University of Birmingham River temperature regimes of England and Wales : spatial patterns, inter-annual variability and climatic sensitivity

: Identi ﬁ cation of the most sensitive hydrological regions to a changing climate is essential to target adaptive management strategies. This study presents a quantitative assessment of spatial patterns, inter-annual variability and climatic sensitivity of the shape (form) and magnitude (size) of annual river/stream water temperature regimes across England and Wales. Classi ﬁ cation of long-term average (1989 – 2006) annual river (air) temperature regime dynamics at 88 (38) stations within England and Wales identi ﬁ ed spatially differentiable regions. Emergent river temperature regions were used to structure detailed hydroclimatological analyses of a subset of 38 paired river and air temperature stations. The shape and magnitude of air and water temperature regimes were classi ﬁ ed for individual station-years; and a sensitivity index ( SI , based on conditional probability) was used to quantify the strength of associations between river and air temperature regimes. The nature and strength of air – river temperature regime links differed between regions. River basin properties considered to be static over the timescale of the study were used to infer modi ﬁ cation of air – river temperature links by basin hydrological processes. The strongest links were observed in regions where groundwater contributions to runoff (estimated by basin permeability) were smallest and water exposure time to the atmosphere (estimated by basin area) was greatest. These ﬁ ndings provide a new large-scale perspective on the hydroclimatological controls driving river thermal dynamics and, thus, yield a scienti ﬁ c basis for informed management and regulatory decisions concerning river temperature within England and Wales. © 2013 The Authors. Hydrological Processes published by John Wiley & Sons, Ltd.


