A multi‐proxy record of abrupt cooling events during the Windermere Interstadial at Crudale Meadow, Orkney, UK

Three clearly defined abrupt cooling events (ACEs) can be observed within Greenland Interstadial (GI)‐1 in the Greenland ice‐core records. However, the spatial variation in amplitude and timing of these ACEs is poorly understood due to the paucity of well‐dated records with quantified temperature reconstructions. This study presents high‐resolution chironomid‐inferred July air temperature (TJul) and oxygen isotope (δ18O) records from Crudale Meadow (Orkney Isles, UK). Three centennial‐scale ACEs punctuate the Windermere Interstadial at Crudale Meadow. The largest ACE shows an amplitude of 5.4 °C and a 1% isotopic decline and is centred on ~14.0 ka bp, consistent with the timing of the GI‐1d event in the Greenland stratigraphy. The two other observed ACEs are of smaller magnitude and are centred on ~13.6 ka bp and ~13.2 ka bp, with these smaller magnitude events tentatively correlated with the GI‐1cii and GI‐1b events, respectively, but lack sufficient chronological constraint to fully assess their timing. When comparing the Crudale Meadow record with other locations in the British Isles a strong relationship can be observed between the magnitude of TJul cooling and latitude, with a reduced signal in more southerly locations, indicating that oceanic forcing may be a key driver of the ACEs.


Introduction
The Last Glacial-Interglacial Transition (LGIT) (~16-8 ka BP) is characterised by a series of abrupt cooling events (ACEs) (Rasmussen et al., 2006(Rasmussen et al., , 2014. The most notable of these is the Greenland Stadial 1 (GS-1) (12 896-11 703 GICC05 b2k), recorded as the Younger Dryas in Europe and locally named the Loch Lomond Stadial (LLS) in Britain. This ACE occurs on a millennial scale and is widely preserved in a range of marine, ice-core and terrestrial archives (Lowe et al., 2008). However, the LGIT is also punctuated by shorter-lived ACEs that operate on centennial timescales (Rasmussen et al., 2014). Such events are clearly seen in the Greenland ice-core records during Greenland Interstadial 1 (GI-1) (14 692-12 896 GICC05 b2k), and also in the Windermere Interstadial (WI), the British correlative of GI-1, where at least three ACEs are recorded (e.g. Lang et al., 2010;Brooks et al., 2012). These centennial-scale ACEs are an important research priority as climate model simulations have demonstrated that northwest Europe will become increasingly susceptible to ACEs as greenhouse gas emissions increase (Gregory et al., 2005;Petoukhov et al., 2005;Hawkins et al., 2011).
The structure and magnitude of these centennial-scale ACEs is poorly understood beyond the Greenland ice cores. This is because: (1) some proxies are insufficiently sensitive to record the full expression of these events (Lowe et al., 1999); (2) many archives do not have sufficient sample resolution to reconstruct them in detail; and (3) some archives have insufficient stratigraphic resolution leading to a relatively poor expression in the sediment sequences of such short-lived climatic events. Even when the above are not an issue, the challenge of developing chronologies with the precision to accurately correlate centennial-scale climatic events has limited any discussions regarding climatic synchronicity (Björck et al., 1998;Lowe et al., 2008).
This study aims to investigate the spatial variations in the magnitude of WI ACEs across the British Isles, with a particular focus on the GI-1d cooling event. This paper presents new chironomid and δ 18 O records from a lacustrine sediment sequence taken at Crudale Meadow, Orkney Isles, Scotland, improving the spatial coverage of temperature proxy records for the northerly reaches of the British Isles. Both chironomid and δ 18 O analyses were applied at high resolution (approximately 20 years/sample resolution) allowing the nature of ACEs across this interval to be understood in greater detail than has been possible for most comparable records from northwest Europe. Both proxies are conducted on the same sequence (CRUM1) which ensures that samples are consistently taken from the same stratigraphic level. Furthermore, the site's robust independent chronology allows comparisons to be made with other archives such as the Greenland ice-core records, which also have δ 18 O records with a resolution of approximately 20 years. The paper concludes with comparisons to a suite of other C-IT and δ 18 O records from across the British Isles, discussing the regional and local variations in the magnitude of different ACEs within the WI.
area of approximately 0.09 km 2 and sits within a topographic depression between Devonian Qui Ayre sandstones and mid-late Devonian Upper Stromness Flagstones (Mykura et al., 1976). Thin tills on the surrounding slopes are partially derived from the carbonate-rich dolomitic siltstones, shales and sandstones of the Middle Old Red Sandstone and Lower Stromness Flagstones (Whittington et al., 2015). The in-filled lake basin currently contains a valley mire dominated by Phragmites australis and Cyperaceae and drains east via a small stream into the Loch of Stenness (Whittington et al., 2015).
The Orkney archipelago currently has a cool-temperate climate and experiences mean daily maximum temperatures of 6.4°C in January, 15.9°C in July and 8°C annually. These temperatures are relatively high when the latitude of the site is taken into consideration but can be explained by the warming influence of the North Atlantic current. Mean annual rainfall is 1038 mm with the majority of the precipitation falling in the winter months (data from the meteorological station at Kirkwall, 21 km east of Crudale Meadow; Met Office data 1981-2010). Further site details can be found in Whittington et al. (2015).
Previous studies obtained sediment records from the Crudale Meadow palaeolake with sequences composed of sediments dating to the WI, LLS and Holocene (Moar, 1969;Bunting, 1994;Whittington et al., 2015), similar to numerous LGIT sequences from other parts of Scotland and the British Isles (Lowe and Walker, 1977;Lowe et al., 1994). Whittington et al. (2015) identified three possible climate oscillations within sediments they believed to be of WI age. Attempts to construct an independent age model for the sequence using radiocarbon dates obtained from bulk sediments and aquatic remains yielded ages that were considered to be too old (Whittington et al., 2015). Proposed correlations were made to the GI-1d, GI-1cii and GI-1b events based on stratigraphic alignment with the Greenland ice-core record, but a consequence of this approach means that the sequence of Whittington et al. (2015) lacks independent verification of the timings of the encountered events. Timms et al. (2018) retrieved a new sediment record from the same palaeolake (CRUM1), which also contained sediments of Dimlington age (c. Late GS-2; pre-14.7 ka BP). Timms et al. (2018) developed a tephrochronology and tephrostratigraphy for the LGIT, using 12 cryptotephra horizons and one visible tephra layer, producing one of the best age-constrained sequences in the British Isles for this period. While there are proxies for landscape stability derived from sediment parameters at this site, palaeoclimatic proxy data supported by an independent chronology are yet to be produced from the Timms et al. (2018) core.

