Extratropical cyclones over the North Atlantic and Western Europe during the Last Glacial Maximum and implications for proxy interpretation

Extratropical cyclones are a dominant feature of the mid-latitudes, as their passage is associated with strong winds, precipitation, and temperature changes. The statistics and characteristics of extratropical cyclones over the North Atlantic region exhibit some fundamental differences between Pre-Industrial (PI) and Last Glacial Maximum (LGM) climate conditions. Here, the statistics are analysed based on results of a tracking algorithm applied to global PI and LGM climate 10 simulations. During the LGM, both the number and the intensity of detected cyclones was higher compared to PI. In particular, increased cyclone track activity is detected close to the Laurentide ice sheet and over central Europe. To determine changes in cyclone characteristics, the top 30 extreme storm events for PI and LGM have been simulated with a regional climate model and high resolution (12.5 km grid spacing) over the eastern North Atlantic and Western Europe. Results show that LGM extreme cyclones were characterised by weaker precipitation, enhanced frontal temperature gradients, and stronger wind 15 speeds than PI analogues. These results are in line with the view of a colder and drier Europe, characterised by little vegetation and affected by frequent dust storms, leading to reallocation and build-up of thick loess deposits in Europe.

differences between PI and LGM conditions (cf. Ludwig et al., 2016;their Figure 2), which also impacts for example the storm track and precipitation (their Figures 4, 6), its main characteristics are generally close to the ensemble average.
Individual extratropical cyclones over the North Atlantic and Europe are identified and tracked based on six hourly mean sea 95 level pressure data with a widely used automatic tracking algorithm (Murray and Simmonds, 1991;Pinto et al. 2005). The resulting cyclone statistics provide information on the lifetime of each identified cyclone, thus enabling the computation of mean cyclone statistics like track density, mean maximum intensity, cyclogenesis, cyclolysis, propagation speed and deepening rates. Cyclone statistics obtained with this method compare well with other methodologies (e.g., Neu et al., 2013;Hewson and Neu, 2015). Following Pinto et al. (2009), cyclones are selected based on the following conditions: (a) cyclone lifetime of at 100 least 24 h hours, (b) a minimum core MSLP value below 1000 hPa (3) a maximum vorticity (approximated by the Laplacian of MSLP) value above 0.6 hPa deg. lat. −2 , and (d) a maximum deepening rate of 0.3 hPa deg. lat. −2 s -1 is achieved at least once during their lifetime. The method is here applied to the MPI-ESM-P data for the extended (ONDJFM) winter season. In order to analyse the characteristics of the most extreme cyclones affecting Europe in more detail, the most intense 30 cyclones (TOP 30) for the PI and LGM periods are selected based on their peak intensity in terms of vorticity and their passage within a pre-105 defined box over the eastern North Atlantic (Figure 1, dashed box). The selection of the box enables the creation of a cyclone ensemble that impacts Western Europe and permits a comparison with terrestrial proxy data e.g. for precipitation and dust.
The Weather Research and Forecast (WRF) model (Skamarok et al., 2008) is used in its version 3.9.1.1 to simulate theses cyclones with a final grid spacing of 12.5 km (including 35 vertical layers up to 30 hPa). To achieve a grid spacing of 12.5 km, a 2-step nesting approach is necessary. Cyclones were initially simulated on a 50 km grid forced by MPI-ESM-P data as 110 initial and boundary conditions with an update frequency of six hours. The final 12.5 km grid spacing is achieved by a second nesting step within the WRF model. An overview of the used physical parametrisations is given in Table 2. For the calculation of wind gusts, a gust parametrisation based on 10 m wind speed and friction velocity (Schulz and Heise, 2003;Schulz, 2008) has been implemented into the WRF model. This gust parametrization shows an overall good agreement with observed wind gusts, particularly over flat terrain (Born et al. 2012). For the WRF simulations, global PI and LGM boundary conditions were 115 adapted considering specifications of the PMIP3 protocol (Braconnot et al., 2012, see also Ludwig et al. 2017). These changes encompass orbital parameters, trace gases (see Table 1), the consideration of ice sheets (extent and height), an associated lowering of the sea level and adaption of land use cover (CLIMAP Project Members, 1984).
