2.1 Observations from automatic weather stations

Through the Thwaites-Amundsen Regional Survey and Network (TARSAN) project integrating atmosphere–ice–ocean processes of the International Thwaites Glacier Collaboration (ITGC), automatic weather stations known as the Automated Meteorology–Ice–Geophysics Observation System (AMIGOS; Scambos et al., 2013) were installed on Thwaites Eastern Ice Shelf at Cavity Camp (75.033^∘ S, 105.617^∘ W) and Channel Camp (75.050^∘ S, 105.4334^∘ W) during a field campaign in austral summer 2019/20 (Fig. 1). Cavity Camp is located on a flat part of Thwaites Eastern Ice Shelf, whereas Channel Camp sits within the surface expression of a basal melt channel (Alley et al., 2016). The AMIGOS at Cavity Camp and Channel Camp provide unique in situ observations of the weather occurring over Thwaites Eastern Ice Shelf by logging hourly temperature, surface pressure, wind speeds and direction, humidity, and snow height, as well as other observations such as GPS position and firn temperature. We utilize these novel observations to characterize the in situ atmospheric conditions – air temperature, surface pressure, wind speed, and wind direction – for the case study of an AR that made landfall over TG on 2 February 2020. The AMIGOS temperature sensors were located about 6 m above the surface during the period of interest in this study, so we refer to AMIGOS air temperatures as “near surface”. The choice not to correct observed temperatures to 2 m above the surface is justified because the atmospheric surface layer is usually neutrally stable and well-mixed during AR events. We use a firn-temperature record from thermistors placed 1 m below the surface at the AMIGOS stations to identify changes in firn temperature associated with percolation of surface meltwater during the case study AR event. Sustained hurricane-force winds of 55–70 m s⁻¹ over Thwaites Eastern Ice Shelf in late September 2021 severely damaged the AMIGOS and have limited the available weather data to a period of just over 1.5 years after installment.

2.2 Reanalysis products: MERRA-2 and ERA5

To characterize the large-scale atmospheric circulation and weather conditions associated with Antarctic AR events, we use the global atmospheric reanalyses Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2; Gelaro et al., 2017) and the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5; Hersbach et al., 2020). Reanalyses assimilate observations (not including the AMIGOS data) with atmospheric modeling to produce regularly gridded, temporally continuous weather and climate data. Due to the spatiotemporally limited observations available for the Southern Ocean and Antarctica, reanalysis products are essential for examining spatial patterns in atmospheric pressure, temperature, moisture, and snowfall over the WAIS, and they are commonly used to examine the synoptic forcing of ARs over Antarctica (Gorodetskaya et al., 2014; Pohl et al., 2021). MERRA-2 features a latitude–longitude grid spacing of 0.5^∘ × 0.625^∘ (Gelaro et al., 2017), which is approximately 56 km × 18 km at the location of TG. ERA5 is the newest global atmospheric reanalysis product from the European Centre for Medium-range Weather Forecasts (ECMWF) and features a latitude–longitude grid spacing of 0.25^∘ latitude (28 km) by 0.25^∘ longitude (7 km) resolution (Hersbach et al., 2020). We primarily use MERRA-2 to analyze the large-scale synoptic conditions and impacts of AR events in West Antarctica, as MERRA-2 explicitly represents ice sheet hydrological and energy budgets and compares best to ice core records of snow accumulation in Antarctica among multiple reanalyses (Gelaro et al., 2017; Medley and Thomas, 2019).

We use the MERRA-2 reanalysis to generate composite maps of the surface pressure and surface pressure anomaly (relative to 1980–2020 climatology) during the times of AR landfalls over coastal West Antarctica, including the Amundsen Sea Embayment and Marie Byrd Land. Over Thwaites Eastern Ice Shelf, we compare MERRA-2 2 m temperatures before and during AR events to examine the impacts of ARs on surface temperatures by season.

We also use MERRA-2 to examine spatial patterns in surface pressure, 500 hPa geopotential height, and the 500 hPa geopotential height anomaly (relative to 1980–2020 climatology) over West Antarctica and the Southern Ocean during the case study AR event in February 2020. We compare time series of both MERRA-2 and ERA5 to the atmospheric conditions observed by the AMIGOS (near-surface temperature, surface pressure, wind speed, and wind direction) on Thwaites Eastern Ice Shelf during the case study event. While AMIGOS observations reflect local conditions at the Cavity and Channel Camp sites, MERRA-2 and ERA5 data represent grid-cell averages, meaning local values for temperature, surface pressure, wind speed, and wind direction can deviate from those grid-cell averages. In the near-surface temperature comparison, MERRA-2 and ERA5 use 2 m temperatures while the observed temperatures are from approximately 6 m above the surface. We include ERA5 in this analysis because there are differences between MERRA-2 and ERA5 in 2 m temperature and snow accumulation during the event, allowing for a comparison between the reanalyses.

