.  3a); outside of the spring/autumn equinoxes, incident shortwave radiation at all sites generally drops below 20% of the summer maximum. Clear inter-site trends are present in the K s * observations. OS recorded the highest average value (6.3 MJ m 2 d −1 ) followed by SNS (4.3 MJ m 2 d −1 ) and CS (2.0 MJ m 2 d −1 ); these inter-site differences were significant (p b 0.01). Clear seasonal trends are also visible in the solar radiation differences between the three sites (Table 3). While SNS and CS receive~67% and 34% of the total radiation of OS during summer,  the difference between OS and SNS/CS is further increased during the winter months (with SNS and CS only receiving 53% and 19% respectively of the radiation at OS). Averaged over the year, net longwave radiation (L s *) is a heat sink at all sites (−2.2, −1.3 and −1.2 MJ m 2 d −1 respectively for OS, SNS and CS; Table 3). Results of the mixed-effects ANOVA indicate that there is no significant difference in mean net longwave radiation across the three sites. However, visual inspection of the data indicates some weak site-dependent seasonal patterns (Fig. 3b). OS is characterised by a pronounced annual cycle, whereby energy loss from L s * is substantially reduced in summer (x̄= − 0.5 MJ m 2 d −1 ) compared to winter (x̄= −4.1 MJ m 2 d −1 ). Conversely, CS and SNS exhibit much reduced seasonal variability. While SNS records a similar pattern of greater L s * losses in winter than in summer (−1.7 (winter) vs −1.2 (summer) MJ m 2 d −1 ), this pattern is inverted at CS, where most energy is lost in summer (−1.4 (summer) vs −1.1 (winter) MJ m2d −1 ). Despite the relative seasonal stability of CS (and to a lesser extent, SNS), absolute differences in L s * among all three sites show clear seasonal trends, with differences maximised around the annual solstices and minimised during the equinoxes.
Latent heat contributes almost exclusively to heat loss (Table 3) at all sites with the exception of isolated occasions whereby Q e -driven energy gains (condensation) occur either due to winter ice cover or during occasions when high relative humidity (Fig. S1b, Table S1) coincides with elevated air-water temperature gradients (Fig. 3c). This is supported by the moderate (significant) correlation between relative humidity and Q e at all sites (R 2 = 0.53, 0.57 and 0.54 at OS, SNS and CS respectively). Over the course of the year, average heat loss due to Q e is greatest at OS (− 1.5 MJ m2d − 1 ), followed by SNS (− 1.0 MJ m 2 d − 1 ) and CS (−0.4 MJ m 2 d −1 ); these differences were found to be significant (p b 0.01). No significant difference in air temperature (T a ) was observed between the three sites, so the inter-site ordering of latent heat losses are likely due to differences in humidity (81 ± 8%, 82 ± 7% and 89 ± 7% for OS, CS and SNS respectively) and wind speed (1.0 ± 0.7 m s −1 , 0.4 ± 0.3 m s − 1 and 0.1 ± 0.1 m s − 1 for OS, CS and SNS respectively; Fig. S1c, Table S1) whereby increased wind speed and decreased humidity enhance evaporation. A strong seasonal cycle is present at all sites, whereby Q e heat losses are maximised in the summer and minimised in the winter. Inter-site differences also track the same annual cycle whereby differences between the three sites are generally low (b±  1 MJ m 2 d −1 ) during October to February but more pronounced during the remaining part of the year. Averaged over the course of the year, SNS and CS lose energy due to sensible heat exchange (−0.2 ± 0.7 and −0.1 ± 0.4 MJ m 2 d −1 respectively) while Q h at OS contributed an almost negligible amount to the overall energy budget (0.0 ± 0.8 MJ m 2 d −1 ). While the differences between the three sites were not found to be significant, there is some (albeit limited) evidence for inter-site variability in annual trends (Fig. 3d), presumably (like Q e ) due to differences in humidity and wind speed. At OS, Q h is generally stationary with little seasonal change in mean. However, standard deviation is decreased in summer compared to winter (1.0 vs 0.4 MJ m 2 d −1 ; Table 3), indicating that the magnitude of daily Q h fluctuations have a strong seasonal component. In addition to this seasonal variability in the magnitude of daily Q h fluctuations, SNS and CS also display a non-stationary trend whereby energy is lost due to sensible heat flux in the winter and gained during summer (− 0.8 vs −0.1 and 0.1 vs 0.1 MJ m 2 d −1 for SNS and CS respectively).