Sedimentology and physical parameters
This study focuses on sediments identified as WI in age (602-645 cm depth) from the CRUM1 sequence recovered by Timms et al. (2018). Sediments were classified using Troels-Smith (1955) and colours identified using the Munsell Soil Colour system. Calcium carbonate (CaCO 3 ) content and the organic component of the sediments were determined using a Bascomb calcimeter (Gale and Hoare, 1991) and by loss-onignition , respectively. The present work has increased the CaCO 3 sampling resolution from 2 cm intervals as used in Timms et al. (2018) to contiguous 1 cm intervals across the WI and 0.5 cm intervals between 629 and 634 cm.

Chronology
The initial age-depth model for the sequence constructed by Timms et al. (2018) was based on the occurrence of 13 tephra layers, six of which correlated to known eruptions, and has been further improved with a single accelerator mass spectrometry radiocarbon date from the later part of the WI. The radiocarbon sample consisted of picked terrestrial plant macrofossil remains including leaves of Dryas octopetala, Empetrum and Betula nana along with fruits of Betula nana from between 614 and 615.5 cm. Analyses were performed at Queen's University, Belfast. Our age-depth model was produced using the P_Sequence function as included in OxCal v 4.4.2 (Bronk Ramsey, 2008 and the IntCal20 calibration curve (Reimer et al., 2020). The age estimates for the Borrobol, Penifiler and Vedde Ash tephra layers are based on remodelling of the data presented in Kearney et al. (2018) utilising the IntCal20 curve and are identical to those report by Timms et al. (in press). All interpolated dates are rounded to the nearest decade to avoid spurious precision and are accompanied by 1σ errors.

Chironomid analysis
The sequence was sampled at 0.5 cm resolution for chironomids between 649 and 597 cm depth where every second sample was analysed for chironomids. Samples were contiguously analysed in intervals where preliminary data revealed notable changes. The sediment properties of the marl meant the standard laboratory approach outlined in Brooks et al. (2007) was not effective in releasing the chironomid head capsules in a clean identifiable state. We experimented with a range of additional laboratory procedures to improve head capsule recovery (see Supplementary information 1). Our final laboratory protocol starts by disaggregating sediment in 10% KOH for 20 min at 75°C followed by sieving over 90 and 212 µm meshes. This initial stage was followed by submerging the samples in a sonic bath for between 2 and 5 s, followed by a second phase of sieving to remove the newly disaggregated marl. The use of the sonic bath increased head capsule yield by up to 300%. The residue was then transferred to a Bogorov sorting tray and chironomid head capsules (HCs) were handpicked from the residue using fine forceps under a Motic SMZ-168-BP stereo microscope at 25× magnification and mounted on permanent microscope slides using Hydromatrix mounting medium. Chironomid HCs were identified using a binocular CX41 microscope at 400× magnification in accordance with Brooks et al. (2007), Rieradevall and Brooks (2001) and Wiederholm (1983).
Several samples with low count sums were amalgamated with adjacent samples with the aim of reaching a minimum count sum of 50 HCs per sample (e.g. Quinlan and Smol, 2001). This resulted in a dataset with 63 fossil samples, of which four samples did not reach the minimum count sum of 50 HCs even after sample amalgamation (samples at 649, 648, 647 and 597 cm; count sum range 39-48 HCs). However, as these samples had a low variety of taxa present, we chose to retain these samples in subsequent numerical analyses, assuming that the count sums of 40-50 are sufficient to provide reliable estimates of past temperatures (Quinlan and Smol, 2001). A percentage diagram was constructed using C2 version 1.7.7 (Juggins, 2007). Zonation of the chironomid diagram was derived using the CONISS clustering algorithm (Grimm, 1987) where statistical significance of the zones was determined using a Broken Stick model (Bennett, 1996). Zonation analyses were conducted in R (R Core Team, 2018) using the vegan (Oksanen et al., 2017) and rioja (Juggins, 2017) packages.
The 274-lake combined Swiss-Norwegian chironomidclimate calibration dataset was used to estimate July air temperatures (T Jul ; Heiri et al., 2011). This calibration dataset has several advantages over the Norwegian dataset (Brooks and Birks unpublished; see Supplementary information 2) which is more commonly applied to British lake records (e.g. Brooks et al., 2012Brooks et al., , 2016: it contains a higher number of taxa, it covers a larger temperature gradient, and it incorporates alkaline carbonate lakes with a chemistry similar to that which is supposed to have been present at Crudale Meadow during the WI. Head capsule preservation in our fossil samples was generally good, allowing identification to a high taxonomic resolution. However, some taxa had to be combined in both the fossil samples as well as in the calibration dataset in order to allow numerical comparisons. Paratanytarsus austriacustype and Paratanytarsus penicillatus-type were combined to Paratanytarsus undifferentiated due to the low number of mandibles preserved attached to HCs (c. <10%), prohibiting the differentiation of these morphotypes with any degree of confidence. Corynocera oliveri-type and Tanytarsus lugenstype were combined and are referred to as COTL-type, as they were hard to distinguish due to the varying degree of overlap between the lateral teeth on the mentum.
A two-component weighted averaging partial least squares (WA-PLS; ter Braak and  regression model was selected for reconstruction purposes as it combines the best predictive power with a root mean squared error of prediction of 1.4°C, r 2 jack of 0.9°C and maximum bias jack of 0.9°C, as estimated using leave-one-out crossvalidation. Sample-specific errors for our reconstructed T Jul values were calculated by 999 bootstrap cycles. The chironomid percentage data were square-root transformed prior to analysis to minimise variances, and 19 outlier lakes were removed prior to analysis. T Jul and sample-specific errors were calculated in C2 (Juggins, 2007). A reconstruction based on the Norwegian calibration dataset is provided for comparison in the Supplementary information.
Several methods were used to test the reliability of the chironomid-based temperature reconstruction (Birks et al., 1990;Engels et al., 2010). The percentage abundance of the identified subfossil chironomids that are absent from or rare in the calibration dataset was calculated for each fossil sample. A taxon was considered rare if it had a Hill's N2 (Hill, 1973) below 5 in the calibration dataset (Birks et al., 1990). Modern analogue technique (MAT) analysis was performed to assess the similarity of fossil assemblages to those in the modern calibration dataset. Cut-levels of the 5th and 10th percentiles of all squared-chord distances in the modern calibration data were used to determine 'no close' and 'no good' analogues and were compared with the distance between an individual subfossil assemblage and its nearest modern analogue (Telford, 2014a). MAT analysis was conducted in R using the analogue package (Simpson, 2020) with squared-chord distance used as a measure of dissimilarity. Samples were not square-root transformed prior to analysis. A goodness-of-fit test was performed using canonical correspondence analysis (CCA) (ter Braak, 1986) with July air temperature as the sole constraining parameter with the subfossil samples added passively in the analysis. Residual distances of the subfossil samples to the first CCA axis were compared with the residual distances of all the modern samples to the first CCA axis. Thresholds of the 90th and 95th percentile of residual distances of modern calibration dataset samples to CCA axis 1 were used to identify subfossil samples with a 'poor' or 'very poor' fit with temperature (Telford, 2014b). CCA analysis was conducted in R using the vegan package (Oksanen et al., 2017). Samples were not square root transformed prior to these analyses.

