Lessons from the 2018 drought for management of local water supplies in upland areas: A tracer‐based assessment

Climate change, combined with industrial growth and increasing demand, could result in serious future water shortages and related water quality and temperature issues, especially for upland and humid areas. The extreme 2018 drought that prevailed throughout Europe provided an opportunity to investigate conditions likely to become more frequent in the future. For an upland rural catchment utilised by the distilling industry in North‐East Scotland, a tracer‐based survey combined discharge, electrical conductivity, stable water isotopes and temperature measurements to understand the impacts of drought on dominant stream water and industry water sources, both in terms of water quantity and quality (temperature). Results showed that water types (groundwater, ephemeral stream water, perennial stream water and water from small dams) were spatially distinct and varied more in space than time. With regards to the drought conditions we found that streams were largely maintained by groundwater during low flows. This also buffered stream water temperatures. Water types with high young water fractions were less resilient, resulting in streams with an ephemeral nature. Although our results demonstrated the importance of groundwater for drought resilience, water balance data revealed these storage reserves were being depleted and only recovered towards the end of the following year because of above average rainfall in 2019. Increased storage depletion under continued trends of extreme drought and water abstraction could be addressed via informed (nature based) management strategies which focus on increasing recharge. This may improve resilience to droughts as well as floods, but site specific testing and modelling are required to understand their potential. Results could have implications for management of water volumes and temperature, particularly for the sustainability of an historic industry, balancing requirements of rural communities and the environment.

electrical conductivity, stable water isotopes and temperature measurements to understand the impacts of drought on dominant stream water and industry water sources, both in terms of water quantity and quality (temperature). Results showed that water types (groundwater, ephemeral stream water, perennial stream water and water from small dams) were spatially distinct and varied more in space than time.
With regards to the drought conditions we found that streams were largely maintained by groundwater during low flows. This also buffered stream water temperatures. Water types with high young water fractions were less resilient, resulting in streams with an ephemeral nature. Although our results demonstrated the importance of groundwater for drought resilience, water balance data revealed these storage reserves were being depleted and only recovered towards the end of the following year because of above average rainfall in 2019. Increased storage depletion under continued trends of extreme drought and water abstraction could be addressed via informed (nature based) management strategies which focus on increasing recharge. This may improve resilience to droughts as well as floods, but site specific testing and modelling are required to understand their potential. Results could have implications for management of water volumes and temperature, particularly for the sustainability of an historic industry, balancing requirements of rural communities and the environment. indicate more frequent periods of hydrological extremes, and relatively lower snow cover in winter (Chan, Falkner, Goldberg, & van Asselt, 2018;Kay, Crooks, Davies, & Reynard, 2014). Through drought periods, groundwater is often the sole source of streamflow (Blumstock, Tetzlaff, Malcolm, Nuetzmann, & Soulsby, 2015;Frisbee, Phillips, Campbell, Liu, & Sanchez, 2011;Gosling, 2014;Winter, 2007). Projected changes in climate could result in reductions to groundwater recharge, and thus lower contributions to streamflow (Cuthbert et al., 2019;Green et al., 2011;Isokangas, Ronkanen, Rossi, Marttila, & Kløve, 2019). This issue is of particular importance for rural upland catchments, such as in the Cairngorms (Scottish Highlands), where groundwater sources involve relatively small, near-surface aquifers (Marsh & Anderson, 2002;Scheliga, Tetzlaff, Nuetzmann, & Soulsby, 2017) which have less resilience to climate change (Hugman, Stigter, Monteiro, & Nunes, 2012;Wright & Novakowski, 2019). With mostly impermeable solid geology, groundwater stores are restricted to fractures and faults or superficial drift deposits (Robins, 2002;Scheliga, Tetzlaff, Nuetzmann, & Soulsby, 2018;Soulsby, Rodgers, Smart, Dawson, & Dunn, 2003).
Climate change predictions also indicate that water temperatures will increase (Capell, Tetzlaff, Essery, & Soulsby, 2014;Jyväsjärvi et al., 2015;Stocker et al., 2013). This will have significant impacts on aquatic and riparian habitats, and their associated organisms (Caissie, 2006;Lake, 2003;Loinaz, Kampp, Butts, & Bauer-gottwein,-2013;Wenger et al., 2011). This issue is even more acute when rural communities and distilling industries depend on such resources. This results in a situation where a balance must be established between potentially increased water demand and maintaining essential ecosystem services (Forzieri et al., 2014;Gosling, Zaidman, Wann, & Rodgers, 2012), in the face of a decreasing resource (Sample, Baber, & Badger, 2016). In this context, decisions made by water managers can have far-reaching consequences.
Various branches of the food and drink industry are reliant on substantial volumes of water meeting specific water quality standards (Crabtree, Macdonald, & Dunn, 2002). This is of particular importance for the distilling industry, as the distillate must be cooled, and changes in temperature of groundwater sources or stream water can have production implications (Freire-González, Decker, & Hall, 2017). Distilleries may use a mixture of private groundwater springs and riverine waters for high-quality process waters and relatively low-quality cooling waters. Abstractions from both surface and groundwater source types are usually regulated to ensure ecological standards are met and the stability of resources is maintained (Scottish Executive Environment Group, 2005), but these regulations often lack catchment-based long-term data and understanding (SEPA, 2019).
