[M-GI29] Data assimilation: A fundamental approach in geosciences

[MGI29-12] Method of effectively incorporating information from observation profiles into the model

★Invited Papers

*Nozomi Sugiura¹ (1.Japan Agency for Marine-Earth Science and Technology)

Keywords:path signature, data assimilation, observation profile

In 4-dimensional variational data assimilation, it is necessary to compare observed profiles with profiles in the model. Conventionally, temperature and salinity of the corresponding vertical points have been compared, or the modes after linear transformation of the profile have been compared. However, these methods do not properly deal with the order information and intrinsic nonlinearity in each profile as a path, and thus it may be difficult to bring them closer flexibly. In this report, we propose a method in which each profile is converted into a sequence of numbers called path signature and then compared. In other words, by setting a cost function of reducing the difference between the observed and model signatures, we can properly handle the order information and nonlinearity of the paths. In particular, we will focus on the mathematical reasons why this method is considered to be effective.