Thermal regimes of groundwater‐ and lake‐fed headwater streams differ in their response to climate variability

Stream thermal regimes are being altered by climate change with consequences for aquatic organisms. Most documented long‐term changes in stream temperature are from large rivers. We know less about water temperature trends for small headwater streams, especially those found in northern landscapes that contain small lakes. We analyzed 36 yr of stream temperature observations from a long‐term watershed study in Ontario, Canada, to understand how headwater streams are responding to climate variability. We found that groundwater‐fed (GWF) and lake‐fed (LF) streams exhibit contrasting responses, as GWF streams warmed in the spring (0.19–0.60°C per decade) and LF streams warmed in the fall (0.39–0.72°C per decade). Both stream types exhibited weak temperature trends in summer and winter. These results highlight that a stream network perspective that includes headwater streams and small lakes, and accounts for seasonal changes in thermal regimes, is important for understanding aquatic ecosystem response to climate variability.

Stream temperature determines habitat suitability for aquatic organisms, such as fish, and influences in-stream biogeochemical rates, ecosystem metabolism, and water resources (McCullough et al. 2009;Van Vliet et al. 2012;Jankowski and Schindler 2019).Climate and land-cover changes are altering stream thermal regimes with potential consequences for aquatic ecosystems (Isaak et al. 2010;Leach et al. 2012;Luce et al. 2014;Menzies et al. 2022).Better understanding of how environmental change impacts stream temperature will help inform effective management, such as identifying and protecting thermal refuges, and contribute to sustaining healthy aquatic ecosystems (Arismendi et al. 2014;Isaak et al. 2015;Armstrong et al. 2021).
Numerous studies have documented long-term (20+ yr) stream temperature trends across North America, Europe, and Asia.Most streams have exhibited increasing stream temperature over the past decades (Webb and Nobilis 2007;Kaushal et al. 2010;Garner et al. 2014;Orr et al. 2015;Arora et al. 2016;Pohle et al. 2019;Ye and Kameyama 2021;Seyedhashemi et al. 2022;Tassone et al. 2023); however, the magnitude and direction of the trends can vary seasonally (Isaak et al. 2012;Michel et al. 2020;Kelleher et al. 2021).Observed stream temperature increases have been linked to changes in climate, land cover, and flow regulation.Some studies have reported long-term decreases in stream temperature, which have also been attributed to changes in streamflow regulation, as well as groundwater influence and reductions in industrial effluent (Kelleher et al. 2021;Ye and Kameyama 2021;Worrall et al. 2022).In addition, trend magnitude and direction can also be sensitive to length of the observation record (Arismendi et al. 2012).
The vast majority of long-term stream temperature datasets are from watersheds with catchment areas that range between 100 and 100,000 km 2 and most of these watersheds have been altered considerably by flow regulations or land cover changes, in addition to climate change (Kaushal et al. 2010;Arora et al. 2016;Michel et al. 2020).These human-induced watershed disturbances can confound the ability to isolate climate change influences on stream temperature (Kaushal et al. 2010;Arismendi et al. 2012;Kelleher et al. 2021).We have far fewer long-term stream temperature observations from small headwater streams (e.g., catchment areas less than 1 km 2 ), despite these streams making up a large portion of total stream network length (Gomi et al. 2002;Bishop et al. 2008).Headwater streams are typically well shaded, have greater groundwater influence and are more sensitive to land cover changes than larger streams and rivers (Richardson 2019); therefore, small streams may exhibit different temperature responses to environmental change than larger streams.
In addition to limited long-term stream temperature records from small headwater streams, we also have a paucity of observations from lake-stream networks (Carlson and Poole 2021).Small lakes are common in northern landscapes and are important components of surface water networks (Downing et al. 2006;Jones 2010;Verpoorter et al. 2014).Small lakes can influence downstream flow and thermal regimes (Spence 2006;Lisi and Schindler 2015;Leach and Laudon 2019;Hudson et al. 2021;Leach et al. 2021), thus providing novel habitats for fish and other aquatic organisms (Richardson and Mackay 1991;Pépino et al. 2017).Numerous studies have documented longterm lake temperature increases associated with climate warming at rates that typically exceed those for streams (O'Reilly et al. 2015;Schmid and Köster 2016;Roberts et al. 2017;Winslow et al. 2017;Li et al. 2022).Indeed, higher rates of long-term warming trends in streams have been linked to sites with greater proportions of upstream lakes (Webb and Nobilis 2007;Michel et al. 2020); however, these results are from large river systems and it is not clear if these findings extrapolate to small lakes within headwater stream networks.
In this study, we document seasonal stream temperature trends using 36 yr  of water temperature measurements made at 13 streams (6 streams sourced from shallow groundwater and 7 streams sourced from lakes) located in a small geographical area with relatively pristine forest cover conditions.All the sites are influenced by similar climate, land cover and geology; therefore, we posit that potential differences in stream temperature trends are associated with position within the stream network.The study site has seen modest increases in air temperature and incoming solar radiation, and decreases in stream discharge, although these trends are seasonally variable.Based on these hydroclimatic trends, as well as findings from previous studies, we hypothesize that (1) stream temperatures will exhibit warming trends across the study area, (2) the warming trends will differ between seasons and we expect more warming in summer and fall compared to winter and spring, and (3) lake-fed (LF) streams will have greater warming rates than groundwater-fed (GFW) streams.