INTRODUCTION
In recent years, there has been an upsurge in river and stream temperature research (Hannah et al., 2008a) as temperature is increasingly recognized as an important and highly sensitive variable affecting biological, chemical and physical processes in flowing waters (Caissie, 2006).Primary research challenges in the field of river temperature include improving understanding of thermal heterogeneity at different spatial and temporal scales, the nature of past variability and likely future trends (Webb et al., 2008).The analysis of spatial and temporal variability in river temperature regimes is vital to (1) elucidate key controls and processes, (2) assess sensitivity to a changing climate and (3) inform management of land use, water resources and freshwater ecosystems (Moore et al., 2005).In part, river temperature has attracted growing attention because water thermal regimes may be highly sensitive to climate change.Instrumented records show significant river temperature warming (e.g.Webb and Walling, 1992;Caissie et al., 2005;Webb and Nobilis, 2007;Kaushal et al., 2010) and cooling (e.g.Arismendi et al., 2012;Isaak et al., 2012) over recent decades, and future projections (Webb and Walling, 1992;Webb and Nobilis, 1994;Mohseni et al., 2003;van Vliet et al., 2011) suggest profound potential impacts on freshwater ecosystems (Ormerod, 2009).Hence, environment managers and regulators urgently need information on space-time variability in river temperatures and a better understanding of the controlling factors as an important first step towards making well-informed decisions concerning the climatic sensitivity of river thermal regimes (Wilby et al., 2010).
Drivers of river temperature dynamics are complex, with multivariate controls and process interactions spatially nested at macro (latitude, altitude and continentality), meso (basin climate and hydrology) and micro (micro-meteorology, channel geometry, riparian shading and substratum conditions) scales (Webb, 1996).Research on river temperature within England and Wales has investigated (1) point-scale heat fluxes that fundamentally control water temperature (e.g.Webb and Zhang, 2004;Hannah et al., 2008b), (2) micro-thermal variability within the water column (e.g.Clark et al., 1999) and riverbed (e.g.Hannah et al., 2009;Krause et al., 2011), (3) the effects of forest canopies and forest clear-felling (e.g.Stott and Marks, 2000;Hannah et al., 2008b) and (4) spatial and temporal dynamics across river temperature networks (e.g.Webb and Walling, 1992).This UK-based research has been restricted to the subbasin scale; thus, no research exists on spatial and temporal variability and controls on river temperature at a larger scale (i.e.inter-basin to region and beyond).
Regimes describe the behaviour of hydroclimatological variables over the annual cycle or hydrological year, and they are useful tools for characterizing spatial and temporal variations in timing and magnitude of seasonal patterns (Bower et al., 2004).The importance of the entire flow regime for maintaining and protecting the integrity of fluvial systems, rather than considering only the maximum, minimum or mean values, is well recognized with regard to the management of river discharge (e.g.Poff, 1996;Harris et al., 2000;Kennard et al., 2010).However, for river temperature, single metrics (particularly maxima, e.g.Picard et al., 2003) and isolated months or seasons are used most commonly.These approaches consider the magnitude of thermal condition, which may limit aquatic stream organisms but ignore other potentially relevant characteristics, particularly the timing and duration of warmer and cooler episodes over the annual cycle (Chu et al., 2010;Arismendi et al., 2013).Spatial variability in multiple aspects of annual regime magnitude has been explored over a single calendar year in the Great Lakes Basin, Canada (Chu et al., 2010), but, typically, consideration of river temperature variability across multiple years has been restricted to trend analyses (e.g.Webb and Walling, 1992;Webb andNobilis, 1994, 2007;Kaushal et al., 2010).Interannual variability/stability in the character of the entire river thermal regime has not been investigated to date.Therefore, there is a clear need to develop methods to assess the key attributes of annual river temperature regimes and their year-to-year dynamics.
To explore linkages between climate and river temperature, air temperature is commonly used as a proxy for net heat exchange at the air-water interface (Webb et al., 2003).Across space, air-water temperature relationships are weaker for (1) upper headwater streams (Brown et al., 2005;Hrachowitz et al., 2010;Kelleher et al., 2012), (2) locations with increased thermal capacity and longer water travel times (Webb and Nobilis, 2007) and (3) sites with major groundwater or anthropogenic inputs (Erickson et al., 2000;O'Driscoll and DeWalle, 2006;Tague et al., 2007;Webb and Nobilis, 2007;Webb et al., 2008;Kelleher et al., 2012).The strength of the relationship between air and water temperatures increases from sub-daily to monthly timescales; but it is weakest for annual samples (Webb et al., 2008) because river temperature is reported to display less year-to-year variability than air temperature (Pilgrim et al., 1998;Erickson et al., 2000;Webb et al., 2003).There is a need to advance methods for rigorous, systematic analysis of dynamic air-water temperature links and to explore the controls on space-time patterns in the climatic sensitivity of river temperature.
Although there is a growing body of river temperature research, there remains limited understanding of large-scale spatial and temporal variability in climate-water temperature associations, and the modifiers of these relationships.Such research is essential for identification of the most temperature-sensitive river waters and to understand the controls on thermal sensitivity.To address these research gaps, this paper aims (1) to provide the first quantitative assessment of spatial patterns and inter-annual variability of the shape (timing) and magnitude (size) of annual river temperature regimes across England and Wales and (2) to assess the climatic sensitivity of river temperature regimes and understand the controls on river thermal sensitivity, including the potential moderating role of static basin properties.In addition to providing a new large-scale, longterm perspective and understanding of English and Welsh river temperatures, this paper seeks to make methodological innovations by testing a classification tool for annual regimes (Hannah et al., 2000) on water temperature and a climatic sensitivity index (Bower et al., 2004) for air-river temperature associations.Notably, this study represents the first application of the classification scheme and sensitivity index to annual river temperature regimes.

STUDY AREA CLIMATE
The countries of England and Wales have a temperate maritime climate.The highest air temperature is observed in July-August with the lowest air temperature in January-February.Air temperature is coldest in northern England and north-west Wales and warmest in South-eastern England, which reflects relief.High seasonality of the annual air temperature regime is observed inland, whereas in coastal locations, air temperature is warmer, but seasonality is reduced (Barrow and Hulme, 1997;Bower et al., 2004).