Energy fluxes at the water-streambed interface
Total heat flux (Q bhf ) between the channel and streambed is several orders of magnitude lower than at the air-water interface (Table 4). While no significant difference was observed in the mean streambed heat flux recorded across the three sites, seasonal patterns vary between sites (Fig. 4a). OS recorded both the highest (mean) and most variable bed heat flux (0.1 ± 0.6 MJ m 2 d −1 ) and also exhibited a moderate seasonal trend whereby the streambed generally acts as a heat source in winter (x̄= 0.5 MJ m 2 d − 1 ) and sink during summer (x̄= − 0.6 MJ m 2 d − 1 ). At CS, despite the considerable loss of data, available records suggest the existence of broadly similar seasonal trends to OS (x̄= 0.1 and −0.3 MJ m 2 d −1 for winter and summer respectively). However, averaged across the year, mean bed heat flux is negative and temporal variability is muted (−0.1 ± 0.3 MJ m 2 d −1 ) in comparison to OS, indicating that bed heat fluxes contribute less to river temperature. Bed heat flux at SNS is more clearly different to OS or CS. Although mean Q bhf at SNS was situated between the other two sites (0.0 MJ m 2 d −1 ), daily variability was substantially reduced (σ = 0.1 MJ m 2 d −1 ). Furthermore, although cursory inspection of the data suggests that like OS/CS, bed heat flux is negative in summer and positive in winter (x̄= 0.1 and −0.1 MJ m 2 d −1 respectively), the magnitude of bed heat fluxes is extremely small.
Net radiative fluxes at the streambed follow broadly similar trends to those at the air-water interface. However, none of the inter-site differences in Q b * (and also K b * and L b *) were found to be significant. In terms of annual cycling in net solar radiation received by the bed, maximum K b * values were observed near to the summer solstice and extremely low values (b0.5 MJ m 2 d − 1 ) were recorded at all sites during the winter months (Fig. 4b). These winter minima result in Table 3 Descriptive statistics (daily mean, minimum, maximum and standard deviation) for variables describing energy fluxes at air-water interface.

Annual
Winter Summer part from persistent ice cover which blocked incoming solar radiation at the stream surface. In terms of seasonal patterns in longwave radiation, OS is characterised by a relatively strong annual cycle, with positive L b * values mainly recorded during the summer months and negative values in the winter (Fig. 4c). CS demonstrates a similar annual trend, but with a lower range. Calculation of L b * for SNS was limited to periods where streambed temperature data at 0.05 m was available (Fig. S2), meaning that it is difficult to discern any clear annual/seasonal cycle.

Total energy available
Owing to the relatively minor contribution of bed heat flux to the overall energy balance and the large loss of bed heat flux measurements at CS, comparison of the total energy available at each site is given in terms of total heat flux at the water surface (Q sn ; Fig. 5a) rather than total heat flux (Q n ). Averaged over the course of the year, all three reaches revealed a net energy gain (Table 3). However, the magnitude of energy fluxes differs significantly (p b 0.01) between the three sites, with OS recording the highest gain (2.6 MJ m 2 d −1 ) followed by SNS (1.9 MJ m 2 d −1 ) and CS (0.3 MJ m 2 d −1 ). In terms of seasonal variability in net energy available at the stream surface, Q sn is strongly positive in summer (x̄= 8.5, 5.6 and 1.9 MJ m 2 d −1 respectively), and negative in winter (x̄= −2.9, −2.2 and −1.4 MJ m 2 d −1 respectively). This result reveals the existence of a clear inversion in site ordering, whereby total energy fluxes are ordered OS N SNS N CS in summer and CS N SNS N OS in winter. Although missing data means that it is difficult to determine exactly where this transition occurs, inspection of Fig. 5a indicates that inter-site differences in net energy fluxes are minimised during mid-March and mid-October. It is therefore likely that the vernal and autumnal equinoxes mark the transition from winter to summer energy balance. Regression analysis allows for the quantification of the intersite similarity between energy fluxes. These data permit a further understanding into how each site 'behaves' from an energy balance point of view. At first glance, Table 5 appears to shows slightly conflicting results in terms of inter-site similarity between the various fluxes. Turbulent (Q e and Q h ) and longwave exchanges indicate that the SNS-CS correlation is generally highest, whereas in terms of solar radiation, the correlation is maximised between OS and SNS. However, given the relative contribution of K s * to the overall energy balance, this has the effect of yielding a clear distinction in terms of total energy flux (Q sn ), whereby the high degree of correlation between OS and SNS indicates that these sites are much more closely matched in terms of energy balance than either OS/CS or SNS/CS. Table 6 highlights seasonal variability in the relative contribution of different heat sources/sinks to the energy balance at each site. In terms of energy gains, K s * is the dominant heat source across all seasons (Fig.  5b-d). However, its relative contribution to the energy balance is lowest in the winter, where Q h contributes the bulk of the remaining positive energy budget (Table 6). This effect is greatest at OS (and to a lesser extent, CS), where freeze-up conditions drive Q h gain; the higher winter water temperatures at SNS explain the lower Q h contribution at this site. In summer, Q h is a relatively minor contributor to energy