Stable isotope analysis
Samples for stable isotope analysis (δ 18 O and δ 13 C) were taken at 0.5 cm intervals. Approximately 0.5 cm 3 of sediment was disaggregated in sodium hexametaphosphate ((NaPO 3 ) 6 ) overnight, then sieved through a 63 µm nylon mesh. This removed any shell fragments from ostracods and molluscs, which are known to strongly fractionate oxygen isotopes and can therefore adversely affect the isotopic signal of the autogenic calcite lake sediments (Candy et al., 2016). The <63 µm fraction, largely autogenic calcite precipitated from the water column, was retained and treated with hydrogen peroxide (H 2 O 2 ) to remove other organic material. Samples were then air-dried and weighed using a Mettler Toledo XP6 microbalance. Samples were digested in phosphoric acid at a temperature of 90°C to produce CO 2 from which the δ 18 O and δ 13 C were then analysed using a ThermoFisher Delta Plus XP mass spectrometer with a Gasbench II preparation system. Three internal standards (BDH) were analysed every five samples and three external standards  were analysed in each machine run. Internal precision produces analytical uncertainties of ±0.07 (δ 18 O) and ±0.04 (δ 13 C).

Sedimentology and chronology
The CRUM1 palaeolake sequence consists of an alternation between two marl units and two minerogenic units (Timms et al., 2018;Fig. 2a). Here we focus on units CM-L1 to -L3 ( Fig. 2b; Table 1). The lowermost unit (CM-L3) is composed of silts and clays (CaCO 3~1 0%) and is succeeded by a marl unit (CM-L2), which is subdivided into eight subunits ( Fig. 2b; Table 1). The CaCO 3 content increases in CM-L2a from~20% to~70% in CM-L2b and remains at this level for the remainder of CM-L2. However, subunits within CM-L2 show falls in CaCO 3 from 67% to 8% (632 cm; CM-L2c) and smaller decreases in CaCO 3 to 60% and 65% in CM-L2e (~620 cm) and CM-L2g (~610 cm), respectively. These decreases in percentage CaCO 3 are normally accompanied by an increase in minerogenic material, rather than an increase in organic content (Fig. 2b). Leaves of Dryas octopetala, Empetrum and Betula nana, along with fruits of Betula nana, were found infrequently throughout CM-L2d to CM-L2h.
Unit 3 is predominantly silty clay with little or no CaCO 3 , but higher organic content is present in CM-L3a (~10%) before this decrease that occurs within CM-L3b.
The silty clay deposition at the base of the sequence shows high levels of mineral in-wash to the basin from the lake catchment. This often indicates relatively low biogenic productivity and enhanced erosion of soils into streams during high discharge events (Palmer et al., 2015;Candy et al., 2016). The change to a system dominated by marl sedimentation is indicative of higher biological activity within the lake and its catchment that allows precipitation of calcite crystals within the lake. Calcite production may also be enhanced by a decrease in the mineral sediment load transported to the basin due to increased plant growth in the catchment stabilising the soil (Palmer et al., 2015;Candy et al., 2016). The transition between mineral-rich sedimentation and marl sedimentation may, therefore, relate to a climatically induced change impacting the lake sedimentation. The lower silty clays are thought to reflect deposition during the colder climatic conditions of the Dimlington Stadial (DS; CM-L1) and the marl sedimentation to reflect warmer climatic conditions of the WI (CM-L2), before the return to colder climatic conditions of the LLS represented by CM-L3. More subtle lithological variability is also recorded within CM-L2 which will be explored further in this paper. This interpretation is supported by the independent age model for CRUM1 (Timms et al., 2018). The marl unit (CM-L2) represents the WI due to the presence of the Borrobol Tephra at 638 cm (14 107 ± 58 cal a BP) and the Penifiler Tephra at 632 cm (14 024 ± 77 cal a BP). The presence of the Vedde Ash at 587 cm (11 997 ± 39 cal a BP) in the following minerogenic unit (CM-L3) means these sediments were deposited during the LLS. The additional radiocarbon sample yielded an uncalibrated age of 11 626 ± 53 14 C a BP, and a calibrated modelled age of 13 491 ± 67 cal a BP, providing further confirmation of the WI age of CM-L2. The radiocarbon date has reduced the age error estimate for the mid-WI by more than 50%.