River temperature
Time series of river water temperature from monitoring stations across England and Wales were extracted from the Environment Agency's Freshwater Temperature Archive.The Archive and its data holdings are described in detail by Orr et al. (2010).Sites were selected from the Archive to provide a robust analysis with optimal spatio-temporal coverage across England and Wales.A total of 88 sites that had water temperature observations in ≥90% of all months over a common 18-year period (1989-2006 inclusive; Figure 1) were identified; water temperature was sampled on average 11.1 times per month over this time span.Monthly mean water temperature (°C) was calculated for each site for each month over the entire record to characterize annual thermal regimes and their year-to-year variability.
The potential influence of sampling frequency on the estimation of monthly means was evaluated systematically using data collected at the site monitored at the highest temporal resolution (i.e.River Dee at Pont Mwnwgl Y Llyn, 15-min intervals over 13 years).The mean was calculated for each possible combination of three samples in each month for all years (> 4.5 M resamples for each month in each year).Results indicated sampling frequency to have minimal impact on shape and magnitude of annual river temperature regimes (Figure 2).

Basin properties
River basin properties were selected to assess the potential role of hydrological controls in moderating spatial river temperature pattern and air-water temperature sensitivity.Properties were derived from two sources: (1) a 25-m resolution digital elevation model of England and Wales (University of Manchester/University of London, 2001) that was used to calculate basin area (km 2 ) and mean basin elevation (metres above sea level) and (2) the British Geological Survey Bedrock Permeability Index of England and Wales that was used to characterize average basin permeability as a measure of basin water storage and hydrological response time (Laize and Hannah, 2010).Basin properties for the 38 rivers selected for analysis of air-river temperature regime associations are summarized in Table I.

METHODS
The analytical procedure was divided into five linked stages: (1) regionalization of long-term average regimes for river and air temperature, (2) classification of annual regimes for each station-year, (3) quantification of interannual regime stability and (4) application of the sensitivity index (SI) to quantify linkage between air and river temperature classes.

Regime classification
As it was important to assess the timing (seasonality) and size of the annual river temperature regime, a hierarchical, agglomerative cluster analysis-based classification approach was used to group intra-annual patterns for river and air temperatures according to two key regime attributes: shape and magnitude.The regime classification procedure was developed by Hannah et al. (2000) and subsequently extended and evaluated for application to annual river flow and climate regimes (e.g.Harris et al., 2000;Bower et al., 2004).The shape classification identified stations (for regionalization) or station-years (to assess inter-annual regime variability) with similar regime forms, regardless of their magnitude.For the regionalization process, shape regimes were determined from long-term mean monthly values (i.e. the mean of observations in each month over the entire study period) standardized separately for each station using z-scores (mean = 0, standard deviation = 1).To classify inter-annual shape regimes, monthly mean values (i.e.means of observations in months for individual years) were standardized for each station prior to classification.
The magnitude classification was based on four indices (i.e. the mean, minimum, maximum and standard deviation), regardless of their timing.For regionalization, the indices were derived from longterm mean monthly values at each station; and stations with similar magnitude regimes were grouped.Each index was z-scored to control for differences in relative magnitudes.Index values for inter-annual magnitude regimes were determined from monthly mean values in each station-year; and station-years with similar magnitude regimes were grouped.Each index was z-scored over the entire study period for each station to control for between-station differences in the indices.Classification of regime shape and magnitude was performed separately for air and river temperatures over the common data period (1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006).This is the first application of these methods to classify annual river water temperature regimes.
It is important to note that the methods applied herein yielded two separate sets of regime classifications: (1) the regionalization procedure grouped stations to examine spatial patterns and (2) the interannual classification grouped annual regimes for each station-year to identify patterns of year-to-year variability.Together, the two classification modes characterized spatial and temporal regime dynamics.The regionalization of long-term regimes provided a basis for structuring analyses of between-region and withinregion patterns in inter-annual regime variability.The long-term average regime for a station was estimated from mean monthly values across all years for all 88 river temperature sites.Annual regimes for each station-year were characterized using monthly mean values for each station-year for a subset of 38 river temperature sites (i.e.closest locations paired with the 38 climate stations; Figure 1).Thus, regime shape and magnitude classes were identified for 702 station-years for both river and air temperatures.It is also important to note that (1) regime classes are not interchangeable between long-term and station-year regime classifications and (2) magnitude classes for regionalization identify absolute differences between stations whereas magnitude classes for station-years identify relative inter-annual variations at a station (Bower et al., 2004).For all classifications performed, (1) inspection of the cluster dendrogram and agglomeration schedule identified the number of classes, and (2) Ward's algorithm yielded the most robust (Kalkstein et al., 1987) and evenly sized classes.