To ensure both future production and appropriate abstraction legislation under climate change projections, there is a need to understand (a) the relative role of different water sources (Isokangas et al., 2019), both in terms of water quantity and quality (specifically, temperature) (b) the resilience of these sources under different climatological conditions (Floriancic et al., 2018), and (c) the role of landscape properties such as geology, soils and land-use (Geris, Tetzlaff, & Soulsby, 2015;Zimmer, Bailey, McGuire, & Bullen, 2013;Zomlot, Verbeiren, Huysmans, & Batelaan, 2015).
In the summer of 2018 a major drought occurred across much of Northern Europe, with wide-reaching consequences resulting from a prolonged period of below-average rainfall and above-average temperatures (Afzal & Ragab, 2019;Brunner, Liechti, & Zappa, 2019;Hänsel, Ustrnul, Łupikasza, & Skalak, 2019). Many rivers in Scotland experienced near-record low flows and high temperatures (CEH, 2019). These included the River Spey catchment which covers one of the most famous and economically important Scotch whisky production areas. Several distilleries were forced to halt production as a result. Nevertheless, the drought provided an opportunity to apply a tracer study to understand how water sources for distilleries are affected by drought conditions that may occur more frequently in the future as a result of climate change (Collet, Harrigan, Prudhomme, Formetta, & Beevers, 2018;Soulsby, Rodgers, et al., 2003;Spinoni, Vogt, Naumann, Barbosa, & Dosio, 2018).
Here, we investigated the impacts of the 2018 drought on the dominant water sources and their temperatures for a small headwater catchment of the River Spey. The site is typical for UK/Scottish headwaters and water is abstracted for the distilling industry. As such, the main aim was to evaluate an economically important small-scale upland water resource in the context of resilience to climate change and the threats it poses to the maintenance of water quantity and quality.
More specifically, our objectives were to: 1. Characterise the 2018 drought by exploring it within a long-term context 2. Use tracers to evaluate variation in catchment water sources and the impacts of drought 3. Investigate the impact of drought on water volumes and temperature 2 | METHODOLOGY

| Study site description
Blairfindy is a 0.9 km 2 sub-catchment of the river Livet, which is one of the main sites from which Glenlivet Distillery obtains cooling water.
Adjacent to Blairfindy, the Castleton and Heather Cottage catchments also provide water to local residents as well as the distillery (Figure 1).
The Blairfindy catchment is characteristic of the Cairngorms, dominated by crystalline bedrock with limited groundwater storage, whereby water movement is mostly restricted to shallow fractures and faults. A more permeable limestone member runs through the Blairfindy catchment (SW-NE) and may act as a deeper groundwater source. Active storage is also likely in shallow drift deposits of gravels, periglacial material and some in till in the valley bottom (Wilkinson, Stutter, & Gunter, 2016). Blairfindy has a mean elevation of 438 m.a.s. l. and is mostly north-facing with winter topographic shading. Mean annual precipitation is 900 mm with most (93%) falling as rain during frequent (60% rainy days per year) low intensity events, typically <10 mm/day. Due to the northerly location, low energy inputs and humid conditions, Potential Evapotranspiration (PET) is relatively low (450 mm/year). Daily temperatures average 6.2 C; maximum average daily temperatures were 18.7 C for July and minimum −1.3 C for December.
On the upper slopes, the humus-iron podzol soils are relatively freely-draining. These support heather shrubs (Calluna and Erica spp), which dominate in the east, and grazed acidic grassland in the west, broken up by a small coniferous woodland (Wilkinson et al., 2016).
The mid-to lower slopes are dominated by peaty podzols and peaty gleys with thicker peats where surface-water flow-pathways occur.
Due to the generally wet soils and impermeable geology, streamflow responds rapidly to rainfall. In a preliminary study, Wilkinson Supplementary Materials S1). For Blairfindy, total annual abstractions (i.e., from streams as well as groundwater wells) equate to approximately 10% of annual precipitation. There is a two-week "shutdown period" at the distillery coinciding with summer maintenance when flows are usually lower and air and stream temperatures higher.
2.2 | Data collection and analysis 2.2.1 | Long term data and analysis for context of study period Precipitation data since 1950 at Ballindalloch (7.5 km north of Glenlivet at 145 m.a.s.l.; Met Office, 2019a) were analysed using the Standard Precipitation Index (SPI) method (Supplementary Materials S2). Briefly, SPI normalises rainfall deficits based on a long term data record for a specified period and data are transformed to normal distribution so that the mean SPI is zero (Tigkas, Vangelis, & Tsakiris, 2015;World Meteorological Organization, 2012). Here, a 4-month running mean of a 3-month SPI was chosen. This captured the impact of the 2018 drought and seasonal deficits over timescales relevant to the distillery which is dependent on groundwater, which may take time to recover from rainfall deficits (Barker, Hannaford, Chiverton, & Svensson, 2016;Svensson, Hannaford, & Prosdocimi, 2017;World Meteorological Organization, 2012).