Chironomids Chironomid assemblages
In total, 51 different chironomid taxa were identified in the 63 samples analysed, with assemblages displaying considerable down-core variation (Fig. 3). Head capsule concentrations are generally high throughout the sequence, except in CM-C1 and CM-C6, where there are less than 30 HCs/g. The majority of taxa present at Crudale Meadow are lacustrine in nature and as most chironomid taxa are indicative of intermediate water depths we suggest that a moderately deep lake (approximately 5-7 m deep) was present at the site throughout the LGIT (Engels and Cwynar, 2011).
The onset of CM-C2 (646.5-633.75 cm; 14 180 ± 150 to 14 050 ± 80 cal a BP) shows a decline in ultra-cold and coldindicating taxa, to nearly 0%, while cool-temperate and temperate taxa increase in abundance to over 95%. The chironomid fauna of CM-C2 are dominated by taxa which indicate well-oxygenated mesotrophic conditions, e.g. Psectrocladius sordidellus-type (Moller Pillot, 2013), Procladius (Brodin and Gransberg, 1993) and Tanytarsus pallidicornis-type (Brodin, 1986). There is a large increase in the range of taxa present from CM-C1 to CM-C2, which typically occurs under warmer conditions (Engels et al., 2020). The presence of Dicrotendipes nervosus-type and Ablabesmyia suggests the development of aquatic macrophytes in the lake (Brodersen et al., 2001;Vallenduuk and Moller Pillot, 2007;Moller Pillot, 2009).  The abrupt increase in Microtendipes pedellus-type to 4% at the onset of CM-C2 suggests relatively warm mesotrophic conditions were established very early on in the WI (Brooks et al., 2012). A centennial-scale cooling event occurs during CM-C3 (633.75-629.75 cm; 14 050 ± 80 to 13 960 ± 160 cal a BP) where assemblages revert to a fauna dominated by coldstenothermic taxa typical of oligotrophic lakes. Temperate taxa associated with mesotrophic conditions and macrophytes virtually all disappear from the assemblage (e.g. D. nervosustype and Ablabesmyia) during CM-C3, while only a few cool-temperate taxa persist in very low numbers (e.g. P. sordidellus-type). The presence of Paratanytarsus suggests aquatic macrophytes may still be present in the lake (Brooks et al., 2007). Paracladius indicates the input of organic detritus which it can exclusively feed on (Moller Pillot, 2013).
Warm conditions continue into CM-C5 (622.75-602.5 cm; 13 730 ± 210 to 12 830 ± 520 cal a BP). However, the notable increase in S. coracina-type, a cold-stenothermic taxon (Brodin, 1986), to an average of 15% suggests marginally cooler conditions during the latter part of the WI. The low abundance but persistent presence of Pseudosmittia at~3% in CM-C5 may indicate enhanced stream input and/or erosion of the landscape as the species within this genus are typically terrestrial or semiterrestrial (Cranston, 1982;Massaferro and Brooks, 2002).
A notable change in taxa occurs at the onset of CM-C6 (602.5-595 cm; 12 830 ± 520 to 12 430 ± 470 cal a BP). A return to very cold oligotrophic conditions is indicated by the dominance of cold stenotherms, e.g. Micropsectra insignilobus-type (Brundin, 1956), Hydrobaenus lugubris-type (Cranston et al., 1983) and Protanypus (Moller Pillot, 2013) contributing up to 25%, 13% and 10%, respectively. Conditions may have been ultra-oligotrophic as inferred from the dominance of M. radialis-type (Brodin, 1986), which forms up to 52% of the assemblages. Another discrete peak of 58% in C. ambigua occurs at the start of this zone indicating another period of instability (Brodersen and Lindegaard, 1999).

Chironomid-inferred July temperature reconstruction
The chironomid-inferred mean July air temperatures record (T Jul ) (Fig. 4) shows a rapid increase from 7.5 ± 1.5°C in CM-C1 to 11.5 ± 1.4°C early in CM-C2. T Jul continues to increase  throughout CM-C2, albeit more slowly, reaching a maximum of 12.8 ± 1.4°C. A strong decline of 5.4°C occurs during CM-C3, with T Jul decreasing to 7 ± 1.4°C, but recovers to 12.7 ± 1.4°C in CM-C4. During CM-C5, two smaller decreases in T Jul occur, decreasing to 7.3 ± 1.4°C at 620.5 cm and 10 ± 1.4°C at 609 cm, a decline of 3.9°C and 1°C, respectively. Between CM-C5 and CM-C6 there is a large decline of 5°C with T Jul falling to 6.1 ± 1.5°C. Sample specific errors range between 1.41°C and 1.59°C. All samples of WI age have a good fit-to-temperature suggesting T Jul is the major driver in chironomid assemblage change (Fig. 4b). It is noted that samples in LLS (CM-C6) and DS (CM-C1) generally have a 'very poor' fit-to-temperature. However, the discussion of the results for the stadial samples is beyond the aim and scope of this paper. The fossil assemblages also generally compare well to those in the modern calibration dataset, with 55 samples having good modern analogues (Fig. 4c). Although Seven samples had 'poor' modern analogues and one 'very poor' analogues in the modern calibration dataset, WA-PLS can still perform very well in nonanalogous situations (Birks, 1998). The vast majority of taxa that occur during the LGIT at Crudale Meadow are well represented in the modern calibration dataset (Fig. 4d and e). Taxa that are rare in the modern dataset (i.e. Hills N2 < 5) remain very low throughout most of the WI, reaching a maximum of 2.2%, and are therefore unlikely to adversely influence the reconstruction. The only taxon not present in the modern calibration dataset is Propsilocerus lacustris-type, but this taxon does not exceed 4.1% in the five samples in which it occurs during CM-C3 (Fig. 4e).

