{"title":"Towards replacing precipitation ensemble predictions systems using machine learning","authors":"Rüdiger Brecht, Alex Bihlo","doi":"10.1002/asl.1262","DOIUrl":null,"url":null,"abstract":"<p>Forecasting precipitation accurately poses significant challenges due to various factors affecting its distribution and intensity, including but not limited to subgrid variability. Although higher resolution simulations are often considered to improve precipitation forecasts, it is crucial to note that simply increasing resolution may not suffice without appropriate adjustments to parameterization schemes or tuning. Traditionally, ensembles of simulations are used to generate uncertainty predictions associated with precipitation forecasts, but this approach can be computationally intensive. As an alternative, there is a growing trend towards leveraging neural networks for precipitation prediction, which offers potential computational advantages. We propose a new approach to generating ensemble weather predictions for high-resolution precipitation without requiring high-resolution training data. The method uses generative adversarial networks to learn the complex patterns of precipitation and produce diverse and realistic precipitation fields, allowing to generate realistic precipitation ensemble members using only the available control forecast. We demonstrate the feasibility of generating realistic precipitation ensemble members on unseen higher resolutions. We use evaluation metrics such as RMSE, CRPS, rank histogram and ROC curves to demonstrate that our generated ensemble is almost identical to the ECMWF IFS ensemble, on which our model was trained on.</p>","PeriodicalId":50734,"journal":{"name":"Atmospheric Science Letters","volume":"25 11","pages":""},"PeriodicalIF":2.0000,"publicationDate":"2024-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/asl.1262","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Atmospheric Science Letters","FirstCategoryId":"89","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1002/asl.1262","RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"METEOROLOGY & ATMOSPHERIC SCIENCES","Score":null,"Total":0}
引用次数: 0
Abstract
Forecasting precipitation accurately poses significant challenges due to various factors affecting its distribution and intensity, including but not limited to subgrid variability. Although higher resolution simulations are often considered to improve precipitation forecasts, it is crucial to note that simply increasing resolution may not suffice without appropriate adjustments to parameterization schemes or tuning. Traditionally, ensembles of simulations are used to generate uncertainty predictions associated with precipitation forecasts, but this approach can be computationally intensive. As an alternative, there is a growing trend towards leveraging neural networks for precipitation prediction, which offers potential computational advantages. We propose a new approach to generating ensemble weather predictions for high-resolution precipitation without requiring high-resolution training data. The method uses generative adversarial networks to learn the complex patterns of precipitation and produce diverse and realistic precipitation fields, allowing to generate realistic precipitation ensemble members using only the available control forecast. We demonstrate the feasibility of generating realistic precipitation ensemble members on unseen higher resolutions. We use evaluation metrics such as RMSE, CRPS, rank histogram and ROC curves to demonstrate that our generated ensemble is almost identical to the ECMWF IFS ensemble, on which our model was trained on.
期刊介绍:
Atmospheric Science Letters (ASL) is a wholly Open Access electronic journal. Its aim is to provide a fully peer reviewed publication route for new shorter contributions in the field of atmospheric and closely related sciences. Through its ability to publish shorter contributions more rapidly than conventional journals, ASL offers a framework that promotes new understanding and creates scientific debate - providing a platform for discussing scientific issues and techniques.
We encourage the presentation of multi-disciplinary work and contributions that utilise ideas and techniques from parallel areas. We particularly welcome contributions that maximise the visualisation capabilities offered by a purely on-line journal. ASL welcomes papers in the fields of: Dynamical meteorology; Ocean-atmosphere systems; Climate change, variability and impacts; New or improved observations from instrumentation; Hydrometeorology; Numerical weather prediction; Data assimilation and ensemble forecasting; Physical processes of the atmosphere; Land surface-atmosphere systems.