At this colloquium, we are happy to welcome:

Michael Groom (CSIRO) - online
Interpretable forecasts of ENSO phase at multi-year lead times using entropic learning

Machine learning, in particular deep learning, has shown great potential in outperforming conventional GCMs at predicting ENSO, providing useful forecast skill beyond the Boreal spring predictability barrier and enabling the possibility of issuing ENSO forecasts at multi-year lead times. However, despite these advancements in forecast skill, much less progress has been made on understanding and interpreting why these models are able to make such accurate predictions. In this work, we show that the recently proposed entropy-optimal Sparse Probabilistic Approximation (eSPA) machine learning algorithm is able to accurately forecast the phase of ENSO (i.e. La Niña, Neutral or El Niño) at lead times that are competitive with state-of-the-art deep learning methods (e.g. up to 24 months), while also being substantially more parsimonious in its formulation. This latter point makes it much easier to obtain important insights into the dynamics of ENSO that are being captured when making successful forecasts at these lead times than would otherwise be possible with a “black-box” deep learning method. The interpretability methods presented include composites of precursor patterns, feature importance maps and case studies of reconstructed vs. true precursors for a given target date, all of which provide a complementary picture of the spatio-temporal signals that are being isolated by eSPA in order to make a correct classification of ENSO phase at a particular lead time.

Jana de Wiljes (TU Illmenau)
Challenges in Data Assimilation and Approaches to Address Them

Data assimilation in applications such as numerical weather prediction faces fundamental challenges arising from the nonlinear and often chaotic nature of the underlying dynamics. The extremely high dimensionality of these models, together with their prohibitive computational cost, forces practitioners to rely on coarse approximations. Despite these limitations, modern data assimilation methods have shown remarkable success in accurately tracking complex signals, while the mathematical theory is still catching up to provide rigorous guarantees for their performance and accuracy.

A key challenge moving forward is the development of more faithful representations of uncertainty, both for reliable state estimation and for optimal design problems involving observation strategies and filtering. In this talk, we will discuss these challenges and present state-of-the-art approaches that seek to bridge the gap between practical performance and theoretical understanding.

Nikki Vercauteren (Universität zu Köln)

Scale interactions in the stably stratified atmospheric boundary layer and impacts on the anisotropy and intermittency of turbulence

The atmospheric boundary layer is the lowest part of the atmosphere and is characterized by highly turbulent flows. This layer is the main sink of momentum and kinetic energy and the main source of water vapor, heat and aerosols for the rest of the atmosphere. Modelling exchange processes in this layer accurately is a necessity for the correct representations of large-scale atmospheric dynamics. Atmospheric boundary layers with thermally stable stratification are the least understood type of boundary layers due to suppressed turbulence and the presence of myriads of processes on multiple spatio-temporal scales that modulate the turbulence. Stable boundary layers (SBLs) are however the norm at nighttime and in cold or Polar environments. 

In such stably stratified environments, turbulence is generated by shear, while its development is inhibited by buoyant forces. Due to this interplay, flow regimes with different physical and dynamical characteristics exist. Fully turbulent stable boundary layers, also coined as weakly stable boundary layers, are rather well described by turbulence theory, but the very stable boundary layer is home to unsteady and intermittent turbulence that is less well understood. At high stability in the atmospheric boundary layer, non-turbulent processes on sub-mesoscales (such as dirty waves, drainage flows, etc) become more important, and the flow becomes highly non-stationary. Multiscale data analyses based on different field measurement campaigns show signs of direct energy transfers between sub-mesoscales and turbulent scales, with impacts on the turbulence characteristics. On the one hand, the scale interactions are linked to anisotropic turbulence; on the other hand, turbulence intermittency becomes important when the energy content of the sub-mesoscales becomes an important percentage of the mean kinetic energy. The seminar will discuss the impacts of such improved physical understanding for atmospheric modelling.