Stable isotopes
For much of the WI the δ 18 O Carbonate values remain relatively stable throughout, oscillating around an average of −4.6% (1σ = 0.3%) (Fig. 5). The exceptions to this are: (1) a number of short-lived falls in δ 18 O Carbonate values (CM-O3, CM-O4b and 4d and CM-06); and (2) uniquely high δ 18 O Carbonate values of −2% in CM-O1, 2.5% higher than any other values in the dataset. The greatest decline in isotope values occurs during CM-O3, where δ 18 O Carbonate decreases from −4.6% to −5.6% at 632 cm, an isotopic shift to more negative values by 1%. Smaller isotopic declines are present in the remainder of the WI with a 0.4% shift to −5% during CM-O4b, a 0.7% shift to −5.3% in CM-O4d and a 0.5% shift to −5.1% in CM-O6. No carbonate was present in CM-L3 and so no bulk isotope data are available for this section of the sequence. The δ 18 O signal of lacustrine carbonate (δ 18 O Carbonate ) sequences in the Lateglacial of northwest Europe is frequently interpreted as a palaeotemperature proxy (e.g. Diefendorf et al., 2006;Van Asch et al., 2012). The rationale for this is that δ 18 O Carbonate is controlled by the δ 18 O of the lake water (δ 18 O lake water ) and the temperature at which the carbonate is precipitated in the water column (Talbot 1990;Leng and Marshal, 2004). In turn, δ 18 O lake water is largely a function of the δ 18 O of rainfall (δ 18 O precipitation ) which is primarily controlled by the prevailing air temperature with other factors also having a small influence (Rozanski et al., 1992(Rozanski et al., , 1993. Such other factors include the amount, effect, water vapour source, seasonality of rainfall, or the distance of air masses from moisture source (Rozanski et al., 1992(Rozanski et al., , 1993. However, it is often assumed that none of these other factors exerts as strong an influence as air temperature on the δ 18 O Carbonate of lacustrine sequence in the Lateglacial in Britain. This can be seen in many Lateglacial sequences, such as that of Hawes Water (Marshall et al., 2002) and Fiddaun (Van Asch et al., 2012) which shows high δ 18 O Carbonate values in the warm early Holocene and WI but low values in the cold LLS.
The δ 13 C Carbonate record from Crudale displays a rapid rise from −0.2% in CM-O1 to 2.7% in CM-O2. During CM-O3 δ 13 C Carbonate briefly declines to -0.9% but recovers to 2.6% by the end of the zone. Throughout CM-O4, −O5 and −O6, δ 13 C Carbonate steadily declines from 2.6% to -0.8%. The δ 13 C of lacustrine carbonates (δ 13 C Carbonate ) is largely controlled by the dissolved inorganic carbon (DIC) of lake waters. DIC is influenced primarily by the carbon isotopic composition of the ground and surface waters recharging the lake, biological activity in the lake basin and the equilibration of CO 2 between the lake water and atmosphere (Leng and Marshall, 2004;Candy et al., 2016). Therefore, δ 13 C Carbonate records a much more local signal than δ 18 O. Many δ 13 C Carbonate records from British carbonate sequences exhibit a gradual decline throughout the WI (e.g. Diefendorf  As with the studies outlined above, we also assume that palaeotemperature is the primary control on the δ 18 O Carbonate signal of the CRUM1 sequence. There is negligible evidence for other factors that may control the δ 18 O Carbonate signal; in particular, evaporation and detrital contamination, which are discussed further in the following paragraphs. Lakes that are strongly influenced by evaporation may not preserve temperature trends clearly within the δ 18 O Carbonate record. This is because the δ 18 O water may be increased due to evaporation prior to carbonate precipitation, disrupting the relationship between air temperature and δ 18 O water (Leng and Marshall, 2004). In such systems there is often a strong relationship between δ 18 O Carbonate and δ 13 C Carbonate as the effect of increasing evaporation is to cause both of these to increase (Talbot, 1990). There is no covariance between δ 18 O Carbonate and δ 13 C Carbonate in the CRUM1 sequence, suggesting that evaporation did not exert a strong influence on the isotopic record at this site (Fig. 6). Furthermore, the δ 13 C carbonate of lake environments is typically between −3 and +3% in an open system (Talbot, 1990) and the values of −0.9 to +2.7% observed for our Crudale Meadow record fall within this range.
In sequences that are dominated by authigenic carbonate, such as CM-L2 with~65% carbonate, the effect of detrital contamination is likely to be minimal because any in-washed fragments of bedrock would have been so heavily diluted by authigenic calcite precipitation that it would not impact the overall δ 18 O Carbonate value. However, in parts of a sequence where authigenic carbonate is low, i.e. CM-O1 and CM-O3, the presence of in-washed limestone particles may strongly  Whittington et al. (2015) in terms of the number and structure of events, although there is less variance in our δ 18 O data. This is potentially due to the differing sample preparation methods used between this study and Whittington et al. (2015), where the latter dry-sieved samples at 100 µm to remove detrital fragments. Despite similarities between the two isotope records, the lack of independent chronological control within the Whittington et al. (2015) data means it is difficult to make robust comparisons between the two datasets.