| Study period and hydrometric monitoring
The study's detailed monitoring period spanned from 01/12/2017 to 15/07/2019 to encompass pre-and post-drought periods. Hydroclimatic data (15-minute intervals) were collected from Glenlivet weather station from 24/05/2018 ( Figure 1b). This included an Environmental Measurements ARG100 tipping bucket rain gauge (0.2 mm precision); temperature and relative humidity sensors; wind speed and direction; a Kipp and Zonen NR-Lite2 net radiometer (with wind speed correction); and air temperature measurement. Snowfall and melt were determined using a time-lapse camera (hourly recording at E1). Stream discharge was also obtained (from 24/02/2018 onwards) for Blairfindy catchment outlet (S3) using 15-minute stage data recorded by an In-situ Rugged TROLL100 level-logger in a rated section (50 discharge gaugings across the full range of stage observations). Office, 2019b), and Livet flow data) were used to extend the Glenlivet hydrometric and climate data for the period prior to installation of the discharge monitoring and weather station on site (scaled, linear regression R 2 values Q = 0.51, PET = 0.86, P = 0.69, Air temperatures = 0.97). To understand temperature variability between water sources, temperature loggers (15-minute recording) were installed at most of the synoptic sampling sites (Figure 1b). Assuming water at these sites was well mixed, the loggers were positioned to be permanently submerged and capture representative temperatures (Folegot et al., 2018). This was of particular relevance for the small surface water reservoirs (maximum depth < 2 m) as their level varied depending on distillery usage and evaporation, as well as discharge.

| Water sampling
To characterise temporal changes in catchment water quality at high frequency, daily stream water samples were collected at Blairfindy outlet (S3; Figure 1b), and bulk daily precipitation samples at the weather station using ISCO 3700 auto-samplers. To characterise temporal changes in water quality at lower frequency, but at higher spatial resolution, synoptic sampling was conducted on a fortnightly basis throughout the study period, with all sites sampled on the same day ( Figure 1b, Table 1). This routine sampling was complemented with opportunistic sampling of higher and lower flow events to capture a fuller range of hydro-climatological conditions. Synoptic sampling captured the internal variability of water sources representing key sources used by the distillery within Blairfindy catchment and its surroundings, as well as other runoff sources ( Figure 1b). Ephemeral streams (E1-4) form on hillslopes, drying out through prolonged periods of low precipitation, and become tributaries to the main stream in average to wet conditions. Stream waters (S1-6) are perennial. Farther downstream of monitoring sites, some of these are canalised and directed to cooling water ponds for use by the distillery. Groundwaters (G1-5) originate from different geological units of the catchment and are abstracted by the distillery.
Finally, waters from small surface reservoirs with Dams (D1-2) were also sampled to assess some of the integrated sources feeding the distillery (Supplementary Materials S1).

| Laboratory analysis
Water samples were analysed for isotopes and a selection of water quality parameters, including temperature. Water stable isotopes of deuterium (δ 2 H) and oxygen-18 (δ 18 O) were measured for the daily (streamflow at site S3 and precipitation) and synoptic samples (all sites) using a Los Gatos DLT-100 laser isotope analyser following standard measurement protocols (precision of ±0.4 ‰ for δ 2 H and 0.1 ‰ for δ 18 O). These data are expressed in δnotation [‰] relative to Vienna Standard Mean Ocean Water. The daily stream (S3) and synoptic samples (all sites) were also analysed for Electrical Conductivity (EC) using a temperature corrected Jenway 4330 m. The fortnightly synoptic samples were analysed for pH and alkalinity using Gran titration to end points pH 4.5, 4.0 and 3.5 (Neal, Hill, Hill, & Reynolds, 1997), and on-site temperature measurements using a HANNA probe. From the synoptic surveys, six occasions were selected based on contrasting hydro-climatological conditions to represent the full variability of the conditions monitored. Table 2

| Data analysis
To understand how source water contribution in Blairfindy catchment changed during the drought, we combined hydrometric data with tracer analyses. First, we used two-component End-Member Mixing T A B L E 1 Landscape characteristics and instrumentation for synoptic sampling sites Note: N/A denotes data not yet available. Abbreviations: Blairfindy Q, Blairfindy outlet (S3) discharge; Livet Q, discharge from the River Livet; P, precipitation; PET, potential evapotranspiration; T, mean daily air temperature for Glenlivet weather station.