2.3 Atmospheric river detection catalogue and attributing precipitation

We use a polar-specific AR detection algorithm produced by Wille et al. (2021) to identify the occurrence and landfall of ARs over the WAIS. The algorithm uses the MERRA-2 atmospheric reanalysis (Gelaro et al., 2017) from 37.5 to 80^∘ S to detect moisture filaments when integrated water vapor (IWV) or meridional integrated vapor transport (vIVT) exceed the 98th percentile of their monthly climatologies. Moisture filaments are classified as ARs if they extend at least 20^∘ in the meridional direction. Because it tracks IWV and vIVT relative to the monthly climatology, this algorithm is uniquely positioned to detect polar ARs. Wille et al. (2021) found that IWV is better suited for identifying ARs that cause surface melting, as high IWV over the AIS is associated with cloud development and high downwelling longwave radiation to the surface. Comparatively, the vIVT-based definition of ARs is better suited for studying snowfall, since the meridional transport of water vapor is linked to atmospheric dynamics that lead to precipitation.

In this study, we primarily focus on AR-driven precipitation, and thus we use the vIVT catalogues with AR detection at 3 h intervals based on MERRA-2 reanalysis. From the catalogues, we generate maps of AR frequency over the AIS. We then identify individual AR events making landfall over the Amundsen Sea Embayment and Marie Byrd Land by counting the number of consecutive time indices of AR landfall within the region. If the time steps indicate a 12 h break or more, we count them as two separate AR events. We do this to account for AR families, which are multiple ARs that make landfall in succession in short periods of time (Fish et al., 2022). Based on the AR events we identified, we quantify their trends, duration, and seasonality from 1980 to 2020. To attribute precipitation to ARs, we integrate MERRA-2 3 h precipitation within the footprint of the AR landfall, based on the vIVT AR catalogue of Wille et al. (2021). We also attribute precipitation that falls up to 24 h after landfall, but within the footprint of the AR, to the AR event.

2.4 SNOWPACK firn modeling

We use observed snow height and temperature from the AMIGOS to force the firn model SNOWPACK (Lehning et al., 2002a, b) to reconstruct accumulation and surface melt during the AR case study event in February 2020. SNOWPACK is a physics-based, multi-layer firn model (Lehning et al., 2002a, b) which has been extensively applied in polar regions (Groot Zwaaftink et al., 2013; Steger et al., 2017; van Wessem et al., 2021; Keenan et al., 2021). The model calculates snow compaction using an overburden formulation and solves the full surface energy balance to provide the upper boundary condition for solving the temperature equation and calculating melt. When snow accumulates, fresh accumulation density is calculated as a function of meteorological conditions, particularly wind and the presence of drifting snow (Keenan et al., 2021; Wever et al., 2022). We configure the SNOWPACK model similar to the “Redeposit”-setup discussed in Wever et al. (2022) while additionally deriving snowfall from the observed snow height: if observed snow height exceeds the simulated snow height, the difference is interpreted as snowfall, when relative humidity, air temperature, and snow surface temperature meet snowfall conditions (Lehning et al., 1999; Wever et al., 2015). This is then combined with fresh accumulation density, parameterized following Schmucki et al. (2014), to convert to accumulated mass.

