At this colloquium, we are happy to welcome:
Tim Hewson (ECMWF)
Automating front identification using gridded NWP data – principles, options and products
Frontal analysis has been a bedrock of operational meteorology in the extratropics since the pioneering work of Bjerknes early in 20th century. The advent of numerical weather prediction brought with it the opportunity to replicate manual procedures with automated identification. The main challenge was finding suitable algorithms to copy the somewhat artistic process of drawing smooth lines (the fronts) to signify discontinuities in thermal and other atmospheric fields. Whilst one should never fully supplant human input, there are huge benefits to be had from such automation. For example, one can plot fronts from every member of any ensemble, at frequent intervals, in a fraction of the time at a fraction of the cost.
In this talk I will discuss current and past operational practice in front plotting, and how that relates to standard NWP output fields, going on to describe how carefully formulated mathematical derivatives can capture the procedures established in the minds of synopticians, and embedded in the physical concepts that underpin atmospheric motion. Reference will be made to different approaches in this domain, including very recent, and somewhat different, AI-based initiatives.
There are many applications of this work, from creating climatologies, to documenting and understanding different 3-dimensional frontal structures, to creating ensemble-based products in different classes for forecasters. Examples of each will be presented, with a particular focus on forecast products created at the UK Met Office and at ECMWF, many of which are freely available on websites in real-time.
Peter Spichtinger (Universität Mainz)
Automatic detection of synoptic-scale weather fronts
Synoptic scale weather fronts are ubiquitous elements of extra-tropical weather. Their connection to severe weather conditions such as extreme precipitation or thunderstorms highlights the importance of being able to reliably detect these structures to conduct analyses surrounding their effect within the atmosphere. However, due to the lack of an unambiguous, physical based definition of fronts, their automatic detection is difficult, and it is still common practise to draw fronts manually for analysis purposes. Numerical algorithms based on classical methods (e.g., gradients of temperature) are quite sensitive to the resolution and are usually limited to frontal lines on two dimensional pressure levels, neglecting their three dimensional structure.
In this work, we present a machine learning method (based on a convolutional neural network), which is capable of locating and classifying atmospheric fronts on the current ERA5 dataset. Our method outperforms a common (classical) numerical approach when detecting fronts within both the North American continent and Western Europe. Our model may also be applied globally, where we can show a high correlation of our detected fronts and extreme precipitation.
Additionally, we propose a novel algorithm for the detection of three-dimensional front structures, being – to the best of our knowledge – the first method to allow automatic detection of 3D frontal structure, which can be used for statistical investigations. We can show that our method exhibits structural features as assumed for different types of weather fronts, e.g., temperature gradients and inclination. We finally provide an improved implementation on a GPU architecture to drastically reduce computing time of the method.
Andreas Beckert (Universität Hamburg)
Detection and Visual Analysis of 3D Atmospheric Fronts using the Interactive Visualization Framework Met.3D
Atmospheric fronts are a widely used conceptual model in meteorology, most encountered as two-dimensional (2D) front lines on surface analysis charts. The three-dimensional (3D) dynamical structure of fronts has been studied in the literature by means of "standard" 2D maps and cross-sections, and is commonly depicted in 3D illustrations of idealised weather systems in atmospheric science textbooks.
This talk highlights the benefits of objective 3D front analysis for atmospheric case studies and forecasting. The gradient-based 3D front detection method, combined with modern and interactive visual analysis techniques, enables rapid analysis of complex weather situations and is applicable to state-of-the-art numerical weather prediction (NWP) models, including convection-permitting, kilometre-scale resolutions.
Additionally, automated front-tracking algorithms based on geometric and physical properties are demonstrated to generate time series of frontal attributes. These time series are clustered using distance metrics and k-means to efficiently compare and analyse different weather scenarios.
Integrated into the interactive 3D visualisation framework, Met.3D, 3D front geometries can be combined with other atmospheric features, such as clouds, jet-stream core lines and trajectories, as well as traditional 2D maps and cross-sections. This facilitates detailed exploration of the spatio-temporal evolution of fronts, supporting rapid analysis of case studies of extratropical cyclones and frontal development during cyclogenesis.