Analysis (EMMA) (Soulsby, Rodgers, et al., 2003) to quantify relative proportions of groundwater and surface water sources in stream water (S3). Although there is well-known uncertainty associated with using EC as a tracer (Benettin & van Breukelen, 2017;Pelizardi, Bea, Carrera, & Vives, 2017), the distinct variation in EC between contributing sources (see below), typical for UK upland catchments (Jarvie et al., 2001), showed that average EC in Blairfindy ephemeral streams (E1) and groundwaters (S1, representing a groundwater source at the head of the stream network and G1 representing a hillslope groundwater source) for end-members would be suitable for a first approximation (Engel et al., 2016). Uncertainty bounds were derived from the upper and lower EC values from different groundwater sampling sites to account for differences between groundwater EC (Robins, 2002). This same method was repeated using alkalinity as a tracer, with similar results (not shown here). Second, we estimated the proportion of relatively young water (<3 months old; Young Water Fraction, YWF) in all sampled sources (except S4, G5 and D1, due to limitations in data availability). This approach used δ 2 H as a tracer in precipitation and all sampling locations. Streams S2, S3 and S6 were flow-weighted using area-scaled discharge from Blairfindy (S3). This followed the method detailed by von Freyberg, Allen, Seeger, Weiler, and Kirchner (2018) whereby uncertainties were expressed as standard errors and used to calculate maximum and minimum estimates of YWFs (Jarvie et al., 2001;Kirchner, 2016aKirchner, , 2016bvon Freyberg et al., 2018).
To investigate the impact of the drought on the water balance and temperature we used hydro-climatological data to track variations in catchment storage (set to 0 at the start of the monitoring period). This was approximated on a daily basis using the water balance approach, accounting daily precipitation inputs against evapotranspiration, dis- 3.2 | Variation in catchment water sources and the impact of drought 3.2.1 | Source water quality varied with source type and spatial location We found distinct differences in the hydrochemistry of different water types. Also, within a water type group, there was generally more variation in space than in time. Overall, differences in geology underlying the drainage of sampling locations and estimated residence times explained trends in EC, isotopes, alkalinity and pH.
EC was lower for ephemeral streams than for other water types, consistent with relatively short residence times, particularly E1 (Table 3). Groundwater average values were higher, although considerable variability reflected differences in geochemical composition of the geology and subsurface mixing of different sources (Table 3;  The pH and alkalinity of water types showed similar patterns to EC (Table 3). Spatial differences between sampling points of the same water type were again apparent, with data from individual groundwater sampling points reflecting different underlying geologies. For example, waters with high EC values in G4 were also high in alkalinity and high in pH. Mean values for groundwaters were on average higher, ephemeral streams the lowest and perennial streams intermediate ( Figure 4; Table 3), and again S1 showed values in the range of deeper groundwaters. Overall, there were clear differences in relative variability between sources (for groundwaters this was low, for streams high, and dams intermediate).
Isotope data showed variation between water sampling type and spatial location, likely due to differences in subsurface mixing and residence times. Ephemeral streams, notably E1, showed the widest range in δ 2 H, with a standard deviation much greater than for other water types, suggesting limited mixing and short travel times, although some mixing must occur as the range covered only 26% of δ 2 H variability in precipitation (Figure 3d). Perennial streams varied less than ephemeral streams (Figure 4), while groundwater samples exhibited far narrower ranges in δ 2 H, suggesting greater subsurface mixing (SD range = 0.4-0.5‰; Table 3). Between groundwater sampling sites, the means differed by 0.5 ‰ demonstrating limited spatial differences.
Water in D1 and D2 reservoirs had δ 2 H characteristics intermediate between streams and groundwaters and relatively low variability, thereby indicating mixing.
Isotopes were also analysed in dual isotope space; Figure 5a shows precipitation isotopes and all water sampling types combined.
The Local Meteoric Water Line (LMWL) plots close to the Global Meteoric Water Line (GMWL). Ephemeral streams and stream waters plotted within a much narrower range than precipitation, and ephemeral streams varied more than stream waters (Table 3; Figure 5b). In comparison, groundwaters covered a very narrow isotopic range There was little evidence of evaporative fractionation as most samples plotted close to both meteoric water lines.
Temperatures of contrasting water sampling methods and water types generally followed relatively similar patterns of variability to δ 2 H being high in ephemeral streams and perennial streams, and low in groundwaters, although S1 at the head of Blairfindy stream and E2 were also particularly stable. D1 and D2 showed greatest variability and highest measured temperatures (Figure 4).

The major ion composition of all synoptic samples was plotted in a
Piper diagram (Figure 6), which showed that the catchment was mostly dominated by calcium and sodium bicarbonate waters. Overlap of stream and dam water between ephemeral and groundwater demonstrated the mixing that occurs in stream waters. Further differences were also highlighted between sources types; ephemeral streams, generally more sodium-chloride dominated, varied in magnesium, calcium and bicarbonate depending on flow; groundwaters were dominant in calcium-bicarbonate; and streams mixed between the two depending on flow.

| Sources of streamflow
Taken together, the hydro-chemical data (alkalinity, EC, δ 2 H and major ions) provided a consistent picture that streamflow at Blairfindy outlet (at S3) was a time-varying mix of groundwater and near-surface water as represented by the ephemeral stream waters. Daily EC values negatively correlated with flow at S3 (linear regression, R 2 = 0.7), with higher values associated with proportionally higher groundwater contributions (Figure 3e and 4). This enabled a simple hydrograph separation to disaggregate contribution of sources to S3 (Figure 7a). We used EC of ephemeral streams (E1; 31 μS/cm) and groundwaters including S1 as groundwater at the head of the stream (range = 115-195 μS/cm; Table 4) as the two end-members for EMMA. This showed that, unlike during rainfall events and wetter periods, S3 during the dry period was dominated (65-100%) by groundwater (Figure 7b). Uncertainty bands reflected the range of groundwater EC values, as groundwater mixing proportions were unknown. However, the absolute volumes of groundwater in S3 remained relatively constant through the year averaging 0.19 mm/day or around 30% of the average annual flow.