We use the MeteoIO library (Bavay and Egger, 2014) to preprocess the meteorological forcing data. Since the measured snow depth from the acoustic sensors exhibited oscillations, seemingly synchronized with the diurnal cycle in temperature and consequently, the speed of sound (e.g., Ryan et al., 2008), we applied a weighted moving average smoothing filter with a centered window of 48 h. Both the Cavity Camp and Channel Camp AMIGOS are represented by the same closest MERRA-2 grid point (Fig. 1). However, we run SNOWPACK individually for each of the AMIGOS. Using the model settings as described in Keenan et al. (2021), we run SNOWPACK for the period from 1980 to the installation date of the respective AMIGOS, using forcing data provided by MERRA-2. We then mark the layer in the SNOWPACK-simulated firn, whose depth corresponds to the measured position of the base plate of the AMIGOS upon installation in the field. In the remainder of the simulation, we track the position of this marked layer, and the AMIGOS snow depth is referenced to this specific layer. We can then use the aforementioned snow-height-driven accumulation approach, while eliminating the settling of the AMIGOS station in the firn column. From the installation date of the AMIGOS forward model, we use the available parameters from the AMIGOS to force the simulation (i.e., snow depth, near-surface air temperature, relative humidity, and wind speed) while taking downwelling shortwave and longwave radiation from the reanalysis. From the installation date of the AMIGOS onwards, we performed two sets of simulations, one forced by MERRA-2 and the other by ERA5.

2.5 Surface height changes via interferometric reflectometry

We supplement the record of surface height change estimates observed by the AMIGOS on Thwaites Eastern Ice Shelf with surface height change measurements from the grounded TG observed with the global navigation satellite system (GNSS) using interferometric reflectometry (IR; Larson et al., 2009, 2015; Roesler and Larson, 2018). Two GNSS receivers have been maintained on TG over the last 10 years. These sites are located 124 km (GNSS Lower Thwaites; 1011 m elevation) and 215 km (GNSS upper Thwaites; 1315 m elevation) from the present grounding zone (Fig. 1; Wilson et al., 2009a, b) and have been used to understand the flow acceleration of the glacier and the influence of lake fill-drain cycles on interior glacier flow speed. These sites can also be used to understand the variability of snowfall over TG using GNSS–interferometric reflectometry (GNSS–IR). The addition of GNSS–IR snow-accumulation records enables us to compare spatial differences in snowfall on Thwaites Eastern Ice Shelf and TG during the AR case study event. We use the GNSS–IR method to determine the height of phase center of the GNSS receiver antenna above the surface using the signal-to-noise ratio of multi-path interference with the direct signal. As snow is deposited at Lower and upper Thwaites receiver sites, the height of the receivers above the surface is reduced. Forward models of reflector height change due to snow loading. Densification combined with inverse methods capable of sampling posterior distributions of model parameters using likelihood estimates that assimilate reflector height data can be used to solve for the joint accumulation and densification history. From these independent records of accumulation and deposition, we can verify the influence of atmospheric river landfall on observed SMB. We show GNSS-derived accumulation with the SNOWPACK-reconstructed accumulation to examine the inland propagation of the case study AR event from Lower TG (AMIGOS and Lower Thwaites GNSS) to Upper TG (Upper Thwaites GNSS). These observations provide additional spatial information on accumulation patterns due to the AR event.

Figure 2(a) Temporal frequencies of AR landfall (number of AR times divided by the total time, as a percentage), (b) precipitation attributed to ARs, and (c) AR precipitation as the percentage of the total annual precipitation in West Antarctica (MERRA-2, 1980 to 2020). The Amundsen Sea Embayment and Marie Byrd Land (region of interest) are outlined in black in all three figures. Gray shading over the interior of the ice sheet marks the 80^∘ S boundary of the Wille et al. (2021) AR detection algorithm.

Figure 3(a) The number of AR events making landfall over the Amundsen Sea Embayment and Marie Byrd Land each year (blue), with the 1980–2020 trend and standard error (gray) and the 1995–2015 trend error (pink). (b) The mean duration of AR events each year (thick blue line), along with the maximum and minimum duration of events (dotted lines) and the standard deviation from the mean (blue shading). (c) The total precipitation attributed to ARs annually from 1980 to 2020.

Download

3.1 Climatology of West Antarctic atmospheric rivers

To determine the frequency of ARs over the Amundsen Sea Embayment and Marie Byrd Land region, we divide the number of AR times by the total time for each year from 1980 to 2020 then take the mean. Our analyses show that ARs exhibit a total frequency of 3.2 % over the whole region from 1980 to 2020 (i.e., on average there is an AR making landfall somewhere in the region 3.2 % of the time). Within the region, localized AR frequencies range from 0.2 % to 0.8 % of the time, with the highest frequencies over the Abbot Ice Shelf and the Getz Ice Shelf (Fig. 2a). Integrated over the entire region, ARs contribute 59 ± (1 standard deviation) 24 Gt precipitation annually (out of 550 ± 63 Gt total annual precipitation, Figs. 2b and c and 3c), and explain 28.7 % of the interannual variability in precipitation (linear trends removed). The correlation between AR precipitation and the total annual precipitation is moderately positive, with the Pearson correlation coefficient r = 0.52 (p < 0.01, Fig. A1).

