{"id":15038,"date":"2026-10-09T12:57:05","date_gmt":"2026-10-09T10:57:05","guid":{"rendered":"https:\/\/wp.unil.ch\/geoblog\/?p=15038"},"modified":"2026-10-09T12:57:08","modified_gmt":"2026-10-09T10:57:08","slug":"distilling-and-validating-novel-mechanistic-insights-on-weather-extremes-with-machine-learning-and-physics-based-counterfactuals","status":"publish","type":"post","link":"https:\/\/wp.unil.ch\/geoblog\/2026\/10\/distilling-and-validating-novel-mechanistic-insights-on-weather-extremes-with-machine-learning-and-physics-based-counterfactuals\/","title":{"rendered":"Distilling and validating novel mechanistic insights on weather extremes with machine learning and physics-based counterfactuals"},"content":{"rendered":"<figure class=\"wp-block-post-featured-image\"><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"1024\" src=\"https:\/\/wp.unil.ch\/geoblog\/files\/2026\/10\/tam.jpg\" class=\"attachment-post-thumbnail size-post-thumbnail wp-post-image\" alt=\"supertyphoon nepartak barreling toward taiwan viewed by nasa&#039;s m\" style=\"object-fit:cover;\" srcset=\"https:\/\/wp.unil.ch\/geoblog\/files\/2026\/10\/tam.jpg 1200w, https:\/\/wp.unil.ch\/geoblog\/files\/2026\/10\/tam-300x256.jpg 300w, https:\/\/wp.unil.ch\/geoblog\/files\/2026\/10\/tam-1024x874.jpg 1024w, https:\/\/wp.unil.ch\/geoblog\/files\/2026\/10\/tam-768x655.jpg 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><\/figure>\n\n\n<div class=\"wp-block-group has-accent-5-background-color has-background has-global-padding is-layout-constrained wp-block-group-is-layout-constrained\">\n<p class=\"has-accent-5-background-color has-text-color has-background has-link-color wp-elements-1 wp-block-paragraph\" style=\"color:#212121;font-size:clamp(14px, 0.875rem + ((1vw - 3.2px) * 0.49), 19px);\"><em>Th\u00e8se en sciences de la Terre, soutenue le 20 octobre 2026 par Iat Hin Tam, rattach\u00e9 \u00e0 l&rsquo;Institut des dynamiques de la surface terrestre (IDYST) de la FGSE.<\/em><\/p>\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Les techniques d\u2019apprentissage automatique ont \u00e9t\u00e9 largement adopt\u00e9es dans les sciences de l\u2019atmosph\u00e8re et ont r\u00e9volutionn\u00e9 la pr\u00e9vision m\u00e9t\u00e9orologique. Toutefois, ces techniques sont opaques quant \u00e0 leurs processus d\u00e9cisionnels, ce qui limite la confiance qu\u2019on peut leur accorder pour de nouvelles d\u00e9couvertes physiques. Il est crucial de comprendre quand et o\u00f9 l\u2019on peut se fier aux connaissances extraites par le&nbsp;ML pour exploiter le potentiel de cette technologie dans la d\u00e9couverte de nouvelles connaissances en sciences de l\u2019atmosph\u00e8re.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">En se concentrant sur deux types d\u2019\u00e9v\u00e8nements m\u00e9t\u00e9orologiques extr\u00eames \u2013 la cyclogen\u00e8se tropicale et les vents extr\u00eames associ\u00e9s aux temp\u00eates europ\u00e9ennes \u2013 cette th\u00e8se vise \u00e0 isoler des structures spatio-temporelles au sein de propri\u00e9t\u00e9s physiques (for\u00e7age radiatif des nuages dans le cas de la cyclogen\u00e8se tropicale) qui permettent de pr\u00e9dire certains ph\u00e9nom\u00e8nes physiques, comme l\u2019acc\u00e9l\u00e9ration de la phase initiale d\u2019intensification des cyclones tropicaux. Le d\u00e9fi majeur de la d\u00e9couverte de connaissances guid\u00e9e par les donn\u00e9es \u2013 \u00e0 savoir d\u00e9terminer si les mod\u00e8les d\u2019apprentissage automatique ont r\u00e9ellement appris des relations de causalit\u00e9- est relev\u00e9 gr\u00e2ce \u00e0 des exp\u00e9riences de perturbation men\u00e9es avec des mod\u00e8les de pr\u00e9vision num\u00e9rique du temps (PNT), permettant ainsi d\u2019\u00e9lever les connaissances extraites du stade de l\u2019association statistique \u00e0 celui de la causalit\u00e9 av\u00e9r\u00e9e. Ce processus en deux \u00e9tapes, combinant apprentissage de structure fond\u00e9 sur les donn\u00e9es et v\u00e9rification par mod\u00e8les de PNT, constitue un flux de travail fiable de d\u00e9couverte de connaissances assist\u00e9e par l\u2019apprentissage automatique.