Towards improved representation of sea ice within Antarctic numerical weather prediction

Antarctic sea ice plays key roles in modulating Southern Ocean weather and climate processes. Accurate representation of sea-ice properties is one of the keys for improving predictive skill in polar atmospheric forecasts. However, sea-ice representation is relatively basic (i.e., static sea-ice properties throughout the forecast period and/or with an unrealistic, prescribed thickness/snow cover) in widely-used global and regional forecast models. Poor representation of sea-ice properties in numerical weather prediction (NWP) models may increase errors in forecast surface and near-surface parameters. Although operational weather forecasting centres are increasingly using global coupled atmosphere-ocean-ice models to replace atmosphere-only models for short-term (10 day) weather forecasting, the influence of sea ice on such forecasting has yet to be fully quantified, especially in the Southern Ocean. To address this gap, a polar-specific version of the Weather Research and Forecasting model (Polar WRF) is implemented within a circumpolar Antarctic domain to investigate the impact of daily updates of sea-ice concentration on short-term weather forecasting. A statistically significant improvement in near-surface atmospheric temperature and humidity is shown from +48 hours to +192 hours when assimilating daily sea-ice concentration into the model. Improvement in model performance is enhanced from July to September, which is the period of late sea-ice advance. Regionally, model improvement is shown to occur in most sea-ice regions, although the improvement is strongest in the Ross Sea and Weddell Sea sectors. The surface heat balance also shows remarkable improvement in outgoing radiative heat fluxes and both sensible and latent heat fluxes after 48 hours. This work demonstrates the nonnegligible effect of including daily updates of sea-ice concentration in numerical weather forecasting and indicates the necessity of a fully coupled atmosphere-ocean-ice model in operational high-latitude southern hemisphere weather forecasting. Global atmospheric reanalyses products have been widely used for climate model validation, analysis of atmospheric phenomena and research into long-term trends. In Antarctica, ice-surface temperature (IST) from atmospheric reanalysis has been used as an indicator of ice melt and climate change. However, the performance of Antarctic IST in atmospheric reanalyses is not fully understood due to the paucity of IST observations over Antarctic sea-ice regions. Here, a bias intercomparison for five atmospheric reanalyses is undertaken using 18-years of NASA MODerate resolution Imaging Spectroradiometer (MODIS) satellite thermal infrared-derived IST observations as the validation dataset. A strong and persistent warm bias is found in all reanalyses examined (European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5) and ERA-Interim (hereafter called ‚ÄövÑv¿ERAI‚ÄövÑvp), the Modern- Era Retrospective analysis for Research and Applications, Version 2 (MERRA2), the National Oceanic and Atmospheric Administration (NOAA), i.e., NCEP-DOE Reanalysis 2 (NCEPR2) and the Japan Meteorological Agency (JMA) 55-year Reanalysis (JRA55)). This warm bias is strongest in ERA5 (6.2 K), MERRA2 (6.1 K) and ERAI (5.6 K). Biases are highest in the summertime and near the ice edge. JRA55 shows the lowest IST bias (1.4 K) due to the binary sea-ice concentration (SIC) representation and relatively high (2 m) sea-ice thickness (SIT), compensating for the overestimated radiative fluxes arriving at the surface from poor cloud representation. Further analysis of the IST warm biases reveals two main culprits: a) incorrect cloud properties, and b) inappropriate sea-ice representation in the reanalysis products. By applying more accurate cloud masks to the comparison to reduce bias caused by unaccounted-for cloud (derived from both observed and modelled cloud probability), MERRA2 shows the largest warm bias reduction of 4.2 K and ERA5 shows a reduction of 3.7 K. However, the relatively low positive bias (1.4 K) of JRA55 changes to a high negative bias (-3.6 K) because in clear sky conditions, the radiative heat flux bias can no longer compensate for the reduced upward heat conduction through the ice due to the overly-simple binary SIC representation. A dynamic downscaling to represent the ice cover of ERA5 and JRA55 using Polar WRF shows that differences in sea-ice representation can contribute a 2.2 K bias difference. This research greatly improves our knowledge by filling a gap in the assessment of Antarctic sea-ice IST using the most widely-used atmospheric reanalysis products. It provides guidance for improving the next generation of global atmospheric reanalysis products. Future reanalyses could be improved by considering a more accurate cloud scheme and a more complex sea-ice surface representation. Future development of forecast models should include not only the atmospheric component, but also ocean, sea ice and their interactions. The importance of including daily SIC updates into NWP models has been demonstrated in Chapter 2. Following this, the importance of appropriate sea-ice representation on IST estimation in atmospheric reanalysis is demonstrated. However, the effects of other sea-ice physical parameters (e.g., sea-ice thickness; snow depth on sea ice) on NWP are less well-understood due to the lack of largescale observations of the distribution of these properties. Here, a series of 10-day and daily sensitivity tests of ice surface and near-surface temperature to SIT and snow depth (SNOWH) have been undertaken using the Polar WRF model for the year of 2018. By validating with snow buoy-derived observations of 2 metre air temperature, the NWP output shows a high sensitivity to SIT and SNOWH during the Antarctic winter. MODIS IST is used for validating the daily-initiated NWP experiments. This suite of forecasts indicates considerable IST sensitivity to SIT and SNOWH, i.e., the difference is largest during May and August. Generally, IST bias reduces with higher SIT. In addition to circum-Antarctic constant values of SIT (0.5 to 3.5 m in 0.5 m increments) and SNOWH (lower bounds of 2, 5, 10, 20 and 30 cm), realistic spatial distributions of SIT and SNOWH from the Global Ice-Ocean Modeling and Assimilation System (GIOMAS) and a high resolution version of the ocean‚ÄövÑv¨sea ice model of the Australian Community Climate and Earth System Simulator (ACCESS-OM2-01) are included into Polar WRF, and the IST bias from these runs is compared. The model runs with a more realistic distribution of SIT and SNOWH show no distinct advantage over those with constant values. The model forced with sea ice properties from ACCESS-OM2-01 shows a warmer bias in summer, and the GIOMAS-forced model shows a wintertime colder bias due to the underestimated SIT and SIC in ACCESS-0M2-01 and overestimated SNOWH in GIOMAS, respectively. This study demonstrates the challenge and potential of variable thickness products run using ocean-ice coupled models forced by reanalysis data as inputs for short-term weather forecasting. An optimisation experiment is implemented to indicate the best SIT and SNOWH parameterisation for Polar WRF, both for the entire circumpolar domain and for different regions of Antarctic sea ice. In addition, a method is designed to simulate the spatial distribution of SIT and SNOWH using Polar WRF and MODIS-derived IST, which suggests that Polar WRF has a good representation of sea-ice thermodynamics, while the parameterisation of the snow layer requires further development. This research has greatly improved our knowledge of the limitations associated with overly-simplified sea-ice representation in NWP models and atmospheric reanalyses, and provides potential useful directions for the development of the next generation of NWP models. It also provides the priority (i.e., assimilating of SIC is more important than SIT and SNOWH) and benchmark for the future Antarctic sea-ice component development for short-term weather forecasting models.