Water isotopes are largely used as a proxy of past climate variations, in particular at Dome C, East Antarctica, where the longest continuous climatic record has been obtained from water isotope measurements in a 3200 m long ice core . Interpreting water isotopic signal in ice cores is however complex due to the impact of the atmospheric water cycle before snowfall and of post-deposition processes. Studying variations of δ18O in atmospheric vapour, precipitations and surface snow is thus critical to understand how large-scale moisture intrusion and local processes interact, to finally produce the signal recorded in ice cores. This is why, since 2011, measurements of the isotopic composition of surface snow and precipitation have been carried out at this location . This motivation also led to the installation of an analyser of water vapour isotopic composition at Dome C for the summer seasons, first in 2014 , and for all summers since December 2018 . The measurement of water vapour isotopic composition is a challenge in such low humidity environment because of the strong influence of mixing ratio on δ18O values .

Between 19 and 23 December 2018, a +17.6 °C temperature anomaly was observed at the Automatic Weather Station (AWS) of French–Italian Concordia station (123.4° E, 75.1° S, 3233 m above sea level (a.s.l.)), located at Dome C, on the East Antarctic Plateau (Fig. ), relative to the December 2018 clear-sky daily cycle averages. On 21 December 2018, the temperature reached -14.7 °C, the second highest value recorded since the installation of the Concordia AWS in 2005 . During this event, humidity levels increased by a factor of three and water vapour δ18O increased by 17 ‰ in 3 d. The primary cause of these anomalous values was the intrusion of warm and moist air from lower latitudes in the form of an atmospheric river (AR), a narrow long band of enhanced water vapour fluxes originating from the sub-tropics and mid-latitudes . This AR was detected using the polar-specific detection algorithm described in , and characterized using 10 d back-trajectory analyses performed every 3 h at 1000 m above ground level (a.g.l.) at Dome C with the FLEXPART Lagrangian Particle Dispersion Model . The event was first identified over Coats Land, west of Dronning Maud Land, on 17 December 2018, before progressing towards the South Pole and Dome C (Fig. ).

ARs are known to have a major impact on the Antarctic surface climate, as they induce surface warming, either through enhanced sensible heat fluxes associated with warm air advection, or via the presence of mixed phase clouds with high supercooled liquid water and ice water content that increase downward longwave radiation towards the surface . In addition to triggering intense surface warming, AR also enable anomalously high inland moisture advection, making them a major contributor to extreme precipitation events over the desert polar plateau . The event of December 2018 recorded at Concordia station is in line with those characteristics, as highlighted the presence of a supercooled liquid cloud at 1500 m above Dome C and of precipitating ice crystals and/or blowing snow 20 December 2018. The impact of atmospheric rivers on water vapour isotopic anomalies has been studied for a specific event recorded in Greenland in July 2012 , during which the large-scale transport of an air mass from Ivittuut to the NEEM drilling site caused a strong positive anomaly in temperature, specific humidity, and δD. However, large-scale transport only accounts for about half of the observed changes in water vapour isotopic composition. At Concordia Station, a +28 ‰ anomaly in water vapour δ18O was observed during the AR event of 15–19 March 2022, which originated from the Indian Ocean sector . While large-scale dynamics appear to be a major driver of surface anomalies in specific humidity and δ18O, this study aims to better understand how local boundary-layer processes shape the surface signal during large-scale events.

In this study, we focus on the Dome C AR event of December 2018, with the aim of disentangling the impact of large-scale advection and boundary layer processes on the surface air vapour isotopic signal. In Sect. , we present our observational dataset, modelling tools and methodology. In Sect. , we describe in detail the December 2018 event, from large scale analysis to local observations at Dome C, and evaluate the LMDZ6iso model during this event. In Sect. , we study the influence of large-scale advection and boundary layer processes on the isotopic signal recorded in the water vapour and surface snow. Finally, we discuss our results in Sect. .

We now use the model to obtain a more detailed understanding of the sources of isotopic variability measured at the surface during the AR event. In the model, variations in specific humidity and water isotopes are governed by the four processes outlined in Sect. : surfaces fluxes, cloud condensation and precipitation sublimation, shallow convection and advection. Our goal is to identify the contributions of these processes to the observed surface variations in specific humidity and δ18O between 19 and 23 December 2018. At Concordia, the diurnal cycle is characterised by maximum solar insolation at 03:00 UTC and minimum insolation at 15:00  UTC (local time at Concordia is UTC+8). The warm temperature anomaly characteristic of the AR event at the surface extends up to 2000 m a.g.l. throughout the study period (Fig. a). This anomaly overlays the diurnal cycle, which is marked by nocturnal surface cooling (Fig. a). For specific humidity, a positive anomaly associated with the AR is observed from 20–21 December within the first 1000 m a.g.l., with a peak intensity on 21 December (Fig. b). Relative humidity exhibits a sharp decrease to approximately 20 % from 18 December, evident above 200 m a.g.l. This is followed by a period of saturation starting on 20 December and lasting until early 22 December, before transitioning back to a drier phase (relative humidity lower than 20 %) on 22 and 23 December, above 300 m a.g.l. (Fig. c).

