Learning model parameters

Numerical models of geophysical phenomena typically contain several tunable parameters. The parameters are not observed and are only crudely known. Traditionally, the numerical values of these model parameters are chosen by manual model tuning, leading to model errors and errors in the prediction. More objectively, parameters can be estimated from observations by the augmented state approach during the data assimilation or by combing data assimilation with machine learning. In this talk, we address the problem of estimating parameters objectively with several approaches, including stochastic neural networks. We show, focusing on the examples inspired by convection permitting numerical weather prediction, that the estimation of parameters combined with data assimilation for the state decreases the initial state errors even when assimilating sparse and noisy observations.

Partecipanti:
Prof. Dr. Tijana Janjić is holding the Heisenberg Professorship for Data Assimilation at the Catholic University of Eichstätt-Ingolstadt in Germany. Her research focuses on advancing data science algorithms, particularly in the field of environmental sciences, where she, for example, incorporates physical conservation laws into complex data assimilation algorithms. This is crucial for improving the accuracy of weather predictions and understanding phenomena like severe weather events.