Abstract:

Operational forecast centers routinely use numerical models to predict various components of the Earth system, ranging from the geosphere to the atmosphere. Despite the diversity in applications, these models share two common characteristics: (1) they generally rely on physical laws to govern the time-rate-of change of prognostic state variables and (2) “data assimilation” guides how environmental measurements inform estimates of initial conditions, boundary conditions, or unknown model parameters. Often, the relative skill of operational models (e.g., comparisons of two or more global weather prediction systems) is dominated by algorithmic choices made during data assimilation, which can be further traced back to assumptions made in priors and likelihoods used when formulating such methods from Bayes’ theorem. This seminar will focus more narrowly on atmospheric applications and discuss major data assimilation obstacles for current high-resolution weather models. In particular, we will discuss the limitations of Gaussian approximations that modern data assimilation methods adopt for prior probability densities and likelihoods and share idealized and real-world examples of where benefits can be achieved through "nonparametric" methods—which do not assume a distribution shape. This seminar will also highlight the many areas where new data assimilation innovations are needed to push current boundaries for environmental prediction.  

Journal article: Poterjoy, J. 2022: Implications of multivariate non-Gaussian data assimilation for multi-scale weather prediction. Mon. Wea. Rev. 150, 1475 – 1493.