Figure 1a shows the location of the observation site used in this study, China's Great Wall Station (62°13^′ S, 58°58^′ W; 10 m a.s.l.). The station is situated on the coastal margin of the Fildes Peninsula on the western part of King George Island, in the South Shetland Islands, northwest of the Antarctic Peninsula. The construction and operation of the station's surface meteorological instrumentation and observing procedures follow the standards of the World Meteorological Organization (WMO) and the China Meteorological Administration (CMA) (Ding et al., 2020). Great Wall Station was established in 1985, and continuous 24 h automatic observations have been conducted since 2002. Atmospheric pressure is measured by a PTB220 barometric pressure sensor installed inside the cabinet of the automatic weather station data acquisition unit. Air temperature and relative humidity are measured by an HMP45D temperature and humidity sensor placed in a Stevenson screen at a height of 2 m above the ground, with relative humidity defined with respect to liquid water. Wind speed and wind direction are measured by an XFY3-1 wind speed and direction sensor installed at a height of 10 m above the ground. Precipitation is measured manually using a standard weighing rain gauge (TQ-SDM6, HY Sounding Inc., Beijing, China), which is installed at a height of 1.5 m above the ground. The 12 h accumulated precipitation is defined as the total precipitation recorded over the preceding 12 h at 00:00 and 12:00 UTC each day. The gauge is not equipped with a dedicated wind shield. Therefore, wind-induced undercatch may occur under windy Antarctic conditions and is acknowledged as a potential source of uncertainty. Visibility is also identified manually by observers in the meteorological room of the scientific building. The meteorological observation field is located northwest of the meteorological room. Since the precipitation observations do not distinguish between solid and liquid precipitation, all cases were carefully screened in the subsequent fitting analysis of Ze–S in Sect. 3.3. Figure 1c1–c3 present, respectively, the daily-mean wet-bulb temperature, daily-mean relative humidity with respect to liquid water (RHw), and daily-mean wind speed recorded by the automatic weather station from 10 March 2024 to 28 February 2026; the wet-bulb temperature was calculated from the measured air temperature in combination with surface pressure and relative humidity.

Deployed at China's Great Wall Station in early 2024, the Micro Rain Radar (MRR; K-band, 24 GHz) is a vertically pointing frequency-modulated continuous-wave (FM-CW) Doppler system (see the lower-left panel of Fig. 1a). During the observation period, the MRR was operated with a native temporal resolution of 10 s, 128 range gates, and a range-gate spacing of 35 m, with a maximum observational height of approximately 4.45 km above ground level. For each time step and range gate, the raw Doppler spectrum was recorded with 64 spectral bins. The maximum unambiguous Doppler velocity was approximately 11.89 m s⁻¹, corresponding to a Doppler velocity bin spacing of about 0.19 m s⁻¹. Because of near-field effects and possible ground contamination, the lowest usable range gate for snowfall statistics was approximately 300 m above ground level.

Figure 1b presents the daily-mean MRR equivalent radar reflectivity (Ze) profiles over the full observation period, providing an overview of precipitation events observed by the radar at Great Wall Station. Together with the daily-mean meteorological variables recorded by the automatic weather station, including wet-bulb temperature, relative humidity, and wind speed (Fig. 1c1–c3), Fig. 1 provides a general view of both precipitation occurrence and the local atmospheric conditions during the observation period. With the exception of two short data gaps caused by an SD-card malfunction (23–27 August and 10 September 2025), indicated by grey hatched areas in Fig. 1b, the MRR record is essentially continuous, yielding 715 d of effective measurements.

All MRR spectra were processed using a Doppler-spectra-based processing framework following Maahn and Kollias (2012), which is designed to enhance performance under low signal-to-noise conditions typical of snowfall. We also identified and removed stationary noise signals by temporally averaging spectral observations as used by Ding et al. (2022a). The procedure applies robust noise suppression and a dynamic dealiasing strategy, thereby improving the retrieval of weak spectral signatures (including subtle vertical-motion signals) and increasing the detectability of light precipitation. For the snowfall statistics, time steps with no valid return at the lowest usable range gate (∼ 300 m above ground level) were treated as snow-free. In addition, profiles exhibiting a bright band associated with the 0 °C melting layer aloft (Li et al., 2020; Li and Moisseev, 2020) were excluded from the snowfall classification and were not counted as snowfall.