T A B L E 3 Summary of mean and standard deviation (SD) values for water quality parameters through total sampling period (01/12/17-15/07/19) for individual sources and mean vales for source type, with number of sampling occasions and % of which sources were dry or frozen (sample was not obtained) Having highlighted the overall volumetric dominance of surface/ near-surface runoff sources in S3 through EMMA, the variations of δ 2 H in precipitation and source waters were used to estimate young water fractions (YWF) (i.e., proportion of water less than 3 months old; Table 3). Ephemeral streams, especially E1, had the highest YWF, with 97% of water less than 3 months old. YWFs at E2 and 4 were much lower (2 and 9%), and in fact lower than stream flow, indicating longer residence times and more mixing (

| The impact of the drought on the water balance and temperature
Basic water balance analyses revealed a storage deficit resulting from low precipitation and high evapotranspiration through the drought period ( Figure 3a,b) and for the 2017-2018 hydrological year overall (Table 4).
Above average precipitation in summer-19 enabled significant but not full recovery of these storage deficits. Figure 3c shows that by the end of the study period (July-19) storage was still 85 mm less than at the start (December-17). Nevertheless, when full hydrological years were considered, most of the storage deficits appeared to be replenished by October-19 (Table 4).
The impact of the drought conditions on water temperature varied with source type as observed in synoptic sampling (Figure 4; Table 3) and in the data from high temporal-frequency loggers, which captured the full extent of the temperature range and variability in sources In our SPI analysis, despite previous drought periods in the mid-1970s, 1990s, 2003the mid-1970s, 1990s, and 2015the mid-1970s, 1990s, (Hänsel et al., 2019Spinoni, Naumann, Vogt, & Barbosa, 2015), only the drought in the mid-1970s was longer and more intense than the cumulative effects of two dry winters preceding the summer of 2018 (Smith, 1977). Whilst the other dry periods (SPI ranging between −2 and −1.5) had heatwaves, low river flows, and impacts on agriculture (Brunner et al., 2019;Fink et al., 2004;Ionita et al., 2017), they were considerably shorter. The 2018 summer period across Central and Northern Europe was a notably "hotter drought," and the UK was identified as one of the drought hotspots (Buras, Rammig, & Zang, 2019). In Scotland this drought also followed a prolonged period with frequent below-average rainfall anomalies (2012-2019), especially in the winters of 2016/17 and 2017/18. Therefore, the impact on and recovery of hydrological systems was more likely extended.
SPI is often a starting point for meteorological drought analysis (CEH, 2019;Hayes, Svoboda, Wall, & Widhalm, 2011;Svoboda & Fuchs, 2017). The resulting drought characteristics depend on averaging and return period choice, plus the length of study (Brunner et al., 2019). In addition, availability of long-term data for site-specific analysis can limit context (Barker et al., 2019). However local conditions at Glenlivet were assessed in context of the last 50 years.
Although multivariate indicators using multiple drought characteristics have become popular (Hao & Singh, 2015), our simple method was well-suited to providing metrics closely associated with impacts of concern to stakeholders at the distillery (Svoboda & Fuchs, 2017).
These included need of water of specific volumes and temperatures for processing and cooling (index focus on precipitation, temperature) F I G U R E 8 High frequency temperature recordings representative of each different source type monitored (a) air and streams, both perennial and ephemeral, with scale of (b) represented in (a) as black horizontal lines showing 3-12 C temperature range and (b) groundwaters and dammed surface water reservoirs. Gaps represent times when sources were drained (D2) or were dry (E1) (Piggott, 2017) and production of discharge waters that need to be accommodated (index focus on River Livet flow) (SEPA, 2017). Additionally, in upland areas with limited catchment storage (such as the Cairngorms), the relationship between meteorological and hydrological drought is usually strongly correlated, with limited storage to buffer drought propagation (Barker et al., 2016;Haslinger, Koffler, Schöner, & Laaha, 2014).

| Use of tracers to evaluate variation in catchment water types
We found distinct differences in water quality characteristics between water source types, which reflected variations in storage and flow paths, both in space and time (Edmunds, Shand, Hart, & Ward, 2003;Hem, 1985). Most prominent differences were found between groundwater and ephemeral streams, while the perennial stream and reservoir water comprised mixtures of the two. Isotope analyses revealed that groundwaters were strongly mixed and had relatively low YWFs (i.e., most water was >2-3 months old) (Blumstock, Tetzlaff, Dick, Nuetzmann, & Soulsby, 2016;Kirchner, 2016b;Scheliga et al., 2017;von Freyberg et al., 2018), whereas signatures in ephemeral streams exhibited most variability and high YWFs (i.e., were dominated by water <2-3 months old) (Zimmer et al., 2013). This is consistent with the high alkalinity in groundwaters, which can be attributed to longer contact time in deeper storage, facilitating greater weathering and ion release (Cresser et al., 2000;Godsey, Kirchner, & Clow, 2009;Soulsby et al., 2007).
