Synoptic and Mesoscale atmospheric features associated with an
extreme Snowstorm over the Central Andes in August 2013

Abstract. We study the synoptic and mesoscale characteristics of a snowfall event over the Bolivian Altiplano in August 2013 that caused severe damage to people, infrastructure and livestock. This event was associated with a cold front episode following the eastern slope of the Andes-Amazon interface and a cut-off low pressure system (COL) over the Pacific Ocean. Large scale analyses suggest a two-stage mechanism: The first phase consisted of a strong cold surge to the east of the Andes inducing low level blocking of southward moisture transport over the SW Amazon basin due to post-frontal high-pressure up to 500 hPa synchronized to a Rossby wave train. The second stage was initiated by the displacement of 500 hPa anticyclone over the Andes due to a Rossby wave passage and a subsequent increase in north-easterly moisture transport, while another cold front along the eastern Andes provided additional lifting. We analyse an analog event (July 2010) to confirm the influence of these large-scale features on snow formation. We conduct a mesoscale analysis using the Weather Research and Forecasting (WRF-ARW) model. For this purpose, we perform a series of high-resolution numerical experiments that include sensitivity studies where we apply orographic and lake Titicaca temperature modifications. We compare our findings to MODIS snow cover estimates and in-situ measurements. The control simulation is able to capture the snow cover spatial distribution and sheds light over several aspects of the snowfall dynamics. In our WRF simulations, daytime snowfall mainly occurs around complex orography whereas nocturnal snowfall is concentrated over the plateau due to a combination of nocturnal winds and complex orography inside the plateau. The sensitivity experiments indicate the importance of the lake and mountain for thermal wind circulation affecting the spatial distribution of snowfall by shifting the position of the convergence zones. The influence of the lake's thermal effect is not evident around the regions surrounding the lake.



Remote sensing data
Snow cover data is obtained from the MODIS Aqua satellite daily snow cover dataset MYD10C1 (Hall and Riggs, 2016).
Because the reflectance of fresh snow is high, this product is able to capture the spatial distribution of snow cover over land under the condition of a clear sky. This conditions is fulfilled after the event and therefore we take the data corresponding to 25th August 2013. TRMM-based precipitation estimates from TMPA 3B42RT are ruled out since this product does not 5 adequately capture snowfall (Beck et al., 2017) In addition to MODIS, we use the Geostationary Operational Environmental Satellite (GOES-12 and 13) in the water vapour spectral and visible bands (6.48 and 0.63 µm respectively) for an spatial assessment about cloud cover.

Global analysis
To asses large-scale circulation and moisture transport, we extract fields from the ERA-interim global reanalysis (Dee et al., Even though we use ERA-interim for large-scale circulation and climatology assessment, we decided to use the Final Analysis (FNL) from the Global Forecast System (GFS) as boundary and initial conditions for the WRF simulations. This choice is made because of the better reproduction of spatial snow-cover distribution by FNL-based simulations with respect to ERA-15 interim based simulations (not shown). Additionally, we note SENAMHI's GFS products utilization for operational forecasts.
Hence they may be able to directly profit from our study.