Study area
The study was conducted at the Turkey Lakes Watershed (TLW; 47 03 0 N, 84 25 0 W) which is a 10.4 km 2 experimental catchment located 65 km north of Sault Ste.Marie, Ontario, Canada (Fig. 1).Forest cover is dominated by hardwood species.Soils in this rugged landscape are shallow (< 1 m depth) and overlay bedrock (Jeffries et al. 1988;Webster et al. 2021).TLW has a continental climate with lake effects from Lake Superior (Creed et al. 2003).The mean annual precipitation over the study period  was 1203 mm and approximately 35% of the total precipitation falls as snow (Beall et al. 2001;Webster et al. 2021).Seasonal snowpacks typically start forming in late November and remain through March-May (Beall et al. 2001;Webster et al. 2021).
We used stream discharge and temperature data collected at 13 hydrometric stations located within TLW: six of these stations are on streams that are sourced from shallow groundwater, and seven stations are on streams that drain small lakes (Hudson et al. 2021).Five lakes within the TLW range in surface area and depth (Supporting Information Table S1).With the exception of Wishart Lake, which is shallow and typically well mixed, the lakes are dimictic and complete ice cover typically extends between December and April (Jeffries et al. 1988).The six GWF stations used in this study were selected based on sufficient data availability, harvest history and catchment area (GWF-c32, GWF-c34, GWF-c35, GWF-c39, GWF-c42, GWF-c46; Supporting Information Table S2).Only catchment GWF-c34 has any recent harvest history, as the lower 70% of the catchment underwent a shelterwood cut in 1997.Analysis by Leach et al. (2022) suggested there were no detectable harvest effects on stream temperature at GWF-c34 and because this site is the largest catchment without a lake, we retained it in the analysis for comparison to similar sized catchments that contain lakes.Most of the GWF streams flow intermittently and usually dry up for a portion of July and August, with the exception of GWF-c34 (Leach et al. 2022).
Station catchment areas and distance downstream from lakes are summarized in Table S2 and illustrated in Fig catchment areas, whereas LF sites tend to be located on streams with larger catchment areas.Exceptions are GWF-c46, GWF-c34, LF-013 and LF-012, which have similar catchment areas (43, 69, 54, and 89 ha, respectively).In addition, the LF sites, with the exception of LF-013, are located on the same stream.We address how these aspects of the study design influence the key results in the discussion section.

Hydrometeorological data
Hydrometeorological observations, including air temperature, precipitation, solar radiation, snow, stream discharge, and lake ice phenology, have been collected at TLW and were used to provide context for the observed stream temperature trends.Meteorological measurements were taken at a 10 m tower located 1.5 km south-east of the TLW boundary (Fig. 1).Snow water equivalent (SWE) on 01 April was modeled using a snow accumulation and melt model informed by historical snow survey measurements (details of the model provided in Leach et al. 2020).Stream discharge was estimated at each hydrometric station based on water level measurements and stage-discharge relationships (Beall et al. 2001;Hudson et al. 2021).Lake ice cover observations were collected from 1980 to 2017.Visits by field staff to the vicinity of the lakes within TLW occurred at least weekly and as often as daily during spring melt.The dates used to characterize ice phenology correspond to permanent, complete ice coverage in the fall and complete disappearance of ice in the spring.

Stream temperature data
Water temperature data were collected during routine station visits over the 1983-2018 study period, except for station LF-013 which was monitored from 1986 to 2018.Technicians measured stream temperature using a handheld thermometer and observations were taken in the shade at the stream surface near the gauging cross section.Stream temperature was measured with a mercury-filled pocket thermometer from 1983 to 2001 and values were typically rounded to the nearest 0.5 C.After 2001, a digital thermometer was used and temperatures were recorded to the nearest 0.1 C. Additional details on the stream temperature observations, including number of observations per site and season, can be found in the Supporting Information (Section S2).