WRF cyclone tracks and intensities were identified manually based on relative vorticity at 850 hPa. For comparison of the PI and LGM cyclone characteristics, and following the methodology from Catto et al. (2010;their Figure 3), each track was 120 rotated so that the cyclones were equally moving in west-east direction, enabling the generation of composites for different atmospheric variables (cf. also Dacre et al., 2012). Composites have been created for peak intensity (0) and 6, 12,18 and 24 hours before peak intensity, and 6 and 12 hours afterwards. For the sake of succinctness, we will primarily discuss the time frames i) 12 hours before peak intensity and ii) peak intensity. The here analysed target variables, based on the 12.5 km WRF https://doi.org/10.5194/cp-2019-139 Preprint. Discussion started: 2 December 2019 c Author(s) 2019. CC BY 4.0 License. simulations, include mean sea level pressure, precipitation, column integrated water vapour, equivalent-potential temperature 125 850 hPa, 925 hPa winds, near-surface wind gusts.

Northern Hemisphere Cyclone Statistics for PI and LGM conditions
In this section, we analyse the general characteristics of cyclones over the North Atlantic and Europe under LGM conditions and compare them to PI climate conditions. Figure 2 shows the cyclone track density for the extended winter for PI and LGM climate conditions. In spite of the lower spatial resolution of MPI-ESM-P, the cyclone track density for the PI is close to 130 cyclone statistics obtained with Reanalysis datasets and CMIP GCMs for recent climate conditions (cp. Pinto et al., 2007;their Figure 1). Still, some regional shortcomings are identified, notably the limited cyclone activity over the Mediterranean basin.
The North Atlantic storm track shows a clear tilt towards Northern Europe and the Arctic Ocean for PI, and its location and orientation are closely related with the eddy-driven jet stream (black contours in Fig. 2a) and the associated upper-air baroclinicity (Hoskins and Valdes, 1990;Pinto et al., 2009). 135 The North Atlantic storm track looks quite different under LGM conditions: the cyclone track density is strongly enhanced over the North Atlantic and more constraint along a corridor close to the ice edge (Fig. 2b). Close to Europe, a bifurcation is found, and cyclones are either deflected northward along the border of the Scandinavian ice sheet or south-eastward towards Central Europe and the Mediterranean (Fig. 2c). In accordance, the eddy-driven jet is stronger under LGM conditions in the MPI-ESM-P, thus establishing more favourable conditions for the occurrence of intense storms affecting Western and Central 140 Europe. For the North Atlantic (70°W -0°, 35°N -70°N), the total number of cyclones for the analysed 30-year period is about 26% larger for LGM than for PI conditions (12.071 vs. 9.541 individual cyclone counts).
Other properties of cyclone activity are depicted in Figure 3 for LGM in terms of absolute values (left column) and their differences to PI (right column). Cyclogenesis is dominant along the North American east coast for LGM (Fig. 3a), and much stronger than in PI (Fig. 3b), which is in line with a much stronger upper level jet stream during glacial conditions. Moreover, 145 cyclogenesis is enhanced south of Greenland and over Western and Central Europe. On the other hand, cyclolysis is enhanced along the borders of the Greenland and Scandinavian ice sheet (Fig. 3c), in strong deviation to the PI conditions (Fig. 3d).
Mean maximum cyclone intensity is typically attained in a region extending from Newfoundland to Iceland and the British Isles, with a secondary maximum over Eastern Europe (Fig. 3e). Compared to the PI cyclones, the LGM cyclones reveal stronger intensities, particularly in an area extending from the south of Greenland to the British Isles, and over most of 150 continental Europe. On the other hand, cyclone intensity close to the North American East coast is considerably lower (Fig.   3f). Deepening rates are particularly stronger for LGM cyclones over the central north Atlantic, and filling rates close to the ice edge / ice shields (Fig. 3g). This is in strong contrast with PI cyclones (Fig. 3h), which comparatively have stronger deepening rates further upstream, and weaker deepening and filling rates. These results point towards different typical https://doi.org/10.5194/cp-2019-139 Preprint. Discussion started: 2 December 2019 c Author(s) 2019. CC BY 4.0 License. development of LGM cyclones compared to their PI counterparts, which occurs either more zonally at lower latitudes towards 155 Central Europe or further downstream closer to the ice edge towards the Arctic. Figure 4 displays the relative frequency distribution of the cyclone intensity over the North Atlantic area (70°W -0°, 35°N -70°N), revealing that LGM cyclones are on average more intense than their PI counterparts. In particular, the number of cyclones exceeding 3 hPa deg. lat. −2 s -1 is twice as large for LGM than for PI. In fact, a small number (18) of LGM cyclones attain intensities exceeding the range identified for PI cyclones. The statistics for the region close to Europe (box) are similar 160 (not shown). All these results document a shift towards stronger intensities for LGM cyclones, both in terms of average numbers and extreme values.

