Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Statistical Learning Approach to Mediterranean Cyclones

L. Roveri, L. Fery, L. Cavicchia, F. Grotto

TL;DR

The paper addresses detecting Mediterranean cyclones by marrying unsupervised topic modeling with supervised localization. It uses Latent Dirichlet Allocation to extract wind- and pressure-configuration motifs from ERA5 wind data, achieving a drastic reduction in dimensionality. A supervised step then localizes cyclones using the LDA topic weights, achieving up to about $82\%$ classification accuracy and high $R^2$ performance for regression with a 4-layer, 800-neuron network when using a 25-topic representation. The approach offers a robust, data-driven tool for cyclone detection that can complement numerical weather forecasts and may extend to other complex geophysical systems.

Abstract

Mediterranean cyclones are extreme meteorological events of which much less is known compared to their tropical, oceanic counterparts. The raising interest in such phenomena is due to their impact on a region increasingly more affected by climate change, but a precise characterization remains a non trivial task. In this work we showcase how a Bayesian algorithm (Latent Dirichlet Allocation) can classify Mediterranean cyclones relying on wind velocity data, leading to a drastic dimensional reduction that allows the use of supervised statistical learning techniques for detecting and tracking new cyclones.

A Statistical Learning Approach to Mediterranean Cyclones

TL;DR

The paper addresses detecting Mediterranean cyclones by marrying unsupervised topic modeling with supervised localization. It uses Latent Dirichlet Allocation to extract wind- and pressure-configuration motifs from ERA5 wind data, achieving a drastic reduction in dimensionality. A supervised step then localizes cyclones using the LDA topic weights, achieving up to about classification accuracy and high performance for regression with a 4-layer, 800-neuron network when using a 25-topic representation. The approach offers a robust, data-driven tool for cyclone detection that can complement numerical weather forecasts and may extend to other complex geophysical systems.

Abstract

Mediterranean cyclones are extreme meteorological events of which much less is known compared to their tropical, oceanic counterparts. The raising interest in such phenomena is due to their impact on a region increasingly more affected by climate change, but a precise characterization remains a non trivial task. In this work we showcase how a Bayesian algorithm (Latent Dirichlet Allocation) can classify Mediterranean cyclones relying on wind velocity data, leading to a drastic dimensional reduction that allows the use of supervised statistical learning techniques for detecting and tracking new cyclones.
Paper Structure (11 sections, 3 equations, 1 figure, 1 table)

This paper contains 11 sections, 3 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Detecting Mediterranean Cyclones
  3. Describing Mediterranean Cyclones
  4. Detecting Coherent Structures in Fluid Dynamics
  5. Looking for Mediterranean Cyclones in a Weather Dictionary
  6. Latent Dirichlet Allocation
  7. Application of LDA to ERA5 wind data
  8. Results
  9. (Supervised) Statistical Learning for Mediterranean Cyclones
  10. Classification approach
  11. Conclusions

Figures (1)

  • Figure 1: The 25 topics identified by LDA as the ones generating Mediterranean weather configurations. Subfigures show the results of our algorithm. The number of motifs is chosen manually based on two principles: optimality with respect to statistical inference techniques for cyclones location given a weather (wind) map, and the fraction of area covered by the topics. Topic 17 is omitted since it is very similar to the first one.