nloa2012: global climate predictionmadrid, 5 july 2012 institut catalÀ de ciÈncies del clima...

NLOA2012: Global Climate Prediction Madrid, 5 July 2012

INSTITUT CATALÀ DE CIÈNCIES DEL CLIMA

Global Climate Prediction: Between

Weather Forecasting and Climate-

Change ProjectionsF. J. Doblas-Reyes

ICREA & IC3, Barcelona, Spain

in collaboration with V. Guémas, J. García-Serrano (IC3), Ch.

Cassou (CERFACS), A. Weisheimer (ECMWF)



Prediction on climate time scales

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 simultaneous comparison with a reference.

Meehl et al. (2009)



Seamless prediction

Courtesy IRI

Illustration of the application of seamless climate and weather information. Example from the IRI-Red Cross collaboration.



Sources of climate predictability

• Both internal and externalo ENSO - large single signalo Other tropical ocean SST - difficulto Remote tropical atmospheric teleconnectionso Climate change - largest for temperatureo Local land surface conditions - soil moisture, snowo Atmospheric composition - difficulto Volcanic eruptions - important for large eventso Mid-latitude ocean temperatures - longer time scaleso Remote soil moisture/snow cover - not well establishedo Sea-ice anomalies - at least local effectso Stratospheric influences - various possibilities

• Unknown or Unexpected



Methods for climate forecasting

• Empirical/statistical forecastingo Use past observational record and statistical methods

o Works with reality instead of error-prone numerical models o Limited number of past cases o A non-stationary climate is problematic o Can be used as a benchmark

• GCM forecastso Include comprehensive range of sources of predictability o Predict joint evolution of ocean and atmosphere flow o Includes a large range of physical processes o Includes uncertainty sources, important for prob. Forecasts o Systematic model error is an issue!



To produce dynamical forecasts

• Build a coupled model of the climate system

• Prepare an ensemble of initial conditions to initial represent uncertainty

o The aim is to start the system close to reality. Accurate SST is particularly important, plus ocean sub-surface. Usually, worry about “imbalances” a posteriori.

• Run an ensemble forecasto Run an ensemble on the e.g. 1st of a given month (the start date)

for several weeks to years.

• Run a set of re-forecasts to deal with systematic error.

• Produce probability forecasts from the ensemble

• Apply calibration and combination if significant improvement is found



May 87 Aug 87 Nov 87 Feb 88 May 88

Assume an ensemble forecast system with coupled initialized GCMs

Ensemble climate forecast systems

Lead time = 3



Equatorial Atlantic: Taux anomalies

Equatorial Atlantic upper heat content anomalies. No assimilation

Equatorial Atlantic upper heat content anomalies. Assimilation

ERA15/OPS

ERA40

• Example for the ocean.

• Large uncertainty in wind products lead to large uncertainty in ocean subsurface.

• Possibility to use additional information from ocean data

• Assimilation of ocean data:

constrain the ocean state

improve the ocean estimate

improve the seasonal forecasts

Data assimilation for initial conditions



Dry Rainy

A farmer is planning to spray a crop tomorrow Consensus

forecast: dry YES

Should he/she?

NO

NO

?

Let p denote the probability of rain from the forecast

Why running several forecasts

The farmer loses L if insecticide washes out.

• He/she has fixed (eg staff) costs

• If he/she doesn’t go ahead, there may be a penalty for late completion of the job.

• By delaying completion of the job, he/she will miss out on other jobs.

• These cost C

Is Lp>C? If p > C/L don’t spray!



Courtesy Ch. Cassou (CERFACS)

Weather regimes and climate anomalies

Observed anomalies for NDJFM 2007-8

Air Temp Precip

Mean frequencies: NAO+ : +62%, AR:+16%,

NAO- : -75%, BL:-3%



Weather regimes and the MJO

Cassou (2008)

Interaction between the Madden-Julian oscillation and the North Atlantic weather regimes

OLR/STRF300



Predicting the NAO and the MJO

Lin et al. (2010)

NAO ensemble-mean correlation from monthly forecasts with the Canadian forecast system (using observed SSTs).