Quantification of inter-annual regime stability and climatic sensitivity
The stability of inter-annual regimes at each site was assessed using the concept of equitability (E), which was a measure (values range from 0 to 1) of the probability of observing each regime shape or magnitude class against all the possible regime classes (Bower et al., 2004) (Equation 1).Higher values indicated greater equitability (evenness).
The SI, which quantifies the strength and direction of associations between air and river temperatures, was adapted by Bower et al. (2004) from an ecological index (Kent and Coker, 1992).From the concept of equitability, the SI considers the conditional probability, P(Y j |X i ), of observing a particular river temperature Y j regime under each air temperature regime X i and also the conditional probability, P(X i |Y j ), of a given air temperature regime prevailing for each river temperature regime.Equations ( 2) and (3) were used to calculate the equitability of regimes as the probability, P(Y j ) and P(X i ), of observing a particular river temperature regime and an air temperature regime, Y j and X i , respectively.
The ratio of E(Y) : E(X) identified one of two scenarios to produce a SI ranging from À1 to +1.Where E(Y) ≥ E (X), Equation (4) (positive scenario) returned a SI value between 0 and +1, which indicated that water temperature was more variable than air temperature.Where E(Y) ≤ E (X), Equation (5) (negative scenario) yielded a SI value between 0 and À1, which indicated that air temperature was more variable than water temperature.Values close to 0 indicated greater river temperature sensitivity to air temperature than those closer to ±1 (Bower et al., 2004). (5)

Influence of basin properties on air-river temperature sensitivity
To assess potential modification of air-river temperature associations by basin properties: (1) the 38 paired water and air temperature stations were grouped by the long-term average river temperature regions, and (2) equitability, SI and basin properties were compared within and between regions.

RESULTS
Results are presented in three sections: (1) regionalization of long-term shape and magnitude regimes for river and air temperatures, to explore spatial patterns and identify regions that structure further analyses, (2) inter-annual river and air temperature shape and magnitude regimes, to assess between-region and within-region regime dynamics and air-river temperature sensitivity and (3) analysis of the influence of basin properties on air-water temperature regime sensitivity between and within regions.