Based on the model's water vapour tendencies, daytime surface sublimation and the turbulent transport of moisture in the first 200 m a.g.l., extending up to 400 m a.g.l. on 21 and 22 December, result in an enrichment of atmospheric vapour in δ18O (Fig. a, b). During nighttime, the temperature decreases, leading to saturation and condensation near the surface, which causes a reduction in δ18O in surface vapour. The analysis of specific humidity and δ18O tendencies in the surface layer (0–6.7 m a.g.l.) confirms the strong influence of surface sublimation and turbulence on the surface signal, both during the typical diurnal cycle and the AR event (Fig. ). Shallow convection mixes this humidity within the mixed layer (from the surface to 200–400 m a.g.l.) during the day, transporting surface moisture upward. This process decreases δ18O at the surface while enriching vapour at higher altitudes (Figs. c, d and ). Cloud condensation leads to a depletion of both moisture and δ18O around 400 m during the peak of the event, as well as during the post-event diurnal cycle (Fig. e, f). In contrast, the sublimation of clouds and precipitation enriches the air in both moisture and δ18O bellow the cloud, around 200 m a.g.l. Finally, advection transports large-scale moisture enriched in δ18O to the surface starting on 21 December, followed by a decrease in both moisture and δ18O on 22 December (Figs. g, h and ).

We now aim to quantify the relative contributions from large-scale processes (advection) and local processes (surface and atmospheric sublimation) to the specific humidity and δ18O signals during the AR. For this, we track the sources of atmospheric vapour through processes with positive tendencies. We consider that atmospheric vapour is the mixing of 3 buckets corresponding to 3 different sources: surface sublimation, large-scale advection, and cloud and precipitation sublimation (Fig. a, b), following the approach described in Sect. . As shallow convection always transports water from the surface towards the upper layers (shallow convection tendencies always negative in the surface atmospheric layer, Fig. ), it does not contribute to the surface specific humidity as a water source. At the peak of the event, the large-scale advection accounts for 30 % of the moisture amount (0.44 g kg⁻¹), while surface sublimation accounts for 69 % (1.01 g kg⁻¹). Cloud and precipitation sublimation plays a minor role, contributing for 1 % of the surface vapour (0.02 g kg⁻¹). To facilitate the interpretation of the total isotopic signal, we analysed the δ18O anomaly relative to the mean vapour δ18O during the clear-sky diurnal cycles (-62 ‰). The total δ18O anomaly (Fig. d) is computed as the weighted sum of each process’s δ18O anomaly value (Fig. c), scaled by its moisture contribution (Fig. b), according to Eqs. () and (). The δ18O associated with surface sublimation exhibits a pronounced diurnal cycle, closely matching the total δ18O, with a maximum of -51 ‰ at the peak of the event (Fig. c). The δ18O related to large-scale advection and cloud and precipitation sublimation shows a significant depletion on 20 December (-94 ‰ and -85 ‰, respectively), followed by a sharp increase, reaching maximum values of -51 ‰ and -40 ‰, respectively, at the event's peak. During the peak of the event, surface sublimation is the primary contributor of surface vapour increase in δ18O, enriching the mean surface vapour by 7.8 ‰ compared to the mean clear-sky diurnal cycle, which accounts for 68 % of the total increase. Large-scale advection contributes for an additional 3.4 ‰, corresponding to 30 % of the increase. Finally, cloud and precipitation sublimation slightly enriches the vapour by 0.2 ‰, contributing to 2 % of the overall increase. The complete period corresponding to Fig. a and d is presented in Fig. S10.

