Influence of landscape and hydrological factors on stream–air temperature relationships at regional scale

Identifying the main controlling factors of the stream temperature (Tw) variability is important to target streams sensitive to climate and other drivers of change. The thermal sensitivity (TS), based on relationship between air temperature (Ta) and Tw, of a given stream can be used for quantifying the streams sensitivity to future climate change. This study aims to compare TS for a wide range of temperate streams located within a large French catchment (110,000 km2) using 4 years of hourly data (2008–2012) and to cluster stations sharing similar thermal variabilities and thereby identify environmental key drivers that modify TS at the regional scale. Two successive classifications were carried out: (a) first based on Ta–Tw relationship metrics including TS and (b) second to establish a link between a selection of environmental variables and clusters of stations. Based on weekly Ta–Tw relationships, the first classification identified four thermal regimes with differing annual Tw in terms of magnitude and amplitudes in comparison with Ta. The second classification, based on classification and regression tree method, succeeded to link each thermal regime to different environmental controlling factors. Streams influenced by both groundwater inflows and shading are the most moderated with the lowest TS and an annual amplitude of Tw around half of the annual amplitude of Ta. Inversely, stations located on large streams with a high distance from source and not (or slightly) influenced by groundwater inflows nor shading showed the highest TS, and so, they are very climate sensitive. These findings have implications for guiding river basin managers and other stakeholders in implementing thermal moderation measures in the context of a warming climate and global change.

within a large French catchment (110,000 km 2 ) using 4 years of hourly data (2008)(2009)(2010)(2011)(2012) and to cluster stations sharing similar thermal variabilities and thereby identify environmental key drivers that modify TS at the regional scale. Two successive classifications were carried out: (a) first based on Ta-Tw relationship metrics including TS and (b) second to establish a link between a selection of environmental variables and clusters of stations. Based on weekly Ta-Tw relationships, the first classification identified four thermal regimes with differing annual Tw in terms of magnitude and amplitudes in comparison with Ta. The second classification, based on classification and regression tree method, succeeded to link each thermal regime to different environmental controlling factors. Streams influenced by both groundwater inflows and shading are the most moderated with the lowest TS and an annual amplitude of Tw around half of the annual amplitude of Ta. Inversely, stations located on large streams with a high distance from source and not (or slightly) influenced by groundwater inflows nor shading showed the highest TS, and so, they are very climate sensitive. These findings have implications for guiding river basin managers and other stakeholders in implementing thermal moderation measures in the context of a warming climate and global change.

K E Y W O R D S
CART method, classification, regional scale, stream temperature, thermal sensitivity
The Tw variability, described by metrics of flow magnitude, frequency, duration, timing, and rate of change, on various timescales (Jones & Schmidt, 2018), is influenced by complex processes related to atmospheric, hydrogeological, geomorphic, and landscape characteristics and anthropogenic pressures, which could interact at multiple spatial scales (Caissie, 2006;Hannah & Garner, 2015). Numerous studies have highlighted the importance of riparian forest and groundwater inflows in moderating Tw variability (Dugdale, Malcolm, Kantola, & Hannah, 2018;Garner, Malcolm, Sadler, & Hannah, 2017;Kelleher et al., 2012;Lalot et al., 2015;Loicq, Moatar, Jullian, Dugdale, & Hannah, 2018). Identifying the main controlling factors of Tw variability remains an important task to target streams sensitive to climate change and to develop mitigation action to preserve aquatic ecosystems (Jackson et al., 2018).
Analyses in most of these studies are carried out on a site-by-site basis, which limits the extent to which broad patterns can be inferred (Laizé, Bruna Meredith, Dunbar, & Hannah, 2017). Some regional-scale studies have used spatial thermal regime classification based on a large set of catchment properties (Chu et al., 2009;Laizé et al., 2017;Maheu, Poff, & St-Hilaire, 2016;Rivers-Moore, Dallas, & Morris, 2013;Tague, Farrell, Grant, Lewis, & Rey, 2007). These studies succeeded in identifying key drivers that influence the thermal regime of streams at the regional scale. Most of these studies use on metrics summarizing the warmest aspects of the Tw regime to examine the threats to cold-water species under climate change.
Several researchers analyse the relationship between Tw and Ta, with Ta taken as a surrogate of the main climatic drivers (Ducharne, 2008;Garner et al., 2014). Ta is a common variable, easily measured on the field, and it is strongly correlated to solar radiation (Bustillo et al., 2014). Kelleher et al. (2012) studied the thermal sensitivity (TS) of streams to represent the relative sensitivity of Tw of a given stream to environmental change. TS is defined as the slope of the regression line between Ta and Tw, which can be linear (or logistic) and can be fitted on data averaged at different timescales (Mohseni & Stefan, 1999;O'Driscoll & DeWalle, 2006;Stefan & Preud'homme, 1993). TS summarizes the cumulative buffering effects of local landscape characteristics on stream temperatures. Although TS may evolve into the future due to the changing drivers considered above, TS computed for a specific period of record gives insight of which streams have the greatest sensitivity to climate based on contemporary conditions, which can be used as a baseline for responsiveness (Kelleher et al., 2012). However, this integrated variable cannot distinguish the cause and effect of groundwater and riparian vegetation shading on Tw variability (Chang & Psaris, 2013;Chu et al., 2009;O'Driscoll & DeWalle, 2006). Understanding the importance of these driving factors is essential to develop appropriate strategies to mitigate and adapt to stream heating under anticipated climate warming.
The aim of our study is (a) to provide a comparison of thermal sensitivity (TS) across a wide range of French temperate streams, based on 4 years of hourly data (2008)(2009)(2010)(2011)(2012), and (b) to identify groups of streams with similar sensitivity and so infer the environmental key factors that control TS at the regional scale. For that purpose, two successive classifications of 127 stations located in the Loire catchment (Beaufort et al., 2016) were carried out: (a) first based on Ta-Tw relationship metrics including TS and (b) second to establish a link between a selection of environmental variables and thermal regimes of stations. Finally, the relative importance of environmental variables on the TS of streams is investigated, and the implication for river management and river restoration is discussed.