Regionalization of long-term regimes
Regime shape and magnitude were classified using long-term  mean monthly river and air temperatures for 88 and 38 stations, respectively.The correspondence between river and air temperatures was explored for 38 paired stations.
Long-term regime shape.One river temperature regime and one air temperature regime were identified across all of England and Wales.Both regimes had similar forms of annual cycle.River temperature regimes peaked in July with minima in January; air temperature regimes exhibited an extended July-August maxima and January-February minima (Figure 3).Air temperature regimes demonstrated less variability between sites than river temperature (cf.composite averaged regime; Figure 3), indicating that air temperature was more spatially conservative than river temperature.
Long-term regime magnitude.Four and three clusters respectively provided a robust classification of the magnitude of air and river temperature regimes.The four river temperature regime classes (R-Tw x ) were characterized by variation in the indices as follows (Figure 4): Long-term regimes: spatial patterns and air-river temperature associations.Regionalization produced only one air class and one river temperature class, indicating no spatial variability in long-term river or air temperature shape regime across England and Wales.Therefore, the focus is on spatial patterns and air-river temperature associations for long-term magnitude regimes.
River and air temperature magnitude regimes displayed clear spatial differentiation across England and Wales (Figure 5); and dynamics of paired air-river temperature stations corresponded with the exceptions of central, southern England and two sites in the north-west.The coldest regimes were observed in the north (Ta 1 and R-Tw 1 , n = 13), whereas those in the south displayed higher means and greater seasonality (Ta 3 and R-Tw 3 , n = 13).Regimes in the south-west were characterized by Three combinations of air and river temperature classes occurred for which air temperature contrasted with river temperature (Figure 5): (1) Ta 2 and R-Tw 2 (n = 8), which was a warm and highly seasonal air temperature with a moderate river temperature, (2) Ta 1 and R-Tw 3 (n = 1), which was a cold air temperature with a warm and highly seasonal river temperature and (3) Ta 1 and R-Tw 4 (n = 1), which was a cold air temperature with a warm and less seasonal river temperature.
Inter-annual regimes: variability and air-river temperature associations Regime shape and magnitude of river and air temperatures were classified using monthly means for each year (1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) across the 38 paired stations (i.e.702 station-years).These regime classes provided the basis for: (1) quantification of inter-annual stability and (2) test application of the SI for linking air and river temperature regimes.The regionalization results (presented earlier) structured analyses of between-region and within-region inter-annual regime variability.As stated in the methodology, the long-term and annual regime classes were not the same; it must be noted that magnitude classes for regionalization were absolute (between stations) whereas magnitude classes for interannual regimes were relative (between years at a station).
Inter-annual shape regimes.Three inter-annual river temperature and four inter-annual air temperature shape regime classes were identified.River temperature shape regime classes (IA-Tw x ) were identified as follows (Figure 6): Associations between inter-annual air and river temperature shape regimes.As the regionalization identified single long-term average river and air temperature shape regimes, stations were not subdivided for analysis of associations between inter-annual classes.Annual frequencies of river and air temperature regime shape classes are summarized in Figure 7.There was no evidence of a trend in the occurrence of either river or air temperature regime over the 18-year study period.Very limited spatial differentiation was observed in patterns of air temperature shape regime occurrence across all sites in each year of the study period.In any given year, one of the four air temperature shape regimes predominated at >90% of stations.Regime IA-Ta C predominated in seven years, IA-Ta D in six years, IA-Ta B in four years and IA-Ta A in one year only.Greater spatial differentiation was observed in patterns of river temperature shape regimes (cf.air temperature), although a predominant river temperature shape regime was identified across >50-92% of stations in Values of E were on average very high with 0.87 ± 0.2 and 0.89 ± 0.5 for river and air temperature regimes, respectively.This indicated that river and air temperature shape regime occurrences were highly variable between years and that all regimes occurred reasonably evenly at each station over the study period.Annual frequencies of river and air temperature shape classes within England and Wales are summarized in Figure 7.The SI quantified the strength and direction of the association between air and river temperature regimes.SI values were positive at all stations, indicating that river temperature regime shape was more variable than air temperature regime shape.The absolute magnitude of SI values averaged 0.35 ± 0.3, which indicated moderate sensitivity of river temperature to air temperature.

