A new tool for identifying boundary conditions in coastal oceanic models : First tryouts with a tidal model

If lateral boundary conditions are crucial for physical modelling of the ocean dynamics, their estimate may lack of accuracy in coastal regions. Data-assimilation has been a long-used tool to improve accuracy, but most of the existing popular methods are difficult to implement. To solve this, we tried a new and an easy-to-implement method to estimate boundary conditions. This method uses data assimilation with a stochastic gradient descent  and successive approximations of the boundary conditions. We  tested it with twin experiments on a tidal model in the lagoon of Ouano, in New-Caledonia. The method was successful and provided robust estimation of the boundary conditions with various settings of subsampling and noise for the pseudo-data. Here we present those results and discuss about how the stochastic gradient descent and the approximations have to be tuned.