gains, but is higher at CS and SNS than OS. Latent heat (Q e ) gains also contribute more to the winter energy balance than in summer, particularly at OS and CS, (presumably due to condensation on ice cover at these sites). In terms of longwave radiation, L s * is a minor contributor to energy gains at CS during the winter, but is negligible at OS or SNS. However, this pattern inverts in the summer, with L s * contributing nothing to energy gains at CS, but a small amount at SNS and OS. L s * is the dominant heat sink when averaged over the course of the year, followed by Q e and Q h . However, closer inspection of the seasonal components reveals considerable sub-annual and inter-site variability in heat loss partitioning. In the winter, L s * is the dominant heat sink at all sites. However, while L s * is responsible for over three-quarters of energy losses at OS, its contribution is much reduced at CS and especially SNS. In the summer, L s * remains the dominant heat sink at CS (accounting for two-thirds of energy losses), but is a substantially smaller term at OS and SNS where it is superseded by Q e as the dominant heat loss term. Indeed, latent energy exchanges contribute substantially to summer losses at all sites, particularly at Q e . In the winter, Q e -driven losses display an inverted pattern, with SNS recording the highest magnitude loss, presumably due to the increased winter stream temperature at SNS. Remaining energy transfers away from the stream are driven by sensible heat processes, the magnitude of which changes considerably between winter and summer. In winter, Q h is a substantial heat loss term at SNS, again presumably due to elevated winter stream temperatures at this site. Q h is slightly lower at CS and more substantially reduced at OS, but still represents a notable heat sink at all sites. However, in summer, Q h contributes extremely little to heat loss at SNS and CS, and while greater at OS, is still a relatively heat sink term.

Stream temperature and energy flux response to different forest cover
This study extends the findings of existing investigations into the effects of riparian woodland on river energy budgets and temperature regimes. While the majority of previous studies focused on comparing  forested environments to open moorland/grassland/clearcut, here we provide an insight into how river temperatures and corresponding heat fluxes are dependent upon the specific type of forest cover. The results of this study provide new information to help refine understanding of river temperatures under contrasting land uses, and have important implications for forest and river management practises in temperate regions (in which these forest types are endemic). This study agrees with previous research (e.g. Broadmeadow et al., 2011;Hannah et al., 2008;Webb and Zhang, 2004) which indicates that deciduous forest cover (SNS) is able to moderate summer stream temperatures in a similar fashion to (potentially denser) conifer forest (CS). However, in this investigation, summer stream temperatures at SNS were found to be significantly warmer than at CS. Indeed, while mean summer stream temperature at CS was 1.0°C cooler than OS, this difference was halved at SNS (ΔT͞ w (SNS,OS) = 0.5°C). Inspection of energy budget data assembled from the AWS observations at each site outlines the reasons for these results. During the summer months, net surface heat flux (Q sn ) was ordered OS N SNS N CS. Because surface heat flux (Q sn ) is a major (generally, the major; Hannah et al., 2008;Webb andZhang, 1997, 1999) driver of summer stream temperatures, it is unsurprising that summer T w followed this same order. Solar radiation is the principal mechanism controlling surface heat flux in summer (Caissie, 2006). These inter-site variations in surface heat flux (and, by extension, water temperature) therefore arise from significant differences in solar radiation received at the stream surface. K s * variability between the three sites is driven by changes in the cover of riparian woodland at each site. This forest cover acts to shade the stream from K s *, effectively blocking a given proportion of solar radiation from being received at the stream surface. CS recorded the smallest energy gain due to the presence of dense conifer vegetation which blocked K s *, limiting the amount of energy received at the air-water interface. The dense forest also reduced wind speeds and augmented humidity in relation to OS and SNS, thus limiting turbulent heat exchange. Nevertheless, the reduction in Q e losses that this entailed (in comparison to OS) was not sufficient to supplant the shortfall in net energy produced by the reduction in K s *. Conversely, OS recorded the highest net energy gain (and hence, highest stream temperature) due to the absence of any riparian vegetation that would otherwise block solar radiation received at the stream surface. Although this, combined with increased wind speed due to the absence of tree cover, drove significantly higher latent heat fluxes at OS in comparison to SNS or CS, the magnitude of these exchanges was insufficient to offset the effect of the increased K s * values on water temperature. At SNS, summer energy gain at the air-stream interface was a composite of that of CS and OS. Like CS, K s * was reduced in comparison to OS due to the presence of forest cover. However, a significantly larger amount of solar radiation was received at the air-water interface than at CS. This increase in K s * indicates that in this study, deciduous vegetation provided a lower degree of shading than the coniferous cover.