The ERA5 atmospheric reanalysis produced by the European Centre for Medium-Range Weather Forecasts is used in this study. ERA5 is generated with the IFS global forecasting system (CY41R2) coupled with a four-dimensional data assimilation framework, providing a wide range of atmosphere–land–ocean variables on an approximately 30 km horizontal grid and a 137-level vertical discretization; although ERA5 records extend to years earlier than 1979, 1979 is commonly regarded as the earliest reliable year for Southern Hemisphere analyses, and ERA5 is therefore widely used as the successor to ERA-Interim. Hourly ERA5 data from March 2024 to February 2026 is analysed. To ensure consistency with radar-derived vertical precipitation profiles, archived surface precipitation is not used directly. Instead, following the approach described in the Supplement (Sect. S3) of Roussel et al. (2023), model-level solid precipitation flux diagnostics defined at the midpoints of the IFS vertical layers are employed to reconstruct height-resolved snowfall amounts, thereby producing snowfall profile products that are directly comparable to the observed vertical structure.

Through the comparative analysis of ERA5 grid points, we found that the grid point nearest to Great Wall Station is the optimal choice for representing the snowfall characteristics at the station, irrespective of its surface type. The detailed topographic map around Great Wall Station (Fig. A1) also shows that there is no obvious elevated terrain or ridge between the station and the nearest ERA5 grid point. Accordingly, in Sect. 3.3, the detailed comparison between the model or reanalysis results and the observations is carried out for this nearest grid point.

In this study, the observed reflectivity (Ze) is used to quantify the height-dependent occurrence frequency of precipitation at Great Wall Station (Fig. 2a). Here, the occurrence frequency is defined as the fraction of all observational time steps at each height for which a given precipitation category is detected by the radar. The vertical binning follows the native range resolution of the MRR (35 m). Figure 2a clearly shows a monotonic decrease in precipitation occurrence with height, with total precipitation occurrence ranging from 6.7 % to 32.0 % across the sampled column. At 0.3 km a.g.l., snowfall occurs 21.3 % of the time, whereas wet snow and rain account for 10.7 %; at 3.0 km a.g.l., the corresponding frequencies decrease to 3.4 % for snowfall and 3.2 % for wet snow and rain. It should be noted that these occurrence frequencies can be affected by the range-dependent sensitivity of the radar. Since the MRR is vertically pointing, increasing range corresponds to increasing height above the instrument. With increasing range, weak echoes are more likely to fall below the detection threshold, and precipitation occurrence at higher range gates may therefore be underestimated. Consequently, the decrease in occurrence at higher altitudes should not be interpreted solely as a physical reduction in precipitation frequency. To reduce the influence of noisy range gates and potential range-dependent underdetection, the occurrence analysis was restricted to the reliable height range of the MRR, excluding the lowest range gates affected by near-field effects and possible ground contamination, as well as the noisiest upper range gates. Within this retained height range, the lower layers are dominated by snowfall. With increasing height, precipitation occurrence decreases, and the reduction is more pronounced for snowfall – the dominant phase near the surface – so that the difference between snowfall and wet snow/rain progressively narrows. This implies that snowfall is more frequent than rainfall near the surface, whereas at higher levels snowfall and rainfall occur with comparable frequency.

To further characterize the snowfall regime, snowfall-only Ze profiles are examined (Fig. 2b). The median reflectivity during snowfall exhibits only weak vertical variability, remaining close to ∼ 9 dBZ throughout most of the profile. A modest enhancement around ∼ 1.25 km is consistent with microphysical growth processes, including aggregation and riming, as well as vapor-deposition growth and sublimation effects (Planat et al., 2021). Below ∼ 0.7 km, a slight decrease is observed; however, unlike the MRR-based statistics reported at DDU (Wiener et al., 2024), the reduction in Ze beneath 0.7 km is not pronounced at Great Wall Station. Because strong low-level reflectivity decreases at DDU have been linked to katabatic-flow-driven sublimation of falling snow (Grazioli et al., 2017a) , the comparatively weak near-surface decrease at Great Wall Station suggests a limited katabatic influence in the Antarctic Peninsula coastal setting. This interpretation is physically consistent with the katabatic-wind mechanism – dry, downslope flow originating from the interior plateau – given that Great Wall Station is not located within a prominent katabatic outflow corridor. Finally, a comparison of the 25th and 75th percentile envelopes with those reported for DDU indicates that Great Wall Station frequently experiences low-topped, long-lasting, and weak snowfall events. The prevalence of these low-intensity cases increases the contribution of weak low-level returns, leading to a downward tendency in the 25th-percentile Ze profile toward the surface, while leaving the median profile relatively invariant with height.

