A longer-term perspective on soil moisture, groundwater and stream flow response to the 2018 drought in an experimental catchment in the Scottish Highlands

The drought of summer 2018, which affected much of Northern Europe, resulted in low river flows, biodiversity loss and threats to water supplies. In some regions, like the Scottish Highlands, the summer drought followed two consecutive, anomalously dry, winter periods. Here, we examine how the drought, and its antecedent conditions, affected soil moisture, groundwater storage, and low flows in the Bruntland Burn; a sub-catchment of the Girnock Burn long-term observatory in the Scottish Cairngorm Mountains. Fifty years of rainfall-runoff observations and long-term modelling studies in the Girnock provided unique contextualisation of this extreme event in relation to more usual summer storage dynamics. Whilst summer precipitation in 2018 was only 63% of the long-term mean, soil moisture storage across much of the catchment were less than half of their summer average and seasonal groundwater levels were 0.5 m lower than normal. Hydrometric and isotopic observations showed that ~100 mm of river flows during the summer (May-Sept) were sustained almost entirely by groundwater drainage, rep-resenting ~30% of evapotranspiration that occurred over the same period. A key reason that the summer drought was so severe was because the preceding two winters were also dry and failed to adequately replenish catchment soil moisture and groundwater stores. As a result, the drought had the biggest catchment storage deficits for over a decade, and likely since 1975 – 1976. Despite this, recovery was rapid in autumn/winter 2018, with soil and groundwater stores returning to normal winter values, along with stream flows. The study emphasizes how long-term data from experimental sites are key to understanding the non-linear flux-storage interactions in catchments and the “ memory effects ” that govern the evolution of, and recovery from, droughts. This is invaluable both in terms of (a) giving insights into hydrological behaviours that will become more common water resource management problems in the future under climate change and (b) providing extreme data to challenge hydrological models.