Meteorological diagnostics
One goal of this study is to analyse the influence of cold surges' position along the Andes on snowfall episodes. There exist 20 several front detection methods, but assessing which method is the most accurate over this region is beyond the scope of this study. Our goal is to focus on the position of the front and with this in mind, we decided to use the equivalent potential temperature θ e gradient at 850 hPa because of its simplicity and its consistency with synoptic features in moist processes (Schemm et al., 2018) Following Sprenger et al. (2012), we aim to relate the cold surges to polar and upper-level influences and therefore we decide 25 to examine the upper-level conditions through a potential vorticity (PV) perspective in PVU units.
We complement the cold surge study with an assessment of large-scale moisture transport, with the calculation of the vertically-Integrated water Vapour Transport (IVT) vector, (1) where q is the specific humidity (gkg −1 ), g = 9.81 ms −2 is the gravity constant, u and v are the zonal and meridional wind in ms −1 and dp is the pressure thickness (hPa). We calculate it starting from surface SF C to the 200 hPa level.
The circulation of cold dry air within the cold surges and the trajectories of moist parcels is studied with a Lagrangian 5 trajectory diagnostic. We use the LAGRANTO tool (Sprenger and Wernli, 2015) for backward and forward trajectory of air masses off-line calculation in WRF outputs.
The mesoscale analysis is performed by studying several diagnosis such as divergence, relative humidity and cross sections over selected areas. The heavy snowfall episode over the Andes occurred between the afternoon of 23 August and the morning of 25 August 2013.
The Bolivian Civil protection office reported more than 1 meter of snow height in Tinguipaya (hereafter TP) and Cocapata IVT analysis further indicates that the moisture is transported southwards from the north-west Amazon towards and over 25 the Altiplano following the Andes slopes. It shows also some signals over the Pacific ocean by occasions ( Fig. 4a-d). At the same time, IVT shows that once reaching the Altiplano, moist air is transported to the south-east following the position of the continental cold front. A climatological study of IVT over two selected grid points located along the Andes-Amazon transition (Points A1 and A2 in Fig. 4a-d) indicates that the moisture transport during this event was extreme for this region; the polar diagrams ( Fig. 4e and 4f) reveal that IVT originating out of the north-west surpassed even the 99 percentile for all seasons.  An upper-level COL developed over the southwest Altiplano and a quasi-stationary cold front was present to the east of the Andes (Fig 5a,d,e). At the same time, southward vapour transport towards the Altiplano intensified (Fig. 5b) once the mid-level high-pressure system was displaced by appearance of the COL (Fig. 5f). The main event occurred between 17 and 18 July and it was located more to the east, following the COL position (Fig. 5c).

Snow cover evolution and station measurements 10
The large-scale circulation analysis is able to explain the large amounts of snow that affected the Altiplano but offers little detail about the spatio-temporal distribution. A first assessment of the snow cover evolution is made with the MODIS MYD10C1 daily snow cover product on 25 August (once the event is over and the reduced cloudiness allows a reasonable snow cover estimate) to validate the snowfall results from the control run.
The snow-covered areas on 25 August (Fig. 6a) were mainly located over complex orography with a particular concentration  As seen in Figure 6b, the WRF CTRL simulation is capable to emulate the timing and spatial distribution of snow cover but not the alleged intensity (The Bolivian Civil protection agency reported up to three meters of snow, while WRF simulates at most one meter). The reasons of this underestimation can range from the choice of the microphysics schemes, the land use models and the accuracy of the reports. Nevertheless, it is possible to numerically investigate the mesoscale circulation and the main dynamical mechanisms that may have led to this historical event.

25
Additional SENAMHI measurements of daily precipitation (we remind that measurements are taken from 1200 UTC every 24h) show a clustering around La Paz city from 23 to 24 August (Fig. 6c) and a more homogeneous distribution from 24 to 25 August (Fig. 6d). The lack of stations in southern Altiplano or even southern Oruro is evident.

Mesoscale dynamics from control run
In the following section we analyse daytime and nocturnal snowfall separately. Taking into consideration the region time zone 30 (UTC-4), we consider as night the period from 0000 to 1200 UTC and as day the period from 1200 to 0000 UTC. Daytime snowfall is mostly concentrated over complex orography (Figs. 7a,c), while night-time snowfall is more homogeneous and extends over much of the Altiplano, including complex orography (Figs. 7b,d).
Hovmoeller diagrams of the snow precipitation over selected cross sections (red lines on Figs. 7a-d) confirm these day-night differences, offering more detail about the hourly variability of the snowfall (Figs. 7e-h). Over the western Cordillera, the firstday snowfall initiates during the afternoon and is maintained until early night, while the second-day snowfall is synchronized to 5 the COL passage (Fig. 7e). Snowfall over southern Altiplano (Fig. 7f) starts in the north around 0000 UTC and then propagates south-eastwards in both days. The north-eastern Altiplano (Fig. 7g) shows snow precipitation after 0400 UTC (local midnight) every day. Finally, snowfall along the Andes-lowlands transition (Fig. 7h) is less homogeneous, but clusters mostly around the night.