Characteristics of Extreme Cyclones over the Eastern North Atlantic
To analyse the characteristics of cyclones for the LGM, a subset of 30 extreme cyclones passing over the Eastern North Atlantic is selected for both the PI and LGM periods based on the highest values for vorticity (Laplacian of MSLP) within the selected 165 box. The trajectories of the selected MPI-ESM cyclones are depicted in Figure 5 and key information are given in Table 3.
Two important facts are clearly identifiable: LGM extreme cyclone trajectories are more zonally orientated and constrained to a narrower corridor (particularly until 15°W) than their PI counterparts, and they achieve higher vorticity (Laplacian of MSLP) values during lifetime (Tab. 3).
For each of these cyclones, WRF simulations down to 12.5 km grid-spacing are performed along the most intense segment of 170 their lifetimes, thus gaining 3D data to analyse the cyclones e.g. in terms of the evolution of their structure, air masses, winds and precipitation. Care was taken that storms agree between the original MPI-ESM-P data and WRF cyclone tracks. Generally, the obtained tracks based on the WRF simulations reveal lower core pressures and higher vorticity than their low resolutions counterparts do. Figure 6 shows a comparison between high and low resolution data for a selected cyclone. The considered cyclone tracks (MPI-ESM LGM cyclone #24) show a good superimposition for MPI-ESM, WRF 50 km and WRF 12.5 km 175 and a good agreement of the position of maximum intensity with a slightly further south-eastward location for WRF 50 km ( Fig. 6a). While the time series, centred at peak intensity, for the development of core pressure show a similar strong pressure drop for all resolutions, the relative vorticity exhibits much stronger values for the high resolution simulation (WRF 12.5 km) in comparison with the coarser realisations. Note that the WRF simulations do not cover the whole trajectory of the cyclones, and thus the time series are shorter. Figure 6c and 6d depict the corresponding precipitation patterns at peak intensity, in this 180 case revealing more small scale structures and higher precipitation values for the 12.5 km WRF simulation.
Based on the top 30 cyclones for PI and LGM and using the composite methodology described in section 2, average cyclone characteristics are investigated for the intensification phase, focussing on the time frames (i) 12 hours before peak intensity and (ii) peak intensity. Figure 7 displays the MSLP fields for PI and LGM for peak intensity; the corresponding panels for 12 hours before peak intensity can be found in Fig. S1. The anomalies compared to the mean MSLP are displayed in colours, in order to analyse MSLP gradients (Fig. 7 a,b; S1 a,b). The core MSLP of the LGM has higher values compared to PI (LGM: 982.7 hPa, PI: 975.2 hPa). This can be explained by the lower sea levels (~120m) for LGM, causing a difference of global mean sea level pressure of about ~13 hPa (PI: 1010 hPa, LGM: 1023 hPa). Taken this into account, the LGM cyclones reach deeper MSLP values compared to the global average MSLP, being consistent with the stronger deepening rates for LGM cyclones identified by the tracking algorithm (Fig.3 g,h). Additionally, the closer isobars south of the cyclone core indicate 190 stronger pressure gradients for the LGM cyclones, which is supported by the LGM -PI differences (Fig. 7c). This is particularly the case on the expected location of the frontal areas. Fig. 7 (d,e) and S1(d,e) displays the anomalies of potential temperature at 850 hPa. Results show that the cross-frontal gradients are particularly intense across the warm front and that the whole development is apparently faster for LGM conditions, indicating a faster occlusion (Fig. 7 f). On the other hand, the total water content is much higher under PI conditions, primarily due to the effect of the higher environmental temperatures 195 ( Fig. 7 g,h; S1 g,h) with differences in the warm sector reaching up to 10 mm and 6-8 mm close to the cyclone core at peak intensity (Fig. 7i).