Trend analysis
We evaluated trends in stream temperature at seasonal scales given strong seasonal patterns in thermal regimes of temperate streams and the potential for annual trends to mask seasonal trends (Isaak et al. 2012;Kelleher et al. 2021).Following the standard meteorological classification for North America (Trenberth 1983), we defined the seasons as follows: Winter (December from the previous year, January, February [DJF]), spring (March, April, May [MAM]), summer (June, July, August [JJA]), and fall (September, October, November [SON]).These seasons also capture key patterns in hydroclimatic and tree phenology in this region, such as ice cover on streams (DJF), freshet (MAM), full tree canopy and high air temperature (JJA), and relatively high rainfall (SON).We calculated mean seasonal stream temperature from the spot measurements taken during a given season and year, and plotted the seasonal means with a linear smoothing line for visual reference.To quantify the change in seasonal mean temperatures over the 36-yr period we used the "trend" R package (Pohlert 2020) to calculate the Sen's slopes (Sen 1968) and associated 50% and 95% confidence intervals (CIs) for each season and site.We multiplied the slopes by 10 to report the change in temperature per decade.We also used visualizations and computed Sen's slopes to assess the long-term changes in seasonal hydroclimatic data and annual changes in 01 April SWE and lake ice phenology.

Hydroclimatic trends
Over the 36-yr study period, annual air temperature exhibited a 0.24 C per decade (95% CI [À1.4,0.67]) trend.Mean air temperature did not show strong seasonal trends in winter, spring, or summer (Fig. 2; Supporting Information Fig. S6).The fall air temperature trend was estimated as 0.46 C per decade (95% CI [À0.06, 0.97]).Incoming solar radiation also had weak seasonal trends for winter (negative), spring, and summer (both positive), but did exhibit an increasing trend during fall.Stream discharge estimated at the LF stations and GWF stations showed similar seasonal trends between groups (Fig. 2; Supporting Information Fig. S3).Stream discharge had a weak increasing trend in winter and decreasing trends in the other seasons.
Ice formation on the TLW lakes has been occurring later in the year (Fig. 2; Supporting Information Fig. S5), with an increasing trend of 2.7-4.1 d per decade.The 95% CIs do not overlap zero for Batchawana South and Wishart Lakes.Day of year when ice cover disappears from the lakes has exhibited weak trends over the study period, with Sen's slopes between À0.81 and 0 d per decade and 95% CIs overlapping zero for all lakes (Supporting Information Fig. S5).The 01 April SWE in the watershed had a decreasing trend of À9 mm per decade, but the 95% CIs overlapped zero (Fig. 2).

Stream temperature trends
Trends based on mean annual stream temperature showed warming across all sites, with generally higher rates for the LF streams (0.02-0.38 C per decade) compared to the GWF streams (0.04-0.23 C per decade; Supporting Information Fig. S6), although the 95% CIs overlap zero for all sites except LF-005.At seasonal scales, stream temperature trends also varied between GWF and LF streams (Figs. 3, 4); however, trends were generally similar within the two stream groupings.
Most GWF streams exhibited winter cooling trends (À0.23 C and À0.03 C per decade), although the 95% CIs overlapped zero for streams GWF-c35, GWF-c39, and GWF-c42.LF streams had winter trends between À0.02 C and 0.06 C per decade and mean temperatures were generally lower than those at GWF streams (0.5 C and 1.4 C, respectively).LF streams also exhibited weak trends in stream temperature during spring (ranging between À0.20 C and 0.07 C per decade).In contrast, GWF streams exhibited warming trends ranging from 0.19 C to 0.60 C per decade during spring, with only GWF-c39 having 95% CIs that overlapped zero.
Neither GWF nor LF stream groups exhibited consistent positive or negative trends in summer stream temperature (range across all sites was À0.31 C and 0.38 C per decade).GWF streams had slight increasing trends in fall (0.08 C and 0.31 C per decade); however, 95% CIs overlap zero in all cases.In contrast, most LF streams had increasing trends in fall stream temperature (0.39 C to 0.72 C per decade) and only LF-012 and LF-007 sites had 95% CIs overlap zero.

