Predictability of summertime

1
Predictability of summertime North American
precipitation
Gilbert P. Compo and Prashant D.
Sardeshmukh NOAA-CIRES Climate Diagnostics Center
To what extent is the actual precipitation skill
smaller than the expected skill?
2
Data for June-August
200 mb and 500 mb height , 500 mb vertical
velocity, and precipitation NCEP MRF9 T40 JJA
integrations Climatological SSTs 135
members 1987 Global SSTs 90 members 1988
Global SSTs 90 members 1950-94
integrations Global SSTs 13 members 30oN-30o
S Pacific SSTs 9 members NCAR
CCM3T42 1950-99 integrations Global SSTs 12
members Tropical SSTs 11 members ECHAM
4.5T42 1950-2002 integrations Global SSTs 24
members NCEP-NCAR reanalysis dataset smoothed
to T42 1950-2002, 500 mb vertical velocity is
chi-corrected GPCP Precipitation smoothed to
T42 1979-2002
3
Local correlation of summertimesummer mean
rainfall and 500 mb w anomalies
X10
Correlations are lower in regions of descent
Long-term summer mean 500 mb w
Observed (1979-2002)
Simulated by NCEP AGCM (1950-1994)
4
Causal Chain Tropical SSTs NH vertical
motion anomalies NH precipitation anomalies
w7 EOFs P7 EOFs
500 mb omega to predict precipitation
w7 EOFs T14 EOFs
Tropical Pacific SST to predict 500 mb omega
P7 EOFs T14 EOFs
Tropical Pacific SST to predict precipitation
5
1988 Drought
Will predictability in specific cases be
substantially different from average skill
suggested? Some evidence in literature Bates
et al. 2001, Hong and Kalnay 2002 Construct 90
member ensembles using NCEP MRF9 AGCM with
specified monthly SSTs for JJA 1987 and
1989. Assess predictability using Signal to
Noise ratio ratio of ensemble mean anomaly
(88-87) to ensemble spread.
6
Predictability of 1988 summer using the signal to
noise ratio
Predictability of 200 mb and 500 mb height does
not accurately reflect precipitation
Predictability of precipitation closely tied to
500 mb omega
7
Verification of 1988 summer
Where S is large, GCM verifies
Small-scale height gradients that lead to
vertical motions not captured
8
Forecast skill as a function of the signal to
noise ratio (mean shift to standard deviation)
from an ensemble of forecasts.

Valid for any forecasting situation at any
lead time.
Perfect model
Imperfect model
Applies to any multivariate forecast distribution.
A large ensemble of forecasts can improve skill.
Sardeshmukh, Compo, and Penland 2000 Kumar and
Hoerling 2000 Rowell 1998, van den Dool and Toth
1990 Model systematic error can negate this
improvement. Sardeshmukh, Compo, and Penland 2000
9
Expected skill of 12-member ensemble with
time-varying systematic error Se
Perfect model
Model with systematic error
10
Case- and GCM-sensitivity of summertime rainfall
predictability
Compute signal as RMS ensemble-mean anomaly over
domain from 12-member ensembles 1979-1999
(CCM3.0) and 1979-2003 (ECHAM4.5). Noise from
independent 135 member ensemble from NCEP MRF9
forced with climatological SSTs. Pattern
correlation of ensemble-mean anomaly with
verification. Bin correlations by the S-values.
North America
western USA
11
Actual skill for CCM3.0 (1979-1999)
Skill is larger for larger values of S. A
time-varying systematic error is present,
particularly in precipitation.
12
Actual skill for ECHAM4.5 (1979-2002)
Skill is larger for larger values of S. Newer
model still has time-varying systematic errors.
13
Actual skill and expected skill r12 for
combinedCCM3 and ECHAM4.5 predicting JJA
anomalies
North America
western USA
14
Conclusions

• Predictability of precipation is largely
  determined by the predictability of 500 mb
  vertical motions, with some enhancements by local
  land-atmosphere feedbacks.
• Case-to-case variations of skill over North
  America are consistent with model-predicted
  precipitation signal-to-noise ratio S.
• 3. A substantial systematic is present in the
  CCM3.0 and ECHAM4.5, preventing actual
  precipitation skill from reaching the expected
  skill.
• 4. Larger ensembles (gt128) are needed to estimate
  case-to-case variations of S reliably.