| INTRODUCTION
Extreme weather outside of previously measured climatic variations is becoming more common in many parts of the world. It is important to understand how these changes force catchment hydrological function and the processes which regulate stream flow generation.
Long-term experimental catchment studies have always been essential for recognizing and contextualizing such extremes, and that importance is likely to increase in the future .
The drought in the summer 2018, which affected much of Northern Europe, resulted in low river flows and threats to water supplies across extensive areas . Elsewhere, similar unusually prolonged and extended droughts have set new records for low rainfall and high temperatures (Wanders et al., 2015), and the concurrent increased risk of catastrophic fires is becoming apparent (Turetsky et al., 2015). Whilst many classic runoff generation studies in catchment hydrology focus on storm events, we have much poorer understanding of how droughts develop and evolve (e.g. Huang et al., 2017). In particular, we need to better understand how the fluxes of "green" water (evaporation and transpiration) during dry and warm hydrometeorological conditions interact with catchment soil moisture and groundwater storage dynamics to affect "blue" water fluxes (to recharge and stream flows) that regulate the runoff generation processes that sustain low flows (Orth & Destouni, 2018). Additionally, there is a need to know how "memory effects"-or the persistent influence of antecedent conditions-affect catchment storage over longer periods preceding droughts and during them (Bales et al., 2018). Often in drought studies, analysis focuses on precipitation and stream flow time-series from national hydrometric networks. However, there is a need to relate these to the catchment context of sensitive soil moisture and groundwater dynamics and other proxies of runoff generation such as tracer dynamics.
The definitions of droughts vary, but relate to extended periods where the effects of below-average rainfall (meteorological drought) become apparent and propagate through hydrological systems (Huang et al., 2017). Droughts can extend over short periods of weeksmonths, but usually have more serious implications when they extend across seasons or even years. The Highlands of Scotland are not a region where extended droughts have historically been a matter of major concern (Spinoni et al., 2018). The country's latitude and proximity to the North Atlantic are associated with a predominance of maritime westerly weather systems and frontal rain, with prolonged periods of dry stable atmospheric conditions being relatively uncommon (Burt and Howden, 2013). Moreover, rainfall is generally high, and energy limited conditions and frequent high humidity supress atmospheric moisture demand and evapotranspiration. That said, short-term monthly or seasonal negative rainfall anomalies can occur throughout the year, but usually positive anomalies soon follow and replenish moisture deficits . Despite the generally wet conditions characterizing the Scottish hydroclimate, it has been shown that slower changes in mesoscale atmospheric circulation and teleconnections with Atlantic storm tracks generally lead to clustering of "wetter" and "drier" years (Werritty & Sugden, 2012). Over the past 50 years, these drier periods have tended to occur at roughly decadal intervals. Under such drier conditions, usual seasonal drawdown of soil moisture and groundwater levels may not be rapidly replenished by late summer/autumn rains. However, rarely are such deficits carried into the following year, unless low summer precipitation is followed by a drier-than-average winter (Soulsby, Birkel, & Tetzlaff, 2016). Even then, deficits are fully replenished by the start of the following spring at the latest . However, recent years leading up to the 2018 drought in the Scottish Highlands resulted in an unusual sequencing of two drier than average winters and two very dry summers, with associated low river flows (Fennell et al., 2020). Climate change projections suggest such changes in seasonal rainfall patterns are likely to be more common in future, with greater distribution of rainfall towards winters and higher summer temperatures (Capell et al., 2014). Additionally, land management pressures in upland areas are changing, with policy drivers creating trends towards increased forest cover which may also increase water use from increased interception evaporation and transpiration Haria & Price, 2000;Soulsby, Braun, et al., 2017).
Consequently, as these environmental and climate changes occur over protracted periods, long-term experimental catchments provide important repositories of information sources and data to examine how droughts affect moisture storage and fluxes and better understand and predict the likely effects of future change . Here, we use long-term hydrometric data from the Girnock Burn observatory in the Scottish Highlands to examine the 2018 drought in terms of the depletion of soil moisture and groundwater stores within the Bruntland Burn sub-catchment. This is supplemented with use of stable isotopes to understand how catchment function changes as the catchment transits between wet and dry conditions. We also estimate the storage changes in relation to output fluxes and contextualize the 2018 drought in the longer term perspective of rainfall variability over the past 50 years. This addresses three research questions.
1. To what extent are extreme droughts identifiable by extreme soil moisture and groundwater anomalies and do these vary spatially?
2. Are isotopes of value in drought studies in terms of identifying changing hydrological pathways?
3. What are the future implications of such droughts and how can catchment monitoring evolve to better understand them?
The latter question is particularly important in the context of the Girnock Burn, as the catchment is an important long-term fisheries monitoring site, particularly for Atlantic salmon (Salmo salar) (Glover et al., 2020). In Scottish rivers, Atlantic salmon is a key biodiversity component which is also economically important as a sports fishing resource. It is also a cold water species that is sensitive to various aspects of river flow and stream temperature and hence future droughts (Glover et al., 2020). Hence, an evidence base to guide conservation management and mitigation measures is urgently required.