Discussion
A challenge with interpreting most stream temperature trend analyses is that data are often from large streams and rivers that may have undergone land cover and flow regulation changes; therefore, it is difficult to directly attribute changes to climate variability (Kaushal et al. 2010;Arismendi et al. 2012;Kelleher et al. 2021;Worrall et al. 2022).In this study, all the sites are influenced by similar climate, land cover, and geology; therefore, it was possible to isolate how thermal regimes of GWF and LF streams differ in their response to climate variability.We found that temperature trends of GWF streams show increases in spring, slight increases or no change in summer and fall, and ) of lake ice-on and ice-off, as well as 01 April SWE.Lake ice-on and ice-off day values are from Little Turkey Lake, which is generally representative of the trends in the other lakes (Supporting Information Fig. S5).Corresponding Sen's slope estimates are provided in Supporting Information Fig. S7.
slight decreases in winter.In contrast, temperature trends of LF streams show increases in fall, but little change in other seasons.
A warming trend for GWF streams in spring could be associated with declines in spring discharge and increases in incoming solar radiation.A decrease in discharge will result in a shallower water column and a corresponding increase in stream temperature if there is a positive input of energy to the stream, all else being equal (Arscott et al. 2001;Leach et al. 2023).Riparian vegetation of TLW streams is dominated by deciduous trees and leaf greening typically occurs in late April and May at this site (Creed et al. 2015); therefore, these streams lack effective shading during this period and increases in solar radiation could translate into temperature increases.Although declines in stream discharge may also be occurring at the LF streams, these sites are not exhibiting strong warming trends during spring.The presence of ice cover on the lakes, which typically extends into May and has shown minimal change over the study period, as well as the generally higher stream discharge for the LF streams (Hudson et al. 2021), appeared to stabilize the temperature of downstream flow during spring (Smits et al. 2020).
In contrast to spring, the fall period was characterized by LF streams showing strong warming trends and GWF stream temperatures having weak positive trends.Solar radiation and air temperature both had a general increasing trend during fall, which likely contributed to lake temperatures staying elevated, which is corroborated by the trend of lake ice formation occurring later.This warmer surface lake water is then advected downstream and contributing to downstream temperature increases (Ploum et al. 2018;Leach et al. 2021).GWF streams will be less sensitive to increased solar radiation during fall since above-stream leaf cover typically persists until late October (Creed et al. 2015) and will provide effective shade for these small streams.Groundwater−fed Lake−fed Fig. 3. Mean seasonal stream temperature for GWF and LF streams (green and blue, respectively).Linear regression line is provided for visual reference.
Note that the number of observations per site varied by season with the lowest average number of observations during summer (7.4 observations per site) and highest during spring (24 observations per site).See Supporting Information Section S2 for more details on the stream temperature observations.Note sites are ordered by catchment area (lowest [left] to highest [right]).
Increased summer temperatures have been documented across numerous other studies (Isaak et al. 2012;Arora et al. 2016;Michel et al. 2020); however, most GWF and LF streams in our study had relatively stable long-term trends during summer.Summer air temperature and solar radiation have exhibited only weak increasing trends; therefore, the lack of trend in summer stream temperatures may not be surprising.For the LF streams, the relatively large volume of water in the lakes may have buffered this variability in air temperature and solar radiation.In addition, the GWF streams are well shaded during this period and would be relatively insensitive to any increases in incoming solar radiation.Summer stream discharge appears to have decreased at most of the sites (Supporting Information Fig. S3), which might be expected to drive increases in stream temperature (Leach et al. 2023).However, the relative changes in stream discharge may be too small to see a temperature response for the LF streams.For the GWF streams, a decrease in mean stream discharge will correspond with a greater occurrence of zero flow conditions for most of these streams.We did not find a clear relationship between frequency of summer zero flow conditions and stream temperature (Supporting Information Fig. S8); however, the intermittent nature of the GWF streams during July and August could further limit the degree to which temperatures are affected by increasing solar radiation, as there would be no surface flow to warm during the hottest periods of the summer (Janisch et al. 2012;Leach et al. 2022).
LF streams show little change in winter temperatures.This is consistent with the influence of seasonal ice cover maintaining surface water temperatures near 0 C. Some of the GWF streams (GWF-c32, GWF-c46, and GWF-c34) showed detectable decreasing trends in winter temperature.This decreasing trend could be associated with increased midwinter runoff events delivering cold melt water to the streams (Leach and Moore 2014) or declining insulation of the subsurface by a shrinking snowpack (Zhang 2005).These potential drivers are consistent with an increase in winter stream discharge and decrease in SWE, respectively, as has been observed at TLW (Fig. 2); however, these trends in winter discharge and SWE are not large, and controls on these winter stream thermal dynamics may warrant further study, as winter can be an important period for fish development (Shuter et al. 2012).
