Understanding the MJO through the MERRA data assimilating model system

1
Understanding the MJO through the MERRA data
assimilating model system
and

• Brian Mapes
• RSMAS, Univ. of Miami
• and
• Julio Bacmeister
• NASA GSFC

2
Outline

1. What is the MJO?
2. Why does it require assimilation-based science?
3. Robust MJO features from 2 active seasons, 2
   longitudes (IO vs. WP), 2 MERRA versions
4. Analysis tendency derived hypotheses about MJO
   mechanisms and model shortcomings
5. Testing the hypotheses improving the model

3
The MJO

• Madden and Julian 1972

4
Eastward moving, 40-50 day period
Wheeler and Kiladis 1999
MJO in OLR data
5
Outline

• Why does it require assimilation-based science?
• Low frequency means small time rate of change.
• many processes (tendencies) small imbalances
  important
• Slow speed
• even weak tropical background flow may be
  important
• Large scale yet longitudinal confinement
• need realistic geography spatially varying
  basic flow
• MJO power does not lie along linear wave theory
  dispersion lines (like the c-c Kelvin wave etc.)
• no reason to believe it is tractable to toy
  modeling
• free-running GCMs dont simulate it realistically
• true understanding implies explaining this fact
  too

6
Models have trouble with this stuffconvection
cloud problems
Obs
Dominant modes MJO, Kelvin, ER, WIG Dispersion
curves correspond to equivalent depth 8, 12, 25,
50, 90m. Larger depth faster phase speed. All
modes 25 m.
Lin et al. 2005
7
Outline

1. What is the MJO?
2. Why does it require assimilation-based science?
3. Pick and study MJOs from 2 active seasons, 2
   longitude sectors, two MERRA versions
4. Analysis tendency derived hypotheses about MJO
   mechanisms and model shortcomings
5. Testing hypotheses improving a model

8
Choosing MJO cases
9
Meanwhile (when I started project)
10
Choosing a case in MERRA streams
best avail
Next (COARE)
11
Satellite OLR 15N-15S, filtered
COARE Dec 1992- Mar 1993
Jan-Apr 1990
12
(No Transcript)
13
MERRA data used

• Scout runs (2 degree) for convenience
• so actually, all other cases are available.
• trying not to make scout an object of research
  though
• Real MERRA (1/2 x 2/3 degree)
• will the parameterized-resolved rain partition
  differ?
• will heating profiles differ in a corresponding
  way?
• convective vs. stratiform

14
Outline

1. What is the MJO?
2. What is assimilation-based science?
3. Robust features from 2 active seasons, 2
   longitudes (IO vs. WP), 2 MERRA versions
4. Analysis tendency derived hypotheses about MJO
   mechanisms and model shortcomings
5. Testing the hypotheses improving the model

15
Incremental Analysis Update (IAU)
i cannot understand this diagram
16
Modeling system integrates
?Z/?t Zmodel Zana ?Z/?t
(Zdyn Zphys) Zana
analyzed variable Z at discrete times
time
through clever predictor-corrector time
integrations
17
is nudging a bad word (or boring)?

• not if we STUDY the analysis tendencies
• (?Z/?t)obs (Zdyn Zphys) Zana
• If state is accurate (flow gradients), then
• Zdyn will be accurate
• and thus
• Zana ? -(error in Zphys)

18
aside hypothesis analyses lie toward free
models bias
Obs vs. analyzed Z
time
19
Outline

1. What is the MJO?
2. Why does it require assimilation-based science?
3. Robust features from two active seasons, two
   longitude belts, two MERRA versions
4. Analysis tendency derived hypotheses about MJO
   mechanisms and model shortcomings
5. Testing the hypotheses improving the model

20
One possible worry

• If the bias is severe, so that realistic states
  are too far off the models solution manifold,
  then Zphys errors could start to be related
  nonlinearly to state variable errors.
• compromising interp. of averages and phase
  composites
• hope not

21
unglorified nudging (no DAS needed!)

• every time step, build Zana (Zobs - Zmodpred)
  /t
• Zobs is NWP analysis, interpolated to model
  timesteps
• really, is DART going to beat EC at raw-data
  assimilation?
• fine, try a few other analyses (MERRA, NCEP) if
  you like
• study Zana as above, and dependence on t
• tiny t will keep model verrrrrry close to
  analysis
• including divergence/omega sounding, despite
  convxn
• moderate values just keep synoptic flow in phase
• try different values for winds vs. thermo
• e.g. let model have its comfort sounding biases
• get at worry of far from manifold nonlinearities

22
Outline

1. What is the MJO?
2. Why does it require assimilation-based science?
3. Robust features from two active seasons, two
   longitude belts, two MERRA versions
4. Analysis tendency derived hypotheses about MJO
   mechanisms and model shortcomings
5. Testing the hypotheses improving the model