| STUDY SITE
The Bruntland Burn (BB) (Figure 1) is a small 3.2 km 2 montane headwater catchment in the Cairngorms National Park in NE Scotland, UK. It is a sub-catchment of the 31 km 2 Girnock Burn, which is a tributary of the larger~2108 km 2 River Dee catchment. The Dee is the main drinking water supply for 300 000 people and is economically critical for Atlantic salmon (Salmo salar) fishing (Tetzlaff et al., 2008).
The predominant climate can be described as transitional between northern temperate and boreal, but with a maritime influence that leads to generally cool, wet summers and mild, wet winters with an annual average temperate of 6 C. Average daily temperatures during summer and winter season are 12 and 1 C, respectively. Half of thẽ 1000 mm annual precipitation (P) falls during frequent, low-intensity events of <10 mm day À1 . There is no major distinct seasonality in the distribution of precipitation events, with heavy rain possible all year round, though winter is usually wetter. The annual potential evapotranspiration (ET) and run-off (R) are around 400 and 700 mm, respectively . Between May and August, ET is usually the dominant water flux out of the catchment (Kuppel et al., 2020). During drier conditions, direct groundwater (GW) inflows to the stream sustain baseflow (Ala-Aho et al., 2017). GW contributions account for~30-40% of the annual discharge, with the remainder coming from storm runoff .
The topography of the catchment is marked by its glacial origin ( Figure 1a). Steep hillslopes up to 61 encompass a wide and flat valley bottom which covers around 20% of the catchment. Mean gradient for the catchment is 14 ; and its elevation ranges from 238 to 539 m above sea level (a.s.l.). Granite dominates the underlying bedrock, fringed by Si-rich and Ca-rich meta-sediments ( Figure 1b).
The solid bedrock is overlain by an extensive low-permeable glacial drift deposit which covers up to 70% of the catchment area (Soulsby et al., 2007). This drift deposit is up to 40 m deep in the valley bottom, in contrast to the steeper hillslope, where shallower (<5 m) more lateral moraines and ice marginal deposits prevail (Soulsby, Bradford, et al., 2016). These deposits have been identified as the main GW source in the catchment .
The local vegetation is typical for the Scottish Highlands: the vegetation on the hillslopes and drier areas is predominantly 30-50 cm high heather shrubs (Calluna vulgaris), while the wetter areas and riparian zone are covered with Sphagnum spp. mosses and grass (Molinia caerulea). Just 11% of the catchment is covered by Scots Pine (Pinus sylvestris) forest to be found mostly on steeper hillslopes . Freely draining peaty podzols prevail on the upper and steeper hillslopes where heather dominates .
The Bruntland burn catchment in the NE of Scotland; (a) topography with the location of the automatic weather station, stream gauge and water sampler (which is also where precipitation is sampled for isotopes); (b) geology of the study site with extent of the drift deposit including the locations of the deeper wells 1-4; (c) soil types distribution with location of the soil moisture sensor in the forest and on the upper hillslope Here, the GW table is often >1 m below the soil surface Scheliga et al., 2018). The catchment interfluves and steeper hillslopes have been identified as the main areas for GW recharge. On the lower hillslopes and valley bottom, peats and peaty gley soils characterize the subsurface with shallower GW tables close (<0.3 m) to the ground surface . These wet conditions are sustained by subsidy from groundwater discharge on the lower slopes and the saturated peaty soils generate overland flow in storm events.
The response of the volumetric soil moisture content (VSMC) in the upper hillslope has been shown to have a more distinct drying and rewetting pattern between storm events compared to very conservative responses in the quasi-permanently saturated riparian zone and lower hillslopes . The hillslopes can contribute rapid storm run-off once they become hydrologically connected to the stream network, via the wetlands, during prolonged storm events . However, it has been shown that during summer, in particular, soils under forest are substantially drier than those under heather  due to higher interception and transpiration losses (Kuppel et al., 2020).

| DATA AND METHODS
This study focussed mainly on the evolution of the 2018 drought, which has been the driest prolonged period since detailed hydrometeorological monitoring started in the Girnock Burn in 2001 (Table 1) (Table 2). For more details on the construction and installation of the DW see Scheliga et al. (2018). The GW levels were monitored with micro-divers which measure the pressure of the water column above itself. The measurements were compensated for atmospheric pressure before deriving the height of the water column using a barodiver near the catchment outlet. The respective ground surface at each DW site was used as height reference point. The recorded GW levels were regularly (every 2 months) verified using manual dip meter measurements.
Daily samples of precipitation and stream water were collected with ISCO 3700 auto-samplers located near the outlet and later T A B L E 1 General precipitation statistics for 2018 and long-term average  (Neal, 2001). The alkalinity was determined on the spatial samples using a three-point acidimetric Gran titration to end points pH 4.5, 4.0, and 3.0 following the guidelines provided by Neal et al. (1997).
The results of the isotopic analysis are reported in the δ-notation The relationship between δ 18 O and δ 2 H in the isotope signal of the global precipitation forms the global meteoric water line (GMWL) (Dansgaard, 1964).The local precipitation can deviate from the GMWL  and forms the local meteoric water line (LMWL) (Equation 1): We further calculated the line-conditioned excess (short lcexcess) defined by Landwehr and Coplen (2006) which defines the unconformity with the local meteoric water line (LMWL) as where a is the slope and b the intercept of the local amount weighted precipitation (a = 7.6‰; b = 4.9‰).
The Standardized Precipitation Index (SPI) is a simple and temporally flexible drought index which was developed for drought detection and monitoring, it solely relies on precipitation data as input (Hayes et al., 1999). However, a long-term data time-series is required (>30 years) and is fitted to a probability distribution. This is then transformed into a normal distribution so that the median SPI for a particular site and period is zero. In other words, half of the previous precipitation amounts are below the median and half above. Positive SPI values are indicative of periods greater than the median precipitation (i.e. wet conditions), and negative values are less (i.e. dry conditions). A dry period anomaly is considered to occur when the SPI value is ≤ À1 and continues until SPI becomes positive (Tigkas et al., 2015). Table 3 shows the drought conditions according to the SPI. The SPI was calculated for each month (for short-term anomalies) and quarterly (for longer-term) during each hydrological year (Oct-Sep) using the Drought Indices Calculator (DRINC; Tigkas et al., 2015).
Additionally, a 6-month running mean for the monthly and quarterly SPI values was calculated to reduce the noise from extreme values.