The role of soil and geological characteristics in determining these patterns has been shown in many tracer-based studies in Scotland Geris et al., 2015;Scheliga et al., 2018;Soulsby et al., 1998), and elsewhere Isokangas et al., 2019;Rinaldo et al., 2011;Tetzlaff et al., 2018;Zuecco, Penna, & Borga, 2018). Differences between sampling locations of the same water type can be explained by such variations. For example, S1 and G4 had much higher alkalinity than other stream and groundwater sites (Table 3), this is likely explained by geological heterogeneity (i.e., the limestone end-member associated with high alkalinity) and/or faults providing deeper groundwater sources (Edmunds et al., 2003). Similarly, variations in ephemeral stream water may also be related to sources, whereby E2 and E4 were more likely connected to shallow groundwater while E1 and E3 were more dominantly related to direct (near) surface runoff.
Overall, water types varied more in space than in time, with differences becoming more distinct during dry conditions as connectivity between end-members was reduced (Blumstock et al., 2015;Darling, Gooddy, Morris, & Peach, 2012). This enabled EMMA to be used as a first approximation assessment of groundwater contributions to S3.
Identifying specific, distinct 'end-members' can be challenging (Barthold et al., 2011) as a result of geological heterogeneity, groundwater mixing, and non-conservative tracer behaviour (Benettin & van Breukelen, 2017;Pelizardi et al., 2017;Soulsby, Rodgers, et al., 2003). However, here use of multiple tracers helped to distinguish differences between sources and associated geological units, which explained results for groundwater contribution to the stream rather than different residence times (e.g., YWFs in groundwater are very similar). This ensured that, although EMMA is approximate, 'end-members' of source types delimited the range of values so that the mixing space of different source contributions to S3 could be understood (Abbott et al., 2016;Isokangas et al., 2019). 4.3 | Impacts of drought on catchment hydrological processes Recharge to shallow, less alkaline groundwaters and deeper, highly alkaline sources in geological fractures was facilitated by infiltration on the more freely-draining upper hillslopes, although the stable composition suggested significant mixing along relatively long flow paths.
More frequent (average) conditions occurred between precipitation events when drainage and ET led to unsaturated hillslope soils, and some ephemeral stream sites dried up (e.g., E1). Other ephemeral streams remained flowing in such conditions likely due to soil and (shallow) groundwater drainage, evidenced by higher EC, alkalinities and lower YWFs (notably E2, 4) than E1. All groundwaters sources continued to contribute. In drought conditions however, precipitation rapidly either evaporated or was transpired with no impact on the stream hydrograph. The other ephemeral streams (E2, 3, 4) also dried up, likely as groundwater levels dropped. Both high and low alkalinity groundwaters continued to contribute to flow. S1 at the head of the main Blairfindy stream also remained a source, and as surface water input decreased through the drought, S3 EC increased to values closer to those observed at S1 showing increasing dominance of groundwater from that of S1.
Our key findings are therefore that, through drought, groundwater (a) continued to maintain streamflow and (b) provided water at cool temperatures that buffered streamflow. This was encouraging in terms of maintaining ecosystem services and industrial abstractions (Folegot et al., 2018;Freire-González et al., 2017). End-member mixing showed that this groundwater contribution was 65-100% of streamflow throughout the drought, consistent with other Scottish studies, albeit not in such extreme droughts (Blumstock et al., 2015;Scheliga et al., 2018;Soulsby et al., 1998). This highlights: (a) the resilience of the streamflow quantity and quality (temperature) to the drought, (b) the importance of groundwater recharge to maintain this flow (Cuthbert et al., 2019;Haslinger et al., 2014;Segura et al., 2019), and (c) the fact that shallow sources with high YWFs are less resilient as they dried up (von Freyberg et al., 2018;Zimmer et al., 2013).
Although storage was sufficient to maintain streamflow throughout the drought, water balance data for 2017-2018 revealed a marked storage deficit. This was replenished due to above-average rainfall in 2018-2019, which included a wetter winter for the first time since 2015/16; though it was not until late summer/early autumn 2019 that storage deficits were replenished. However, the water balance estimates in the catchment are uncertain and previous investigation into the nature and role of deeper groundwater sources (e.g., G4) is also limited. One way to usefully complement our F I G U R E 9 Conceptual Diagram derived from tracer-based methods and hydrological understanding of catchment, illustrating the presence and activation of flow pathways of different types of water as the catchment transitions from (1) a wet, surface-water dominated system to (2) an average, mixed surface/groundwater system to (3) drought conditions in groundwater dominated state. Light-blue arrows suggest younger water (precipitation/surface waters), darker blue arrows suggest older water (well-mixed groundwater). Black triangle denotes water table and Limestone member representative of different geological features with more/less permeable zones in crystalline geology enabling groundwater circulation with faults and groundwater springs to surface preliminary work could be by geophysical surveys to characterise subsurface stores . Nevertheless, while our data suggested that the water balance for Blairfindy catchment at the end of the year 2017-2018 would have been negative regardless, this could have been mitigated with reduced industry abstractions. As drought events are likely to become more frequent, this suggests there is need for more informed water management, to improve sustainability of the resources available in future.