Synthesis: temperature reconstructions at Crudale Meadow
The Crudale Meadow chironomid-inferred temperature record (T Jul ) and oxygen isotope record (δ 18 O) provide independent indicators of summer and annual temperature variability, respectively, across the WI. Overall, there is good agreement between both proxies, broadly showing three centennial-scale cooling events separated by four warm intervals (Fig. 7). While chironomid-based climate reconstructions provide quantified July temperature estimates, quantified temperature values cannot be produced from the δ 18 O record. This is because temperature is only one of the factors that controls this proxy.
It should be noted that there are short-lived declines in δ 18 O Carbonate values that have either no or negligible expression in the T Jul record, i.e. CM-O4b. This does not mean that these are not climatic events but could simply reflect that, at times, the δ 18 O Carbonate and the T Jul are recording different parts of the climate system. Although lake carbonates precipitate primarily in the late spring/early summer, they are precipitating from lake waters that frequently reflect the δ 18 O of groundwater which is in turn composed of homogenised rainfall from across the year (Blockley et al., 2018). Also, small-medium sized open lake systems with shifts in δ 18 O of a few permillage points, such as the Crudale Meadow palaeolake, often have residence time of ≥1 yr (Leng and Marshall, 2004). This means that the δ 18 O Carbonate value reflects an approximate average of the δ 18 O value of annual rainfall and, therefore, an approximate average of annual air temperatures. The chironomid-based reconstruction, as outlined above, records July temperatures. Consequently, if an ACE occurs and winter temperatures decline but summer temperatures are relatively unaffected, the ACE may have an expression in the δ 18 O Carbonate values but not in T Jul (Blockley et al., 2018). It is, therefore, likely that some of the climatic events seen in the CRUM1 record affect winter temperatures more strongly than summer temperatures. Here we examine the magnitude and timing of the observed temperature reconstructions at Crudale Meadow. To help synthesis the proxy data, a number of climate zones have been formed and are denoted as CM-CZn.  The lowermost zone, CM-CZ1, is characterised by a very cold climate with T Jul~8°C , dominated by minerogenic deposits and dated to the DS (Fig. 8). The δ 18 O values of approximately −2% are outliers in this record caused by detrital contamination from geological carbonate. Crudale Meadow represents only the second chironomid-based temperature reconstruction to register the pre-WI temperature conditions in Scotland, with Whitrig Bog in southern Scotland being the other (Brooks and Birks, 2001). Whitrig Bog displays a temperature of~6°C during the DS, broadly consistent with the reconstruction from Crudale Meadow given the estimated errors and differing training sets used.
T Jul shows a strong rise into the start of the WI (CM-CZ2), reaching 11.5 ± 1.4°C at 14 170 ± 140 cal a BP. Inferred temperatures then continue to rise to 12.8 ± 1.4°C at 14 070 ± 70 cal a BP. T Jul rose faster than CaCO 3 , most likely due to a lag in landscape stability leading to inhibited authigenic carbonate precipitation.
Early Windermere Interstadial ACE (CM-CZ3)~14 070 ± 70 to 13 960 ± 160 cal a BP The onset of the Early WI event around 14.0 ka BP does not appear to be synchronous between both temperature proxies. Both T Jul and δ 18 O start to decline at 14 070 ± 70 cal a BP. However, T Jul only shows a very minor decline from the previous sample of 0.4°C, well within the errors of the WA-PLS model, whereas δ 18 O displays a much more substantial decline. T Jul only starts to show a stronger decline in the following centimetre which has an age of 14 050 ± 80 cal a BP. As both proxies were sampled from the same core and same stratigraphic level, age-model errors cannot explain the difference between the two proxies. Therefore, approximately 20 years had elapsed between the decrease in δ 18 O and the decline in chironomid-inferred T Jul . This pattern is supported by the bulk lithological data which also show a decline in percentage CaCO 3 prior to a change in T Jul . Palaeoecological proxies may display a lag in response to changes in environmental variables due to migration delay as well as physiological and metabolic processes of organisms buffering against environmental change (Słowiński et al., 2017). This possible delay in chironomid response may have only been seen due to the application of high-resolution sampling (20 years/sample) at Crudale Meadow.
It is potentially possible that δ 18 O does not record the full magnitude of temperature change. The isotopic value at 633 cm of -4.15% appears anomalously high and its position on the downward temperature trend may mean that the sample was affected by detrital contamination as vegetation broke down and landscape erosion increased. Therefore, it could be that minimum δ 18 O values were attained at 633 cm, the same time as T Jul , but were obscured due to detrital contamination. This would have consequences for the timing of when minimum temperatures were reached in both proxies. However, this cannot be explored further with the available data.
Later part of Windermere Interstadial and other ACEs (CM-CZ4 to CM-CZ7)~13 960 ± 160 to 12 830 ± 500 cal a BP During CM-CZ4, there was a decline in δ 18 O with a magnitude of around 0.4%. However, there was no corresponding decline in T Jul . Several factors may explain this. First, summer temperatures declined, but there was no response in the chironomids, either because the change in summer temperatures was not large enough to impact on chironomid ecology and/or ecological thresholds were not crossed. Second, winter temperatures may have declined, but no corresponding decline occurred in T Jul , i.e. there was an increase in seasonality. Finally, factors other than temperature may have influenced the δ 18 O signal, such as a change in moisture source (Rozanski et al., 1992(Rozanski et al., , 1993 or a shift in the seasonality of precipitation . Additional measurements would be needed for such factors to be fully understood, e.g. n-alkanes or δD.
In CM-CZ5, a mid-WI ACE is detected with both proxies showing a decline in inferred temperatures. Minimum temperatures were attained synchronously at 13 670 ± 200 cal a BP, taking into account the sampling resolution employed. A late WI ACE occurs during CM-CZ7 with T Jul and δ 18 O showing a decline of 1°C and 0.5%, respectively. As with the mid-WI ACE, the timing of event onset between both proxies is synchronous with respect to the sampling resolution, both showing declines at 13 330 ± 320 cal a BP. Minimum values are, however, attained first in the δ 18 O record at 13 220 ± 400 cal a BP and later at 13 170 ± 420 cal a BP in the T Jul record.
End of the Windermere Interstadial and onset of Loch Lomond Stadial (CM-CZ8 and -CZ9)~12 990 ± 500 cal a BP Temperatures increase briefly in CM-CZ8 before declining into CM-CZ9, with T Jul falling to 6 ± 1.5°C. The lack of CaCO 3 precipitation also suggests a very cold climate. Fragments of terrestrial moss and herbaceous material are found at 602 cm/ 12 800 ± 500 cal a BP, indicative of landscape erosion and in-wash of material.
In summary, the T Jul and δ 18 O records from the CRUM1 sequence indicates that multiple ACEs occurred during the WI in the northern British Isles. The clearest and most extreme of these is seen in CM-Z3, centred around~14.0 ka BP, which has the strongest expression of any ACE in both proxies and is characterised by approximately 5.4°C worth of cooling and 1% of isotopic decline. While the middle and late part of the WI are clearly characterised by other abrupt events, centred around~13.6 ka BP and~13.2 ka BP, respectively, neither of these are as pronounced as~14.0 ka BP ACE.

