Deriving South America seasonal rainfall from - eurobrisa

Deriving South America seasonal rainfall
from upper level circulation predictions
Caio A. S. Coelho
Centro de Previsão de Tempo e Estudos Climáticos (CPTEC)
Instituto Nacional de Pesquisas Espaciais (INPE)
[email protected]
1st EUROBRISA workshop, Paraty, 17-19 March 2008
Conceptual framework
Data Assimilation
“Forecast Assimilation”
p( y i | x i ) p( x i )
p( x i | y i ) 
p( y i )
p( x f | y f ) p( y f )
p( y f | x f ) 
p( x f )
Stephenson et al. (2005)
Calibration and combination experiment
y: observed rainfall (GPCP, Adler et al. 2003)
x: predicted upper level (200 hPa) circulation
Use two EUROSIP model predictions:
• ECMWF System 3
• UK Met Office
Common hindcast period: 1987-2005 (19 years)
Target season: DJF
Start date: November (i.e. 1-month lead predictions for DJF)
NCEP/NCAR Reanalysis (Kalnay et al. 1996) is used
to verify predicted upper level circulation
Upper level circulation

V  ( u , v)
v
u
u: zonal wind
v: meridional wind
: stream function
d
d
v

u
dx
dy
    '
 : zonal mean of 
 ': perturbed (eddy)
stream function
Observed perturbed stream function
ECMWF perturbed stream function
UKMO perturbed stream function
Perturbed stream function verification
Observed perturbed stream function anomaly
El Niño
Observed perturbed stream function anomaly
La Niña
Observed perturbed stream function anomaly
El Niño
La Niña
ECMWF perturbed stream function anomaly
El Niño
ECMWF perturbed stream function anomaly
La Niña
ECMWF perturbed stream function anomaly
El Niño
La Niña
UKMO perturbed stream function anomaly
El Niño
UKMO perturbed stream function anomaly
La Niña
UKMO perturbed stream function anomaly
El Niño
La Niña
Perturbed stream function anomaly
verification
Calibration and combination procedure:
Matrices
Forecast Assimilation
X : n p
Stephenson et al. (2005)
p( X | Y ) p(Y )
p(Y | X ) 
p( X )
X: circulation predictions (ECMWF + UKMO)
Y: DJF rainfall
Prior:
Y : nq
Yb : 1 q
C : qq
S : p p
Ya : n  q
D : qq
Y ~ N(Yb , C)
Likelihood: X | Y ~ N(G(Y  Yo ), S)
Posterior:
Y | X ~ N (Ya , D)
Forecast assimilation uses first three leading MCA modes of the matrix YT X.
Forecast Assimilation: First MCA mode
Forecast Assimilation: Second MCA mode
Forecast Assimilation: Third MCA mode
Correlation between predicted and observed anomalies
ECMWF
UKMO
Forecast Assimilation
 Upper level circulation derived predictions obtained with forecast
assimilation have comparable level of skill to indiv. model predictions
Issued: November, Valid for DJF, Hindcast period: 1987-2005
Ranked Probability Skill Score (tercile categories)
ECMWF
UKMO
Forecast Assimilation
Issued: November, Valid for DJF, Hindcast period: 1987-2005
Gerrity Score (tercile categories)
ECMWF
UKMO
Forecast Assimilation
Issued: November, Valid for DJF, Hindcast period: 1987-2005
ROC Skill Score (positive or negative anomaly)
ECMWF
UKMO
Forecast Assimilation
Issued: November, Valid for DJF, Hindcast period: 1987-2005
Reliability diagram (positive or negative anomaly)
ECMWF
UKMO
Forecast Assimilation
 Forecast assimilation improves prediction reliability
Issued: November, Valid for DJF, Hindcast period: 1987-2005
ROC plot (positive or negative anomaly)
ECMWF
UKMO
Forecast Assimilation
Issued: November, Valid for DJF, Hindcast period: 1987-2005
Summary
•
Forecast assimilation is a useful framework for exploring
atmospheric teleconnections in seasonal forecasts
•
ENSO atmospheric teleconnections is the main source of
skill for South America rainfall predictions
•
Combined and calibrated circulation derived predictions
obtained with forecast assimilation have comparable level
of skill to single model rainfall prediction
•
Additional skill improvements can be investigated by
including humidity predictions in the forecast assimilation
procedure