On average, there are 17 ± 5 AR events over the Amundsen Sea Embayment and Marie Byrd Land each year (Fig. 3a). From 1980 to 2020, there is a positive trend in AR events of +0.12 ± 0.06 events per year squared (p = 0.055), similar to the results from Wille et al. (2021), which also showed an increasing trend in AR frequency from 1980 to 2018 over the WAIS region. From 1995 to 2015, there is a marked trend of +0.32 ± 0.16 events per year squared (p = 0.059), indicating multi-decadal variability in the number of AR events embedded within the longer-term positive trend (this 20-year period is selected based on its high trend, low p value, and low standard error of 0.16 events per year squared). The mean duration of AR events in this region is 16.2 ± 13.9 h (Fig. 3b), which indicates that there is large variability in the duration of AR events. Throughout the study period, there were many short events that passed the AR detection threshold for only 3 h, while the longest detected durations after AR landfall occurred in 1985 and 2010, lasting 84 h. There is no statistically significant correlation between the number of AR events per year and their maximum, or mean, duration. Furthermore, there is no statistically significant seasonality to the number of AR events nor their duration over the Amundsen Sea Embayment and Marie Byrd Land.

Figure 4MERRA-2 composite maps of (a) surface pressure and (b) surface pressure anomaly during AR landfall in the region of interest (the Amundsen Sea Embayment and Marie Byrd Land) which is outlined in black. Stippling on the surface pressure anomaly indicates regions of statistically significant anomaly, where the anomaly exceeds one half of the standard deviation from the mean surface pressure.

Figure 5MERRA-2 2 m temperature difference by season between the average temperature 24 h before AR landfall and the average temperature 24 h from AR landfall (post- minus pre-event) from MERRA-2 reanalysis over Thwaites Eastern Ice Shelf. Starting from the bottom, austral summer (December–January–February or DJF), fall (March–April–May or MAM), winter (June–July–August or JJA), and spring (September–October–November or SON). From left to right on each box plot: minimum, first quartile, median, third quartile, maximum. Outliers are represented as circles on the plot.

Download

From the average surface pressure maps during AR events, we can identify that ARs in this region are driven by large-scale synoptic patterns involving the coupling of a blocking high and a low-pressure system (Fig. 4). In the surface pressure composite of all AR events on TG, we find that AR events are associated with a climatological low-pressure system, commonly referred to as the Amundsen Sea Low, located west of TG (Raphael et al., 2016). There is also a high pressure ridge extending southward from South America towards the Antarctic Peninsula. In the surface pressure anomaly composite, we see statistically significant high-pressure anomalies over the Antarctic Peninsula, signaling the presence of a blocking high during AR landfall on TG. We also observe statistically significant low-pressure anomalies in the region of the Amundsen Sea Low, west of TG. The blocking high serves to inhibit the zonal flow of low-pressure systems and directs circulation towards the Amundsen Sea Embayment and Marie Byrd Land. Furthermore, the surface pressure anomaly composite shows a zonal wave tree-like structure of alternating high- and low-pressure anomalies surrounding the AIS (Raphael, 2004).

To further examine the impacts of AR landfalls on TG surface conditions, we calculate the change in surface temperatures on Thwaites Eastern Ice Shelf during AR events (Fig. 5). To do this, we take the difference between the mean MERRA-2 2 m temperature 24 h before landfall and the mean 2 m temperature 24 h after landfall. Atmospheric river events are associated with a temperature increase of 1.4 K (first quartile) to 7.1 K (third quartile), with a median of 3.8 K over Thwaites Eastern Ice Shelf over all seasons. In austral summer (December–January–February), the median temperature increase is the smallest at 1.5 K. In fall (March–April–May), winter (June–July–August), and spring (September–October–November), the median temperature increases associated with AR landfall are 4.3, 6.3, and 4.3 K, respectively. The increase in surface temperatures is associated with temperature advection and increased downwelling longwave radiation during AR landfall (Wille et al., 2019). Here, we see the largest increases in temperature associated with the landfall of winter AR events. There are many more summer events where 2 m temperatures exceed the melting point of 273.15 K (six events in 1980–2020) than in fall (two events), winter (one event), and spring (0 events). While winter AR events are associated with the largest median and maximum temperature increases during AR landfall, they are also associated with the largest standard deviation in temperature changes (± 5.0 K), and some winter events exhibit a decrease in temperature from the mean 24 h before landfall to the mean 24 h after landfall.