While previous studies have indicated that light transmission is normally lower through deciduous trees than through coniferous vegetation (e.g. Konarska et al., 2014;Lintunen et al., 2013), this investigation recorded significantly higher K s * at SNS than at CS. This apparent contradiction is presumably due to the fact that forest cover at SNS was much lower than at CS. Hemispheric photography of the riparian canopy at SNS and CS supports this theory, whereby forest cover at CS (quantified here using Gap Light Analyzer; Frazer et al., 1999 as the proportion of light in the above-stream hemisphere that is blocked by vegetation; see Garner et al., 2014) was found to be substantially higher than at SNS (80% vs 66%). This lower forest cover relative to CS also explains why wind speed and humidity were reduced (and hence, evaporative heat losses increased) in comparison to the conifer site. Taken together, and given the lack of significant inter-site differences between any of the other energy budget terms recorded in this investigation (e.g. longwave radiation, sensible heat flux), these results clearly demonstrate that summer stream temperature differences between forest treatments are driven by inter-site variability in K s * resulting from differing levels of riparian shading (rather than by variability in the other components of the stream energy budget). Furthermore, our findings suggest that while both deciduous and coniferous forest types are able to moderate summer water temperatures, their effectiveness in doing so differs significantly, likely as a function of their contrasting stand densities (i.e. no. of trees per m 2 ), forest structure (e.g. shape and size of the canopy and branches; Lim et al., 2003) and canopy architecture (e.g. leaf area index and leaf inclination; Welles et al., 1991).
The long-term perspective offered by this dataset also allows for an insight into how winter stream temperatures and energy budgets vary between forest treatments. Substantial differences in winter stream temperature were observed between the three sites, and it is pertinent to examine the extent to which this variability was driven by the effects of forest cover on radiative energy transfers. Solar radiation (K s *) followed the same trends as during the summer, with lowest values recorded at CS and highest values at OS. While this trend again emphasises the role of forest cover in preventing solar radiation from reaching the stream surface, the picture is slightly more complex in winter. During these colder months, net radiation (Q*) was ordered CS N SNS N OS because the presence of riparian canopy at CS (and to a lesser extent, bare tree limbs at SNS) acted to limit radiative losses by retaining and re-emitting longwave energy towards the stream surface. Given the magnitude of these L s * fluxes, net surface heat flux (Q sn ) was also consequently ordered CS N SNS N OS. However, stream temperatures did not mirror this result, with SNS instead recording substantially (significantly) higher temperatures than CS or OS. This departure from the expected pattern warrants detailed investigation. It is likely that these anomalously warm temperatures are due to groundwater exfiltration, but the lack of significant differences in bed heat flux between the three sites (or bed temperature data indicative of cool groundwater inflows at SNS during summer) means that any groundwater inputs into the SNS reach must occur upstream of the AWS. The increased winter stream temperatures recorded at SNS are therefore most plausibly due to heat advected from warm groundwater inputs upstream of the study reach. Energy advected towards SNS also accounts for the increased losses from sensible and latent heat during the winter because of the increased air-water temperature gradient in comparison to CS or OS. Overall, this apparently contradictory result at SNS has important implications for understanding the effects of forest cover on stream temperature and energy fluxes. In winter, the lower magnitude of net radiation (when compared to the summer) means that advective processes can have a larger impact on stream temperature than radiative energy transfers. However, in the summer, the substantial increase in radiative energy means that its effect on stream temperatures (and thus, the relative impact of forest cover) is considerably elevated in comparison to other energy fluxes. These results give credence to the findings of Garner et al. (2014) which also emphasise the importance of considering the influence of advective heat on stream temperatures, in addition to radiative fluxes. While it might be expected a priori that energy fluxes at SNS and CS behave similarly (given the presence of riparian vegetation at both sites and the implications of this energy flux variability), regression analysis suggests that this was not necessarily the case. Instead, OS and SNS were found to be more closely correlated in terms of net surface heat flux (Table 5). This result is presumably a function of both a) the increased transmission of K s * at SNS during the summer (in comparison to CS), b) the relative similarity of energy fluxes at OS and SNS during the winter, owing to the lack of foliage at SNS. Furthermore, the gridlike commercial plantation layout at CS results in an uneven diel shortwave flux cycle, whereby unlike at SNS (where shortwave radiation is uniformly reduced throughout the day), solar radiation at CS is substantially reduced by the dense forest stand until the sun reaches its zenith (midday), at which point there is sufficient space between individual trees for solar radiation to strike the water surface almost unimpeded. These potentially unexpected findings further highlight the fact that that the impacts of forest cover on river energy budgets are not fixed, and that the interaction between K s * and riparian forests varies across multiple axes of forest structure, canopy architecture and stand density.