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Gr\u00e2ce \u00e0 ce flux de travail, nous d\u00e9montrons qu\u2019une anomalie de chauffage par rayonnement de grande longueur d\u2019onde situ\u00e9e \u00e0 moyenne altitude \u2013 un \u00e9l\u00e9ment n\u00e9glig\u00e9 dans les analyses composites classiques- joue un r\u00f4le d\u00e9terminant dans la gen\u00e8se du Typhon Haiyan (2013). Cette anomalie thermique humidifie le c\u0153ur de la temp\u00eate et favorise le d\u00e9veloppement d\u2019une convection profonde, ce qui acc\u00e9l\u00e8re ensuite la gen\u00e8se du Haiyan par l\u2019agr\u00e9gation de tourbillon de basse altitude. La v\u00e9rification \u00e0 l\u2019aide de mod\u00e8les PNT est cruciale pour valider le lien de causalit\u00e9 associ\u00e9 \u00e0 cette structure thermique, identifi\u00e9e comme importante par les mod\u00e8les IA. Ces travaux ouvrent la voie \u00e0 une d\u00e9couverte de connaissances fiable et fond\u00e9e sur les donn\u00e9es dans le domaine de la m\u00e9t\u00e9orologie tropicale.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Outre la v\u00e9rification&nbsp;<em>a posteriori&nbsp;<\/em>des mod\u00e8les de PNT, cette th\u00e8se vise \u00e9galement \u00e0 \u00e9laborer des strat\u00e9gies d\u2019entra\u00eenement permettant d\u2019extraire de mani\u00e8re fiable des informations g\u00e9n\u00e9ralisables. En situant ce travail de d\u00e9veloppement de mod\u00e8les dans le contexte exigeant de la d\u00e9couverte d\u2019\u00e9quations appliqu\u00e9es aux temp\u00eates europ\u00e9ennes \u2013 un domaine caract\u00e9ris\u00e9 par la raret\u00e9 des donn\u00e9es \u2013 la th\u00e8se a permis de d\u00e9finir des strat\u00e9gies d\u2019entra\u00eenement garantissant que les \u00e9quations extraites sont physiquement interpr\u00e9tables et g\u00e9n\u00e9ralisables \u00e0 des temp\u00eates n\u2019ayant pas servi \u00e0 l\u2019entra\u00eenement.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Th\u00e8se en sciences de la Terre<\/p>\n","protected":false},"author":47,"featured_media":15039,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_seopress_titles_title":"","_seopress_titles_desc":"","_seopress_robots_index":"","_seopress_robots_follow":"","_seopress_robots_imageindex":"","_seopress_robots_snippet":"","_seopress_robots_primary_cat":"","_seopress_robots_breadcrumbs":"","_seopress_robots_freeze_modified_date":"","_seopress_robots_custom_modified_date":"","_seopress_robots_canonical":"","_seopress_social_fb_title":"","_seopress_social_fb_desc":"","_seopress_social_fb_img":"","_seopress_social_fb_img_attachment_id":0,"_seopress_social_fb_img_width":0,"_seopress_social_fb_img_height":0,"_seopress_social_twitter_title":"","_seopress_social_twitter_desc":"","_seopress_social_twitter_img":"","_seopress_social_twitter_img_attachment_id":0,"_seopress_social_twitter_img_width":0,"_seopress_social_twitter_img_height":0,"_seopress_redirections_value":"","_seopress_redirections_enabled":"","_seopress_redirections_enabled_regex":"","_seopress_redirections_logged_status":"","_seopress_redirections_param":"","_seopress_redirections_type":0,"_seopress_analysis_target_kw":"","_seopress_news_disabled":"","_seopress_video_disabled":"","_seopress_video":[],"_seopress_pro_schemas_manual":[],"_seopress_pro_rich_snippets_disable_all":"","_seopress_pro_rich_snippets_disable":[],"_seopress_pro_schemas":[],"footnotes":"","_links_to":"","_links_to_target":""},"categories":[49465],"tags":[],"class_list":["post-15038","post","type-post","status-publish","format-standard","has-post-thumbnail","category-theses-soutenues"],"_links":{"self":[{"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/posts\/15038","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/users\/47"}],"replies":[{"embeddable":true,"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/comments?post=15038"}],"version-history":[{"count":1,"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/posts\/15038\/revisions"}],"predecessor-version":[{"id":15041,"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/posts\/15038\/revisions\/15041"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/media\/15039"}],"wp:attachment":[{"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/media?parent=15038"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/categories?post=15038"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wp.unil.ch\/geoblog\/wp-json\/wp\/v2\/tags?post=15038"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}