Figure 6Three ARs make landfall on TG in short succession on (a, d) 2 February, (b, e) 4 February, and (c, f) 7 February 2020. (a–c) MERRA-2 500 hPa geopotential heights (contours) and height anomaly (colors) during each AR landfall. (d–f) MERRA-2 surface pressure (colors) during each AR landfall. In each panel, the AR detection catalogue by Wille et al. (2021) indicates the spatial footprint of the ARs in pink. The location of shortwave troughs (geopotential height) and associated nodes in the low-pressure system (surface pressure) are indicated in each panel. Trough 1 (T1) drives Low 1 (L1), T2–L2, T3–L3, and T4–L4.

3.2 Case study: atmospheric river surface impacts on Thwaites Glacier

The geometry and orientation of TG render it highly susceptible to synoptic flow-induced snow storms caused by ocean air masses that are driven from the Southern Ocean into the Amundsen Sea sector. When air masses rise up the slope of TG, the water vapor in marine air masses condenses, leading to orographic intensification of precipitation. With two AMIGOS installed on Thwaites Eastern Ice Shelf in January 2020, we are now able to examine unique in situ atmospheric conditions during AR landfalls on Thwaites Eastern Ice Shelf. Case studies such as the February 2020 family of ARs can bring insight into the hydroclimate dynamics responsible for the West Antarctic AR climatologies and quantify examples of SMB impacts.

In February 2020, a family AR event occurred over TG over a period of 5 d, involving the landfall of three distinct ARs in rapid succession. The event began on 2 February, when the first AR made landfall on the western flank of the glacier for 6 h (Fig. 6a and d). A second AR made landfall shortly after, from 3 to 5 February, propagating from west to east across TG (Fig. 6b and e). A third AR made landfall from 7 to 8 February, on the eastern flank of TG (Fig. 6c and f). This AR family event was associated with a persistent high-pressure ridge extending from South America to the Antarctic Peninsula, coupled with a broad low-pressure system northwest of TG. This blocking high acted to keep the low pressure in place, preventing its zonal migration.

During this time, several distinct sea-level pressure minima (nodes) developed within the broader surface low-pressure system, which were associated with the passage of short-wave troughs in the middle troposphere (Fig. 6a–c). Short-wave troughs promote enhanced divergence in the middle troposphere and the subsequent development of localized sea-level pressure minima. These nodes facilitated the formation of ARs along the eastern edge of the broader surface low by enhancing the meridional transport of marine air masses southward. In particular, the first AR is driven by two nodes embedded within the broader surface low-pressure system, L1 and L2, which were associated with short-wave troughs T1 and T2. Subsequently, L1 and T1 dissipate, with the second AR driven by L2 and L3. The development of L3 can be tied to a separate short-wave trough (T3) that propagated around the broader upper-level trough prior to the second AR. The third AR is driven exclusively by L4, which is associated with another short-wave trough (T4) that can be tracked from off the coast of East Antarctica on 2 February to TG on 7 February. Coincident with this family of ARs, a persistent ridge of high pressure, known as a blocking high, forms downstream of the broader surface low-pressure system. The coupling of the high- and low-pressure systems channels each AR into the Amundsen Sea Embayment and onto TG, leading all three ARs to make landfall at the same location. The release of latent heat due to condensation and deposition within each AR amplifies the mid-tropospheric geopotential height anomalies, reinforcing the blocking high in a positive feedback cycle. This evolution permits multiple AR events to occur in rapid succession, as seen in this case study of the February 2020 AR family event.

On Thwaites Eastern Ice Shelf, we observe distinct signals in the AMIGOS surface pressure, as well as near-surface air temperature, wind direction, and wind speed during the AR family event (Fig. 7). From the onset of the first AR, near-surface temperatures increase to the melting point and remain consistently at or above the melting point in observations and ERA5 (MERRA-2 remains below the melting point) for the entire duration of the AR family event. The observed surface pressure repeatedly drops during the AR events, indicating the onset of new low-pressure system nodes associated with and reinforcing each AR. The wind direction experiences rapid changes before the first AR event, and after the third, and shifts to the north during the events. This reflects the synoptic pattern of northerly flow onto TG between the low-pressure system and blocking high. The wind speed accelerates during the first and second events, then drops and rises once more during the third AR landfall.

