An empirical model for probabilistic decadal prediction: global attribution and regional hindcasts

Emma B. Suckling, Geert Jan van Oldenborgh, Jonathan M. Eden, Ed Hawkins

Research output: Contribution to journalArticle

11 Citations (Scopus)
20 Downloads (Pure)

Abstract

Empirical models, designed to predict surface variables over seasons to decades ahead, provide useful benchmarks for comparison against the performance of dynamical forecast systems; they may also be employable as predictive tools for use by climate services in their own right. A new global empirical decadal prediction system is presented, based on a multiple linear regression approach designed to produce probabilistic output for comparison against dynamical models. A global attribution is performed initially to identify the important forcing and predictor components of the model. Ensemble hindcasts of surface air temperature anomaly fields are then generated, based on the forcings and predictors identified as important, under a series of different prediction ‘modes’ and their performance is evaluated. The modes include a real-time setting, a scenario in which future volcanic forcings are prescribed during the hindcasts, and an approach which exploits knowledge of the forced trend. A two-tier prediction system, which uses knowledge of future sea surface temperatures in the Pacific and Atlantic Oceans, is also tested, but within a perfect knowledge framework. Each mode is designed to identify sources of predictability and uncertainty, as well as investigate different approaches to the design of decadal prediction systems for operational use. It is found that the empirical model shows skill above that of persistence hindcasts for annual means at lead times of up to 10 years ahead in all of the prediction modes investigated. It is suggested that hindcasts which exploit full knowledge of the forced trend due to increasing greenhouse gases throughout the hindcast period can provide more robust estimates of model bias for the calibration of the empirical model in an operational setting. The two-tier system shows potential for improved real-time prediction, given the assumption that skilful predictions of large-scale modes of variability are available. The empirical model framework has been designed with enough flexibility to facilitate further developments, including the prediction of other surface variables and the ability to incorporate additional predictors within the model that are shown to contribute significantly to variability at the local scale. It is also semi-operational in the sense that forecasts have been produced for the coming decade and can be updated when additional data becomes available.

Original languageEnglish
Pages (from-to)3115-3138
Number of pages24
JournalClimate Dynamics
Volume48
Issue number9-10
Early online date4 Jul 2016
DOIs
Publication statusPublished - May 2017
Externally publishedYes

Fingerprint

prediction
temperature anomaly
greenhouse gas
surface temperature
sea surface temperature
persistence
air temperature
calibration
climate
ocean
comparison
trend
forecast

Keywords

  • Decadal prediction
  • Empirical modelling
  • Hindcast skill

ASJC Scopus subject areas

  • Atmospheric Science

Cite this

An empirical model for probabilistic decadal prediction : global attribution and regional hindcasts. / Suckling, Emma B.; van Oldenborgh, Geert Jan; Eden, Jonathan M.; Hawkins, Ed.

In: Climate Dynamics, Vol. 48, No. 9-10, 05.2017, p. 3115-3138.

Research output: Contribution to journalArticle

Suckling, Emma B. ; van Oldenborgh, Geert Jan ; Eden, Jonathan M. ; Hawkins, Ed. / An empirical model for probabilistic decadal prediction : global attribution and regional hindcasts. In: Climate Dynamics. 2017 ; Vol. 48, No. 9-10. pp. 3115-3138.
@article{73d86a17645d48bda6d4b064a76cb43c,
title = "An empirical model for probabilistic decadal prediction: global attribution and regional hindcasts",
abstract = "Empirical models, designed to predict surface variables over seasons to decades ahead, provide useful benchmarks for comparison against the performance of dynamical forecast systems; they may also be employable as predictive tools for use by climate services in their own right. A new global empirical decadal prediction system is presented, based on a multiple linear regression approach designed to produce probabilistic output for comparison against dynamical models. A global attribution is performed initially to identify the important forcing and predictor components of the model. Ensemble hindcasts of surface air temperature anomaly fields are then generated, based on the forcings and predictors identified as important, under a series of different prediction ‘modes’ and their performance is evaluated. The modes include a real-time setting, a scenario in which future volcanic forcings are prescribed during the hindcasts, and an approach which exploits knowledge of the forced trend. A two-tier prediction system, which uses knowledge of future sea surface temperatures in the Pacific and Atlantic Oceans, is also tested, but within a perfect knowledge framework. Each mode is designed to identify sources of predictability and uncertainty, as well as investigate different approaches to the design of decadal prediction systems for operational use. It is found that the empirical model shows skill above that of persistence hindcasts for annual means at lead times of up to 10 years ahead in all of the prediction modes investigated. It is suggested that hindcasts which exploit full knowledge of the forced trend due to increasing greenhouse gases throughout the hindcast period can provide more robust estimates of model bias for the calibration of the empirical model in an operational setting. The two-tier system shows potential for improved real-time prediction, given the assumption that skilful predictions of large-scale modes of variability are available. The empirical model framework has been designed with enough flexibility to facilitate further developments, including the prediction of other surface variables and the ability to incorporate additional predictors within the model that are shown to contribute significantly to variability at the local scale. It is also semi-operational in the sense that forecasts have been produced for the coming decade and can be updated when additional data becomes available.",
keywords = "Decadal prediction, Empirical modelling, Hindcast skill",
author = "Suckling, {Emma B.} and {van Oldenborgh}, {Geert Jan} and Eden, {Jonathan M.} and Ed Hawkins",
year = "2017",
month = "5",
doi = "10.1007/s00382-016-3255-8",
language = "English",
volume = "48",
pages = "3115--3138",
journal = "Climate Dynamics",
issn = "0930-7575",
publisher = "Springer Verlag",
number = "9-10",