The December 2018 AR event observed at Concordia, which originated from the Atlantic sector and crossed the Antarctic continent, illustrates the interaction between local processes and large-scale mechanisms that control the isotopic signature of surface water vapour. This is the first atmospheric river event observed at Dome C with available isotopic vapour observations, and the only one recorded over the 2009–2019 period originating from the Atlantic sector . The processes governing the surface vapour δ18O signal are presented in Fig.  for clear-sky conditions (left) and for the AR event (right). Under clear-sky conditions, the diurnal cycle of δ18O is primarily controlled by the surface sublimation and condensation cycle. During the day, surface snow sublimation enriches surface vapour, which is mixed within the boundary layer by shallow convection. At night, cooling of the air leads to condensation both in the atmosphere and at the surface, resulting in a depletion in the vapour δ18O. Additionally, the strong atmospheric stratification at night leads to a decoupling between surface and higher-altitude atmospheric layers. At the peak of the AR event, surface sublimation is identified as the primary source of both specific humidity and vapour δ18O, accounting for approximately 70 % of the total contribution. The second major contribution comes from large-scale moisture advection associated with the AR, accounting for approximately 30 %. These results emphasize that the observed isotopic signal cannot be attributed solely to large-scale atmospheric transport but requires a detailed consideration of local boundary layer processes. Thus, this study deepens our understanding of the local and synoptic processes that shape the vapour isotopic signal in polar regions. We note, however, that in the model there is currently no isotopic fractionation during surface sublimation. While numerous field studies have demonstrated that the sublimation flux is fractionating , the introduction of a fractionation scheme in the model could reduce the contribution of sublimation. Nevertheless, the analysis of the total water budget suggests that surface sublimation makes a substantial contribution to the observed specific humidity anomaly.

To extend our surface analysis, we also performed a water and isotope budget analysis across the entire mixed layer (Fig. S11). Integrating processes over a variable mixed layer height introduces a contribution from the entrainment with upper layers, as explained in Sect. . Applying the same process attribution as for the surface, we find that 49 % of the specific humidity content at the AR peak is attributed to surface sublimation and turbulence, while 38 % is due to advection and 13 % to entrainment (Fig. S12a, b). Since the entrainment term represents a combination of water originating from surface sublimation and advection, we observe that at least half of the mixed layer vapour content comes from surface sublimation, aligning with the surface-level analysis. For δ18O, the mixed layer analysis indicates that 48 % of the contributions originate from advection, 39 % from surface sublimation, and 13 % from entrainment compared to the mean clear-sky diurnal cycles (-62 ‰) (Fig. S12d). By attributing all entrained water to advection, we obtain a contribution to total vapour of 61 % by advection and 39 % by sublimation, which we consider to be the upper bound for the contribution of advection to the mixed layer isotopic signal.

Finally, we perform a process-based analysis for positive and negative tendencies to estimate not only the sources of vapour, but also the processes driving the variability of specific humidity and δ18O during the event. To this end, we compared the unperturbed diurnal cycle prior to the AR event (1–17 December), during which we computed the mean cumulative temporal change in specific humidity associated with each process (Eq. ; Fig. S13a) and in δ18O associated with each process (Eqs.  and ; Fig. S13c). We then examined the anomaly during the AR event (19–24 December) relative to this reference diurnal-cycle period, for both specific humidity and δ18O (Fig. S13b, d). For specific humidity, the first 3 d of the event are characterized by a progressive increase in surface sublimation, to which advection contributes from the third day onward, explaining the main humidity peak on 21 December (Fig. S13b). During the night on 21 December, the surface condensation is partly compensated by increased advection. On 22 December, the combination of surface sublimation and advection explains the second peak in specific humidity. Throughout the event, shallow convection and cloud condensation act to reduce the amount of water in the near-surface atmosphere. For δ18O, we find a slight positive anomaly in surface sublimation during the first 2 d of the event compared with the typical diurnal cycle, followed by a negative surface-flux anomaly from the fourth day onwards. The negative contributions to δ18O from condensation and shallow convection are weaker than during unperturbed diurnal cycles. In addition, advection decreases the vapour δ18O in the model during the peak of the event. Because the amplitude of vapour δ18O during surface sublimation and condensation cycles is overestimated under unperturbed conditions, as shown here and previously discussed in and , the interpretation of isotopic anomalies during the event remains limited, underscoring the need to correct this bias in the LMDZ6iso model (Fig. S13d).

This study demonstrates that the LMDZ6iso model performs well in representing surface meteorological variables during the AR event. With regard to vapour isotopes, the model accurately captures the maximum vapour δ18O during the event, however the overestimated amplitude of vapour δ18O outside the event, discussed above, raises doubts as to whether this agreement reflects the correct physical processes. This overestimated amplitude may result either from excessive daytime enrichment during sublimation and/or from excessive nighttime depletion during condensation. The analysis of humidity and isotopic tendencies using the LMDZ6iso model proves particularly effective in disentangling and quantifying the contributions of processes influencing the simulated isotopic signal. This methodology offers a promising perspective for better understanding the interactions between local mechanisms and large-scale processes within the surface boundary layer, as well as for more effectively interpreting the isotopic signature of atmospheric rivers in water vapour.