| Basin description
The Loire basin ( Figure 1) comprises a hydrographical network of 88,000 km and drains a catchment area of 117,000 km 2 . It is characterized by varying climates between the upstream and the downstream (annual rainfall between 550 and 2,100 mm/year and annual air temperature between 6 C and 12.5 C), landform (10% of the basin area >800 m; mean altitude = 300 m), and lithology (metamorphic, magmatic, and sedimentary rocks). The percentage of riparian vegetation, defined on a buffer zone of 10 m on both sides of the streams, is globally greater in the southern basin where the altitude is the highest (mean ratio of the riparian vegetation = 75%; dark green; Figure 1c). Streams located in the central part of the basin, mainly composed of sedimentary rocks, benefit more from groundwater contributions (Figure 1d). The main aquifers are found in the sedimentary rocks in the centre of the basin. The Beauce formations (12,700 km 2 ) are composed of many semipermeable aquifers (Mohseni & Stefan, 1999) with numerous groundwater inflows located at the north of the Loire basin. Some streams are very directly connected to this aquifer, and their flow depends on the level of the Beauce water table (Baratelli, Flipo, & Moatar, 2016). The hourly Ta was taken from the SAFRAN (Système d'analyse fournissant des renseignements atmosphériques à la neige) reanalysis data (grid 8 km) at hourly time step between 2008 to 2012 (Quintana-Seguí et al., 2008;Vidal, Martin, Franchistéguy, Baillon, & Soubeyroux, 2010). Ta is extracted from the SAFRAN mesh (64 km 2 ) overlapping the station. The mean annual Ta of these stations ranges from 6.4 C to 12.5 C. The coldest temperatures are observed in the mountainous part of the basin (mean annual Ta < 10 C), whereas the warmest temperatures are observed in the west and in the sedimentary plain (mean annual Ta > 11 C).

| Field monitoring
Both hourly Ta and Tw have been averaged over the day and over the week in the next section.