Near-surface wind speed may influence low-level snowfall reflectivity profiles through turbulence, enhanced sublimation, changes in the particle size distribution of falling snow, and possible blowing-snow processes. Previous studies have noted that blowing-snow effects become non-negligible for wind speeds exceeding 7 m s⁻¹ when deriving Ze–S relationships (Scarchilli et al., 2020; Wiener et al., 2024). Accordingly, wind speed was binned into four regimes: 0–7 m s⁻¹ (Fig. 2c1), 7–10 m s⁻¹ (Fig. 2c2), 10–13 m s⁻¹ (Fig. 2c3), and ≥ 13 m s⁻¹ (Fig. 2c4), in order to examine the variations in snowfall Ze profiles under different near-surface wind-speed conditions. Under the two lower-wind regimes (Fig. 2c1–c2), the snowfall Ze profiles closely resemble the aggregate snowfall statistics, and no pronounced change is observed with increasing wind speed. By contrast, for wind speeds of 10–13 m s⁻¹ (Fig. 2c3), a marked enhancement of low-level reflectivity emerges. When wind speed exceeds 13 m s⁻¹ (Fig. 2c4), a pronounced reduction in Ze occurs below 1 km, accompanied by an abrupt decrease in the relative frequency of snowfall occurrence at 1 km – from about 70 % in Fig. 2c3 to roughly 60 % – and, toward lower altitudes, a noticeably steeper vertical gradient in the occurrence profile. We also added Appendix B to provide additional analyses, including reflectivity statistics stratified by season and wind-speed regime. The seasonal statistics of reflectivity under different wind-speed conditions show similar features (Fig. B1), especially in summer. Furthermore, the two-dimensional distribution of wind speed and visibility for all snowfall cases (Fig. B2) shows that wind speeds ≥ 13 m s⁻¹ are generally associated with reduced visibility. Together with the wind-speed dependence of the reflectivity profiles, these results provide a possible explanation for the decrease in low-level Ze under wind speeds ≥ 13 m s⁻¹, this feature may be related to blowing-snow conditions under strong winds. Stronger winds may enhance low-level sublimation and turbulent mixing, thereby reducing the abundance of larger falling snow particles. In addition, smaller snow particles that have previously been deposited may be resuspended by strong winds, increasing the contribution of low-reflectivity echoes in the lower layers and shifting the distribution of radar returns toward weaker Ze. However, this interpretation remains subject to uncertainty, as MRR measurements below 300 m were excluded from the statistics due to their limited reliability, while seasonal differences in surface snow conditions and the availability of mobilizable snow particles may also affect the wind-speed range over which blowing snow occurs. Therefore, the decrease in low-level Ze under wind speeds ≥ 13 m s⁻¹ is interpreted here as a feature possibly related to strong-wind and blowing-snow-related processes.

