Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions

Abstract We explore the potential of feed‐forward deep neural networks (DNNs) for emulating cloud superparameterization in realistic geography, using offline fits to data from the superparameterized community atmospheric model. To identify the network architecture of greatest skill, we formally opti...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Griffin Mooers, Michael Pritchard, Tom Beucler, Jordan Ott, Galen Yacalis, Pierre Baldi, Pierre Gentine
Formato:	article
Lenguaje:	EN
Publicado:	American Geophysical Union (AGU) 2021
Materias:	climate modeling cloud superparameterization convection parameterizations machine learning Physical geography GB3-5030 Oceanography GC1-1581
Acceso en línea:	https://doaj.org/article/4930a8d8d4d04a43ac88af26b715590b
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:4930a8d8d4d04a43ac88af26b715590b
record_format	dspace
spelling	oai:doaj.org-article:4930a8d8d4d04a43ac88af26b715590b2021-11-24T08:11:41ZAssessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions1942-246610.1029/2020MS002385https://doaj.org/article/4930a8d8d4d04a43ac88af26b715590b2021-05-01T00:00:00Zhttps://doi.org/10.1029/2020MS002385https://doaj.org/toc/1942-2466Abstract We explore the potential of feed‐forward deep neural networks (DNNs) for emulating cloud superparameterization in realistic geography, using offline fits to data from the superparameterized community atmospheric model. To identify the network architecture of greatest skill, we formally optimize hyperparameters using ∼250 trials. Our DNN explains over 70% of the temporal variance at the 15‐min sampling scale throughout the mid‐to‐upper troposphere. Autocorrelation timescale analysis compared against DNN skill suggests the less good fit in the tropical, marine boundary layer is driven by neural network difficulty emulating fast, stochastic signals in convection. However, spectral analysis in the temporal domain indicates skillful emulation of signals on diurnal to synoptic scales. A closer look at the diurnal cycle reveals correct emulation of land‐sea contrasts and vertical structure in the heating and moistening fields, but some distortion of precipitation. Sensitivity tests targeting precipitation skill reveal complementary effects of adding positive constraints versus hyperparameter tuning, motivating the use of both in the future. A first attempt to force an offline land model with DNN emulated atmospheric fields produces reassuring results further supporting neural network emulation viability in real‐geography settings. Overall, the fit skill is competitive with recent attempts by sophisticated Residual and Convolutional Neural Network architectures trained on added information, including memory of past states. Our results confirm the parameterizability of superparameterized convection with continents through machine learning and we highlight the advantages of casting this problem locally in space and time for accurate emulation and hopefully quick implementation of hybrid climate models.Griffin MooersMichael PritchardTom BeuclerJordan OttGalen YacalisPierre BaldiPierre GentineAmerican Geophysical Union (AGU)articleclimate modelingcloud superparameterizationconvection parameterizationsmachine learningPhysical geographyGB3-5030OceanographyGC1-1581ENJournal of Advances in Modeling Earth Systems, Vol 13, Iss 5, Pp n/a-n/a (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	climate modeling cloud superparameterization convection parameterizations machine learning Physical geography GB3-5030 Oceanography GC1-1581
spellingShingle	climate modeling cloud superparameterization convection parameterizations machine learning Physical geography GB3-5030 Oceanography GC1-1581 Griffin Mooers Michael Pritchard Tom Beucler Jordan Ott Galen Yacalis Pierre Baldi Pierre Gentine Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions
description	Abstract We explore the potential of feed‐forward deep neural networks (DNNs) for emulating cloud superparameterization in realistic geography, using offline fits to data from the superparameterized community atmospheric model. To identify the network architecture of greatest skill, we formally optimize hyperparameters using ∼250 trials. Our DNN explains over 70% of the temporal variance at the 15‐min sampling scale throughout the mid‐to‐upper troposphere. Autocorrelation timescale analysis compared against DNN skill suggests the less good fit in the tropical, marine boundary layer is driven by neural network difficulty emulating fast, stochastic signals in convection. However, spectral analysis in the temporal domain indicates skillful emulation of signals on diurnal to synoptic scales. A closer look at the diurnal cycle reveals correct emulation of land‐sea contrasts and vertical structure in the heating and moistening fields, but some distortion of precipitation. Sensitivity tests targeting precipitation skill reveal complementary effects of adding positive constraints versus hyperparameter tuning, motivating the use of both in the future. A first attempt to force an offline land model with DNN emulated atmospheric fields produces reassuring results further supporting neural network emulation viability in real‐geography settings. Overall, the fit skill is competitive with recent attempts by sophisticated Residual and Convolutional Neural Network architectures trained on added information, including memory of past states. Our results confirm the parameterizability of superparameterized convection with continents through machine learning and we highlight the advantages of casting this problem locally in space and time for accurate emulation and hopefully quick implementation of hybrid climate models.
format	article
author	Griffin Mooers Michael Pritchard Tom Beucler Jordan Ott Galen Yacalis Pierre Baldi Pierre Gentine
author_facet	Griffin Mooers Michael Pritchard Tom Beucler Jordan Ott Galen Yacalis Pierre Baldi Pierre Gentine
author_sort	Griffin Mooers
title	Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions
title_short	Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions
title_full	Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions
title_fullStr	Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions
title_full_unstemmed	Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions
title_sort	assessing the potential of deep learning for emulating cloud superparameterization in climate models with real‐geography boundary conditions
publisher	American Geophysical Union (AGU)
publishDate	2021
url	https://doaj.org/article/4930a8d8d4d04a43ac88af26b715590b
work_keys_str_mv	AT griffinmooers assessingthepotentialofdeeplearningforemulatingcloudsuperparameterizationinclimatemodelswithrealgeographyboundaryconditions AT michaelpritchard assessingthepotentialofdeeplearningforemulatingcloudsuperparameterizationinclimatemodelswithrealgeographyboundaryconditions AT tombeucler assessingthepotentialofdeeplearningforemulatingcloudsuperparameterizationinclimatemodelswithrealgeographyboundaryconditions AT jordanott assessingthepotentialofdeeplearningforemulatingcloudsuperparameterizationinclimatemodelswithrealgeographyboundaryconditions AT galenyacalis assessingthepotentialofdeeplearningforemulatingcloudsuperparameterizationinclimatemodelswithrealgeographyboundaryconditions AT pierrebaldi assessingthepotentialofdeeplearningforemulatingcloudsuperparameterizationinclimatemodelswithrealgeographyboundaryconditions AT pierregentine assessingthepotentialofdeeplearningforemulatingcloudsuperparameterizationinclimatemodelswithrealgeographyboundaryconditions
_version_	1718415794617974784

Assessing the Potential of Deep Learning for Emulating Cloud Superparameterization in Climate Models With Real‐Geography Boundary Conditions

Ejemplares similares