Description:

Teleconnections from low-frequency modes of climate variability modulate extreme events across the globe. Understanding the structure and evolution of these teleconnections is important for sub-seasonal to seasonal (S2S) to decadal climate predictions as well as preparing for future climate scenarios. However, teleconnection patterns and their impacts on extremes can often be difficult to detect, and thus predict, on various spatial and temporal scales in large climate datasets. Major factors contributing to this challenge are related to the strong intrinsic atmospheric variability as well as the diversity within the low-frequency climate modes like ENSO. A recent example of great prominence is the failed prediction of strong California precipitation during the strong 2015-16 El-Nino event that evolved as a Central Pacific El Nino instead of the canonical (Eastern Pacific) event.

Traditionally, spatial and temporal characteristics of climate modes of variability are quantified using linear approaches like empirical orthogonal functions (EOFs). Limited observational datasets imply large sampling errors for the estimated EOFs, making it a challenge to accurately identify EOF patterns, particularly as they do not necessarily relate to a physical phenomenon. Analyses of teleconnections are also generally limited to capturing only linear dependencies to modes of variability and underlying non-linear relationships often get ignored. Furthermore, extreme events, particularly compound extremes, are often identified using ad-hoc methods largely based on human expertise and lack a universal definition.

Recent advances in AI and ML techniques have shown great promise in the area of pattern recognition, spatial analysis, and tracking problems, especially in large and complex datasets. Studies have demonstrated the usefulness of various ML methods, including convolution neural networks (CNNs), artificial neural networks (ANNs), self organizing maps (SOM), and autoencoders, to detect, track and predict extreme weather events or large-scale teleconnections in observational and modeled climate data. These approaches have also shown advantages in identifying relationships between modes of climate variability and regional climate extremes over other traditional methods.

We invite all submissions related to the use of ML techniques for a broad range of topics including extreme event detection, quantifying relationships between modes of climate variability, teleconnection pattern recognition, and identification of the relationships between large-scale climate modes and regional extremes, leading to advances in S2S and decadal prediction capabilities.