Data Assimilation Research Team Events / Media

We explore the definition and modeling of location uncertainty of coarse resolution models based on the snapshots of of the coarse resolution solution. We show the connection and difference between this definition of location uncertainty and Etienne Memin's model of location uncertainty.

Neural ODE is a neural network model which parameterizes the time derivative or dynamics of the hidden state. Neural ODEs have an advantage over discrete-time recurrent neural networks in incorporating irregular input data. The adjoint method is used to compute gradients in Neural ODEs, like four-dimensional variational (4D-Var) data assimilation. We discuss the possible use of Neural ODEs in a model bias correction problem within the data assimilation context.