Comparison with other records
The trends observed in the T Jul and δ 18 O records from Crudale Meadow are similar to those seen in other records around the North Atlantic region, with three centennial-scale ACEs punctuating the WI (e.g. Diefendorf et al., 2006;Lang et al., 2010;Brooks et al., 2012) (Fig. 8). Many T Jul and δ 18 O records, including Crudale Meadow, have particularly good agreement with the pattern of temperature changes seen in the Greenland ice-core records (Lowe et al., 2008;Rasmussen et al., 2014). It is, therefore, tempting to correlate these cold oscillations observed in terrestrial records to the GI-1d, GI-1cii and GI-1b events; here we examine the validity of this notion. Brooks and Langdon (2014) have previously synthesised T Jul records from northwest Europe using time slices based on the GI-1 subdivisions with averaged reconstructed temperatures for each period. Here we build on their work by including more recent T Jul reconstructions from Muir Park Reservoir (Brooks et al., 2016) and Tirinie (Abrook et al., 2020), and expanding the analysis to include δ 18 O information. We restrict our analysis to reconstructions from the British Isles as these tend to provide high sample density reconstructions and robust chronological control. Particular attention is paid to the ACE centred on~14.0 ka BP, where radiocarbon and tephra data provide a robust link between the records.
Early WI ACE~14.0 ka BP (≈GI-1d) The presence of the Borrobol and Penifiler Tephras at Crudale Meadow means that the large cooling event during CM-C3, the largest of the WI ACEs, with a cooling of 5.4°C, can be constrained to~14.0 ka BP with a high degree of certainty. This broadly correlates with the GI-1d cooling event expressed in Greenland (14 075 ± 169 to 13 954 ± 165 b2k; Rasmussen et al. 2014).
The Borrobol and Penifiler Tephras are also present in many other Scottish records and allow robust correlation of this event between sites. The majority of Scottish C-IT records also show this early WI ACE as being the largest ACE in terms of magnitude to punctuate the WI. However, there are large spatial differences in the magnitude of T Jul for the~14.0 ka BP event and a strong correlation with latitude is observed (Fig. 8a). Sites further north experience a far greater magnitude of cooling, e.g. Crudale Meadow and Loch Ashik (4.4°C, 57°N, Brooks et al., 2012), than those sites further south, e.g. Muir Park Reservoir (3.5°C, 56°N, Brooks et al., 2016). Sites in England (Hawes Water; 53°N; Lang et al., 2010) and Ireland (Fiddaun; 54°N; Van Asch et al., 2012) have less chronological certainty during the early WI as they do not contain the Borrobol and Penifiler Tephra. However, there is still evidence of an early WI ACE, which, based on the site chronologies, is also thought to occur at~14.0 ka BP. Fiddaun has two radiocarbon dates constraining an event at~14.0 ka BP with a magnitude of 1°C. Hawes Water relies on climatostratigraphy for dating events during the WI; however, an event early on in the WI has a magnitude of 2°C and is the largest event present in the WI.
It is difficult to evaluate whether similar latitudinal trends were present in δ 18 O records at~14.0 ka BP due to the paucity of δ 18 O records arranged along a north-south transect with sufficient sampling resolution, methodological controls, and chronological constraints. However, three records which lie between 54 and 52°N -Loch Inchiquin (Diefendorf et al., 2006), Hawes Water (Marshall et al., 2002) and Fiddaun (Van Asch et al., 2012)all show δ 18 O shifts of -1.3%, -1.5% and -1.4%, respectively, for the event broadly constrained tõ 14.0 ka BP. Further north at Tirinie, which lies at 56°N, the δ 18 O record displays a much larger isotopic shift of~2.32% (Candy et al., 2016). The isotopic shift at Crudale Meadow of 1% is much smaller than that seen at the sites mentioned above and may be due to: (1) the westerly position of the site causing reduced seasonality; (2) the 'actual' minimum isotopic value being obscured by detrital contamination with the potential of in-washed material at 633 cm; or (3) the most northerly latitudes of the British Isles may have been affected by changes in moisture source and/or changes in winter temperatures may have been more subdued.
Additional spatial trends in the data were explored (altitude and longitude) but were found to not have significant correlations with the magnitude of the events identified. Therefore, for the~14.0 ka BP event at least, a forcing mechanism must be acting that exerts a stronger influence from south to north than it does from west to east. To further our understanding of the spatial expression in the magnitude of temperature changes for the~14.0 ka BP event, a greater density of records is needed over longer north-south and east-west transects.

