Weather Type Reconstruction Using Machine Learning Approaches - A Daily CAP9 Series Back to 1728

Aug 3, 2026

Weather types characterize large-scale synoptic patterns and contain valuable information on day-to-day variability, shifts in atmospheric circulation, and related surface impacts. Many reconstructions remain limited in time span and affected by methodological constraints. This study evaluates multiple machine learning approaches for station-based weather type reconstruction over Europe using the nine-class CAP9 clustering. A feedforward neural network achieves the best performance, reconstructing daily CAP9 types back to 1728, with strong validation versus previous statistical methods and close agreement for multiple climatological analyses.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

AccessFree to download

Introduction

Weather Types and Existing Reconstruction Limits

Machine Learning Approaches for Station-Based Reconstruction

CAP9 Weather Type Classification Framework

Results and Validation

Applications and Guidance for Other Classifications

Weather Clim. Dynam., 6, 571–594, 2025 [https://doi.org/10.5194/wcd-6-571-2025](https://doi.org/10.5194/wcd-6-571-2025)[ ](https://doi.org/10.5194/wcd-6-571-2025)© Author(s) 2025 . This work is distributed under the Creative Commons Attribution 4 .0 License. Weather type reconstruction using machine learning approaches Lucas Pﬁster 1,2 , Lena Wilhelm 1,2 , Yuri Brugnara 1,2,a, Noemi Imfeld1,2 , and Stefan Brönnimann 1,2 1 Oeschger Centre for Climate Change Research, University of Bern, Bern 3012, Switzerland 2Institute of Geography, University of Bern, Bern 3012, Switzerland anow at: Empa, Dübendorf 8600, Switzerland Correspondence: Lucas Pﬁster (lucas.pﬁ[ster@unibe.ch](ster@unibe.ch)) Received: 6 May 2024 – Discussion started: 21 May 2024 Revised: 16 December 2024 – Accepted: 28 January 2025 – Published: 21 May 2025 Abstract. Weather types are used to characterise largescale synoptic weather patterns over a region. Long-standing records of weather types hold important information about day-to-day variability and changes in atmospheric circulation and the associated effects on the surface. However, most weather type reconstructions are restricted in their temporal extent and suffer from methodological limitations. In our study, we assess various machine learning approaches for station-based weather type reconstruction over Europe based on the nine-class cluster analysis of principal components (CAP9) weather type classiﬁcation. With a common feedforward neural network performing best in this model comparison, we reconstruct a daily CAP9 weather type series back to 1728. This new reconstruction constitutes the longest daily weather type series available. Detailed validation shows considerably better performance compared to previous statistical approaches and good agreement with the reference series for various climatological analyses. Our approach may serve asa guide for other weather type classiﬁcations. 1 Introduction Weather type (WT) or circulation type classiﬁcations are a widely used tool to characterise the prevailing large-scale synoptic weather patterns over a speciﬁc region (Philipp et al., 2010) . In regions such as Europe, where daily weather is largely governed by transient high-and low-pressure systems, such classiﬁcations prove particularly useful to describe the prevailing atmospheric conditions. WT time series yield important information about variability and changes in atmospheric patterns (Jones et al., 2014 ; Rohrer et al., 2017 ; Kuerová et al., 2017) and the surface effects associated with them (Paegle, 1974 ; O'Hare and Sweeney, 1993 ; Kostopoulou and Jones, 2007 ; Lorenzo et al., 2008 ; Jones and Lister, 2009 ; Casado et al., 2010 ; Küttel et al., 2011) . Various studies have assessed the links between WTs and extreme events such as droughts (Fleig et al., 2010), temperature extremes (Hoy et al., 2020 ; Sýkorová and Huth, 2020), or extreme precipitation and ﬂoods (Mináˇrová et al., 2017 ; Petrow et al., 2009) . Moreover, WT classiﬁcations are applied to evaluate weather forecast model outputs (Stryhal and Huth, 2019 ; Weusthoff, 2011) or for forecasting in the renewable energy sector (Wang et al., 2022 ; Drücke et al., 2021 ; Liet al., 2020), among other uses. The ﬁrst WT classiﬁcations were created by experienced meteorologists, who classiﬁed the atmospheric situation, employing manually drawn weather charts derived from station observations (Hess and Brezowsky, 1952 ; Lamb, 1972 ; Schüepp, 1979) . While these subjective classiﬁcations represent real synoptic features, they are often subject to inconsistencies and ambiguities (e.g. James, 2007 ; Cahynová and Huth, 2009 ; Jones et al., 2014 ; Wanner et al., 2000) . In more recent decades, hybrid (mixed) or objective (automatised) WT classiﬁcations have been introduced that classify atmospheric patterns numerically using various statistical approaches, such as clustering algorithms, class attribution based on a distance measure, or

FAQs

What are weather types, and why do they matter for climate research?

Weather types describe large-scale synoptic patterns over a region. Their long records help quantify day-to-day variability, changes in atmospheric circulation, and associated surface effects.

How does the study reconstruct weather types using machine learning?

The work tests several machine learning approaches for station-based reconstruction over Europe using the nine-class CAP9 weather type classification derived from principal components clustering.

How far back does the new reconstruction extend, and how is it validated?

The best-performing feedforward neural network reconstructs a daily CAP9 series back to 1728. Validation shows substantially better performance than prior statistical approaches and good agreement with the reference series for multiple climatological analyses.

Aug 3, 2026

Upload by River Wang