IA-Tw
Inter-annual magnitude regimes.Five inter-annual river temperature and four inter-annual air temperature magnitude regime classes were identified.Inter-annual IA-Ta 2 Moderatemoderate mean, moderate maximum, high minimum and moderate standard deviation (158 station-years, 23%).IA-Ta 3 Warmhighest mean, high maximum, high minimum and moderate standard deviations (182 station-years, 26%) IA-Ta 4 Warm with greatest seasonalityhigh mean, high maximum, low minimum and highest standard deviation (164 station-years, 23%) Associations between inter-annual air and river temperature magnitude regimes.Regionalization identified four long-term average river temperature regions (R-Tw 1 -R-Tw 4 ) for which stations were pooled for analysis of associations between inter-annual regimes of river and air temperature magnitudes.Annual frequencies of river and air temperature magnitude classes in each region are summarized in Figure 9.There was no apparent trend in either river or air temperature regime magnitude for any region.
For air temperature, all inter-annual magnitude regimes occurred in each region; equitability was high and varied little between regions (cf.river temperature regimes).Equitability was greatest within R-Tw 3 (0.88 ± 0.14), followed by R-Tw 4 (0.86 ± 0.13), R-Tw 1 (0.81 ± 0.27) and R-Tw 2 (0.74 ± 0.33).Differences in the frequency of RIVER TEMPERATURE REGIMES OF ENGLAND AND WALES regimes were observed between regions (Figure 9) but were not as pronounced as differences for river temperature.In R-Tw 1 , regime IA-Ta 1 occurred most frequently, followed by IA-Ta 2 , IA-Ta 4 and IA-Ta 3 (Figure 9).The frequency of inter-annual air temperature regime occurrence was similar in regions R-Tw 2 and R-Tw 3 (Figure 9) because stations within these regions formed the same long-term average air temperature region (Figure 6); IA-Ta 1 occurred most frequently followed by IA-Ta 3 , IA-Ta 2 and IA-Ta 4 .In R-Tw 4 , interannual regime IA-Ta 1 occurred most frequently, followed by IA-Ta 2 , IA-Ta 3 and IA-Ta 4 .
The strength and direction of inter-annual associations between river and air temperature regime magnitudes were quantified using the SI and the number of synchronous air-river temperature regime switches.All stations within regions R-Tw 1 , R-Tw 2 and R-Tw 3 were associated with negative SI values; therefore, river temperature was not as variable year to year as air temperature.Within region R-Tw 2 , 12 stations had negative SI values, but two stations had positive SI values; thus, at a minority of stations, river temperature was more variable than air temperature between years.
Region R-Tw 3 was associated with the greatest absolute SI values (Figure 10) and the least synchrony of air-river temperature regime switches from year to year, five on average.Region R-Tw 2 had a SI of 0.68 and displayed an average of nine synchronous air-river temperature switches.Regions R-Tw 1 and R-Tw 4 had the lowest SI values and the most synchronous air-river temperature regime switches; on average, stations within R-Tw 1 had a SI value of 0.56 and displayed ten synchronous switches.Stations within R-Tw 3 had a SI value of 0.52 and displayed 11 synchronous switches.
Influence of basin properties on air-water temperature regime sensitivity.Mean basin permeability (a measure of basin water storage and response time), basin area and mean basin elevation were compared across the four long-term average river temperature regions for magnitude (Figure 9).For basin area, stations within R-Tw 4 were characterized by the smallest basins and lowest sensitivity to air temperature.Region R-Tw 2 was associated with the most permeable geologies and was the second least sensitive to air temperature.R-Tw 4 was situated on the second most permeable geologies.Regions R-Tw 1 and R-Tw 3 were

DISCUSSION
This paper has quantified the space-time links between the shape and magnitude of air and river temperature regimes within England and Wales and identified the role of basin properties in modifying these associations.Static basin properties were not found to influence river temperature shape regimes; therefore, the discussion of the role of the river basin in modifying river temperature is confined to regime magnitude.

Shape regimes
No spatial differentiation of regime shape occurred within England and Wales as only one river regime and one air temperature regime were identified in the regionalization process.Broad temporal correspondence of air and river temperature dynamics was observed both intra-annually and inter-annually.For long-term regimes, river and air temperatures displayed maxima in July and minima in January, but air temperature maxima (minima) continued into August (February).The observed discrepancy between the timing of maximum and minimum regime features is attributable probably to the dominance of summer river flow by baseflow (i.e.groundwater) contributions (Marsh et al., 2007;Tague et al., 2007;Payn et al., 2012).The thermal dynamics of groundwater are slightly influenced by intra-annual variability in air temperature, whereas runoff is influenced by intraseasonally variable meteorological conditions (O'Driscoll and DeWalle, 2006;Tague et al., 2007;Herb and Stefan, 2011).Hence, maximum air temperature continued through August, whereas river temperature declined from its annual maximum in July potentially because of changing hydrological sourcing of river flow.
Inter-annually, river temperature regimes varied more than air temperature regimes, and the SI quantified moderate links between air and river temperatures.Moderate correspondence suggests that basin controls modified links between air and river temperatures, but because the regionalization process did not discern regional-scale variability in the long-term average regime, it is likely that these controls were basin specific and that the strength of their influence was variable between years (i.e. they were not static), explaining also why river temperature varied more than air temperature.Interannually variable discharge and hydrological sourcing of river flow (i.e. from runoff or groundwater) would generate varied thermal capacity and initial water temperature for atmospheric warming/cooling (Poole and Berman, 2001).Therefore, responsiveness of river temperature to air temperature would be varied between years.A lack of previous research on the variability of river temperature seasonality and links with air temperature hampers the comparison of these results for regime shape with studies conducted elsewhere.