| Metrics of air-water temperature relationship
We used four metrics to characterize the relation between air and water temperature. (a) Two of these metrics, the thermal sensitivity (TS) and intercept (b), provide information on the link between weekly Tw and Ta over the year. Weekly linear regressions were selected on the basis of the best mean R 2 fitted for the 127 stations in comparison with daily or logistic regressions.
For each station, a linear regression is fitted between the weekly Tw (Tw7D) and the weekly Ta (Ta7D) and the distribution of slopes, hereafter called thermal sensitivity (TS), and intercept (b) were analysed (Equation 1; Kelleher et al., 2012;O'Driscoll & DeWalle, 2006).
(b) Two others metrics, ΔT Jan and ΔT Aug , are based on the seasonal difference between monthly Tw and Ta. For all stations, the monthly Tw (MTw) is the coldest in January and the warmest in August. To account for the relative sensitivity of Tw during extreme months, we introduced two metrics, which are the differences between the monthly Ta (MTa) and Tw in January (ΔT Jan ) and in August (ΔT Aug ) averaged between 2012 and 2016: where ΔT is the mean difference between monthly Ta (MTa) and Tw (MTw) calculated in January or August, MTa Jan (i) and MTa Aug (i) are respectively the monthly Ta in January and August of the year i, MTw Jan (i) and MTw Aug (i) are respectively the monthly Tw in January and August of the year i, and Ny is the number of year where monthly Ta and Tw are both available (1 ≤ Ny ≤ 4, see Section 2.1). values between 0 and 1. Details on calculation can be found in Gustard, Bullock, and Dixon (1992). Low values are related to catchments with no storage capacity and also to catchments exposed to very high climate variability resulting in severe low-flow and quick run-off in response to rainfall events. High values are observed where artificial reservoirs, large aquifers, and storage in snow packs moderate the variability of daily flow. In our study, BFI is considered as a proxy of groundwater influence. The discharge Q was not monitored at Tw station, and each Tw station was coupled to the nearest gauging station (distance between both stations ranges from 10 m to 15 km).

| Explanatory variables
The matching is based on two criteria: (a) The gauging stations has to be located in the same or nearby streams and (b) the difference of catchment area between the location where Tw was measured and the location where Q was measured was kept to a maximum of ±20%.
The daily discharge was extracted from the French river flow monitoring network (HYDRO database, http://www.hydro. eaufrance.fr/).
(b) The average specific discharge in August (Q Aug ) is calculated at each station between 2008 and 2012. The goal is to measure the capacity of the catchment to produce a flow in summer, when precipitation is low. The specific discharge is the ratio between the discharge and the corresponding catchment area (in L s −1 km −2 ) and is used to standardize discharge for basin area.
Two climatic variables were determined from the Safran reanalysis data: (a) the mean summer cumulated precipitation (P in mm) and (b) the mean summer potential evapotranspiration (PET in mm) both calculated between June 1 and September 30 of each year between 2008 and 2012 for the entire upstream area of each monitoring station. Streams with wetter basin (high P and low PET) are expected to have higher water yields and more groundwater contributions that should cool streams (Isaak et al., 2017).
One variable was determined to characterize the riparian vegetation. A shading factor (SF), corresponding to a coefficient of reduction of the overall incident radiation, was estimated by Valette et al. ing is at its annual seasonal maximum for the North Hemisphere. The model of Li, Jackson, and Kraseski (2012) was implemented in its simplest version, that is, considering rectangular trees, located at the edge of the bank, without overhang.
where H is the tree height (assumed to be 20 m everywhere), W is the stream width, estimated using the ESTIMKART empirical model (Lamouroux et al., 2010), Ψ is the solar elevation angle, δ is the angle between solar azimuth and the mean azimuth (0-180 ) of the river reach, and vc is the vegetation cover (%).

| Thermal regimes clustering
To identify natural thermal regimes of stations sharing similar Ta-Tw relationship, an agglomerative hierarchical clustering (AHC) has been used. The AHC is based on the four metrics described above (TS, b, ΔT Jan , and ΔT Aug ). The Euclidean distance is used to measure the dissimilarity, and clusters are found with the Ward's minimum variance method. The stability of clusters is assessed through a bootstrap approach with the R package "fpc" (Hennig, 2019), and the similarity between each new cluster set and initial cluster was assessed with the Jaccard index (Hennig, 2007;Maheu et al., 2016). The Jaccard coefficient ranges from 0 to 1, and a cluster with a coefficient larger than 0.75 can be considered as stable (Maheu et al., 2016). Each thermal regime identified is described in terms of magnitude (mean Tw over a month) and amplitude (differences between the maximum and minimum values of MTw) and compared with MTa.

| Identification of environmental drivers in thermal sensitivity
A CART is used to examine the relationship between TS and the set of explanatory variables described above. CART analysis (Breiman, Friedman, Stone, & Olshen, 1984) is nonparametric and non-linear and does not introduce an a priori structure of the link between explanatory variables and the variable to be explained contrary to generalized linear models implicit assumption (Breiman et al., 1984;Ripley, 1996).
CART recursively partitions observations in a matched data set, consisting of TS (response) and the eight explanatory variables, into progressively smaller groups (De'ath & Fabricius, 2000 (Breiman, 2001), which is constituted by selecting randomly 80% of the observations (80% of 127 stations × eight explanatory variables × TS), and the test set consists of the remaining 20%. We used the implementation in the R package "randomForest" (Liaw & Wiener, 2002). The explanatory variable importance is given directly by the "randomForest" algorithm, which determines how much the mean square errors in prediction increases when that covariate is randomly permuted within the tree.
The random selection is performed 100 times, and the explanatory variables importance for each test set was then averaged.

