[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-514020-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-514020-en":131},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","uncertainty-quantication-for-data-driven-weather-models-large-scale-deep-learning-for-the-earth-system-workshop","Uncertainty Quantiﬁcation for Data-driven Weather Models - Large-scale Deep Learning for the Earth System Workshop","","Presentation content covers uncertainty quantification for data-driven weather forecasting, contrasting deterministic predictions with probabilistic, calibrated alternatives. It motivates generating ensembles via randomly perturbed initial conditions and discusses post-hoc methods to transform deterministic outputs into probabilistic forecasts. The material details a setup using FourCastNet (and PrecipNet), sub-domain constraints, and an ensemble benchmark. It further explains isotonic distributional regression via EasyUQ and neural network-based distributional regression, using losses such as CRPS and reporting preliminary results for multiple target variables.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/uncertainty-quantication-for-data-driven-weather-models-large-scale-deep-learning-for-the-earth-system-workshop/514020/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/uncertainty-quantication-for-data-driven-weather-models-large-scale-deep-learning-for-the-earth-system-workshop/514020.png","ImageObject",300,407,{"name":42,"@type":43},"hdkawk","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-11","2026-10-06",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",17,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"Why does the presentation focus on probabilistic forecasts instead of deterministic ones?","Question",{"text":63,"@type":64},"It argues that efforts in data-driven weather models have mostly produced deterministic forecasts, and that such deterministic outputs are insufficient without quantified uncertainty. The goal is to produce calibrated probabilistic forecasts.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How are ensemble members generated for the data-driven weather model setup?",{"text":68,"@type":64},"An ensemble of initial conditions is created by adding random perturbations derived from differences between two randomly chosen atmospheric analyses within the same month. These perturbations are scaled using a tuning parameter and energy-norm-based formulation.",{"name":70,"@type":61,"acceptedAnswer":71},"What does EasyUQ (isotonic distributional regression) do in this workflow?",{"text":72,"@type":64},"EasyUQ transforms deterministic model outputs into calibrated predictive distributions using training pairs of outputs and outcomes. It relies on a monotonicity assumption in stochastic order and avoids selecting tuning parameters.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},514020,1791403273,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,120,123,127],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":116,"doc_module":4,"doc_module_name":25,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":25,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":124,"doc_module":4,"doc_module_name":25,"category_name":125,"show_sort_weight":124,"slug":126},10,"Lifestyle","lifestyle",{"id":128,"doc_module":4,"doc_module_name":25,"category_name":129,"show_sort_weight":97,"slug":130},19,"General","general",{"code":4,"msg":82,"data":132},{"doc_id":79,"user_id":133,"nickname":42,"user_avatar":134,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":140,"language":141,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":12,"update_tm":145,"read_time":146,"preview_page_count":33,"preview_urls":147},3985747859855,"https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c","Uncertainty Quantiﬁcation for Data-driven Weather Models  \nLarge-scale Deep Learning for the Earth System Workshop  \nSebastian Lerch, based on joint work with Christopher Bülte, Nina Horat and Julian Quinting | September 2023  \nKIT – The Research University in [the Helmholtz Association](the Helmholtz Association www.kit.edu)[ www.kit.edu](the Helmholtz Association www.kit.edu)  \nFrom deterministic to probabilistic forecasts  \nMotivation: Efforts in data-driven weather models have mostly focused on deterministic forecasts, but:  \n“Deterministic forecasts are useless.”   \nUncertainty quantiﬁcation approaches:  \n (probabilistic (e.g. generative) data-driven weather models)  \n generation of an ensemble of predictions by running a deterministic data-driven model based on  \n randomly perturbed initial conditions  \n initial conditions of an NWP ensemble model  \n suitably generated initial conditions based on past data  \n statistical/AI methods for post-hoc uncertainty quantiﬁcation to generate probabilistic forecasts from deterministic predictions  \n isotonic distributional regression / EasyUQ  \n neural network-based probabilistic post-processing  \nSetup  \n FourCastNet (+ PrecipNet for precipitation) forecasts, up to a lead time of 192h in 6h steps, 0 .25􀀎 grid  \n 51-member ensemble with initial condition based on random (Gaussian) perturbations as in the original publication; and the initial conditions of the IFS ensemble  \n operational ECMWF ensemble as a benchmark  \n restriction to sub-domain (ECMWF European Grid) due to limited disk space and computing resources  \n 2018–2022 period, with 2018–2021 as training data for the (post-hoc) UQ methods and 2022 as test set  \n six target variables  \nRandom ﬁeld perturbation-based initial conditions  \nAn ensemble of “data-driven” initial conditions is generated by adding random perturbations based on the difference between two randomly chosen atmospheric states (i.e. analyses):  \n􀀋  ad1 􀀀 ad2  ;  \nj ad1 􀀀 ad2 j Etot  \nwhere adi is the state vector of the analysis from the date di , j 􀀁 j Etot is the total energy norm, and 􀀋 is a tuning parameter.  \nHere, we set 􀀋 = 0:5 􀀁 107 and select d1 ; d2 randomly from the same month within the 2018–2021 period.  \nMagnusson, L., Nycander, J. and Källen, E. (2009) . Flow-dependent versus ﬂow-independent initial perturbations for ensemble prediction, Tellus A, 61A, 194–209 . DOI:10.1111/j . 1600-0870 . 2008 . 00385 . x.  \nIsotonic distributional regression (IDR) / EasyUQ  \nEasyUQ, a special case of isotonic distributional regression (IDR; Henzi et al. 2019, arXiv:190903725), aims at transforming real-valued model output into calibrated statistical distributions, based solely on training data of model output-outcome pairs (xi ; yi); i = 1 ; :::; n.  \nKey assumption: conditional distributions of the outcome Fx (y ) = P (Y 􀀔 yjX = x) are increasing in stochastic order, i.e. , Fx (y ) 􀀕 Fx0 (y ) if x 􀀔 x0 .  \nThen, statistically optimal (in ﬁnal samples) and calibrated predictive distributions are obtained via  \n^Fxj (y ) = k1 :n: :;j lj ;a::n l~~ ~~􀀀~~ ~~1k~~ ~~+~~ ~~1 Xik Ifyi 􀀔 yg;  \nwithout requiring the choice of any tuning parameters.  \nWalz, E.-M., Henzi, A., Ziegel, J. and Gneiting, T. (2022) . Easy Uncertainty Quantiﬁcation (EasyUQ): Generating  \nPredictive Distributions from Single-valued Model Output, Working paper, arXiv:2212.08376 .  \n5 Sebastian Lerch: Uncertainty Quantiﬁcation for Data-driven Weather Models  \nµ  \nσ  \nStation information  \nStation  \nembeddings  \nStationbased NWP  \npredictions  \nStation ID  \n1  \nLatent features  \n2  \n537  \nInput: deterministic FourCastNet predictions, grid point characteristics  \nOutput: Distribution parameters 􀀖; 􀀛  \nEmbeddings generate local adaptivity  \nCRPS as a mathematically principled loss function  \nDistributional regression networks (DRN) for probabilistic post-processing  \nRasp, S. and Lerch, S. (2018) . Neural networks for post-processing ensemble weather forecasts, Monthly Weather Review, 146, 3885–","cbCaie41dn8mln7L","https://ap.wps.com/l/cbCaie41dn8mln7L","pdf",4199447,14,"English","# From deterministic to probabilistic forecasts\n## Motivation: deterministic forecasts and the need for uncertainty\n## Uncertainty quantification approaches\n# Setup\n## FourCastNet/PrecipNet ensemble configuration and benchmarks\n## Random field perturbation-based initial conditions\n# Isotonic distributional regression (IDR) / EasyUQ\n## Core goal and key assumption\n# Distributional regression networks (DRN) for post-processing\n## CRPS-based calibration and preliminary results","[{\"question\":\"Why does the presentation focus on probabilistic forecasts instead of deterministic ones?\",\"answer\":\"It argues that efforts in data-driven weather models have mostly produced deterministic forecasts, and that such deterministic outputs are insufficient without quantified uncertainty. The goal is to produce calibrated probabilistic forecasts.\"},{\"question\":\"How are ensemble members generated for the data-driven weather model setup?\",\"answer\":\"An ensemble of initial conditions is created by adding random perturbations derived from differences between two randomly chosen atmospheric analyses within the same month. These perturbations are scaled using a tuning parameter and energy-norm-based formulation.\"},{\"question\":\"What does EasyUQ (isotonic distributional regression) do in this workflow?\",\"answer\":\"EasyUQ transforms deterministic model outputs into calibrated predictive distributions using training pairs of outputs and outcomes. It relies on a monotonicity assumption in stochastic order and avoids selecting tuning parameters.\"}]","Uncertainty Quantiﬁcation for Data-driven Weather Models - Large-scale Deep Learning for the Earth System Workshop | PDF",1791295200,35,[148,152,156,159],{"page_no":22,"url":149,"width":150,"height":151},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/ff88a513bc439f2b67b85f7fef526bcf.webp",983,552,{"page_no":26,"url":153,"width":154,"height":155},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/824ff4921eb0685a4baff302c71b6acc.webp",982,551,{"page_no":30,"url":157,"width":158,"height":155},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/fdcadb83a0af1a8451cb829ab84f951d.webp",981,{"page_no":33,"url":160,"width":161,"height":162},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-17/60e69f4b5fd5df637c5f284c716c4cc2.webp",980,550]