Single-column model to global reconstruction: theory, nuance, and applications in sea ice data assimilation

Abstract

The assimilation of sea ice observations into numerical sea ice and climate models has garnered increasing interest in the last decade. Appropriately constraining estimates of sea ice concentration and sea ice thickness is a key motivator, from both prediction and state estimate applications. Improved forecasts and more spatiotemporally complete estimates of the sea ice state are desired to better plan for the future and understand the recent past in an environment that is rapidly changing. A growing body of literature explores the use of various data assimilation (DA) methods to appropriately constrain sea ice models with observations, typically of sea ice concentration (SIC) or sea ice thickness (SIT). While other types of observations, including those of sea ice freeboard (FB), are available, few studies have explored beyond SIC and SIT. Even for these commonly assimilated types of sea ice observations, interpreting the results and determining the accuracy of the assimilation analysis is often difficult. First, many modern sea ice models evolve along an ice thickness distribution. The sea ice state is thus described within the DA algorithm in a categorized structure that is notably different from the form of most types of sea ice observations. Second, sea ice is a physically bounded material; as most DA algorithms represent modeled and observed quantities using unbounded Gaussian distributions, there may be intrinsic errors in sea ice DA analyses when these traditional DA algorithms are used to assimilate sea ice observations. The inefficiency of sea ice DA development in large models has slowed progress in exploring and addressing these issues, and likely also limited the degree to which sea ice DA has been applied in novel, but relatively attainable, scientific contexts. This dissertation presents a new sea ice DA framework (CICE-SCM-DART) that couples a single-column sea ice model (Icepack) to an ensemble data assimilation software (DART) and demonstrates its efficacy for sea ice DA development. First, CICE-SCM-DART is used to illustrate the variable impact across sea ice regimes of two commonly assimilated types of observations (SIC and SIT) and an underutilized third type (FB). Next, the framework allows for the testing and verification of new approaches to sea ice DA that acknowledge the bounded and categorized nature of modeled sea ice. Finally, the knowledge accrued from working with CICE-SCM-DART is utilized to produce a novel daily sea ice thickness reconstruction in both the Arctic and Antarctic hemispheres by assimilating never-before-used FB observations into a multi-category sea ice model. The reconstruction highlights the positive impact of FB observations on modeled SIT, particularly in Antarctica, and raises important comparative considerations for the application of sea ice DA in each hemisphere. Though much remains to be explored, the collective work presented in this dissertation makes multiple technical and scientific contributions to the field of sea ice DA by highlighting both the nuances of applying data assimilation to sea ice specifically and the achievable avenues for further development and experimentation.