Comparisons with other studies
4.2.1. Impacts of riparian vegetation on stream temperature.
In broad terms, the results of this study agree with other literature that demonstrates stream temperature reductions under forest cover when compared to open terrain (e.g. Brown et al., 2010;Hannah et al., 2008;Malcolm et al., 2008;Moore et al., 2005a;Webb and Crisp, 2006). However, detailed inter-comparison of the results of this study with similar investigations is difficult. Although a large amount of literature exists comparing stream temperatures and energy fluxes under (predominantly conifer) forested and open terrain, complications arise from a) their use of varying temperature metrics, b) the bias towards limited-length study periods that only encompass summer data  and c) the high degree of geographic variability among studies which encompass locations as disparate as New Zealand (e.g. Rowe et al., 1994), western Canada (e.g. Moore et al., 2005b) and the south-west United Kingdom (e.g. Webb and Zhang, 2004).
The vast majority of previous studies (particularly in the pre-2005 literature) comparing stream temperatures under forested and open terrain derive from the West Coast of North America (see Caissie, 2006;Moore et al., 2005a for review). Most of these studies only comprise summer data, and generally examine differences in maximum recorded temperature between conifer-lined reaches and agroforestry clear-cuts. Although some recent studies report relatively small differences in summer maximum temperature between forested and open terrain in the Pacific Northwest (~1-2°C; Guenther et al., 2014;Janisch et al., 2012;Kibler et al., 2013;Pollock et al., 2009), differences in maximum summer temperatures on the order of 4-9°C are more common (e.g. Dunham et al., 2007;Kreutzweiser et al., 2009;Leach et al., 2012;Moore et al., 2005a). Indeed, increases in summer maximums of up to 13°C have occasionally been recorded following clearcutting of commercial conifer plantations (Moore et al., 2005a). In this study, the largest daily difference in summer maximum temperature between CS and OS were towards the lower end of this scale (ΔT w,max (OS,CS) = 3.9°C), presumably due to the fact that that a substantial upstream area of the OS catchment is forested and water entering the study reach was already reasonably cool. This may explain the reduced temperature difference in comparison to studies from North America where clearcutting covers much larger areas, but the considerable differences in latitude, prevailing hydroclimatic conditions and catchment hydromorphology between our study site (west coast of Scotland) and the Pacific Northwest of North America mean that limited inferences can be drawn from this result.
Instead, it is more relevant to compare this study to previous investigations conducted in the United Kingdom, minimising differences arising from hydroclimatology and hydromorphology. Many previous studies from the UK incorporate longer term datasets (e.g. Garner et al., 2015;Hannah et al., 2008;Webb and Zhang, 2004) than those from other regions, providing additional information about stream temperature response to forest conditions during different seasons. Furthermore, rather than focusing solely on coniferous forest types, several UKbased studies have also examined stream temperature responses to deciduous vegetation, meaning that a fuller comparison of the results of this investigation can be achieved. These studies often record temperatures in terms of daily (or seasonal) means rather than maximums, further facilitating comparison with this investigation. In terms of mean annual temperature differences between forest treatments, Hannah et al. (2008) reported that stream temperature was 0.14°C warmer in deciduous woodland than in open moorland. However, studies examining annual temperature patterns between conifer-covered reaches and open terrain reported opposite trends, with forested reaches 0.4-0. 9°C cooler than open/clear-cut sites (Brown et al., 2010;Stott and Marks, 2000;Crisp, 1997;Webb and Crisp, 2006). While the results of this study demonstrate similar trends (deciduous site warmer and conifer site cooler than open site), the magnitude of differences between OS and SNS (− 0.8°C) was considerably higher than that reported by Hannah et al. (2008), presumably because of the (postulated) winter advective flux at SNS. However, the mean annual difference between OS and CS (+0.3°C) was closer to that recorded by previous studies examining the effects of coniferous cover (e.g. Crisp, 1997;Stott and Marks, 2000;Webb and Crisp, 2006).