Magnitude regimes and their modification by basin properties
For regime magnitude, spatially distinct regions of longterm average air and river temperature dynamics were observed.Bower et al. (2004) and Chu et al. (2010) also observed variability of air and river temperature magnitudes at regional scales.Intra-annual dynamics of air and river temperature regimes corresponded broadly within most regions, although some exceptions occurred.Air temperature regimes were warmer and varied less between seasons across a north to south-west gradient within England and Wales (as observed by Bower et al., 2004) in response to reducing altitude and continentality (Barrow and Hulme, 1997).River temperature regimes became warmer and varied less between seasons across a north to south-west gradient too, with the exception of moderate regimes within R-Tw 2 (located in the south-east) that were observed under warm and highly seasonal air temperature regime Tw 2 .Stations within R-Tw 2 were located on the most permeable geologies (i.e.predominantly on the Chalk in central southern England;Marsh et al., 2000) and received larger influxes of groundwater.Groundwater contributions in these regions would have contributed cooler water to river flow during summer and warmer water during winter (Story et al., 2003;Hannah et al., 2004;O'Driscoll and DeWalle, 2006;Tague et al., 2007;Chu et al., 2010;Kelleher et al., 2012) and so dampened the magnitude and inter-seasonal variability of the long-term average annual river temperature regime.
Inter-annually, clear regional spatial differentiation was observed in the occurrence of river temperature magnitude regimes and in the strength of links between air and river temperatures.River temperature regimes were least stable between years and displayed the strongest (yet weaker compared with shape) links with air temperature in regions R-Tw 1 and R-Tw 4 , where stations were situated on the least permeable and also largest basins.River temperature regimes were most stable and displayed the weakest links with air temperature in regions R-Tw 2 and R-Tw 3 , where stations were located on the most permeable geologies and smallest basins, respectively.These results were consistent with studies conducted on North American and continental European rivers.Runoff sourced from groundwater in Pennsylvanian and Oregon streams, USA, was less variable and less sensitive to variability in air temperature in comparison with those sourced from shallow sub-surface flows (O' Driscoll and DeWalle, 2006;Tague et al., 2007).Reduced sensitivity of headwater streams to air temperature was observed in the Aberdeenshire Dee, Scotland (Hrachowitz et al., 2010), and River Danube, Austria (Webb and Nobilis, 2007), and small Pennsylvanian streams were shown to be less sensitive to changes in air temperature than larger streams (Kelleher et al., 2012).The thermal dynamics of headwater streams were similar to those of groundwater because they were likely to be located closer to the river source and water had insufficient exposure time to equilibrate with the atmosphere (Edinger et al., 1968;Poole and Berman, 2001;Tague et al., 2007;Kelleher et al., 2012).Furthermore, stations on small headwater catchments may be forested (e.g.Hrachowitz et al., 2010), so that downstream warming may have been reduced or interrupted (Poole and Berman, 2001;Moore et al., 2005).
Although more sensitive to air temperature than smaller basins, larger basins did not exhibit strong links between thermal dynamics and air temperature as would be expected if they had reached atmospheric equilibrium.Only moderate air-river temperature links were observed, for which a number of causes may be hypothesized: (1) dynamic basin properties (e.g.discharge and changing hydrological sourcing of runoff; see Discussion sub-section on Shape regimes) varied the strength of air-river temperature between years; (2) larger basins within England and Wales (i.e.predominantly 'mesoscale basins', 10 2 -10 3 km 2 in size; Cappel et al., 2012) were smaller than so-called large river basins in continental Europe (e.g.Webb and Nobilis, 2007) and North America (e.g.Kelleher et al., 2012), and thus because of shorter travel times (Mohseni & Stefan, 1999), river temperature may not have enough time to fully equilibrate with the atmosphere; (3) thermal capacity was greater at stations in larger basins owing to higher discharge (Poole and Berman, 2001), and thus, response to air temperature variations was weakened (as demonstrated by Webb et al., 2003 in the Exe basin, UK).