Collins et al. (2010)

ENSO: a seasonal prediction forcing



ROC area of the (left column) statistically downscaled and (right column) original DEMETER seasonal predictions for several events where the

years verified are segregated depending on the ENSO phase. Only values statistically significant with 90% confidence level are shown.

Extra-tropical links of ENSO

Frías et al. (2010)



Sources of predictability: snow cover

Correlation of System 3 MAM temperature in 1981-2005 wrt to GHCN temperature (adapted from Shongwe et al., 2007).

ROC area for anomalies above the upper quartile

ROC area for anomalies below the lower quartile



DJF

T2m

JJA

T2m

JJA

Prec

DJF

Prec

Correlation of System 3 seasonal forecasts of temperature (top) and precipitation (bottom) wrt GHCN and GPCC over 1981-2005. Only values significant with 80% conf. plotted.

How good are the seasonal forecasts?



EC-Earth 2-5 year near-surface air temperature ensemble-mean predictions started in November 2011. Anomalies are computed with

respect to 1971-2000.

Initialised Uninitialised

Difference

Decadal predictions



CMIP5 decadal predictions

(Top) Near-surface temperature multi-model ensemble-mean correlation from CMIP5 decadal initialised predictions (1960-2005); (bottom)

correlation difference with the uninitialised predictions of 2-5 year (left) and 6-9 year (right) wrt ERSST and GHCN.

Init ensemble-mean

correlation

Init minus NoInit

ensemble-mean

correlation difference

Doblas-Reyes et al. (2012)



But there are systematic errors

• Model drift is typically comparable to signal Both SST and atmosphere fields

• Forecasts are made relative to past model integrations Model climate estimated from e.g. 25 years of forecasts (1981-2005), all

of which use a 11 member ensemble. Thus the climate has 275 members. Model climate has both a mean and a distribution, allowing us to estimate

eg tercile boundaries. Model climate is a function of start date and forecast lead time.

• Implicit assumption of linearity We implicitly assume that a shift in the model forecast relative to the

model climate corresponds to the expected shift in a true forecast relative to the true climate, despite differences between model and true climate.

Most of the time, the assumption seems to work pretty well. But not always.



Averaged precipitation over 10ºW-10ºE for the period 1982-2008 for GPCP (climatology) and ECMWF System 4 (systematic error) with start dates of

November (6-month lead time), February (3) and May (0).

GPCP climatology

ECMWF S4 - GPCP (MAY)ECMWF S4 - GPCP (FEB)ECMWF S4 - GPCP (NOV)

Model drift: West African Monsoon

(NOV)

(FEB)(MAY)



Multi-model benefits: Reliability

Reliability for T2m>0, 1-month lead, May start, 1980-2001

System 1 System 2 System 3 System 4

System 5 System 6 System 7



Control-GPCP

Stochastic physics-GPCP

Precipitation bias (DJF, 1-month lead, 1991-2001, CY29R2)CASBS reduces the tropical and blocking frequency biases

Error reduction: stochastic physics

control CASBS ERA40

Blocking frequency

Berner et al. (2008)



System with the best skill score for one-month lead seasonal probability predictions (Brier skill score). The predictions have been formulated with

a multi-model (MME), a system with stochastic physics (SPE) and a system with parameter perturbations (PPE) over 1991-2005.

Skill improvement: model inadequacy

Weisheimer et al. (2011)



SST (with North Atlantic Current path) and North Atlantic zonal wind bias, plus winter blocking frequency for HadGEM3 using the (left) standard (1º)

and (right) high-resolution (0.25º) ocean components.

Increase of model resolution

Scaife et al. (2011)



Summary

● Substantial systematic error, including lack of reliability, is still a fundamental problem in dynamical forecasting and forces a posteriori corrections to obtain useful predictions. Don’t take model probabilities as true probabilities.

• Initial conditions are still a very important issue.

• Estimating robust forecast quality is difficult, but there are windows of opportunity for reliable skilful predictions, and there is always the anthropogenic warming.

• There is potential in methods that deal with model inadequacy (multi-model ensembles, stochastic physics).

• Model development is a must. And many more processes to be included: sea ice, anthropogenic aerosols, chemistry, …

nloa2012: global climate predictionmadrid, 5 july 2012 institut catalÀ de ciÈncies del clima...

Documents