| Distribution of thermal sensitivity and link with catchment size
The R 2 values for weekly Ta

| Cluster classification analysis
The AHC yielded four clusters of station corresponding to four thermal regimes: • WarmHighVar-warm and high variability (47 sites-37%): stations characterized by low b (<3 C) and high TS (>0.8). At these stations, MTw is higher than MTa in January and August with a median difference of 1.5 C (Table 1) • ColdHighVar-cold and high variability (44 sites-35%): Stations have MTw higher than MTa in January by 2 C (Table 1) The thermal regimes named "WarmHighVar" and "ColdLowVar" were stable clusters and had a Jaccard coefficient larger than 0.7.

| Drivers of thermal sensitivity (TS)
The CART model output leads to develop dichotomic tree plots to better visualize the effects of main drivers (Figure 7). The three most important explanatory variables used by the model to cluster stations as a function of their TS are SF, D (distance from source), and BFI ( Figure 6). This is consistent with the RF model output where D, SF, and BFI are identified as the main environmental variables to explain the TS of streams (variable importance >15%; Figure 6). The variables Q Aug and S are also used to differentiate clusters in the CART model and obtained a moderate importance close to 8% with RF. Elevation (E) is identified as the fourth relevant variable with RF model (variable importance = 11%; Figure 6) but is not used by the CART model for  Figure 7). The 11 stations having these characteristics belong to the thermal regime ColdLowVar (Table 2).
• C2 and C3-low and moderate TS with high SF: Streams with an SF higher than 30% and a BFI less than 0.8 belong mostly to the thermal regime ColdHighVar from AHC results (Figure 7). Q Aug has also an important influence, and we can see contrasts in terms of TS within this class. The TS was lower for the 17 stations located on streams with a Q Aug value higher than 5 L s −1 km −2 (mean TS = 0.67; C2) than for the 29 remaining stations in C3 with a Q Aug value less than 5 L s −1 km −2 (mean TS = 0.76).
• C4-moderate TS with low SF, low D, and high BFI: The six stations with SF less than 30%, a D less than 120 km, and a BFI greater than 0.8 have a moderate TS (mean TS of 0.71, C4) and belong to the two thermal regimes WarmLowVar and ColdLowVar (Table 2).
• C5 and C6-moderate and high TS with low SF, low D, and low BFI: Stations located on small and medium streams (S < 120 km) with a BFI lower than 0.8 obtained moderate and high TS. The TS of the 13 stations located on streams with a higher slope (S > 2.5 m km −1 ) F I G U R E 5 Representation of (a) the deviation from the mean annual Tw, (b) the deviation from the mean annual Ta (Table 2).
• C7-high TS with low SF, low D, and high BFI: Stations located on streams with low SF (SF < 30%) and a high D (D > 120 km) have the highest TS (mean TS of 1; C7; Figure 7). The 23 stations having these characteristics belong to the thermal regime WarmHighVar (Table 2).

| Regression robustness and comparison with other studies
In our case study, best correlations between Ta and Tw were obtained  daily (mean R 2 = 0.86; SD = 0.05) and weekly (mean R 2 = 0.93; SD = 0.03) logistic regressions. The weekly time step is more accurate because this time step filters out the lag time between Ta and Tw peaks, which can be of several days. In contrast to other studies (e.g., Kelleher et al., 2012), taking into account a non-linear relationship between Ta and Tw did not improve the performance of the regressions. This is probably explained by the fact that the Loire basin is not subject to Ta (2006). The negative correlation between TS and b is also consistent with previous studies. Streams controlled by groundwater inflows are characterized by intercepts closer to the regional groundwater temperature and low slopes. Inversely, streams more sensitive to climate conditions have steeper slopes and lower intercepts closest to Ta.
Our TS and b range were consistent with other studies results for linear regression models using a weekly time scale (Table 3). These TS and b values were close to those found by Webb (1992), Stefan and Preud'homme (1993), and Morrill et al. (2005) except that we observe no negative b and the range of our TS and b is slightly higher ( Figure 8). This can be explained by a higher number of streams used in our study and by the larger size of the watershed compared with other studies (Table 3)