CONCLUSIONS AND IMPLICATIONS
This study is innovative in presenting: (1) an assessment of large-scale spatial and temporal variability of the shape (timing) and magnitude (size) of annual river temperature regimes within England and Wales, (2) a quantification of their associations with air temperature regimes and (3) the identification of basin controls that modified the strength of air-river temperature links.
The application of a regime classification methodology (after Hannah et al., 2000) and sensitivity index (after Bower et al., 2004) proved to be useful tools for identifying spatial and temporal patterns in annual air and river temperature regimes and assessing the strength of air-river temperature links.Observed patterns of, and associations between, river and air temperature regimes within England and Wales were explained by physically meaningful basin controls, which modified the climatic signal in similar ways to those observed in North American (e.g.Tague et al., 2007;Kelleher et al., 2012) and continental European rivers (e.g.Webb and Nobilis, 2007).Thus, the methods applied herein to annual river temperature regimes have wide potential applicability for the assessment of largescale hydroclimatological interactions.
Future changes in river temperature are anticipated in response to a changing climate (Webb and Walling, 1992;Webb and Nobilis, 1994;Mohseni et al., 2003;van Vliet et al., 2011).This study represents an important first step in identifying the locations within England and Wales and dynamics of annual river temperature regimes, which may be impacted most (i.e.those most sensitive to air temperature change).The results suggest that the regime shape will be most sensitive to a changing climate, followed by regime magnitude in the largest and least permeable basins.Regime magnitude in the smallest and most permeable basins is anticipated to be least sensitive.The outcomes of this study contribute new knowledge to the scientific basis for making informed regulatory and management decisions regarding river temperature within England and Wales.

Figure 1 .
Figure 1.Maps of England and Wales showing (a) water temperature stations during the 1989-2006 common period, (b) water temperature monitoring stations within the common period with temperature samples in >90% of months and (c) paired air-water temperature monitoring stations

Figure 4 .
Figure 4. Box plots of annual (a) mean, (b) minimum, (c) maximum and (d) standard deviation for long-term average river and air temperature magnitude regime classes A Extended June-August maximum and January minimum with gradual warming and rapid autumn cooling (157 station-years, 22%) IA-Tw B July maximum and January minimum with rapid warming and gradual cooling (270 station-years, 38%) IA-Tw C August maximum and December minimum with gradual warming and rapid cooling (275 station-years, 39%) Four inter-annual air temperature shape regimes (IA-Ta x ) were identified as follows (Figure 6): IA-Ta A Extended June-August maximum and November-February minimum with a very rapid rate of cooling (37 station-years, 5%) IA-Ta B July maximum and December minimum (240 station-years, 34%) IA-Ta C August maximum and December minimum (269 station-years, 39%) IA-Ta D August maximum and January minimum (156 station-years, 22%)

Figure 5 .
Figure 5. Maps of England and Wales showing distribution of long-term average (a) river temperature regime classes, (b) air temperature regimes classes and (c) associated air-river temperature regime classes at paired air-river temperature stations

Figure 9 .
Figure 9. Percentage frequency of occurrence of each inter-annual magnitude regime in each study year for river temperature in (a) R-Tw 1 , (b) R-Tw 2 , (c) R-Tw 3 and (d) R-Tw 4 and air temperature in (e) R-Tw 1 , (f) R-Tw 2 , (g) R-Tw 3 and (h) R-Tw 4

Figure 10 .
Figure 10.Box plots showing (a) SI strength, (b) mean basin permeability, (c) basin area and (d) mean basin elevation for for river temperature stations in each region

Table I .
Properties of 38 river basins selected for analysis of air-water temperature associations [River basins are listed in alphabetical order.] RIVER TEMPERATURE REGIMES OF ENGLAND AND WALES © 2013 The Authors.Hydrological Processes published by John Wiley & Sons, Ltd.Hydrol.Process.(2013)