| Rainfall and discharge dynamics
The year-round distribution of rainfall in the Scottish Highlands dictates that prolonged dry spells of no rainfall for a few weeks are extremely rare, were notable for a lack of such high rainfall events, even in the winter, despite the overall number of rain days remaining high (Figure 3).
Streamflow in the catchment is closely coupled to rainfall; with even small (<5 mm) rainfall events usually initiating a runoff response from the valley bottom wetlands (Figures 3 and 4). Streamflow is generally seasonal and is usually highest in winter, though this more reflects the effects of higher ET and greater soil water storage availability in summer-before reaching field capacity (Kuppel et al., 2020).

| Soil moisture and groundwater levels
The two long-term soil moisture sites showed quite different storage regimes due to different landscape positions and landcover (Figure 4b,c).

| Stable water isotope and alkalinity dynamics
Stable isotopes complemented the hydrometric data by providing useful insights into how the drought affects water flows paths  For better spatial assessment of isotope patterns and the influence of groundwater inflows under drought conditions, two synoptic surveys were undertaken (in August and November 2018) along the main stem of the BB downstream of the confluences of its three main headwater tributaries (Figure 9). During the dry August sampling, the headwaters showed small differences in δ 2 H, with HW1 and HW3 being similar, but HW2 being depleted by almost 3‰ in comparison (Figure 9a). At the sample point immediately downstream of the confluence of HW3 (BB17), the stream isotope signal showed a~1‰ increase moving towards that of HW3, which is the largest headwater catchment. However, by BB16, δ 2 H signaturesfell by~2‰. They then increased slightly between B15-B13 before falling to that of B16, and remained constant until B4 where after they increased by~1‰ at the catchment outlet. The lc-excess remained positive, with only B17 falling to zero (Figure 9b). Interestingly, the Gran alkalinities (Figure 9c) were high and showed a general increase downstream of the HW1 confluence. This overall pattern is consistent with depleted groundwater inflows in the valley bottom sustaining low flows, with limited inflow from wetlands that would increase δ 2 H and decrease the lcexcess and Gran Alkalinity.
In contrast, sampling when the catchment was re-wetted on 28th November 2018 showed higher and more consistent δ 2 H (betweeñ À56 to À55‰) moving in the direction of recent rainfall, more positive lc-excess and a consistently low Gran Alkalinity (<100 μeq L À1 ) along the entire stream network. In contrast to the drought conditions, this clearly showed that surface and near surface flow paths dominated runoff generation, and groundwater influences were overprinted in the tracer signals.

