Developing and advancing Seasonal Predictability of Arctic Sea ice

The overall goal of SPARSE is to examine whether more accurate and reliable sea ice information will improve the predictability of Arctic sea ice on seasonal time scales. We use three different methods to explore this question: 1) by using the regional coupled ice-ocean model (ROMS-CICE) forced by atmospheric fields from a global seasonal forecast model, 2) by using an earth system model (NorESM), and 3) by a statistical modeling. Field observations are used to improve some critical physical processes in the models. A regional seasonal forecast system, METROMS Arctic-20km, has been developed based on the coupled ROMS-CICE model METROMS (Kristensen et al., 2017). To improve the sea ice predictability, the combined optimal interpolation and nudging (COIN; Wang et al., 2013) has been developed to assimilate sea ice concentration (SIC) for the multi-category sea ice. The comparison between COIN and ensemble Kalman filter (EnKF) shows comparable improvement in the assimilated results and short-term forecast (Fritzner et al., 2018), but uses about 5% computational resources as the EnKF. The METROMS Arctic-20km participated in the 2019 and 2020 September sea ice extent (SIE) predictions in the Sea Ice Outlook (SIO; https://www.arcus.org/sipn/sea-ice-outlook), denoted as METNO-SPARSE there. The August predictions are quite close to the observations, with predicted SIE of 4.22Mkm2 and 4.3 Mkm2 against observed SIE of 4.30 Mkm2 and 3.92 Mkm2. However, both of the June predictions show a significant overestimate of SIE. From these two years predictions we see that more accurate and reliable sea ice information does improve the prediction skill. However, assimilation of SIC tends to improve the forecast skill by about one month, for longer time the prediction skill drops markedly (Wang et al., in prep.). To further improve sea ice predictability for a longer time, we have been developing the assimilation of sea surface temperature (SST) and sea ice thickness (SIT) during 2020. An optimal merging method has been developed to merge multi-sensor SIC and SIT (Wang et al., 2020). However, due to the pandemic of COVID-19, the assimilation development was delayed, and these results were not integrated in the SIO 2020 sea ice prediction. From the preliminary model experiments, we see the overall ocean mixed layer temperature is considerably corrected by SST assimilation. It is expected that such improvements will enhance the seasonal predictability of Arctic sea ice. The predictability of assimilating SIC and SST is currently under thorough assessment (Wang et al., in prep.). The new prediction will be further evaluated through attending the SIO 2021. A linear regression statistical model has been developed to forecast the monthly SIE. Due to the COVID-19, we were not able to participate in the Sea Ice Outlook 2020. We plan to participate in the Sea Ice Outlook 2021. Snow plays an important role in the sea ice mass balance in the Arctic Ocean. Heavy snow load may hinder ice growth due to its high reflectance and thermal insulation, but it can also increase ice thickness due to snow-ice formation. The impact of snow on Arctic sea ice was investigated using an Ice Mass Balance Buoy deployed on fast ice off the northeast coast of Greenland (Wang et al., 2020), and a pan-Arctic study using reanalysis and ice motion vectors with a 1D sea ice model (Merkouriadi et al., 2020). The first year ice is the main contributor for potential snow-ice formation in the Arctic. With the Arctic sea ice transition from multi-year ice to younger, first year ice, snow-ice as a sink of snow would play a big role in the Arctic sea ice, and shall be considered when predicting the year-round Arctic sea ice. The simulated solar radiation partition using the delta-Eddington radiation scheme has been validated against the observations (Wang et al., 2014, 2016). The simulations successfully reproduced the observed variation before mid-June. However, it markedly underestimated the summer albedo, likely due to the inadequate precipitation in the forcing data (Wang et al., 2019). The simulated transmittance is comparable with the observation. One parameterization is proposed for the transmittance relating to the albedo, snow depth and ice thickness, which agrees well with the observed and simulated transmittance (Wang et al., in preparation).

Three seasonal forecast systems have been developed, i.e. a regional coupled ocean-sea ice model (ROMS-CICE), a statistical model, and a general circulation model (NorESM). The ROMS-CICE model participated in the 2019 and 2020 September Arctic sea ice extent prediction. Its August predictions are close to the observations. A multi-source optimal data merging method was developed, and the combined optimal interpolation and nudging (COIN) method was further developed to assimilate sea ice concentration for multi-category sea ice models. The methods have been incorporated in the high resolution (2.5km) Barents Sea operational sea ice forecast system (https://projects.met.no/ocmod/sea_ice_forecast/barents/). The statistical model can make reasonably accurate predictions for year-round monthly mean Arctic sea ice extent.

Arctic sea ice plays a key role in the global climate system. Sea ice significantly reduces heat and mass exchange between the atmosphere and ocean, and contributes to the global ocean circulation through releasing freshwater and salt. Due to its high reflectance, much of the incoming solar radiation is reflected back to space, inhibiting the primary production within and below the ice. The rapid changes in the Arctic sea ice in recent decades (reduction in extent and thickness, and shifting toward a thinner seasonal ice pack) have increased the accessibility to Arctic waters. This has raised the interest from communities, industry and governments in Arctic fishing, transportation and resources exploration, and led to a rising demand for seasonal prediction of Arctic sea ice. The overall aim of SPARSE is to investigate whether more accurate and reliable information about the sea ice state will improve the predictability of Arctic sea ice on seasonal (1-12 months) time scales. The problem will be addressed with three different approaches: by using a regional ice-ocean model (ROMS-CICE) forced with atmospheric input from a global seasonal forecast model (IFS), by "perfect model" experiments with a global general circulation model (NorESM), and by statistical modeling. In addition, three field campaigns are planned focusing on melt ponds and snow melt observations. The data will be analyzed together with previous observations to form a new sea ice albedo parameterization scheme to represent the evolution of melt ponds and snow melt. This new parameterization scheme will be implemented in CICE model to improve the representation of these processes in the model. The results will be a comprehensive integrated understanding of the predictability of seasonal forecast of Arctic sea ice. Also, a statistical model and a stand-alone ice-ocean model (ROMS-CICE) will be available for operational seasonal forecast of Arctic sea ice.