| Groundwater influence on TS
In theory, groundwater influence is more visible on smaller streams because the volume of water is small and the travel time of the water from the source is short and not sufficient to equilibrate Tw with the atmosphere (Beaufort et al., 2016;Mohseni & Stefan, 1999). Groundwater inflow is a heat source during winter and a heat sink during summer resulting in little seasonal variation in Tw and a low TS (Hannah, Malcolm, Soulsby, & Youngson, 2004;Kelleher et al., 2012).
Stations with a thermal regime ColdLowVar have low TS (median TS = 0.5) and a high intercept (median b = 5.7 C). In other studies, low TS could be due to the upstream influence of reservoirs or impoundments (Erickson & Stefan, 2000;Morrill et al., 2005) or to high groundwater contribution (Kelleher et al., 2012). groundwater influences (C1 and C4; Figure 7). The decrease of TS is accentuated when a high BFI is combined with an SF higher than 30% as on the 11 stations in C1 (mean TS = 0.5). The shading of riparian vegetation leads to increase the thermal moderation of surface water in summer by shading from solar radiation. The BFI appears as a very influential variable in TS ( Figure 6). However, in the Loire basin, TS remains greater than 0.4 even when the BFI is higher than 0.8 and when b is higher than 6 C. In other studies, TS values are close to 0 when the BFI is close to 1 (Kelleher et al., 2012). It could be suspected that the temperature of groundwater inflows feeding streams follows a seasonal trend correlated with Ta and more marked than those observed in the literature (Kelleher et al., 2012;Krider et al., 2013;O'Driscoll & DeWalle, 2006). This could also explain the high residuals of slope intercept regression for stations having a TS lower than 0.6 ( Figure 3a).
The variable Q Aug is the mean specific discharge during the warmest month and represents the sustainability of low flows. It is a moderately influential variable in TS ( Figure 6). It can be assumed that a stream with a high Q Aug , in the case of natural flowing, benefits from groundwater inflows and/or of important contribution of its tributaries allowing it to maintain a sufficient depth to moderate Tw in summer. CART analysis results showed that streams with a Q Aug value higher than 5 L s −1 km −2 , associated to an SF less than 30% and a BFI less than 0.8, have a lower TS than others stations (C2 vs. C3; Figure 7), which seems to confirm our assumption. However, the importance of Q Aug remains applies to a subset of stations and the BFI remains the main variable representing the influence of groundwater inflows in our dataset.

| Riparian shading influence on TS
Shortwave ( Sinokrot . The riparian vegetation captures solar radiation and leads to reduced Tw resulting in a decrease of TS. This effect is particularly visible in summer when the solar radiation is the strongest and represents the main source of energy inputs (e.g., Hannah, Malcolm, Soulsby, & Youngson, 2008). In addition, the riparian vegetation of the Loire basin is mainly composed of deciduous trees, which considerably limit the effect of shading in winter.
In our study, we tried to differentiate the effects of shading and of  (Table 1). Between thermal regimes Cold-HighVar and WarmHighVar, their ΔT Jan is similar (ΔT Jan = −1.6 C), but Tw is clearly lower than Ta during August for stations in thermal regime ColdHighVar (median ΔT Aug = 1.9). The influence of the riparian vegetation shading is suspected. CART model results seems to confirm this assumption because all stations in ColdHighVar were identified with an SF higher than 30% (C2 and C4; Figure 7). The effects of shading could be accentuated when the specific discharge in August is higher than 5 L s −1 km −2 (C2; TS ≈ 0.67) because the thermal inertia of the streams is increased.