| Drought dynamics and storage changes
The dynamics of water storage in the Bruntland Burn prior to, and during, the 2018 drought are highly instructive for understanding how the effects of periods of moisture stress propagate through the catchment's hydrological compartments. The humid hydroclimate of Scotland dictates that storm runoff generation processes and floods have been much more extensively researched than droughts (Werritty, 2019).
However, the widespread impacts of the 2018 drought and future climate change forecasts, with more marked seasonal distribution of rainfall and warmer drier summers, call for a new focus on drier periods and their hydrological impacts (Capell et al., 2014;Spinoni et al., 2018).
In many environments, the increased dominance of green water fluxes, relative to blue water fluxes, occurs as catchments are drying down, and this is also the case in the Scottish Highlands (Kuppel et al., 2020). With increased evapotranspiration in the summer, the F I G U R E 9 Results of the synoptic sampling along the Bruntland burn (BB) main channel in August 2018 and November 2018; (a) deuterium composition; (b) lcexcess and (c) gran alkalinity; (d.1) sampling locations along the Bruntland burn stream, (d.2) full extent of the Bruntland burn catchment with the red rectangle indicating the extent of (d.1). Please note that the gap between outlet and location BB 1 is due to inaccessibility to the stream channel due to a deer fence upper soils begin to dry as atmospheric demand is met, and transpiration of the shallow-rooting heather and Scots Pine vegetation is sustained (Haria & Price, 2000). The resulting cessation of drainage through the soil profile, even though the organic, upper soil profile may retain moisture storage, restricts recharge, which is then mainly focused during winter . As a result, groundwater sustains stream flows and water tables fall as groundwater storage becomes depleted. In most years, recharge in the late autumn and early winter replenishes soil moisture deficits and groundwater depletion. However, in more severe dry spells with insufficient winter precipitation, "memory effects" of the previous year(s) can result in moisture deficits before summer green fluxes increase (Soulsby, Birkel, & Tetzlaff, 2016).

| Insights into runoff generation and storage changes during droughts through isotopes
The use of isotopes to better understand the drought evolution was highly informative. Traditionally, the rapid changes in the isotopic composition of precipitation and stream flow during storm events have been used to understand runoff generation processes . Isotopic changes are less obvious for droughts and for slower responding parts of catchment hydrological systems. However, some recent studies, at a range of scales, have used isotopes to assess green water fluxes and runoff generation. These vary from assessing sources of plant water uptake in droughts at the plot scale in Mediterranean environments (e.g. Barbeta et al., 2015) to assessment of groundwater inputs to larger catchments under low flow conditions in geographical regions as diverse at the Italian Alps (Chiogna et al., 2018), Costa Rica  and upland areas in central India (Noble & Arzoo Ansari, 2019). In the BB, the trajectory to more stable, intermediate isotopic ratios was consistent with the dominance of groundwater inflows, which increase in the valley bottom . This was confirmed by the spatial plots, which also underlined the value of using other tracers in conjunction with isotopes as the Gran Alkalinity data supported the influx of deeper waters downstream of the confluence with the main headwater tributaries.
The high lc-excess values were also consistent with the composition of groundwater sources which were mainly recharged by winter rainfall. Any evaporative signal from the upper soil horizons is lost in mixing as water percolates through the usually wet winter soils (cf. Tetzlaff et al., 2014). Runoff storm event runoff coefficients in the catchment are typically 10-40%, with higher values only occurring in winter storms . New water fractions are typically <10%, so precipitation isotope and lc-excess signals are not transferred directly to streams due to mixing processes in the wetlands . Consequently, in summer, absence of low lc-excess in stream water indicates that surface drainage from peats has stopped . Water in peaty pools usually evaporates in summer and affects the stream isotope composition, but the cessation of inflows of peat waters indicated a very dry catchment state in the drought of 2018, fed only by groundwater.

