Predictability of Monthly Mean Temperature and Precipitation: Role of Initial Conditions

1
Predictability of Monthly Mean Temperature and
Precipitation Role of Initial Conditions
Mingyue Chen, Wanqiu Wang, and Arun Kumar Climate
Prediction Center/NCEP/NOAA Acknowledgments
Bhaskar Jha for providing the AMIP simulation data
33Rd Annual Climate Diagnostics Prediction
Workshop October 20-24, 2008, in Lincoln, Nebraska
2
Monthly outlook is one of CPCs official products
Temperature - Sep 2008
0.5 Lead (Official)
0 lead (update)
Observation
Off. Up. Change All
Stations 2.4 7.3 4.9 Non-EC
14.9 17.4 2.5 Cov 16.0 42.2
26.2
3
How is current monthly outlook produced?(Ed O
Lenic et al. 2008)

• 0.5-month lead 1-month outlook
• CCA, OCN, SMLR, and CFS
• 0-lead 1-month outlook
• CCA, OCN, SMLR, CFS, and GFS 1-14 day daily
  forecasts, etc.

Sources of predictability

• Initial atmospheric and land conditions, and SSTs
• An initialized coupled atmosphere-land-ocean
  forecast system, such as CFS, is needed to
  harness this predictability

4
Issues to be discussed

• What is the predictability (prediction skill)
  because of initialized observed conditions?
• What is the lead-time dependence?
• How does the predictability due to
  atmospheric/land initial conditions compare with
  that from SSTs?

Analysis method

• Assess lead-time dependence of prediction skill
  of monthly means in CFS hindcasts
• Compare CFS with the simulation skill from the
  AMIP integrations to assess predictability due to
  SSTs, and to assess on what time scale influence
  of initial conditions decays

5
Models and data

• Retrospective forecast
• CFS (5 member ensemble)
• AMIP simulations
• GFS (5 member ensemble)
• CCM3 (20 member ensemble)
• ECHAM (24 member ensemble)
• NSIPP (9 member ensemble)
• SFM (10 member ensemble)
• Variables to be analyzed
• T2m
• Precipitation
• The analysis is based on forecast and simulations
  for 1981-2006

6
Assessment of CFS monthly mean forecast skills
with different lead times
7
Definition of forecast lead time
30-day-lead
20-day-lead
10-day-lead
0-day-lead
1st day
1st day
11th day
21st day
Target month
8
CFS T2m monthly correlation skill

• High CFS skill at 0-day lead time
• Dramatic skill decrease with lead time from 0-day
  lead to 10-day lead and more slow decrease
  afterwards
• Large spatial variation

9
CFS T2m monthly correlation skill (global mean)

• High CFS skill at 0-day lead time
• Dramatic skill decrease with lead time from 0-day
  lead to 10-day lead and more slow decrease
  afterwards

10
CFS T2m monthly forecast skills with different
lead time(zonal mean)
20
10
0
30
40
50

• Little change with lead time over tropics
• Quick decrease in high latitudes

11
CFS T2m monthly forecast skills with different
lead time(zonal mean, DJF, MAM, JJA, SON)

• CFS forecast skill decays vary seasonally
• Skills are higher in winter spring over N. high
  latitudes
• Less changes over tropics

12
CFS Prec monthly forecast skills with different
lead time

• The monthly prec useful skills are at 0-day-lead
  forecast
• No useful skill at lead time long than 10 day for
  most regions
• Prec skill much lower than T2m skill

13
Question What is the source of remaining skill
for longer lead-time forecasts?
A comparison of CFS hindcasts with GFS AMIP
simulations
14
CFS T2m monthly correlation skill vs. GFS AMIP

• The AMIP skill in high-latitudes is low
• The GFS AMIP is similar to CFS in the tropics.

15
CFS T2m monthly correlation skill vs. GFS
AMIP(global mean)
CFS forecast
GFS AMIP

• Globally, the AMIP skill is comparable to CFS
  skill at 20-30-day lead

16
T2m monthly correlation skill (CFS vs. GFS
AMIP)(zonal mean)
0
10
20
40
30
50
GFS AMIP

• Similar skills in CFS GFS AMIP near the equator
• In N. lower latitudes (5N-35N), CFS skill higher
  at lead time shorter than 20 days
• Over N. high latitudes (35N-80N), CFS skill
  higher at lead time shorter than 20-30 days

17
CFS T2m monthly forecast skills vs. AMIPs MME

• The skills are different among 5 AMIPs
• GFS AMIP is comparable to 20-30 lead CFS
• The AMIP MME is almost comparable to 10-day lead
  CFS

Similar to AMIP MME, coupled MME may have
potential to improve.
18
CFS T2m monthly forecast skills vs. AMIP GFS
MMEzonal mean
AMIP MME
AMIP GFS

• The AMIP MME skills are better than the single
  GFS over all the latitudes.
• Similar to AMIP MME, coupled MME may have
  potential to improve.

19
Conclusions

• For monthly forecasts, contribution from the
  observed land and atmospheric initial conditions
  does lead to improvements in skill.
• The improvement in skill, however, decays
  quickly, and within 20-30 days, skill of
  initialized runs asymptotes to that from SSTs.
• A simple average of multi-model AMIP runs shows a
  positive increase of the skill of monthly
  simulation, indicating room for further
  improvements with the MME coupled forecasts.

20
Thanks!