Other events in the Windermere Interstadial
The Crudale Meadow record, as well as other T Jul and δ 18 O records from Scotland and northern England, display several other ACEs in the middle and later parts of the WI. The comparison of these ACEs between sites is problematic for two reasons. Firstly, the lack of tephra marker layers and radiocarbon dates means definitive correlation of ACEs is not possible. Age models have error estimates of up to ±400 years which are as large or larger than the duration of the known ACEs in this period (Rasmussen et al., 2014). Our chronological precision does not permit detailed comparisons of the timing of onset, duration and recovery from these later events, while in some cases the magnitudes of change are in the order of 1-3°C, which is less than the cooling experienced at 14.0 ka BP and similar to the predictive powers of the WA-PLS model. Despite these limitations, we suggest that broad-scale comparisons between records provide useful information on mid-to late WI variability.
A mid-WI event (often correlated to GI-1cii; 13 660-13 600 GICC05 yrs) is recorded in many records as the smallest of the WI ACEs (Brooks and Birks, 2000;Lang et al., 2010;Brooks et al., 2012). However, at Crudale Meadow, the mid-WI event is the second largest ACE with a magnitude similar to that seen in the~14.0 ka BP event with T Jul displaying a decline of 3.9°C and δ 18 O decreasing by 0.7%. Currently, the large magnitude of this ACE in the Crudale Meadow sequence is difficult to reconcile and requires further testing at other locations to determine whether this is a local phenomenon specific to Crudale, the Orkney Islands, or whether it may be observed more widely. It may be that this event is very short-lived which has meant that it has been neglected due to the lower sampling resolution employed in previous chironomid studies missing this change. Alternatively, it may reflect a lower stratigraphic resolution in previous studies which has led to a mixing of adjacent warm and cold taxa, seemingly reducing its magnitude. A later WI event (often correlated with GI-1b; 13 311-13 099 GICC05 yrs b2k) is infrequently recognised in isotope and T Jul records across Britain with its best expression and dating seen at Tirinie (Candy et al. 2016;Abrook et al. 2020), where an event dated to between 13.45 and 13.02 cal a BP is defined by a decline of 1.8°C T Jul and an isotopic depletion of -2.9%. Other sites around Scotland, e.g. Whitrig Bog (Brooks and Birks, 2000), Loch Ashik and Abernethy Forest (Brooks et al., 2012) and Muir Park Reservoir (Brooks et al., 2016), record a late WI drop in T Jul temps ranging from 2 to 3.5°C and if these are registering the same event it would indicate that Crudale does not support this degree of cooling. This event appears to have a duration of~450 years (at Tirinie) so it seems unlikely that low sampling resolution is a factor in its identification or characterisation. Therefore, where this late WI event is poorly represented or absent, it would suggest that the individual site records have suffered from either: (1) low stratigraphic resolution; (2) an alteration to the proxy signal from sediment mixing or reworking; and/or (3) overriding influence of site-specific factors.

Forcing mechanisms
At present, this study can only present the pattern of variability seen in ACEs in WI/GI-1; it is not possible to confidently ascribe a single forcing mechanism to them. However, the fact that such a large latitudinal gradient in response to the 14.0 ka BP/GI-1d oscillation existed and indicates that whatever drives this event has a greater influence at higher latitudes. As such, it is likely that ocean forcing plays a key role in triggering the cooling (McManus et al., 2004). Changes in the strength of Atlantic Meridional Overturning Circulation (AMOC) are likely to be more strongly expressed in higher latitudes as these regions are more dependent upon the heat delivered by this circulation to maintain temperatures (e.g. Zhang, 2010). If a decline in AMOC strength was the driver of this event then it would, therefore, be reasonable to expect the northern parts of the British Isles to be most affected. While it is beyond the scope of this study to prove such mechanisms, the construction of transects of sites that span key latitudinal belts, as shown above, is a key means of generating empirical data against which model results can be validated.

Conclusion
The new high-resolution T Jul and δ 18 O records covering the WI from Crudale Meadow provide important insights into temperature changes for northern Britain. The data generated in this study allow the following conclusions to be drawn: (1) An ACE centred on~14.0 ka BP is clearly present in the CRUM1 sequence with a magnitude in the order of 5.4°C. This event is bracketed by the Borrobol and Penifiler Tephras which suggests that it is broadly correlative to the GI-1d oscillation observed within the Greenland ice-core records.
(2) Although in a spatially restricted area of northwestern Europe, the magnitude of the temperature change of the 14.0 ka BP event across the northern British Isles suggests that latitude plays an important role. The correlation between summer temperature and latitude that existed at the peak of this event could potentially occur in future ACEs caused by the melting of the Greenland ice sheet due to anthropogenic global warming. (3) Two later ACEs centred around~13.6 ka BP and~13.2 ka BP, of smaller magnitude than the~14.0 ka BP oscillation, are present at Crudale Meadow. These seem to broadly equate with GI-1cii and GI-1b seen in Greenland. However, the chronological uncertainty across these parts of the sequence means these attributions are not certain. Additional chronological control is required to fully explore these events at Crudale and other locations in northwestern Europe. (4) Crudale Meadow is the second record from Scotland to quantifiably record the onset of warming during the WI. It, and Whitrig Bog in southern Scotland, demonstrate ice-free conditions at their respective locations prior to WI warming; something not seen in other locations. Both sites have considerable potential to constrain climatic events over the WI and demonstrate the climatic differences recorded between two geographically near locations (Fig. 9).

Supporting information
Additional supporting information may be found in the online version of this article at the publisher's web-site. Supporting information 1chironomid sample preparation procedure tests Supporting information 2chironomid-inferred temperature reconstruction and validation statistics using the Norwegian calibration dataset