To this end, a first step is to improve the representation of isotopic processes in climate models. Recent studies have demonstrated that isotopic fractionation occurs during sublimation , challenging the assumption of non-fractionating fluxes. Given that local sublimation fluxes dominate over large-scale advection in contributing to surface vapour isotopic variations, incorporating isotopic fractionation during sublimation, as in and , is a promising perspective to improve the simulation of water vapour isotopes over Antarctica, as it is expected to reduce the amplitude of isotopic diurnal cycle variations by decreasing vapour enrichment during sublimation. Consequently, this could lower the isotopic contribution of surface sublimation during the AR. Additionally, improving the representation of snow in the LMDZ6iso model could further advance the accuracy of vapour isotopic composition simulations. Currently, the model averages snowfall from the start of the simulation into a single snow bucket, yielding a long-term mean isotopic value that obscures variability associated with recent events. Consequently, fresh precipitation, which directly influences the isotopic composition of vapour through sublimation, is not adequately represented. To address this limitation, future developments will aim at implementing a more detailed snow bucket scheme capable of capturing the isotopic impact of recent precipitation on surface water vapour.

In parallel, a more detailed investigation of snowfall and surface snow evolution during AR events is also needed. Surface snow is expected to reflect both the isotopically enriched snowfall associated with ARs and wind-blown snow, as wind speed exceeded 10 m s⁻¹ during this specific event. At such intensities, drifting snow can be lifted several hundred metres above the surface , where it may readily sublimate, potentially changing the isotopic composition of atmospheric water vapour . Consequently, the sublimation of airborne snowflakes may exhibit a distinct isotopic signature than that of sublimation flux from the surface. As this process is not currently represented in LMDZ6iso, understanding its contribution to the surface isotopic signal would be a valuable next step.

To assess whether our analysis is representative of the interactions between large-scale and local processes, it would be valuable to analyse additional AR events, both at Concordia and other locations, to determine whether the contributions of local and large-scale processes vary depending on the event and location. In terms of detection algorithms, this trans-Antarctic AR highlights the need for an AR detection framework that is not restricted to 85° S. Future developments in polar AR detection will hopefully address this limitation. In our study, the AR crosses the Antarctic Plateau for 3 d, from its arrival in western Dronning Maud Land on 17 December to its passage over Concordia on 19 December. During this period, the AR moves through an extremely dry atmosphere, where its passage induces a strong temperature anomaly, leading to significant surface sublimation. This process may have progressively modified the isotopic composition of the AR. As a result, the AR could gradually reach an isotopic composition similar to that of surface snow, thereby weakening the contribution of advection to the observed δ18O anomaly at the surface in Concordia. To test this further, similar studies should be conducted on other AR events with different moisture origins and at stations located at varying distances from the coast. The March 2022 atmospheric river would be an excellent case study, as a +28 ‰ anomaly in water vapour δ18O was measured at Concordia , with the moisture originating from the Indian Ocean. This would help determine whether the isotopic composition of an AR equilibrates with that of surface snow as it travels across the Antarctic Plateau. Moreover, an isotopic analysis of these events using a Lagrangian approach, combined with a process-based decomposition similar to that of , would provide a complementary perspective on the processes involved during the transport of atmospheric rivers.

Finally, it would be valuable to gain a more detailed understanding of the isotopic composition of water vapour throughout the vertical structure of the atmosphere. In this study, we have highlighted the strong contribution of surface sublimation to the isotopic anomaly at the surface and within the boundary layer. However, evaluating our model at higher altitudes remains crucial. For example, using observational data, it would be necessary to evaluate daytime shallow convection, as this process mixes the boundary layer and transports moisture from the surface to higher altitudes. Recently, used fixed-wing uncrewed aircraft to obtain isotopic observations (δ18O and δD) up to 1500 m a.g.l. over the northeastern Greenland Ice Sheet during summer 2022. This innovative approach revealed an “inverted-C” vertical structure for δ18O and δD under cloudy conditions and a “C-shaped” structure for d-excess between the surface and 1500 m altitude. This pattern reflects the dual influence of local-scale processes, such as surface sublimation and large-scale advection. Reproducing this methodology under clear-sky conditions and during AR events in Antarctica and Greenland would provide a valuable extension to our surface and mixed layer analysis. Furthermore, isotopic measurements at higher altitudes would help better constrain the microphysics of mixed-phase clouds, which are frequently observed over the Antarctic Plateau .