23
Satellite observed OLR 1990 Jan-Apr
15NS 10NS
24
MERRA analysis models OLR
25
15NS u850 NCEP 10NS
26
15NS u850 MERRA 10NS
27
Wheeler - Hendon RMM1-RMM2 view
(Difficulties WH removed a mean seasonal cycle,
and the preceding 120-day mean at each time. I
just removed the 120-day time means of my 4-month
dataset, and offset RMM1 by 0.5 to match their
figure more closely)
28
time-height sections 60-110E, 130-180E
29
60-110 130-180
Rather noisy - next lets make a phase composite
30
MJO phase definition
9
5
0
0
31
1990 MJO phase in time-lon space
WP
IO
9
0
5
32
1992-3 MJO phase in time-lon space
WP
IO
COARE Dec 1992- Mar 1993
9
0
5
33
COARE1992-3
34
COARE 1992-3
35
Line checks 1990 OLR vs. satellite
IO
WP
MERRA biased high 10-20W in active phase
misses 10W IO-WP difference
36
Rainrate compared to SSMI (SSMI is over water
only)
too rainy here
x 10-4 mm/s
MERRA
0
37
PW MERRA has humid bias, too little IO-WP
difference
WP
1990 SSMI
IO
1990 MERRA
IO too humid especially here
38
LWP MERRA too low by half
39
1990 wind speed vs. SSMI good (assimilated)
SSMI over water only
MERRA
40
Total rainconvectiveanvillarge-scale
cloud
41
u200 and u850
1990
1992-3
42
1990 1992-3 COARE
-5
-5
-50
-50
43
tilt more obvious in div maybe
44
1990 (not same color bars!) 1992-3
45
(No Transcript)
46
1990 T 1992-3 COARE
47
IO WP
1990 1992-3
48
1990 RH 1992-3 COARE
60
60
lt40
lt40
lt40
lt40
60
60
49
1990 1992-3 COARE
0.5
0.45
50
1992-3 COAREperiod in MERRA
COARE OSA qv lag regression (Mapes et. al. 2006
DAO)
?
51
1990 qcond 1992-3
52
CloudSat echo (2006-8 cases)thesis of Emily
Riley same MJO def. techniques
25
lt 7.5
-6
7
53
MERRA Cloud fraction
Cloudsat echo coverage
50
25
from Emily Riley MS thesis
-15
15
-6
7
54
MERRA Cloud fraction
from Emily Riley MS thesis
55
1990 cloud 1992-3
55
60
56
Outline

1. What is the MJO?
2. Why does it require assimilation-based science?
3. Robust features from two active seasons, two
   longitude belts, two MERRA versions
4. Analysis tendency based hypotheses about MJO
   mechanisms, and model shortcomings
5. Testing the hypotheses improving the model

57
MERRA has a Dry bias at 850, humid bias at 600
qv DJF 1990 minus JRA typical of MERRA vs.
all others
58
Analysis tendencies oppose humidity bias(with a
little MJO dependence too)
DJFM 1992-3 COARE
1990 JFMA MJOs
zonal mean qv bias
Zana ? -(error in Zphys)
59
Bias stripes correspond to Moist Phys tend.


-
-

Zana ? -(error in Zphys)
-
60
Beyond the bias phase dependence

• Moist physics Qv tendencies MINUS MEAN

61
1990 1992-3 COARE
analysis Qv tend.
62
Benedict and Randall schematic
63
(No Transcript)
64

• Hypothesis model convection scheme acts too deep
  too soon in the early stages of the MJO.
• (Hypothesis for improving it is another seminar)

65

• Hypothesis model convection scheme acts too deep
  too soon in the early stages of the MJO.
• (Hypothesis for improving it is another seminar)
• Might be entangled with the mean state biases.
• Improving the model must consider both

66
MERRA Temperature biases (DJF)

• 2 different years, 3 different reference
  reanalyses

-JRA
-NCEP2
-ERA
----------------------cool lt 200
mb----------------------
-----------------------warm 550-------------------
----
------------------------cool 700------------------
-----------
67
1990 1992-3
----------------------cool lt 200
mb----------------------
-----------------------warm 550-------------------
----
------------------------cool 700------------------
-----------

• Again analysis tendencies fight the bias

68
T budget DYN-PHYS balance
mostly MST
sharp shelf in moist heating profile may be
bias source. Again the shallow to deep convection
transition issue?
69
(No Transcript)
70
easterly mean analysis tendencies
westerly analysis tendencies
1990 1992-3
westerly biases
easterly biases
zonal mean u biases wrt 3 other reanalyses
71
Outline

1. What is the MJO?
2. Why does it require assimilation-based science?
3. Robust features from two active seasons, two
   longitude belts, two MERRA versions
4. Analysis tendency based hypotheses about MJO
   mechanisms, and model shortcomings
5. Testing hypotheses / improving the model

72
closing the loop

• Adjust model based on hypotheses
• convection scheme formulations
• after learning them (what im here for)
• Re-run in assimilation mode
• or replay
• ? advice ?
• Remake diagrams and evaluate
• mean AND variability
• will interplay make results inscrutable?
• Focus on improved aspects, declare victory.
• Refine hyp., go to 1. Progress, if not
  victory...

73
Poor mans version

• Learn inside scoop on model convection
• what im here for
• Make diagrams for existing alt. model versions
• starting with ½ deg regular MERRA
• e.g. does resolution matter, via conv-LS
  partitioning?
• Publish speculative interpretations