Data Assimilation: From an Eventful Past to a Bright Future

Data Assimilation: From an Eventful Past to a Bright Future

Michael Ghil, Ecole Normale Supérieure and PSL University, Paris, France, and University of California at Los Angeles, Los Angeles, CA, USA

Vilhelm Bjerknes first described weather prediction as an initial-value problem in 1904. As John von Neumann and associates started using computers to implement this idea immediately after World War II, it quickly became apparent that the requisite initial data available then were incomplete. The appearance of weather satellites in the 1960s led further on to the concept of time-continuous assimilation of remote-sensing data. Nowadays, data assimilation (DA) is being applied across all the areas of the climate sciences and much beyond.

This presentation traces the evolution of DA methodology from the successive corrections and polynomial interpolation of the beginnings through the development of sequential-estimation (“Kalman filtering”) and control-theoretical (“variational” or “adjoint”) methods to today’s machine-learning–aided methods. Key concepts, such as information transfer between variables and between regions, as well as parameter estimation and new areas of applications are emphasized. Cutting-edge developments covered in the presentation touch upon the application of concepts and tools from non-autonomous and random dynamical systems theory, as well as upon combining machine learning with DA and with knowledge-based models for weather and climate prediction.