Low-level circulation and moisture transport
10 Analysis of the 850 hPa equivalent potential temperature (θ e ) gradient indicates the position of the cold front. We combine this information with the IVT and the relative humidity in order to assess the full column moisture circulation over the Altiplano and the cloud cover. Figure 8a indicates a high initial southward IVT along the eastern Andean slopes. Once the cold surge enters the domain, moisture is advected towards the Altiplano along intra-andean valleys. At the same time the relative humidity increases follow-15 ing this circulation until reaching saturation (Fig. 8b). The cold front position stays quasi-stationary during the next 24 hours and the IVT is quite homogeneous all over the Altiplano and Pacific Ocean while still having some preference over the eastern Andean valleys. IVT appears to be unaffected by orography but additional analysis suggest that synoptic transport is dominant over low level fluxes (not shown). The atmospheric water vapour content is high at this moment everywhere for 12 more hours ( Fig. 8b-c), consistent with satellite imagery (Fig. 1c). The end of the event is characterized by near-saturated mid-troposphere 20 atmosphere propagating eastwards following the COL appearance (Fig. 8d). The IVT is still high, but increasingly directed south-eastwards. 500 hPa circulation analysis on the afternoon prior the start of the event (Fig. 9a) shows convergence over complex orography with an inside Altiplano prolongation from the west mountains. The convergence zone over the western Cordillera appears to propagate eastward over the Altiplano during the night (Fig. 9b-c). On the other hand, convergence over complex orography 25 near Titicaca lake and La Paz city is still present. The first night was characterized by a relatively weak 500 hPa winds. Satellite image on August 24 at 0000 UTC (Fig. 9d) agrees with the convergence band shown in Fig. 9c that starts over to the lake surface. We compare this two variables since the simulated outgoing long-wave radiation (OLR) overlaps to the convergence zones (not shown) The same analysis for the second night (25 August) shows an enhancement of 500 hPa winds (Fig. 9e-f). Therefore the 30 convergence zones tend to be slightly less closely aligned with local topography and more affected by the synoptic circulation, although still showing a clear preference over complex orography. The satellite image on August 25 at 0000 UTC (Fig. 9g) also agrees with CTRL run and shows a more generalized cloud cover. The increase in wind speed culminates with the arrival of the COL over the Altiplano, accompanied by dry and strong winds (see next section). Additional 3-day back-trajectory analyses over the domain D02, indicate that the north-western Amazon Basin serves as the main air income source. Selected places located along the Andes-Amazon transition received air masses from the southwestern Amazon basin. There are also some trajectories that indicate the northward propagating cold surge east of the Andes, although this air is unlikely to be associated with major moisture transport as these air masses tend have very low moist static energy (Hurley et al., 2015) (Fig 10a-b). More central Altiplano located regions shows a mix of sources coming generally from 5 the north-west (Fig 10c-d). The first night (24 August at 0600 UTC, Fig. 10a,c) is characterized by a slow-moving circulation with air masses origination at low levels and following frontal and thermal up-slope motion preferably through the valleys, while upper-level air comes mostly from the upper atmosphere located over north-west Bolivia (Fig. 10a). Over the central Altiplano, arriving air masses also follows thermal circulation but coming mostly from the Pacific Ocean and from north-west Altiplano (Fig. 10c). The circulation during the second night (25 August at 0600 UTC, Fig. 10a,c) shows a similar pattern as 10 during the previous night, but generally stronger winds.
Concerning the vertical structure, we show cross-sections over the red solid lines plotted in Figures 7b (Fig. 11a-c) and 7c ( Fig. 11d-e) corresponding to three days at 0600 UTC. The western region shows rapid surface cooling from 23 to 24 August ( Fig. 11a-b), staying cold until 25 August (Fig. 11c). Vertical motion on the 24 August follows complex orography and cold parcels (Fig. 11b) while on the 25 August it follows mountain gravity waves (Fig. 11c). The vertical motion over the eastern 15 Altiplano follows mostly cold air regions propagating from the west, while the lake doesn't seem to exert a significant impact on atmospheric instability ( Fig. 11d-f)

COL passage over the Andes
Cross sections over the Pacific Ocean show an atmospheric vertical structure consistent with synoptic COL development. WRF simulations shows that the cyclone reached the surface over the Altiplano on 25 August at 1800 UTC, accompanied by strong 20 and dry winds. This COL passage marked the end of the event and its characteristics are consistent with previous studies of such events in the region (Fuenzalida et al., 2005;Garreaud and Fuenzalida, 2007;Garreaud et al., 2003a). Thus, we decided to not to include these results here.

