Progress with decadal predictions
Francisco J. Doblas-Reyes
Rome, 22 October 2018
EUCP
1
• From a policy-relevant perspective predictions should contribute to the formulation of statements about climate variations for the next 30 years; but 30 years involves merging predictions and projections
• Variability, predictability and prediction are closely linked – Systems aim at predicting the relevant variability up to the
level that predictability estimates suggest is possible – Predictability is non-observable and changes across time
scales, variables and regions • Climate predictions, initialised ensemble simulations up to 10 years,
allow to both phase in the internal variability and correct the forced model response
• Close links exist with climate services (GFCS) as many stakeholders make decisions on interannual to interdecadal time scales
• Predictability and prediction rely on scientific coordination from WCRP’s CLIVAR, WGSIP, DCPP, GC-NTCP, as well as increasingly from WMO’s CBS
Some initial ideas
2
Progression from initial-value problems with weather forecasting at one end and multi-decadal to century projections as a forced boundary condition problem at the other, with climate prediction (sub-seasonal, seasonal and decadal) in the middle. Prediction involves initialization and systematic comparison with a simultaneous reference.
Climate prediction time scales
Adapted from Meehl et al. (2009)
Initial-value driven
Boundary-condition driven
Time
Weather forecasts
Subseasonal to seasonal forecasts (2 weeks-18
months) Decadal forecasts (18
months-30 years) Climate-change
projections
3
Bodegas Torres (a Spanish winery) is looking for new locations for its vineyards (and it’s not the only one doing it). Land is being purchased closer to the Pyrenees, at higher elevation. They are considering acquiring land in South America too, in areas where wine is currently not produced. Bodegas Torres needs local climate information (including appropriate uncertainty assessments) for the vegetative cycle of the vine, which lasts 30-40 years. The user needs to make the decision now.
User requests for the near term
4
Global mean radiative forcing (Wm–2, dashed) and effective radiative forcing (solid) with 1850 as baseline. There is little difference between the RCPs before 2040.
Relative scenario independence
IPCC AR5 WGI (2013)
5
Seasonal-mean air temperature change for the RCP4.5 scenario over 2016-2035 (wrt 1986-2005). Stippling for significant changes, hatching for non-significant. The meridional gradient decreases (it increases at the tropopause).
Near-term projections
IPCC AR5 WGI (2013)
8
Global mean near-surface air temperature over the ocean (one-year running mean applied) from CMIP5 hindcasts. Each system is shown with a different colour. NCEP and ERA40/Int used as reference. Shock and drift is the norm (mainly in full-field initialisation).
Decadal climate predictions
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015
IPCC AR5 WGI (2013)
9
CMIP5 decadal predictions. Global-mean near-surface air temperature and AMV against GHCN/ERSST3b for forecast years 2-5.
CMIP5 decadal predictions
Predictions Historical
simulations
Observations
Atlantic multidecadal variability (AMV)
Global mean surface air temperature (GMST)
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015
Doblas-Reyes et al. (2013, Nat. Comms.)
Initialised simulations reproduce the temperature tendencies and some of the AMV and suggest that initialization corrects the forced model response and phases in internal variability.
10
The AMV
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015
García-Serrano et al. (2014, Clim. Dyn.)
Atlantic multidecadal variability (AMV) pattern from ERSST data (left) and regression of the AMV index on the GPCC precipitation (right) over 1960-2010 using four-year averages.
12
Reliability diagrams of (left) initialised and (right) uninitialised multi-model simulations for basin-wide accumulated cyclone energy (ACE). The results are for 2-9 year averages above the climatological median over 1961-2009. Statistically significant values are in bold. Some of the added value of the predictions is their better management of uncertainty, which leads to increased credibility.
The key role of reliability
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015 Caron et al. (2015, GRL)
13
(Top) Time series of 8-year running mean JJA sub-polar North Atlantic (60-10W, 50-65N) temperature. The right panel shows the residual of the observations and decadal predictions regressed against the non-initialised simulations. (Bottom) Impact of initialisation measured by the difference between initialised and uninitialised correlation for years 2-9 JJA near surface temperature.
A new view at decadal prediction skill
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015 Smith et al. (in prep)
14
Multi-model rolling five-year global-mean air temperature from non-initialised and initialised simulations and probability of exceeding 1.5ºC for either a month (light green) and year (dark green) with respect to pre-industrial (1850-1900) levels.
Decadal prediction and the SR1.5
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015 Smith et al. (2018, GRL)
15
• The climate prediction and climate modelling community benefit from working closer together. But there is a language problem.
• Decadal prediction has skill. • Most of the skill in temperature is forced by both natural and
external forcings. • Hindcasts should be started every year. • It is very important to use the same model configurations as for
other climate experiments; learn about your model, use the hierarchy that is likely available.
• Drift, initial shock and systematic error are a hard reality we have to live with. Anomaly initialisation is not a solution yet.
Some of the CMIP5/AR5 lessons
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015
16
• The contribution to CMIP6 via DCPP • The real-time forecast exchange promoted by the WCRP Grand
Challenge on Near-Term Climate Prediction • The operationalisation of decadal prediction by the WMO
Commission of Climatology; Copernicus Climate Change Service (C3S) might play a role in this
The post-CMIP5 scene
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015
17
• The CMIP6 decadal contribution is managed by DCPP (WGSIP, WGCM, DCVP-CLIVAR).
• Three components: hindcasts, forecasts and predictability exercise
• DCPP benefits from joining CMIP6: – Better understanding of model
error – Control runs for predictability – Infrastructure
• Other MIPs benefit from DCPP: – Reduction of model errors by
understanding drift sources – Forecast quality assessment
CMIP6 decadal prediction
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015
13 18 17
13 18
8 10 10
16 13
9 14 13 12 14
3 12
10 17
7 7 7 9
2 11
1
0 10 20 30
AerChem…C4MIPCFMIPDAMIP
DCPPENSOMIP
FAFMIPGeoMIPGMMIP
HighResMIPISMIP6LS3MIPLUMIP
OCMIP6OMIP
PDRMIPPMIP
RFMIPScenario…SolarMIP
VolMIPCORDEXDynVarGDDEX
SIMIPVIAAB
Participating
Not Participating
Don't Know Yet
18
The multi-model real-time decadal prediction exchange is a research exercise that guarantees equal ownership to the contributors. BSC is one of the four centres recognised as global producers of decadal climate predictions by WMO-CCl.
Real-time decadal climate prediction
19
EC-Earth GLOBAL ORCA12-T1279 (ocean and atmosphere at ~10 km!)
SST NEMO
ORCA12
T2m IFS T1279
High-resolution global modelling
20
• Requests for climate information for the next 30 years comes from a broadening range of users and should be addressed from a climate services perspective. What forecasters provide is still far from what some users demand (even in the absence of skill) and is only part of a complex story.
• Decadal prediction shows signs of providing useful information for some regions and events.
• Different tools are available to provide near-term climate information (global and regional projections, decadal predictions, empirical systems, etc.). Merging all this information into a reliable, unique source is a problem still not solved.
• Dynamical models still have substantial errors to be understood and communicated and have to deal with a substantial drift.
• None of this will materialize without appropriate investment in observational networks, increased collaboration and reduction of all aspects of model error.
Summary
• C3S Climate Projections Workshop: Near-term predictions and projections, 21 April 2015