In terms of mean summer temperatures, previous studies report differences of 2-3°C between conifer and open reaches (e.g. Brown et al., 2010;Webb and Crisp, 2006), whereas differences between deciduous and open reaches are generally on the order of 1-2°C (e.g. Broadmeadow et al., 2011;Garner et al., 2015;Hannah et al., 2008;Imholt et al., 2010;Malcolm et al., 2004;Malcolm et al., 2008). Although summer mean differences reported in this study are approximately half of these values for both the conifer and deciduous forest treatments (1.0 and 0.5°C respectively), they follow the same general trend of deciduous woodland moderating summer temperatures to a lesser extent than coniferous forest. The lower values presumably result from a combination of factors including differences in basin hydromorphology and geology, but are also likely due in part to the fact that Loch Ard is situated on the west coast of Scotland, where maritime conditions provide a different hydroclimatic context to previous UK-based studies (e.g. Broadmeadow et al., 2011;Brown et al., 2010;Hannah et al., 2008).

Impacts of riparian vegetation on energy budgets and radiative fluxes.
Energy budget partitioning in this study is generally consistent with similar studies that also highlight net radiation as a dominant source of energy at the air-water interface (Caissie, 2006;Evans et al., 1998;Hannah et al., 2008;Leach and Moore, 2010;Webb and Zhang, 1997. However, while an inter-comparison of the relative magnitude of net or solar radiation received under different forest treatments would be useful, there remain very few similar paired-site investigations to which the findings of this study can be contrasted. However, in one notable example, Hannah et al. (2008) observed that over the course of the year, a stream running through deciduous forest received approximately 41% of the solar radiation of the open (moorland) site.
Meta-analysis of similar data from Broadmeadow et al. (2011) indicates that mixed deciduous and coniferous woodland received an average of 33% of the radiation of open terrain. Both of these reductions are considerably higher than that recorded in this study, which indicates SNS re-ceived~67% of the mean annual solar radiation of OS. However, the difference in K s * received between CS and OS (32%) is broadly similar to that reported by Broadmeadow et al. (2011).
Inconsistencies in the reduction of solar radiation under deciduous forest cover between this study and previous research again likely results from differences in the forest environment present. Indeed, Webb and Zhang (1997) highlight the fact that the effect of tree cover on solar radiation attenuation can be highly variable. An examination of the absolute magnitude of energy fluxes recorded in this study reveals potential reasons for the lower-than-expected temperature reductions under forest treatments (when compared to other studies). Although it would be fallacious to compare absolute energy flux values between studies conducted at substantially different locations, Hannah et al. (2008) was conducted at a broadly similar latitude. A comparison of annual mean values between Loch Ard and the Girnock Burn (Cairngorms National Park, Scotland) site detailed in Hannah et al. (2008) is therefore relatively valid. Comparing OS to a similarly open site at Girnock Burn, there was little difference in the mean annual K s * observed (6.3 vs. 6.3 MJ m 2 d −1 respectively). However, L s * and Q e losses observed by Hannah et al. (2008) were indeed greater than in this study (−2.2 vs − 3.9 for L s * and −1.5 vs − 2.3 for Q e at Loch Ard and Girnock Burn respectively). The lower Q e losses reported here are presumably due to the decreased wind speed observed in this study in comparison to Hannah et al. (2008;1.0 vs. 2.3 ms −1 ), while the lower magnitude L s * loss is partially attributable to the cloudier maritime conditions at Loch Ard. The considerably increased annual precipitation at Loch Ard (approximately 4 × that of Girnock Burn) is also likely to have reduced L s * and Q e due to increased input from cool runoff. Conversely, when comparing energy fluxes at SNS to an analogous deciduous site at Girnock Burn, K s * was substantially higher in this study (4.3 vs 2.6 MJ m 2 d −1 respectively), while all other fluxes were relatively similar. This increased solar radiation at SNS compared to Hannah et al. (2008) clearly explains why the stream temperature reduction at SNS was lower than previous studies. Indeed, inspection of the results shows that in Loch Ard, it is actually CS (coniferous woodland) that most closely resembles the deciduous woodland reported in Hannah et al. (2008) in terms of energy fluxes. These results explain why the summer stream temperature difference between CS and OS was closer to that reported by Hannah et al. (2008) for deciduous woodland (~1°C ). Indeed, Imholt et al. (2013) note similar occurrences, whereby deciduous woodland in Girnock Burn had a larger impact on stream temperatures than conifer woodland at another nearby location. They attributed the differences to variability in stem density and tree coverage between the two sites. These results therefore lend further credence to the fact that the moderating effect of forest cover on water temperature is not solely a function of forest type or species but the result of a combination of factors including stand density and canopy architecture.