Various studies have highlighted a general trend of increasing lake temperatures associated with climate change (Roberts et al. 2017;Winslow et al. 2017;Woolway et al. 2020).Warming temperature trends of streams downstream of lakes have been found to exceed warming trends for streams that lack extensive upstream lakes (Robinson and Matthaei 2007;Webb and Nobilis 2007;Michel et al. 2020).We found that LF streams show an overall higher warming trend than GWF streams, which is consistent with these other studies, and may reflect small lakes being more responsive to atmospheric conditions during the open water season compared with shallow groundwater and well-shaded small streams.
In general, the LF streams have larger catchment areas than the GWF streams (Supporting Information Table S2).Due to the observational nature of this study and the correlation between catchment area and stream class (GWF vs. LF), it is difficult to isolate the influence of water source on stream temperature trend over the influence of catchment area; however, observed patterns are consistent with expected differences associated with stream water being primarily sourced from either lakes or groundwater.For example, contrasting trends between GWF and LF streams during both spring and fall appear to reflect key differences in water volume, residence time, surface area, and ice cover, as discussed above.Two of the GWF streams (GWF-c34 [69 ha] and GWF-c46 [43 ha]) have comparable catchment areas to two of the LF streams (LF-012 [89 ha] and LF-013 [54 ha]).The seasonal differences in stream temperature trends for this subset of GWF and LF sites are similar to the differences seen across all sites, regardless of catchment area.For these reasons, it seems plausible that the seasonal differences between GWF and LF sites are primarily due to dominant water source and not entirely due to catchment area, although additional research to disentangle these influences is warranted.Since catchment area and lake cover are often correlated (Janssen et al. 2014), one possible avenue for further research is the use of virtual experiments using models (Leach et al. 2021) to isolate the influence of catchment area and dominant water source on stream temperature trends.
A further limitation of this study is that, other than station LF-013, all the LF sites are located along the same stream; therefore, these stations are flow-connected and their thermal regimes may not be independent of each other.This is clearly the case for stations LF-009 and LF-005 which are not located directly at lake outlets, but rather downstream of stations LF-012 and LF-006, respectively.Therefore, it is not surprising that these station pairs (LF-009 and LF-012, and LF-005 and LF-006) show similar temperature trends.In contrast, LF-006, LF-007, LF-008, and LF-012 are located immediately downstream of their respective lakes.For this reason, it may be reasonable to treat each lake outlet independently since estimated water renewal times for these lakes range between 2 and 16 months (Jeffries et al. 1988) and lake energy budgets are primarily dominated by energy exchanges at the lake surface, especially during the open water period outside of snowmelt (Spence et al. 2003;Nordbo et al. 2011;Smits et al. 2020).Therefore, each lake could be considered a thermal reset within the stream network; however, this assumption requires further investigation (Jones 2010).
The overall increasing stream temperature trends documented in this study for headwater streams are generally similar in magnitude to those reported for larger river systems (see summary tables in Arora et al. 2016;Seyedhashemi et al. 2022).Although not the case everywhere (Arismendi et al. 2012;Orr et al. 2015;Kelleher et al. 2021), most of those studies reported the largest warming trends in summer.Summer stream temperature is important for aquatic organisms, especially as this is the time of the year when maximum temperatures are reached and mortality thresholds can be exceeded (Mayer 2012;Hinch et al. 2021).However, others have highlighted the critical importance of stream temperature on organism life cycles during periods outside of summer (Holtby 1988;Baldock et al. 2016;Armstrong et al. 2021).Our study suggests that for headwater streams, the largest stream temperature responses to climate variability may be occurring during spring or fall rather than summer.In addition, when and how much these streams respond to climate change depends on their water source.A stream network perspective that includes headwater streams and small lakes, and accounts for potential changes outside of summer, is therefore important for understanding how aquatic ecosystems respond to climate change.
Fig. 1.Map of the TLW with the seven LF (blue) and six GWF (green) stream hydrometric stations and their delineated catchment boundaries (black lines).Light blue lines show the stream network.The inset map shows the location of the TLW in Canada.Elevation is given in meters above sea level (masl).

Fig. 2 .
Fig.2.Seasonal trends in air temperature, incoming solar radiation, and streamflow at sites LF-005 (example LF stream) and GWF-c34 (example GWF stream).Annual trends in day of year (d.o.y.) of lake ice-on and ice-off, as well as 01 April SWE.Lake ice-on and ice-off day values are from Little Turkey Lake, which is generally representative of the trends in the other lakes (Supporting Information Fig.S5).Corresponding Sen's slope estimates are provided in Supporting Information Fig.S7.
Fig. 4. Seasonal Sen's slope estimates for stream and air temperatures.The 50% CI (thick line) and 95% CI (thin line) estimates are also provided.