The strong difference in available water content again leads to a large difference in terms of accumulated precipitation, which is clearly larger for the PI composites (Fig. 8 a,b; Fig. S2 a,b). While some of the smaller deviations can be potentially attributed to slightly different developments, the total precipitation is considerably lower for LGM extremely cyclones (up to 1.7 mm h -200 is strongly increased for LGM cyclones (Fig. 8 d,e; Fig. S2 d,e), where wind speeds over large areas south of the cyclone exceed their PI counterparts by 10-12 m/s, particularly for 12 hours before peak intensity (Fig. 8 f; S2f). Strong differences are also revealed for near-surface wind gusts (Fig. 8 g-i; S2g-i). In this case, the wind gusts are particularly enhanced along the expected location of the cold front at peak intensity, with deviations exceeding 5 m/s. The above described patterns remain 205 true for the cyclone characteristics 12 hours after peak intensity (Figs S3, S4). For example, stronger wind gusts remain dominant in the area south of the cyclone core (Fig. S4i).
In summary, LGM cyclones display steeper MSLP gradients (in agreement with higher intensity in terms of circulation), larger temperature gradients between the air masses, weaker precipitation and stronger wind gusts than their PI counterparts do.

Discussion with available proxy-based climate reconstructions 210
The above analysis shows that extreme cyclones under LGM conditions were typically more intense than PI extreme cyclones.
Likewise, they were associated with larger frontal temperature gradients, stronger winds and reduced precipitation. In this section, we analyse in how far these cyclone characteristics can help to explain the occurring climate in Western and Central Europe under LGM conditions and the frequent occurrence of dust storms in that area.  Ludwig et al (2017Ludwig et al ( , 2018 show that the wet bias over Iberia and Western Europe can strongly be reduced by considering more realistic boundary conditions (particularly in terms of sea surface temperatures, land use and vegetation cover) in high-resolution RCM

simulations. 225
The present results enable a more detailed evaluation of the above hypotheses and interpretations. Under current climate conditions, precipitation in the mid-latitudes is largely associated with the passage of extratropical cyclones, where extreme cyclones have a comparatively larger contribution to total/extreme precipitation (Pfahl and Wernli, 2012;Hawcroft et al. 2012). Hawcroft et al. (2016) have provided evidence that GCMs typically have some shortcomings in representing the precipitation associated with mid-latitude cyclones. It is reasonable to assume that this caveat may be exacerbated at lower resolutions, e.g. 230 those typical of the PMIP3 models. An idealised study by Pfahl et al. (2015) had shown that precipitation increases (decreases) disproportionately in considerably warmer (colder) climate conditions. With the help of the high resolution WRF simulations, we could provide evidence that extreme cyclones under LGM conditions indeed induce considerably less precipitation than their PI counterparts (~22% within a 10° radius around the cyclone centre). Even though this effect could be somewhat compensated by the moisture advection embedded in the (stronger) westerly large-scale flow, particularly for areas where 235 orographic precipitation dominates (e.g. upward slopes of mountain ranges / glaciers), this lower precipitation associated with extratropical cyclones is consistent with the hypothesis of a drier Western and Central Europe, and thus with the dominant land cover types (polar desert close to the glaciers, forest steppe over Southern Europe and steppe in-between) estimated from proxy data (Ray and Adams, 2001) or statistical reconstruction based on temperature and precipitation (Shao et al., 2018).