Implications for river management and future research
The results of this study shed further light on the role of riparian woodland in moderating temperature extremes in river environments. As such, they have important implications for the management of streams and riparian woodlands in order to limit the exposure of aquatic species to summer temperature extremes. Although this study echoes similar previous investigations that clearly highlight the ability of riparian forest to moderate high stream temperatures, the clear difference in stream temperature and energy budget between different forest treatments raises new questions regarding the efficacy of particular forest types and configurations in moderating summer temperature extremes.
While current riparian tree-planting initiatives often consider species or forest type (i.e. coniferous, deciduous) as a factor when determining planting strategies, emphasis is usually placed on the selection of forest types that satisfy anthropogenic demands (e.g. the preferences of landowners and farmers; Lawrence and Dandy, 2014), contribute to the re-establishment of native species (e.g. EA, 2011;Lowe et al., 2012) or conform to arboricultural practicalities (i.e. tree type is governed by what will grow well in a specific environment; Withrow-Robinson et al., 2011). These compromises mean that the resultant forest configuration may indeed be sub-optimal for the generation of shade suitable for moderating stream temperatures. While previous research (e.g. Broadmeadow and Nisbet, 2004) has documented strategies for successfully negotiating these regulatory and scientific requirements when designing effective riparian buffer strips, the goal of such work is often the enhancement of bank stability or the increased input of nutrients to the stream, rather than the specific creation of shading. As a result, Orr et al. (2015) highlight the clear need for more information into the combinations of tree size and buffer spacing that are most effective in generating shade to moderate water temperature, a research gap that recent studies (e.g. Garner et al., 2017) are slowly starting to address.
Given this pressing research need, and in light of the results of this study, we advocate the development of numerical models that are able to simulate tree shading (and the resulting stream temperature response through appropriate energy budget parameterisation) under different combinations of tree species, stand density and canopy architecture. Although many deterministic stream temperature models contain routines that are currently capable of simulating the impacts of tree shading on stream temperature patterns (e.g. Bartholow, 1989;Boyd and Kasper, 2003;Chen et al., 1998), none of the currently-available models are able to differentiate between the subtleties of different forest types or configurations. Furthermore, while an investigation by Imholt et al. (2013) revealed clear statistical linkages between forest allometric features and river temperature patterns, the study was largely empirical in nature and did not consider the effects of forest configuration on radiative transfers. As such, the next generation of stream temperature models should aim to incorporate ray tracing or radiosity approaches common to computer vision or remote sensing (e.g. Bittner et al., 2012;García-Haro et al., 1999;Jones and Vaughan, 2011) in order to more accurately simulate radiative transfers through specific riparian vegetation types and forest structures. The outputs of such models would allowing river managers to optimise riparian plantation strategies in order to maximise summer stream temperature reductions.
Recent research by Garner et al. (2017) simulating stream temperature responses to varying amounts of riparian shading, channel orientations and flow velocities represents a step in the right direction, and illustrates the utility of deterministic models for quantifying the riparian shading configurations that will have the largest water temperature 'impact'. However, there remains considerable uncertainty about how best to upscale the non-radiative energy fluxes recorded at spatiallyisolated weather stations to the quasi-continuous streamwise data required by river temperature models. While simulations of shading and or channel orientation can be used to scale radiative energy fluxes from isolated locations to whole-river extents, there are currently no deterministic models that compute the spatially-explicit impacts of riparian vegetation on latent or sensible heat exchange (through changes to wind speed or relative humidity). Although (as seen here), these energy flux components represent a smaller part of a river's overall energy budget and therefore contribute a smaller amount to stream temperature processes, future deterministic models should also look to simulate the impacts of riparian vegetation on these non-radiation energy fluxes. One possible way in which this could be achieved is through the development of empirical coefficients that could be applied to the latent/sensible heat flux components in a similar way to the 'shading factor' currently used to scale shortwave radiation; data necessary for informing these empirical coefficients could be derived from experiments using portable weather stations stationed throughout a variety of different riparian forest types.
In addition to the development of models capable of optimising riparian shading solutions for moderating stream temperatures, future research must also address the secondary impacts of riparian forest plantations on stream ecosystems. In addition to providing shade, riparian vegetation can (for example) modify sediment erosion/deposition regimes (e.g. Broadmeadow and Nisbet, 2004), impact invertebrate communities (both negatively; Rouquette and Thompson, 2005 and positively;Rios and Bailey, 2006) and distribution (through acidification pressure; Malcolm et al., 2014), and influence stream water quality and nutrient loading (e.g. Dosskey et al., 2010;Hefting et al., 2005). Indeed, while coniferous vegetation may produce a greater reduction in summer stream temperature than deciduous forest cover, broadleaf vegetation produces higher quality leaf litter (resulting in increased productivity; Broadmeadow and Nisbet, 2004). All aspects of the river's physico-chemical and biological environment must therefore be considered when devising plantation strategies in order to ensure that in searching for optimal shading results, the introduction of riparian vegetation does not actually result in a negative ecosystem response. Any further research into the comparative effects forest cover on stream temperature should therefore aim to take an holistic approach, comparing the simultaneous impacts of different riparian forest treatments on both stream temperature and ecosystem function.