In Fig. 3a, a site-specific Ze=aSb relationship for the Great Wall Station MRR is derived by pairing radar equivalent reflectivity Ze (mm⁶ m⁻³) with 12 h snowfall accumulation S (mm per 12 h) obtained from manual observations using a standard precipitation gauge, and by estimating the parameters a and b using a differential-evolution (DE) optimization algorithm. This approach is widely adopted for retrieving snowfall rate from radar reflectivity and has been applied in numerous Antarctic studies (Grazioli et al., 2017b; Scarchilli et al., 2020; Schoger et al., 2021; Souverijns et al., 2017; Wiener et al., 2024). For the fit, Ze at the lowest usable radar range gate (approximately 300 m a.g.l.) is used and converted to linear units (mm⁶ m⁻³). However, not all observations are suitable for constructing a robust Ze–S relation. Snowfall case selection was based on routine gauge precipitation observations, in which precipitation was recorded twice daily at 00:00 and 12:00 UTC, representing the accumulated precipitation over the preceding 12 h, without distinguishing between rainfall and snowfall. To identify cases containing snowfall only, the following screening was applied. For each 12 h time window (00:00–12:00 and 12:00–24:00 UTC), a case was retained only when the corresponding radar file was available, the 12 h gauge accumulation was valid and no less than 1.0 mm, and the radar temporal coverage within the 12 h window was at least 80 %. In addition, to minimize contamination by liquid precipitation, the 12 h accumulation window is required to contain no time steps with wet-bulb temperature Tw>0 °C. Cases containing any valid radar sample with Tw>0 °C were excluded. Only radar samples satisfying both valid reflectivity and Tw≤0 °C were used in the subsequent analysis. In particular, strong horizontal winds can resuspend previously deposited snow, leading to biased gauge accumulations and, consequently, biased regression. Previous studies have noted that blowing-snow effects become non-negligible for wind speeds exceeding 7 m s⁻¹ when deriving Ze–S relationships (Scarchilli et al., 2020; Wiener et al., 2024). Therefore, the cases used in Fig. 3a (black circles) are restricted to snowfall periods with wind speeds of 0–7 m s⁻¹. Applying these filters yields the subset shown by the black points in Fig. 3a. The resulting best-fit parameters are a=166.64 and b=1.76, i.e., Ze=166.64S1.76, plotted as the solid blue line. Following Wiener et al. (2024), we estimated the uncertainty associated with the fitted Ze–S relationship using the regression RMSE in log space. Specifically, the residuals between the observed ln⁡(Ze) values and the fitted ln⁡(Ze) values were used to calculate the regression RMSE. The blue shaded region in Fig. 3a represents the range ln⁡(Ze,fit)±RMSE, converted back to linear Ze units. This uncertainty envelope illustrates the substantial scatter in the Ze–S fitting and provides a visual estimate of the uncertainty of the locally derived relationship. For reference, several published Ze–S relations are also overlaid in Fig. 3a using distinct colors. To assess fitting performance and potential bias, Fig. 3b compares 12 h snowfall accumulations estimated from MRR reflectivity using each Ze–S relation against the corresponding 12 h accumulations recorded by standard precipitation gauge; the color coding matches that in Fig. 3a. Although the locally fitted relation yields a correlation coefficient R=0.63, it achieves the smallest root-mean-square error (RMSE = 0.78) and the lowest bias among the tested relations. Among the published relations, those of Scarchilli et al. (2020), Ze=54S1.15 (yellow), and Schoger et al. (2021) Ze=77.61S1.22 (green), perform comparatively well, both producing R=0.65 with RMSE values of 1.98 and 1.17, respectively.

Building on the fitted relationship Ze=166.64S1.76, together with the power-law relations proposed by Scarchilli et al. (2020), and Schoger et al. (2021), cumulative snowfall profiles at Great Wall Station were computed for 10 March 2024 to 28 February 2026 (Fig. 3c) and compared with the cumulative snowfall profile derived from ERA5 (Fig. 3c). Figure 3c indicates that ERA5 substantially underestimates cumulative snowfall relative to the MRR-based estimates: the ERA5 maximum occurs near 300 m a.g.l. and reaches only ∼ 380 mm, which is not only smaller than the MRR-derived accumulation at the same height but also lower than the surface accumulation recorded by the automatic weather station (∼ 570 mm). In contrast, the MRR-derived accumulation at the lowest usable gate (∼ 300 m a.g.l.) exceeds the gauge total for all tested Ze–S relations; using the locally fitted relation yields an accumulation of ∼ 620 mm at 300 m. This result highlights the value of profiling radar observations for snowfall quantification, as the vertically resolved radar retrieval provides a more robust estimate of snowfall accumulation than surface gauge records that are susceptible to wind-related biases. ERA5 reproduces the general increase in cumulative snowfall from the top of the profile down to approximately 300 m, consistent with the MRR-derived profiles, but the increase is weaker, i.e., the vertical gradient is insufficiently steep. Moreover, ERA5 exhibits a decrease in cumulative snowfall below ∼ 300 m, consistent with strong near-surface depletion associated with sublimation, whereas the MRR-based profiles do not show a comparable near-surface reduction, suggesting a limited imprint of katabatic-flow-driven sublimation at Great Wall Station. This contrasts with previous work at Dumont d'Urville (DDU), where ERA5 cumulative snowfall profiles typically peak at ∼ 0.8–1.0 km before decreasing toward the surface (Grazioli et al., 2017a; Roussel et al., 2023; Wiener et al., 2024). At Great Wall Station, by comparison, the ERA5 maximum is attained much lower – near ∼ 300 m – before any decrease is simulated. Taken together with the snowfall reflectivity statistics discussed in Sect. 2.3.1, these results support the conclusion that katabatic influence on near-surface snowfall at Great Wall Station is comparatively weak.