Sub-seasonal to Seasonal Arctic Summer Sea Ice Forecasts Using Dynamical Downscaling with the Regional Arctic System Model

The Arctic region has been warming at a rate significantly faster than the global average, leading to an accelerated decline in sea ice. This trend is expected to continue, potentially resulting in a "low-ice regime," which could make sea ice conditions more unpredictable. Anticipating changes in Arctic sea ice and climate states is therefore crucial for guiding various human activities, from natural resource management to risk assessment decisions. While global climate and Earth system models project continuous sea ice decline over decadal time scales, achieving reliable seasonal forecasts remains challenging. To address this, we apply dynamical downscaling with the state-of-the-art Regional Arctic System Model (RASM), which enables us to forecast Arctic sea ice on time scales ranging from weeks to six months. RASM is a fully coupled regional climate model that integrates components for the atmosphere, ocean, sea ice, and land, interconnected through the flux coupler of the Community Earth System Model. In our study, we simulate RASM at a horizontal resolution of 1/12 degree (approximately 9 km) for both the ocean and sea ice, with 45 vertical levels in the ocean and five thickness categories for sea ice. The atmosphere is configured on a 50-km grid with 40 vertical levels, dynamically downscaled from the NOAA/NCEP Climate Forecasting System version 2 (CFSv2) at 72-hour intervals for the upper half of the atmosphere. Monthly ensemble forecasts extending up to six months are generated using initial conditions derived from a fully-coupled RASM hindcast simulation without bias correction and assimilation. This presentation highlights results for September sea ice predictions initialized on April 1, May 1, June 1, July 1, August 1, and September 1, covering pan-Arctic and regional sea ice spatio-temporal conditions from 2012 to 2021. Specifically, we examine how lead time and initial conditions affect the quantitative skill of seasonal predictability for Arctic sea ice and demonstrate skillful predictions of September sea ice up to six months in advance. Overall, our study underscores that enhancing model physics and obtaining more realistic initial conditions are crucial for achieving skillful sub-seasonal to seasonal predictions.