Thèse Vers une Amélioration des Prévisions Infra-Saisonnières des Rivières Atmosphériques en Europe et au Sud de l'Amérique du Sud. H/F Doctorat.Gouv.Fr

Objectifs : Cette thèse vise à développer un cadre de prévision infra-saisonnière à saisonnière (S2S) pour les RA en Europe et en Amérique du Sud, deux régions où leur impact est majeur mais peu étudié à cette échéance, par rapport à d'autres régions comme l'Amérique du Nord. L'enjeu est d'améliorer la fiabilité des prévisions probabilistes en exploitant :

-L'agrégation temporelle des données (fenêtres de 48 h à hebdomadaires) pour mieux capturer la variabilité des RA.
- Les téléconnexions climatiques (comme l'Oscillation de Madden-Julian, le Mode Annulaire Austral ou l'Oscillation Nord-Atlantique) et les anomalies infra-saisonnières de circulation atmosphérique des moyennes latitudes pour identifier des périodes propices à une prévisibilité accrue.
-L'intelligence artificielle (réseaux de neurones) pour affiner les prévisions en intégrant ces signaux climatiques à large échelle.

Impact attendu : Ce travail fournira les premières climatologies et benchmarks de prévisibilité S2S pour ces régions, ainsi que des outils innovants pour anticiper les événements extrêmes liés aux RA. Les résultats contribueront à des collaborations scientifiques internationales (ECMWF, WCRP) et établiront les bases d'applications opérationnelles pour la gestion des risques hydrométéorologiques. Atmospheric rivers (ARs) are elongated, narrow corridors of intense horizontal water vapor transport in the lower troposphere that play a dual role in the Earth's water cycle: they are simultaneously the primary mechanism for extreme precipitation and flooding in mid-latitude coastal regions and a critical driver of water resource replenishment (Ralph et al., 2019). Over the Chilean and Argentinean Andes, ARs account for 45-60% of annual precipitation along the subtropical coast (32°-37°S) and 40-55% along the mid-latitude west coast (37°-47°S), averaging 35-40 landfalling days per year between 38° and 50°S (Viale et al., 2018). Over Europe, North Atlantic ARs are the principal drivers of extreme precipitation and flooding along the Iberian Peninsula, British Isles, and western France (Lavers et al., 2016). Recent analysis by Scholz and Lora (2025) confirms that AR frequency has increased broadly across mid-latitudes over the 20th century, with hotspots in the North Atlantic and Southern Ocean, underscoring the growing societal need for reliable extended-range AR outlooks in both regions.

Short-range (0-7 day) AR forecasts have improved substantially over the past two decades (DeFlorio et al., 2018; Lavers et al., 2016). However, the subseasonal-to-seasonal (S2S) horizon (~weeks 2 through 6) remains a significant frontier. DeFlorio et al. (2018) established, using the ATRISK object-based framework applied to ECMWF S2S hindcasts, that daily AR prediction skill saturates at 7-10 days, with skill 15-20% higher over Northern Hemisphere ocean basins during boreal winter. DeFlorio et al. (2019) took a decisive next step, demonstrating experimentally that multi-model S2S systems (ECMWF, NCEP, ECCC) show isolated but statistically significant skill for AR activity over the western United States at week-3, particularly for the absence or extreme presence of AR conditions. Zhang et al. (2024), using the GFDL SPEAR system, further showed that aggregating AR activity into weekly frequencies (3-7 day/week) dramatically improves S2S predictability relative to daily snapshots, and identified three dominant predictable modes of AR variability in the North Pacific sector at extended leads. Critically, the role of tropical-extratropical teleconnections-specifically the Madden-Julian Oscillation (MJO)-as sources of S2S AR predictability has been demonstrated over the North Pacific and California coast (Mundhenk et al., 2016; Baggett et al., 2017), but remains largely unexplored for Europe and Southern South America. In the Southern Hemisphere, the Southern Annular Mode (SAM) is an additional low-frequency driver of storm-track variability and moisture transport over the Andes, yet its potential as a predictor of AR activity at S2S leads has not been assessed.

This PhD project addresses a clear and pressing scientific gap: no equivalent S2S AR prediction framework exists for Europe or Southern South America. This project advances the state of the art relative to the foundational work of DeFlorio et al. (2018, 2019), which focused on daily AR snapshots over the North Pacific and US West Coast.
The first and most fundamental departure concerns the temporal aggregation of AR metrics. DeFlorio et al. (2018) characterized skill using daily AR snapshots, which becomes increasingly uninformative at extended leads where S2S models lose phase accuracy but may still capture the probabilistic envelope of multi-day AR activity. This project instead evaluates 48 h, 72 h, and weekly aggregated AR frequency and Integrated Vapor transport (IVT) persistence metrics, following the evidence from Zhang et al. (2024) that higher-frequency AR activity (3-7 days per week) is substantially more predictable at S2S leads than individual daily events. Whether this principle transfers to the North Atlantic and southeastern Pacific corridors remains an open question that motivates the core diagnostic work here. In addition, the treatment of tropical-extratropical teleconnections is expanded. DeFlorio et al. (2018, 2019) considered ENSO as a seasonal modulator but did not systematically exploit intraseasonal drivers. Mundhenk et al. (2016) demonstrated that MJO phase alone enables skillful empirical subseasonal AR landfall prediction over the US West Coast; Baggett et al. (2017) extended this to include the QBO. This project applies both drivers to the European and South American domains and adds the Southern Annular Mode (SAM) as a predictor specific to the Southern Hemisphere. SAM modulates extratropical storm-track variability and moisture transport over the Southern Ocean and Andes at intraseasonal timescales, yet its role in AR predictability at S2S leads has not been systematically assessed, representing a genuine scientific gap.
The geographical focus shifts to two understudied hotspots. Despite hosting some of the highest AR landfall frequencies on Earth (Viale et al. 2018)-and despite the direct socioeconomic dependence of Andean water resources on AR-driven precipitation-neither southern South America nor Europe has been subjected to the kind of systematic S2S AR prediction assessment performed by DeFlorio et al. for North America. Scholz and Lora (2025) have recently confirmed that both the North Atlantic and Southern Ocean sectors are experiencing significant upward trends in AR frequency, making the development of reliable extended-range outlooks for these regions increasingly urgent.
The main goal is to develop and evaluate a prototype S2S atmospheric river prediction framework for Europe and Southern South America, capable of providing probabilistic AR activity outlooks at 2-6 week leadtimes Data will be sourced from ERA5 reanalysis (0.25° resolution, 1940-present) and ECMWF and CNRM subseasonal reforecasts. The ATRISK algorithm will be adapted to detect ARs over aggregated time windows. Probabilistic scores will assess model performance, while conditional analysis will stratify results based on the phases of variability modes. Finally, AI models will be trained to improve forecast accuracy.