Assessing the simulation of snowfall at Dumont d'Urville, Antarctica, during the YOPP‐SH special observing campaign

Abstract

The performance of a set of atmospheric models and meteorological reanalyses in the representation of precipitation days in Antarctica is assessed using ground-based observations such as a precipitation gauge and a Micro Rain Radar during the Year Of Polar Prediction Special Observing Period at Dumont d'Urville (November 2018-February 2019), East Antarctic coast. The occurrence of precipitation is overall well predicted, but the number of days and intensity with snowfall are overestimated by the models. This is reflected by high values of bias, probability of detection, and false alarm ratios, in particular for reanalyses, due to too frequent simulated precipitating days. The Heidke skill score shows the overall great contribution of the models in the forecasting of precipitating days, and the best performances are achieved by numerical weather prediction models. The chronology is better represented when the models benefit from the data assimilation of in-situ observations, such as in reanalysis or weather forecasting models. Precipitation amounts at the surface are overestimated by most of the models. In addition, data from a ground-based radar make it possible to evaluate the representation of the vertical profiles of snowfall rate.We can show that an excessive sublimation in the atmospheric boundary layer can compensate for overly strong precipitation flux in the mid and low troposphere. Therefore, the need to expand the measurement of precipitation across the atmospheric column using radars is highlighted, in particular in Antarctica where the cold cloud microphysics is poorly known and observations are particularly rare.

A 7-year record of vertical profiles of radar measurements and precipitation estimates at Dumont d’Urville, Adélie Land, East Antarctica

“…On the other hand, the radome induces some attenuation of the radar transmitted and reflected electromagnetic waves and thus some reduction of the sensitivity. This is in particular discussed in Grazioli et al (2017a) that describes the first year of MRR data, Durán-Alarcón et al (2019) that compares 2 years of the DDU MRR to another MRR deployed at the Princess Elizabeth station (longitude : 23.35, latitude : -71.95), and lastly in Roussel et al (2023) that presents an analysis of precipitation at DDU during the YOPP (Year Of Polar Prediction) southern hemisphere special observing period. A consequence of good operating conditions at DDU station is that, to the authors' knowledge and albeit not the first MRR deployed in Antarctica, the DDU MRR offers the longest quasi continuous data series for the Antarctic region so far.…”

“…The model profiles are precipitation threshold-sensitive, as they produce a large number of very small precipitation events (below 0.01 mm hr −1 ). Palerme et al (2014) proposed a threshold of 0.07 mm per 6 hours to optimize the comparison of ERA-Interim precipitation rates with CloudSat observations, which was also used in Roussel et al (2023) for models evaluation. Converted to mm hr −1 , the models' threshold is 0.012 mm hr −1 .…”

“…This model has already been used for various studies in Antarctica(Krinner et al (2019),Vignon et al (2018)) and in particularLemonnier et al (2021) that opens the way to the evaluation of the representation of the Antarctic precipitation in the model.We use a simulation ranging from November 2015 to December 2021 included with a 96x95 horizontal grid and 95 vertical levels. The grid is refined around Adélie Land, resulting in a resolution of approximately 50 km in the zoom center (see Fig.S1ofRoussel et al (2023) for a map of the grid). The run was nudged with wind, temperature and humidity by the ERA5 reanalysis outside of the zoom area.…”

“…We evaluate the model physics used for CMIP-6, which has not been specifically adaptated and calibrated on Antarctic precipitation. Hourly precipitation profiles are simply extracted at the grid cell nearest DDU of coordinates[140.4, -66.63], as the most accurate representation of snowfall amounts is that of the closest grid point regardless of the surface type(Roussel et al (2023)).…”

“…derived for a two-month period in austral summer have been used (e.g.,Lemonnier et al (2019),Jullien et al (2020),Roussel et al (2023)). The Ze-S relation ofScarchilli et al (2020) Ze = 54 S 1.15 (although for an integration time of 5 minutes) in dashdot yellow on Fig.7fairly well fits the data, although its R 2 score is lower (0.20) as it does not take into account the outliers located in the lower right corner of the scatter plot Souverijns et al (2017).…”