The high groundwater contributions (with stable temperatures) to the stream also helped buffer stream temperatures through the warm drought period (Snyder, Hitt, & Young, 2015). This was evident through comparison between datasets of high-frequency temperature recordings capturing full variability, rather than synoptic sampling, as this avoided issues with comparison between different approaches.
The low variability in the stream compared to air temperatures, suggest that the narrow, short, north facing stream was relatively sheltered from radiation inputs (Dick, Tetzlaff, & Soulsby, 2015;Isokangas et al., 2019;MacDonald, Boon, Byrne, Robinson, & Rasmussen, 2014) allowing the imprint of groundwater temperatures to be maintained.
This groundwater imprint was also observed in the 0.66 magnitude slope of the linear regression between air and S3 temperatures, which is relatively low given that this involves a small and shallow upland stream, and values up to 1 were observed by Arismendi et al. (2014) in Californian streams draining areas several orders of magnitude larger than Blairfindy at S3. However, a full assessment of the effects of factors such as altitude, aspect, vegetation shading, and riparian saturation (Dick et al., 2015) on stream temperatures would need energy balance modelling.

| Limitations
In this study we employed complementary hydrometric and temperature data with tracer analyses to investigate drought impacts on dominant water sources and their temperatures. However, certain limitations remain, though these are typical for most small-scale rural catchment studies (Soulsby, Tetzlaff, Dunn, & Waldron, 2006). Some of the issues relate to data availability. For example, we used hydrometeorological data from nearby stations to provide long-term context and to extend the time series pre-installation of the Glenlivet monitoring network. Although we scaled the data to consider variations in microclimate, topography and catchment size, some uncertainty remains, though the general picture is likely accurate.
Additionally, despite the sampling throughout the period of interest being conducted at a relatively high temporal and spatial resolution; observations of all water balance components are inherently associated with uncertainty (Beven & Westerberg, 2011;McMillan, Krueger, & Freer, 2012). In particular, the abstraction data were obtained from aggregated distillery records, so that short-term variations in daily amounts were unknown. Even though this did not affect the overall annual water balance calculations, short-term uncertainty is likely. Higher frequency characterisation of variability could have been useful for several other parameters too. For example, high-frequency (i.e., sub-hourly) sampling has recently become more widely available for EC and isotopes (von Freyberg et al., 2017) and can provide additional insights, for example into source variations within an event (Kirchner, Feng, Neal, & Robson, 2004). However, as our aim was to capture the longer-term impacts and recovery of a catchment to drought, the longer study period was prioritised over high frequency sampling campaigns. We used EC as a tracer in EMMA with two groundwater end-members, which had relatively broad uncertainty bands. Nonetheless, the results from our complementary analyses have shown that this still provided a reasonable first approximation of groundwater contributions to the stream.

| Implications
The 2018 drought was an opportunity to study the implications of prolonged dry and warm conditions on water quantity, quality and temperature. Although we found that groundwater buffered stream flows and moderated water temperatures, projected trends to drier and warmer summers in Scotland may result in depletions of catchment storage becoming more common (Afzal & Ragab, 2019;Cuthbert et al., 2019;Gosling, 2014;Spinoni et al., 2018). Although these projections are also expected to be coupled with higher and more intense winter precipitation (Chan, Kahana, Kendon, & Fowler, 2018;Kay et al., 2014), the impacts of cumulative reoccurring dry periods could be significant, especially if groundwater recharge is affected (Cuthbert et al., 2019;Taylor et al., 2012;Van Lanen, Wanders, Tallaksen, & Van Loon, 2013). Generally, deeper aquifers take longer to respond to climatic fluctuations and may offer short-term resilience to low precipitation, maintaining streamflow with water of more stable temperatures longer than near-surface stores (Bovolo, Parkin, & Sophocleous, 2009;Shah, 2009). However, these longer response times also mean that groundwater takes longer to be replenished following droughts (Kundzewicz & Döll, 2009). In addition, if air temperatures continue to increase, particularly in the winter recharge period, groundwater temperatures may also increase (Figura, Livingstone, Hoehn, & Kipfer, 2011;Jyväsjärvi et al., 2015), reducing the potential for stream temperature buffering in future.
The gradual drying of catchments, climate change or anthropogenically induced, can affect water resource availability (Allen, Breshears, & McDowell, 2015;Buras et al., 2019;Seneviratne et al., 2010) and may lead to further cascades in ecological change and degradation (Allen et al., 2015;Stocker et al., 2019). Drought-induced soil deterioration such as the breakdown of organic matter in peat may lead to extensive carbon release (Cooper, Thoss, & Watson, 2007;Green et al., 2019;Worrall, Burt, & Adamson, 2006), and mobilisation to streams (Fenner & Freeman, 2011). Deterioration of soils may also affect physical properties linked to infiltration and recharge (Hueso, García, & Hernández, 2012;Toberman, Freeman, Evans, Fenner, & Artz, 2008). In drought, higher temperatures, changed biogeochemistry and longer hydraulic residence times (Figure 9), mean that deterioration of water quality is highly likely (Mosley, 2015). Aquatic ecosystem response to these multiple drivers can be difficult to predict (Brennan & Collins, 2015), but such ecological changes need to be assessed conjunctively while maintaining water abstractions (Caissie, 2006;Ormerod, 2009). In regions such as Speyside in Scotland, environment and industry are highly dependent on maintaining these balances for the benefit of salmon and the production of whisky (Fabris, Malcolm, Buddendorf, & Soulsby, 2018), so likely changes need to be assessed to provide an evidence base for management (Isaak, Wollrab, Horan, & Chandler, 2012;Jackson, Fryer, Hannah, Millar, & Malcolm, 2018).