Orographic features 25
Reducing the topography of the Cordilleras surrounding the Altiplano (experiment RTA) resulted in a smaller amount of snowfall over the high mountains and a larger amount further to the south during the first day of snowfall. The second day event showed an opposite tendency and was accompanied by larger quantities of snowfall over the Altiplano and less snow to the south. The influence of the reduced topography on the resulting snow cover is small over the Altiplano, but it becomes important at higher elevation, where the snow cover is reduced everywhere by around 20 %. A similar experiment, but with 30 the elevation of the Altiplano also reduced (experiment RT80), confirms the importance of the mountains for influencing the local circulation over the Altiplano, but it also reaffirms the role of the Altiplano's height itself for determining the type of precipitation. Not surprisingly, the RT80 experiment shows a general snowfall reduction all over the Altiplano and Cordilleras (Supplementary figure S1)

Titicaca lake influences
The removal and surface temperature modification of the lakes over the Altiplano leads to a change in the spatial distribution of precipitation. We only show results for the 24. August snowfall during the night, because the most important thermal circulation 5 influences occur at that time.
A colder lake (experiment LK-5, Fig 12a) restrict the airflow from the Amazon and weakens the nocturnal breeze towards the lake. The impact is to shift snowfall away from the lake along a NW-SE oriented band in the direction of more complex orography inside the Altiplano. At the same time, cross sections ( Fig. 12b-c) show a limited influence of the lake for thermal convection. The restriction of the Amazon airflow over the eastern edge of the Altiplano provokes an increase in snowfall over 10 CP by around 10 mm (Fig 12a).
The NOLAKE experiment has a similar influence as the LK-5 experiment inside the Altiplano, albeit the air blocking from the Amazon, which now also affects the eastern Andean ridge and CP. The suppression of the nocturnal lake breeze influences the spatial distribution of snowfall by displacing the main snowfall band eastwards (Fig. 12d). Associated with this shift is the decrease of snowfall over the Andes-Amazon transition by around 10 mm. Cross sections (Fig. 12e-f) shows similar features 15 than in the cooler lake experiment.

Discussion
The snowfall event documented here occurred within the context of 2013 being a particularly cold winter in South America.
A high number of snowfall days (11) were observed in southern Brazil (Mintegui et al., 2018) and several of the coldest air intrusions in the twenty-first century were documented (Lanfredi and de Camargo, 2018).

20
The synoptic analysis suggests that a high atmospheric water vapour content over the north-western Amazon basin primed this event several days prior to its occurrence. The onset of the event was characterized by the establishment of a quasistationary Rossby wave over the Pacific Ocean and the appearance of a persistent 500 hPa high-pressure system over the Andes. At the same time, over the Amazon, there were remnants of a prior cold surge that suppressed the low-level jet. During this phase, the southward water vapour transport was blocked by the frontal activity over the Amazon, resulting in the formation 25 of an important atmospheric water vapour reservoir. This results are consistent with the findings of Sicart et al. (2015), who identified the Amazon region as humidity source for cloudinness over the eastern Cordillera.
The second phase included the southward water vapour transport and discharge phase; the IVT climatology analysis showed an intense water vapour transport towards and along the eastern Andean slope. This phase was initiated by the displacement of the Andes' anticyclone synchronized with the Rossby wave displacement and thermal cyclogenesis over the Andes' eastern 30 slope due to weak frontal activity. During this phase, we observed extreme amounts of IVT over the Andes that resulted in two days of heavy snowfall.