Figure 7Observations of (a) 2 m temperature, (b) wind direction, (c) surface pressure, and (d) wind speed during the AR family event from AMIGOS at Cavity Camp (blue) and Channel Camp (pink) sites, and MERRA-2 (gray) and ERA5 (black) reanalyses. Pink shading indicates the timing of the AR events, based on the Wille et al. (2019) AR detection algorithm.

Download

Figure 8Reconstructed accumulation at Cavity Camp (blue) and Channel Camp (pink) sites on Thwaites Eastern Ice Shelf during the AR family event. AMIGOS observations of snow height and atmospheric conditions are used to force SNOWPACK with radiation provided by MERRA-2. Light purple and dark purple lines show GNSS–IR-derived accumulation at Lower Thwaites and Upper Thwaites, respectively. We also show accumulation only from MERRA-2 (gray) and ERA5 (black). Pink shading indicates the timing of the AR events.

Download

Next, we examine the impacts of the February 2020 AR family event on the SMB of Thwaites Eastern Ice Shelf (Fig. 8). From our SNOWPACK firn modeling analysis, which converted observed snow heights into accumulation associated with the AR family event, we find a total accumulation of 110 kg m⁻² (or millimeters water equivalent) at Cavity Camp and 97 kg m⁻² at Channel Camp. A rapid increase in accumulation occurs at the onset of the first AR, and continues through the third AR. During this event, Cavity Camp exhibits earlier and higher total accumulation than Channel Camp, possibly due to local topographical differences between the two sites on Thwaites Eastern Ice Shelf. The SNOWPACK-reconstructed accumulation at each site is very similar when using MERRA-2 or ERA5 radiative forcing.

When we compare the SNOWPACK-reconstructed accumulation with the accumulation in MERRA-2 and ERA5 reanalyses, we find that both reanalyses slightly underestimate the total accumulation during this time by 10 %–20 %. MERRA-2 accumulation is 88 kg m⁻², ERA5 accumulation is 87 kg m⁻². The total precipitation in 2020 is 1005 kg m⁻² in MERRA-2 and 983 kg m⁻² in ERA5, so this AR family event represents 9 % of the total annual precipitation in both reanalysis products. While the snow height observations at Cavity Camp and Channel Camp and represent point locations, the accumulation in the reanalyses represents a grid-cell average. Therefore, the reanalyses may partly underestimate local accumulation on Thwaites Eastern Ice Shelf, particularly during extreme events, due to the lower spatial resolution.

Based on the SNOWPACK analysis, we also find small amounts of surface melting associated with the AR family event (not shown). Alongside higher near-surface temperatures in ERA5 compared to MERRA-2 during the event, we find slightly higher amounts and longer duration of surface melt when we use ERA5 radiative forcing. When using MERRA-2 radiative forcing, we find 2 kg m⁻² of surface melt at Cavity Camp and 1 kg m⁻² of total melt at Channel Camp. When using ERA5 radiative forcing, we find 5 kg m⁻² at Cavity Camp and 3 kg m⁻² of total melt at Channel Camp. We find that the melt occurs for 34 h spread across 6 d at Cavity Camp with MERRA-2 radiative forcing and for 59 h spread across 11 d with ERA5 radiative forcing. At Channel Camp, the melt occurs for 26 h over 3 d with MERRA-2 forcing and for 46 h over 7 d with ERA5 forcing. Furthermore, thermistors placed 1.5 m above the battery box at the AMIGOS stations indicate a sub-surface increase in temperatures from −6 ^∘C to the melting point coinciding with the second and third ARs in the AR family event. The rise in firn temperatures indicates surface meltwater percolation through the firn (Fig. A2). Overall, surface melt is nearly 2 orders of magnitude lower than the snowfall, indicating that the primary impact of this AR family event is to contribute snowfall to TG.

Throughout the AR family event, accumulation is consistently higher at the Lower Thwaites GNSS site than at the Upper Thwaites GNSS site (Fig. 8). Lower Thwaites experiences 184 kg m⁻² total accumulation, and Upper Thwaites experiences 94 kg m⁻² total accumulation. The difference in accumulation between the two sites may indicate that Lower Thwaites experiences more orographic precipitation than Upper Thwaites (the sites are 129 km apart and 304 m in vertical distance), and local topographical variation may contribute to the difference as well.