| Landscape factors influence
The distance from the source (D) is a key driver of TS ( Figure 6).
CART model results showed that stations with a D higher than 120 km obtained the highest TS (TS = 1; C7; Figure 7). D is highly positively correlated with the drainage area, and several studies identified this driver as playing an important role in the TS of rivers (Chang & Psaris, 2013;Garner et al., 2014;Hrachowitz et al., 2010;Imholt et al., 2013). Some others studies have also identified the Strahler order, which is correlated to D (R 2 = 0.6), as a strong influence factor of TS (Chang & Psaris, 2013;Ducharne, 2008;Kelleher et al., 2012;Wehrly, Wiley, & Seelbach, 1998). Streams with a high D and a large drainage area are weakly dependent on upstream conditions, and the travel time of the water body between upstream and downstream allows Tw to equilibrate with Ta (Mohseni & Stefan, 1999), leading to increase TS. Also, a longer D and a larger F I G U R E 8 Representation of the range of TS and b found in reviewed publications for linear regression models of weekly Ta-Tw relationship catchment area corresponds to lower topographical slopes, slower flow velocities, and greater regional residence time, which allow more time for Tw to adjust to local Ta (Mayer, 2012).
Stations located on small and medium streams, not influenced by shading and groundwater inflows (SF < 30%; BFI < 0.8; and D < 120 km) belonging to cluster C5 and C6 (Figure 7), obtained a TS less than those of large rivers in C7. There is an influence of S because stations located on streams with a high slope (S < 2.5 m km −1 ) had a mean TS of 0.8 (C5; Figure 7), whereas others had a mean TS of 0.88 (Cluster C6; Figure 7). The stream slope is mostly linked to elevation (R 2 = 0.65). A higher slope increases the flow velocity, and the elevation influences Tw over the adiabatic lapse rates of Ta (Hrachowitz et al., 2010) and also through snow and glacier meltwater inflow (Arora, Toffolon, Tockner, & Venohr, 2018;Morrill et al., 2005), which may contribute to decrease TS. P and PET are not relevant in CART model, which may be explained by the relative climatic homogeneity of the study area (Cfb = temperate oceanic climate, Table 3).

| Implication for river management and river restoration
The study of streams TS makes it possible to identify the most sensi-  (Wawrzyniak, Piégay, & Poirel, 2012) and be preserved by limiting advective thermal mixing (Kurylyk et al., 2015) or activated by geomorphological restoration of streams (Eschbach et al., 2017;Loheide & Gorelick, 2006). On small and medium streams, it is necessary to preserve and/or favour the presence of riparian vegetation to moderate TS (Fabris, Malcolm, Buddendorf, & Soulsby, 2018). The effects will be most pronounced, in comparison with large streams, because of their smaller width, but investments have to be made strategically (Isaak et al., 2017;Johnson & Wilby, 2015). From a watershed management perspective, stream shading would be less effective in streams where Tw is already strongly moderated by groundwater inflows but more effective along losing reaches or stream reaches distant from groundwater inflows (O'Driscoll & DeWalle, 2006 Isaak et al., 2017). In order to limit these warmings and preserve ecosystems, it seems important to identify streams constituting cold-water thermal refuges (with low TS) and to restore and preserve thermal diversity in the hydrographical network (Torgersen, Ebersole, & Keenan, 2012). However, the main factors limiting TS (BFI and SF) could change in the future, and several streams could become much more sensitive to environmental change (Leach & Moore, 2019). For example, the loss in groundwater inflows would result in greater meteorological controls increasing the annual amplitude of Tw (O'Driscoll & DeWalle, 2006). Limiting water abstraction during lowflow periods may avoid a disconnection of groundwater/surface water exchanges and ensure environmental flows during the summer (Elmore, Null, & Mouzon, 2016). Some cooling strategies proposed to reconnect streams to floodplains and to facilitate greater lateral and hyporheic flow exchanges (Beechie et al., 2012;Daniel Caissie & Luce, 2017;Kurylyk et al., 2015) but need to be tested at a regional scale.
To apply efficient and effective actions, river managers have to focus on small and medium streams and can use the environmental variables identified in our classification results as indicators to assess the climate sensitivity of unmonitored streams.

| CONCLUSION
In this study, we proposed a framework to compare thermal sensitivity (TS) for 127 stations located on temperate streams between 2008 and 2012 and to cluster stations sharing similar natural thermal regimes, not influence by anthropogenic effects. On the basis of weekly Ta-Tw relationships, four thermal regimes were identified with differing annual Tw in terms of magnitude and amplitudes in comparison with Ta. We linked each cluster to different environmental controlling factors as inferred by TS. This highlighted that shading from riparian vegetation, groundwater inflows, and the distance from the source of streams were the main drivers of the moderation of streams located in the Loire catchment. Streams influenced by both groundwater inflows and shading are the most moderated with the lowest TS and an annual amplitude of Tw around half the annual amplitude of Ta. Inversely, stations located on large streams or on streams slightly or not influenced by groundwater inflows and/or shading showed the highest TS and are very climate sensitive. Their Tw amplitude and magnitude were very close to those of Ta; consequently, these rivers are deemed the most sensitive to the effects of future climate change.
The Tw metrics and the environmental variables remain simple to determine and can easily be applied in others catchments at a regional scale. One of the perspectives to this work would be to explore if main controlling factors of the Tw variability identified here are the same in different climate and physiographical regions elsewhere. We observe that almost invariability streams studied in reviewed publications for linear regression models of weekly Ta-Tw relationship (