Single-precision in the tangent-linear and adjoint models of incremental 4D-VAr

The use of single-precision arithmetic in ECMWF’s forecasting model gave a 40% reduction in wall-clock time over double-precision, with no decrease in forecast quality. However, using reduced-precision in 4D-Var data assimilation is relatively unexplored and there are potential issues with using sin...

Бүрэн тодорхойлолт

Номзүйн дэлгэрэнгүй
Үндсэн зохиолчид: Hatfield, S, McRae, A, Palmer, T, Düben, P
Формат: Journal article
Хэл сонгох:English
Хэвлэсэн: American Meteorological Society 2020
_version_ 1826299829516500992
author Hatfield, S
McRae, A
Palmer, T
Düben, P
author_facet Hatfield, S
McRae, A
Palmer, T
Düben, P
author_sort Hatfield, S
collection OXFORD
description The use of single-precision arithmetic in ECMWF’s forecasting model gave a 40% reduction in wall-clock time over double-precision, with no decrease in forecast quality. However, using reduced-precision in 4D-Var data assimilation is relatively unexplored and there are potential issues with using single-precision in the tangent-linear and adjoint models. Here, we present the results of reducing numerical precision in an incremental 4D-Var data assimilation scheme, with an underlying two-layer quasigeostrophic model. The minimizer used is the conjugate gradient method. We show how reducing precision increases the asymmetry between the tangent-linear and adjoint models. For ill-conditioned problems, this leads to a loss of orthogonality among the residuals of the conjugate gradient algorithm, which slows the convergence of the minimization procedure. However, we also show that a standard technique, reorthogonalization, eliminates these issues and therefore could allow the use of single-precision arithmetic. This work is carried out within ECMWF’s data assimilation framework, the Object Oriented Prediction System.
first_indexed 2024-03-07T05:07:54Z
format Journal article
id oxford-uuid:da8fef1e-4ab8-4cae-8ff3-9a53cbbf862c
institution University of Oxford
language English
last_indexed 2024-03-07T05:07:54Z
publishDate 2020
publisher American Meteorological Society
record_format dspace
spelling oxford-uuid:da8fef1e-4ab8-4cae-8ff3-9a53cbbf862c2022-03-27T09:04:07ZSingle-precision in the tangent-linear and adjoint models of incremental 4D-VArJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:da8fef1e-4ab8-4cae-8ff3-9a53cbbf862cEnglishSymplectic ElementsAmerican Meteorological Society2020Hatfield, SMcRae, APalmer, TDüben, PThe use of single-precision arithmetic in ECMWF’s forecasting model gave a 40% reduction in wall-clock time over double-precision, with no decrease in forecast quality. However, using reduced-precision in 4D-Var data assimilation is relatively unexplored and there are potential issues with using single-precision in the tangent-linear and adjoint models. Here, we present the results of reducing numerical precision in an incremental 4D-Var data assimilation scheme, with an underlying two-layer quasigeostrophic model. The minimizer used is the conjugate gradient method. We show how reducing precision increases the asymmetry between the tangent-linear and adjoint models. For ill-conditioned problems, this leads to a loss of orthogonality among the residuals of the conjugate gradient algorithm, which slows the convergence of the minimization procedure. However, we also show that a standard technique, reorthogonalization, eliminates these issues and therefore could allow the use of single-precision arithmetic. This work is carried out within ECMWF’s data assimilation framework, the Object Oriented Prediction System.
spellingShingle Hatfield, S
McRae, A
Palmer, T
Düben, P
Single-precision in the tangent-linear and adjoint models of incremental 4D-VAr
title Single-precision in the tangent-linear and adjoint models of incremental 4D-VAr
title_full Single-precision in the tangent-linear and adjoint models of incremental 4D-VAr
title_fullStr Single-precision in the tangent-linear and adjoint models of incremental 4D-VAr
title_full_unstemmed Single-precision in the tangent-linear and adjoint models of incremental 4D-VAr
title_short Single-precision in the tangent-linear and adjoint models of incremental 4D-VAr
title_sort single precision in the tangent linear and adjoint models of incremental 4d var
work_keys_str_mv AT hatfields singleprecisioninthetangentlinearandadjointmodelsofincremental4dvar
AT mcraea singleprecisioninthetangentlinearandadjointmodelsofincremental4dvar
AT palmert singleprecisioninthetangentlinearandadjointmodelsofincremental4dvar
AT dubenp singleprecisioninthetangentlinearandadjointmodelsofincremental4dvar