The fact that we have selected an area close to the Iberia peninsula for the selections of the cyclones leads to the strong 240 assumption that largely drier conditions must also have been found for Southwestern Europe, in agreement with the proxies (e.g. Bartlein et al., 2011, Moreno et al., 2014. Mineral dust plays an important role in our climate system (Shao et al., 2011). Dust emissions are typically initiated by wind stress on land surfaces with little to no vegetation cover and easily erodible soils (e.g., Prospero et al., 2002). Such areas were very common in Europe under LGM conditions (e.g., Ray andAdams, 2001, Ugan andByers, 2007), when the global dust 245 cycle is estimated to have been more intense than in recent times (Maher et al., 2010). A large number of loess deposits over between 34-17 ka on several sites in France, Germany and Belgium, with particularly high sedimentation rates, and attributed these to numerous and intense dust storms in periods with adequate large-scale circulation and reduced precipitation. 250 Furthermore, the high accumulation rates of loess in the middle and lower Danube basin indicate cold, dry and windy conditions during the LGM in south-eastern Europe (Fitzsimmons et al, 2012), consistent with increased storm activity over Central Europe (Fig. 2c). These and other findings document a more intense (global) dust cycle for LGM conditions (e.g. Albani et al., 2016;Újvári et al., 2017;Albani and Mahowald, 2019). The occurrence of dust storms over Western and Central Europe has been conceptually associated with the passage of intense extratropical cyclones penetrating deep into the European 255 continent (Antoine et al., 2009;their Figure 12). Following on previous studies (e.g. Laine et al., 2009;Hofer et al. 2012;Ludwig et al., 2016), the present results provide evidence for the first time that individual LGM cyclones would be indeed capable of triggering such dust events: their frequent tracks over Western/Central Europe and strong wind speeds could easily trigger dust emissions and transport it over short (for coarse grain) and large (for fine grain material) distances (cp. Shao et al. 2011). As the precipitation associated with these cyclones (which could potentially hamper the triggering of dust storms) is 260 considerably reduced, it should be a less relevant limiting factor under LGM conditions. In addition to the role of the westerlies and embedded cyclones into generating dust storms in Europe, there is evidence that situations with persistent easterlies associated with anticyclonic flow triggered by a strong anticyclone over the Scandinavian ice sheet may have also played a significant role for loess deposition not only over Eastern Europe but also over Central Europe (Újvári et al., 2017;Schaffernicht et al., 2019). 265

Summary and Conclusions
The statistics and characteristics of extratropical cyclones over the North Atlantic and Western Europe were analysed for timeslice experiments for PI and LGM conditions. First, the statistics of the climatologies of PI and LGM cyclones are analysed and compared based on global MPI-ESM-P simulations. Second, the characteristics of extreme LGM cyclones over the Eastern North Atlantic were analysed in detail based on high-resolution simulations (12.5 km grid-spacing) with the regional climate 270 model WRF and compared to their PI counterparts. The results were discussed with available proxy reconstructions of climate parameters and vegetation types. The main conclusions are as follows:  The North Atlantic storm track was more intense under LGM conditions, featuring an enhanced number and more intense synoptic systems than under PI conditions. One of the downstream branches brought more often extreme cyclones towards Western and Central Europe and the Mediterranean area. 275  LGM cyclones were more intense due to stronger baroclinicity and apparently less influenced by diabatic processes (lower rainfall, lower temperatures, lower water vapour content). In particular, LGM cyclones benefit from a stronger https://doi.org/10.5194/cp-2019-139 Preprint. Discussion started: 2 December 2019 c Author(s) 2019. CC BY 4.0 License. and extended jet stream. The development was typically faster, with deepening rates and peak intensities exceeding those from PI cyclones.
 LGM extreme cyclones were characterised by lower precipitation, enhanced frontal temperature gradients, and 280 stronger mean wind speeds and wind gusts than PI analogues.
 These characteristics are in line with the view of a colder and drier Europe, characterised by steppe/tundra land types and affected by frequent dust storms, leading to reallocation and build-up of thick loess deposits.
Given that this study is based on a single GCM, a single tracking method and a single RCM, it should be regarded as a preliminary analysis as the uncertainties may be considerable. Still, the identified differences between PI and LGM (extreme) 285 cyclones are unequivocal, are consistent with idealised studies, demonstrate the potential of the approach and may become instrumental to facilitate a better interpretation of LGM proxy data. In particular, this study provides new understanding of the relationship between the large scale mean cyclone activity and short term variability on the regional scale and thus may help to reduce numerical interpretative uncertainties (Harrison et al. 2016).
Even though the added value of regional climate models in paleoclimate applications is still controversially discussed 290 (Armstrong et al., 2019), there is a general call for improvements towards a new generation of reliable regional projections (e.g., Harrison et al., 2015;Kageyama et al., 2018). The present and other studies (see Ludwig et al.,