Measurement uncertainty
We do not believe that measurement uncertainty impacted the findings of this study. Nevertheless, it is necessary to address potential sources of error within the dataset. In addition to instrument uncertainty associated with variables that were measured directly with instrumentation installed at the AWSs (i.e. air/water temperature, wind speed, relative humidity, shortwave and longwave radiation, bed heat flux), a further degree of uncertainty arises from instrument placement. For example, in the case of the pyranometers/net radiometers used to quantify radiative heat flux, small changes in instrument position can produce measurement variability, especially in shaded conditions where light transmission can vary considerably over a small area. Every attempt was therefore made to ensure that pyranometers/net radiometers were placed in a location where light transmission was consistent and representative of the location; we therefore believe that any uncertainty arising from instrument placement will be substantially less than the (considerable) differences recorded between the various sites. Instrument placement may also drive uncertainty in the streambed heat flux data. Numerous articles have documented substantial spatial variability in bed heat flux and bed temperature profiles (e.g. Birkel et al., 2016;Caissie and Luce, 2017;Gariglio et al., 2013;Leach and Moore, 2014), and given that bed heat flux was only measured in one location at each of our study sites, it is possible that the data recorded here does not adequately represent true bed heat flux or temperature. Nevertheless, given that bed heat flux was found to be several orders of magnitude lower than surface exchanges, uncertainty arising from spatial variability in bed heat flux is unlikely to alter the findings of this study.
Another source of uncertainty in the heat flux calculations arises from the use of equations to compute the turbulent (i.e. latent and sensible) heat fluxes. These fluxes are rarely measured directly, and the vast majority of previous studies (e.g. Hannah et al., 2008;Leach and Moore, 2010;Webb and Zhang, 1997) have, like this investigation, derived these fluxes mathematically using semi-empirical or physicallyderived equations. Although this means that it is difficult to ascertain the exact uncertainty associated with the computed latent and sensible heat flux values presented here, a comparative study of various heat flux equations (Ouellet et al., 2014) demonstrated that the Dalton-style equations used in this study to estimate turbulent heat flux performed among the best of the mathematic approaches documented in the literature. Although future studies should aim to improve these estimates using real measures of evaporation derived from mini evaporation pans (e.g. Maheu et al., 2014), we believe that the trends in turbulent heat flux data presented within this study are of sufficient accuracy to support our conclusions regarding the relative impacts of tree shading on stream energy budgets.

Conclusion
Future climate change is expected to result in raised summer stream temperatures in many temperate river basins, negatively impacting a wide range of fish species. This is likely to have serious consequences both for stream ecosystems and also for communities who benefit from freshwater fisheries. This investigation adds to the growing body of literature demonstrating the efficacy of riparian tree planting for the moderation of these summer water temperatures, and sheds new light on the response of stream temperature (and above-stream energy budgets) to different forest treatments. While related studies (e.g. Garner et al., 2015) note that the effectiveness of riparian planting will vary as a function of both current and future climatic conditions, our results demonstrate that the efficacy of riparian planting is also highly dependent upon the type and structure of forest stands. Future investigations into the effects of climate change on stream temperature must therefore consider not only future climatic/hydrological variability, but must also account for land-use with particular attention paid to the impacts of differing riparian vegetation. Coupled to the increasing sophistication of climate-hydrological-water temperature models, the recent proliferation of monitored riparian planting schemes across the UK and elsewhere means that there is increasing scope to study and analyse the real-world results of riparian planting on stream temperatures. Our results also highlight the complexity of the energy balance response to differing riparian vegetation types, not only in terms of the impact of riparian tree cover on shortwave radiation, but also its influence on all energy fluxes at the air-water interface. Data such as those presented here will be increasingly important for the adequate parameterisation of process-based stream temperature models. However, as data from existing tree-planting schemes becomes increasingly available, it is hoped that it will also be possible to use this to further refine stream temperature models to ensure they are able to fully account for the impacts of different riparian forest types on stream temperature. Such advances will not only add to the growing body of evidence regarding the impacts of climate change on rivers, but will also contribute to efforts to mitigate the effects of such changes on sensitive river environments.