Abstraction and catchment management options have the potential to mitigate such impacts if designed to adapt to projected increases in precipitation, and reduce the impact of extended drought periods (Bouwer, 2002;Hewett, Wilkinson, Jonczyk, & Quinn, 2020).
Storage and attenuation features, shown to affect flood peak discharge (Nicholson, O'Donnell, Wilkinson, & Quinn, 2020;Wilkinson, Quinn, & Welton, 2010) and groundwater recharge (Escalante et al., 2007), could help build resilience to flooding and droughts using both traditional engineering approaches (Volpi, Di Lazzaro, Bertola, Viglione, & Fiori, 2018) or nature-based solutions (NBS) (Nesshöver et al., 2017). NBS are "actions that aim to help societies address a variety of environmental, social and economic challenges in sustainable ways. They are actions which are inspired by, supported by or copied from nature" (Bauduceau et al., 2015). Applicable in rural or urban contexts, they are designed to address specific issues (e.g., fluvial or surface-water flooding) but often achieve multibeneficial outcomes such as physical changes in water volumes, temperature, sediment and pollution control, but also improved water but may also be suitable for drought mitigation (Hartmann & Slavíková, 2019;Sapiano, Schembri, & Brincat, 2013). Measures may include the building of bunds or barriers to temporarily store water allowing time for infiltration, more commonly applied in small-scale managed aquifer recharge projects (Escalante, Sebastián Sauto, & Gil, 2019;Glendenning & Vervoort, 2011), or cross-slope planting to capture snow during winter (Kravčík et al., 2012). Furthermore, water temperature management may include localised riparian or hillslope planting (Jackson et al., 2018) offering shading of water sources. The efficacy of NBS however is likely very dependent on site conditions and location (Bouwer, 2002). Although the body of evidence is accu- At Blairfindy, NBS may provide one option for management and will benefit from the detailed knowledge of the role of soils, geology and hydrology reported here. With relatively responsive hydrology ( Figure 9), opportunity for infiltration into soils lies mostly in the upper hillslopes (less peat, more rankers) where recharge is greatest. This could encourage further recharge into the active storage zone (till aquifer), and to connections with fractures and deeper groundwater sources. Our results have shown these sources provide a more stable, reliable contribution to streamflow. The water balance analysis has shown groundwater would benefit from informed management to improve resilience through drought periods. Hence, managing abstractions and potentially encouraging recharge to these sources through NBS could improve resilience to climate change in future.

| CONCLUSION
Drought conditions in summer 2018 had a significant impact on the quantity and quality (temperature) of water supplies to the Glenlivet distillery. Both in terms of duration and intensity, the drought provided an opportunity to investigate hydrological conditions that are likely to become more frequent in future. A tracer-based survey combined continuous measurements at the catchment outlet with synoptic sampling to identify dominant water sources and determine the impacts of the drought on them. Spatially distinct water sources could be identified with distinct hydrological and hydrochemical differences between water source types. End-member mixing analysis and young water fraction analyses showed that older groundwater maintained 65-100% of flow and more stable temperatures through the summer low flow period in the main stream.
Water balance analysis revealed that, although groundwater maintained streamflow throughout the drought, catchment storage was being depleted, and the hydrological year 2017-2018 ended with substantial water deficits. This was mostly replenished towards the end of 2018-2019 as a result of the years above average rainfall.
With projected changes in climate suggesting more frequent drought events and warmer summer conditions, the effect on catchment storage dynamics will be a key relationship to maintaining water supplies.
Although our short study has limitations, it highlighted the importance of groundwater in maintaining resilient water supplies during an extreme drought period. Moreover, the results show potential opportunities to maintain water volumes and temperatures through informed groundwater management: this would involve combining storage and infiltration for groundwater recharge, through the use of NBS to improve resilience to hydro-climatic extremes. Although the evidence base for NBS is developing, pilot interventions and modelling are needed to develop site-specific evidence, to understand impacts at different locations and spatial scales. This research may be essential in improving the sustainability of industry in rural communities in upland catchments.
Dimitrova-Petrova and Eva Loerke are thanked for assistance with fieldwork, while we acknowledge Dr Aaron Neill for his advice on young water fraction analyses. Trevor Buckley and staff at the Glenlivet Distillery are thanked for on-site assistance and supply of data and abstraction records. We thank Audrey Innes, Dr Bernhard Scheliga, and Dr Ilse Kamerling for their support with the laboratory isotope analysis.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.