Pré-Publication, Document De Travail Année : 2024

Distance Learning for Analog Methods

Résumé

Analogs are similar states of a system, occurring at remote times within independent numerical simulations or previous observations. This concept has been developed in atmospheric sciences, and was further used in atmospheric and ocean sciences for forecasting, downscaling, upscaling, extreme event attribution, and many other applications. The distance used to find and rate analogs is a key feature of analog methods. Most studies are based on the Euclidean distance or other pre-defined metrics. In this investigation, we leverage distance learning algorithms originally designed for classification and regression and adapt them for statistical forecasting objectives, using in particular the continuous-ranked probability score as a loss function. Our algorithm allows to jointly optimize three key hyperparameters of analog methods: the feature space, the distance, and the number of analogs used. In particular, this algorithm allows to reduce the feature space dimension while keeping analog ensemble performances as high as possible, a key requirement for small and medium-sized datasets. We test our algorithm on an idealized chaotic system and on a small-size tropical cyclone dataset from meteorological agencies. Our experiments suggest that the optimal distance strongly depends on the forecast horizon and the number of available data, and that our algorithm allows for reasonable performances of analog ensemble methods even for small-size datasets. Our approach is not limited to forecasting and can assist the search for optimal hyperparameters of any analog method, enhancing exploration possibilities and improving overall performances.

Fichier principal
Vignette du fichier
Distance_Learning_for_Analog_Methods-preprint.pdf (3.6 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-04841334 , version 1 (16-12-2024)

Licence

Identifiants

  • HAL Id : hal-04841334 , version 1

Citer

Paul Platzer, Arthur Avenas, Bertrand Chapron, Lucas Drumetz, Alexis Mouche, et al.. Distance Learning for Analog Methods. 2024. ⟨hal-04841334⟩
246 Consultations
271 Téléchargements

Partager

  • More