targeted sensitivity studies with new parameterizations can reveal weaknesses of other parameterizations. These simulations make it possible to analyze the sensitivity of simulation results to some of the decisions made in model development. The weather community may be able to contribute lessons learned for conducting sensitivity studies (Box 3.4).
As the Arctic ice cover transitions to a predominantly thinner state, the decadal model sensitivities among variables affecting ice growth, melt, and movement may change relative to those of the past. Targeted sensitivity studies would help identify which variables and processes are critical to simulating a realistic ice cover in the new state, including the mean climatology, simulated variability of and response to changes in external forcing, and where uncertainties inherent in model parameterizations have the largest influence on simulated sea ice processes. Through collaborative efforts with the model development community, this should feed back into improvements in the physics of the models.
A particularly important issue to address via the model sensitivity studies and intercomparison activities is climate model drift from an observationally-based initialized state, which contaminates predictions on seasonal to decadal timescale. This drift results from systematic biases in coupled climate simulations. Improvements in model simulations are required to address this issue. Research on data assimilation methods and alternative methods of model initialization, for example, by using anomaly fields, will need to be considered. Additionally, various mechanisms to “de-drift” predictions need to be assessed in retrospective studies to determine their utility in realizing useful predictive skill.
Related to the call for targeted model sensitivity studies, there is a need for enhanced capabilities in numerical models to provide useful information on key variables of interest. For example, many climate models do not distinguish between land-fast ice and other sea ice, yet the behavior of land-fast ice is of keen interest to a variety of stakeholders. The date of spring breakup is a particularly influential event for coastal infrastructure and operations, but most models lack sufficient resolution and the specific processes that govern evolution of land-fast ice. Many of these requirements call for predictions that have sufficient spatial detail to resolve the highly varying ice characteristics that occur near the coastline. The nature of seasonal sea ice prediction demands accuracy within a few hundred kilometers of shore and within the marginal ice zone. In addition, forcing from tides and ocean waves may play an important role in sea ice evolution on seasonal to decadal timescales. These factors are not typically considered in large-scale numerical models. Model enhancements that incorporate these and other relevant processes would allow for investigations of their role in sea ice prediction and ultimately result in better predictive skill and more useful information for stakeholders.
One way in which model capabilities can be enhanced is by finer resolution. Recent studies have shown that models with higher horizontal and vertical resolution are able to more realistically simulate certain processes