}

TY - JOUR

T1 - An empirical model for probabilistic decadal prediction

T2 - global attribution and regional hindcasts

AU - Suckling, Emma B.

AU - van Oldenborgh, Geert Jan

AU - Eden, Jonathan M.

AU - Hawkins, Ed

PY - 2017/5

Y1 - 2017/5

N2 - Empirical models, designed to predict surface variables over seasons to decades ahead, provide useful benchmarks for comparison against the performance of dynamical forecast systems; they may also be employable as predictive tools for use by climate services in their own right. A new global empirical decadal prediction system is presented, based on a multiple linear regression approach designed to produce probabilistic output for comparison against dynamical models. A global attribution is performed initially to identify the important forcing and predictor components of the model. Ensemble hindcasts of surface air temperature anomaly fields are then generated, based on the forcings and predictors identified as important, under a series of different prediction ‘modes’ and their performance is evaluated. The modes include a real-time setting, a scenario in which future volcanic forcings are prescribed during the hindcasts, and an approach which exploits knowledge of the forced trend. A two-tier prediction system, which uses knowledge of future sea surface temperatures in the Pacific and Atlantic Oceans, is also tested, but within a perfect knowledge framework. Each mode is designed to identify sources of predictability and uncertainty, as well as investigate different approaches to the design of decadal prediction systems for operational use. It is found that the empirical model shows skill above that of persistence hindcasts for annual means at lead times of up to 10 years ahead in all of the prediction modes investigated. It is suggested that hindcasts which exploit full knowledge of the forced trend due to increasing greenhouse gases throughout the hindcast period can provide more robust estimates of model bias for the calibration of the empirical model in an operational setting. The two-tier system shows potential for improved real-time prediction, given the assumption that skilful predictions of large-scale modes of variability are available. The empirical model framework has been designed with enough flexibility to facilitate further developments, including the prediction of other surface variables and the ability to incorporate additional predictors within the model that are shown to contribute significantly to variability at the local scale. It is also semi-operational in the sense that forecasts have been produced for the coming decade and can be updated when additional data becomes available.

AB - Empirical models, designed to predict surface variables over seasons to decades ahead, provide useful benchmarks for comparison against the performance of dynamical forecast systems; they may also be employable as predictive tools for use by climate services in their own right. A new global empirical decadal prediction system is presented, based on a multiple linear regression approach designed to produce probabilistic output for comparison against dynamical models. A global attribution is performed initially to identify the important forcing and predictor components of the model. Ensemble hindcasts of surface air temperature anomaly fields are then generated, based on the forcings and predictors identified as important, under a series of different prediction ‘modes’ and their performance is evaluated. The modes include a real-time setting, a scenario in which future volcanic forcings are prescribed during the hindcasts, and an approach which exploits knowledge of the forced trend. A two-tier prediction system, which uses knowledge of future sea surface temperatures in the Pacific and Atlantic Oceans, is also tested, but within a perfect knowledge framework. Each mode is designed to identify sources of predictability and uncertainty, as well as investigate different approaches to the design of decadal prediction systems for operational use. It is found that the empirical model shows skill above that of persistence hindcasts for annual means at lead times of up to 10 years ahead in all of the prediction modes investigated. It is suggested that hindcasts which exploit full knowledge of the forced trend due to increasing greenhouse gases throughout the hindcast period can provide more robust estimates of model bias for the calibration of the empirical model in an operational setting. The two-tier system shows potential for improved real-time prediction, given the assumption that skilful predictions of large-scale modes of variability are available. The empirical model framework has been designed with enough flexibility to facilitate further developments, including the prediction of other surface variables and the ability to incorporate additional predictors within the model that are shown to contribute significantly to variability at the local scale. It is also semi-operational in the sense that forecasts have been produced for the coming decade and can be updated when additional data becomes available.

KW - Decadal prediction

KW - Empirical modelling

KW - Hindcast skill

U2 - 10.1007/s00382-016-3255-8

DO - 10.1007/s00382-016-3255-8

M3 - Article

VL - 48

SP - 3115

EP - 3138

JO - Climate Dynamics

JF - Climate Dynamics

SN - 0930-7575

IS - 9-10

ER -