| Past and future context of the 2018 drought
Clearly, the drought of 2018 was an unusually widespread event throughout Europe (e.g. Imbery et al., 2018). In the Scottish Highlands, the prolonged period of lower-than-average rainfall over the two preceding years with two dry winters created antecedent conditions that forced the catchment into a very low state of storage. The degree to which summer soil water and groundwater levels dropped across the BB catchment was not previously seen in the years of monitoring. In the context of the 50 year long-term record, the drought was not unprecedented, with the mid-1970s being similarly characterized by a run of dry years and warm summers between 1975-1977. Although recovery was rapid following high autumn rainfall in 2018, a major concern is the increasing likelihood that such events become more frequent in future given the severity and implications of climate change impacts forecast for northern Europe in general, and the Scottish Highlands in particular (Capell et al., 2013).
The effects of drought-induced storage deficits on water fluxes, especially for stream flow, potentially have profound implications for upland Scottish rivers. Low summer flows reduce the thermal capacity of streams via lower volumes to heat which renders them more vulnerable to warming (as a result of lower volumes of water to heat), particularly in the context of climatic change (Fabris et al., 2018;Garner et al., 2014). Many streams in the Scottish Highlands support important cold-water ecosystems, most notably sustaining Atlantic salmon, which locally is the focus of an economically-important sports fishery. Salmon suffer growth inhibition and sub-lethal stress at temperatures >22.5 C and lethal effects when temperatures >32 C (Fabris et al., 2018). In recent years in the Girnock Burn, maximum temperatures have exceeded 22.5 C, though such headwater streams are especially vulnerable to warming through climate change (Hrachowitz et al., 2010). Reduced summer flows and the resulting lower thermal capacity could reduce both available physical habitat in stream channels from reduced wetted areas and increase thermal stress.
Drought-induced storage deficits may also be compounded by land use policies that are seeking to encourage tree planting both in general terms of increasing timber and biofuel production, but also to specifically mitigate against increased temperature effects from climate change (Scottish Government, 2019). While trees reduce short wave radiation reaching streams and wetlands, they also increase green water fluxes such as interception losses and transpiration, again at the expense of groundwater recharge and stream flow generation (Soulsby, Braun, et al., 2017;Wang et al., 2017) This is a particular risk as moorlands are often burned under management regimes to create habitat for grouse or red deer populations which are managed for game hunting (Davies et al., 2016). It is increasingly recognized that burning moorlands in headwaters can have wide-ranging effects on downstream river systems including altered hydrological regimes (Brown et al., 2015). Increased risk of managed fires getting out of control, or accidental fires in summer, which can affect extensive areas is increasingly recognized as a climate change effect in Scotland and beyond (Turetsky et al., 2015).
After burning, carbon losses can occur, and the loss of vegetation cover decreases green water fluxes, increasing runoff (Brown & Holden, 2020). It is clear that climate change will force a fundamental rethink of land and water management in the Scottish Highlands and elsewhere. As ever, long-term catchments provide a vital evidence base to inform decision making.

| Wider implications
Experimental catchments with long data time-series are valuable resources for drought assessment and will be crucial in monitoring the effects of climate change in the coming decades. There is potential in focusing more data collection on catchment storage dynamics, rather than flux measurements. This can be facilitated by new, more integrated tools for monitoring storage over a larger footprint using, for example, cosmic ray neutron sensors (Dimitrova-Petrova et al., 2020), hydrogeophysics (Dick et al., 2018) and micro-gravity changes (Kennedy et al., 2014). In addition, such tools can be used in catchments to assess the spatial distribution of storage in relation to landscape position (e.g. Soulsby, Bradford, et al., 2016), identification of important recharge zones, discharge areas and regions of important dynamic storage change to understand the non-linear responses of stream flows in both wet and dry periods . Such new insights can complement more traditional hydrometric monitoring in providing new data times series of extreme conditions such as droughts. Such data are also invaluable for challenging environmental models, as often these perform well under "average" conditions. In most spatially hydrological modelling, correctly capturing re-wetting after dry periods is usually the most challenging aspect (e.g. Soulsby et al., 2015), but crucial to understand the effects of droughts and their recovery periods. Thus, data streams from long-term catchments have a key role to play in improving models and making better projections for future climate change and land use scenarios (Ala-Aho et al., 2017). In particular, the combination of hydrometric and isotope data has the potential for improved representation of water storageflux-age dynamics (A. Smith et al., 2021).

| CONCLUSION
The 2018  Although these deficits were modest and rapidly replenished in the wet autumn of 2018, they resulted in low river flows generated from groundwater drainage. Such conditions are likely to become more common in future, as drier, warmer summers for Scotland are a predicted consequence of climate change, though winters are expected to remain wet. This raises questions over future land and water management as flows from headwater catchments provide water that sustains a wide range of ecosystem services, from public water supplies to economically important Atlantic salmon fishing.
Moreover, land management trends in the Scottish Highlands, like increasing forestry, increase green water fluxes and reduce blue water fluxes, so may exacerbate climate effects during low flows. Additionally, drier soil conditions may increase fire risk in dry peaty soils and require more careful approaches to moorland burning. The study underlines the value of long-term data in experimental catchments, both to contextualize and more fully understand hydrological function and provide data from extreme events to challenge modelling.

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