Toward understanding
the MJO through the MERRA
data-assimilating model
Brian Mapes, U. Miami
Stefan Tulich, CIRES
Julio Bacmeister, GSFC
and
37 years of studying the MJO: Progress in description, but still no widely accepted theory
Madden and Julian 1972 Benedict and Randall 2007
37 years of studying the MJO: Progress in description, but still no widely accepted theory
Madden and Julian 1972 Benedict and Randall 2007
Outline1. Previous GCM studies of moisture
preconditioning & the MJO
2. Using novel MERRA data-assimilating model to study this and other MJO science issues
3. Structure of the MJO in MERRA Not new, but shows model biases
“Analysis tendencies” provide a new aspect to the problem
4. Future work: Model improvement as a path towards understanding
One of the first GCM moisture preconditioning experiments
• Tokioka et al. (1988): The equatorial 30-60 oscillation and the Arakawa-Schubert cumulus parameterization (J. Meteor. Soc. Japan)
Control No non-entraining plumes
One of the first GCM moisture preconditioning experiments
• Tokioka et al. (1988): The equatorial 30-60 oscillation and the Arakawa-Schubert cumulus parameterization (J. Meteor. Soc. Japan)
Control No non-entraining plumes
This modification also improves the MJO in the CAM 3.1
Maloney (2009)
This modification also improves the MJO in the CAM 3.1
Maloney (2009)
Still the model is not perfect
Even worse when looking at rainfall variance
Maloney (2009)
Improvements are also model dependent
Lee et al. (2009; in press)
How do we proceed further?
• Standard approach: Tinker with the model physics, run long time integration, diagnose model performance/feedbacks, repeat – Drawback: Time-consuming, tedious, feedbacks may
impact other aspects of the simulation in unintended ways
• Our alternative: Assimilation-based science to study the MJO in global models (illustration of concept here)
MERRA
• Modern Era Reanalysis for Research and Applications (GEOS-5 based)
• NASA’s new atm. reanalysis, 1979-present
• Still running (3 streams), ~90% available
• Attractive features:
- nowOpenDAP access (you needn’t download)
- many budget terms, not just state variables
- “analysis tendencies” available
time
analyzed variable
Z at discrete
times
free model solution: Żana= 0 (biased, unsynchronized, may lack oscillation altogether)
initialized free model
ΔZ/Δt = Żmodel + Żana
ΔZ/Δt = (Żdyn + Żphys) + Żana
use piecewise constant Żana(t) to make above equations exactly true in each time interval*
Modeling system integrates:
*through clever predictor-corrector time integration
Learning from analysis tendencies
(ΔZ/Δt)obs = (Żdyn + Żphys) + Żana
• If state is accurate (including flow & gradients), then (ΔZ/Δt)obs and advective terms Żdyn will be accurate
• and thus
Żana ≅ -(error in Żphys)
Choosing MJO cases
good(COARE)
MJO amplitude index
MERRA data available when I started
MERRA stream 2
bestavail
MERRA stream 3
Satellite OLR 15N-15S& MJO-filtered (contours) – used as reference lines below
Filtered OLR courtesy G. Kiladis eastward wavenumbers 0-9, 30-96 days
I averaged this over 15N-15S
15N-15S
GIBBS image archive
MJO phase definition
05
excluded
IO WP
Objective MJO phase categories
PHASE
10 phases relative to Benedict and Randall (2007)
9 8 7 6 5 4 3 2 1 0 ‘back (W)’ ‘front (E)’
5 = filtered OLR min.
Benedict & Randall 2007
MERRA rainrate compared to SSMI (SSMI over water only)
MERRA
SSMI
0
x 10-4 mm/s
too rainy phase 1-2
MERRA’s rain:
convective:
anvil:
large-scale cloud:
premature rain in phase 2 is mainly convective
deep Mc
Phase dependent mass flux
9 8 7 6 5 4 3 2 1 0 ‘back’ (W) ‘front (E)’
5 = filtered OLR min.
Model seems to be choking on the shallow-to-deep transition (even
with Tokioka modification)
Impact? Look at analysis tendencies
Phase dependent part of qv analysis tendency
1990 1992-3
Blame the convection scheme!
• seems to act too deep too soon in the early stages of the MJO.
• Analysis qv tendency has to compensate with moistening
Future work: Improving the model as path towards understanding
• Convection parameterization seems to be too insensitive to low- and mid-level moisture (even with Tokioka modification)
• Question: can we somehow further tighten/adjust the Tokioka limiter to reduce model errors?
Strategy: perform short assimilation runs; does Żana get smaller?
If so, something scientific learned from this technical activity.
Future work: Use analysis tendencies to develop a better forecast tool?
Consider MJO index of Wheeler and Hendon (2004):
Future work: Use analysis tendencies to develop a better forecast tool?
Idea: First, composite model analysis tendencies in this phase space
Future work: Use analysis tendencies to develop a better forecast tool?
Idea: First, composite model analysis tendencies in this phase space
Next, perform multi-day forecasts with these composite tendencies added during runtime.
Forecast improved?