Results and Status of the MIIDAPS 1DVAR Integration in GSI

Kevin Garrett, Eric Maddy and Krishna Kumar

Riverside Technology, Inc., JCSDA

Sid BoukabaraJCSDA

13th Annual JCSDA Technical Review and Science WorkshopCollege Park, MD

May 15, 2015

2MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Quality Control-Consistent algorithm for all sensors to determine data quality• Data quality depends on DA system capability• Quality control algorithm flexibility for variety of capabilitiesPreprocessing-Characterize satellite observations before assimilation• Surface characterization: surface properties (ocean/non-ocean)• Atmosphere characterization: clear, cloudy, precipitation• Background update/state linearization

Streamline quality control and preprocessing for all satellite radiance observations in data assimilation

MotivationWhy Consider a 1DVAR Preprocessor?

Ultimate goal: Facilitate assimilation of all-sky/all-surfaces/full spectrum satellite radiance observations

3MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Outline

Overview of the MIIDAPS 1DVAR

Integration Status

Preliminary Forecast Impacts

Next Steps

4MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Overview of the MIIDAPS 1DVAR

Integration Status

Preliminary Forecast Impacts

Next Steps

5MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Assimilation/Retrieval All parameters retrieved simultaneously Valid globally over all surface types Valid in all weather conditions Retrieved parameters depend on information content from sensor frequencies

1DVAR PreprocessorMulti-Instrument Inversion and

Data Assimilation Preprocessing System (MIIDAPS)

MIIDAPS

S-NPP ATMS

DMSP F16 SSMI/SDMSP F17 SSMI/SDMSP F18 SSMI/S

GPM GMI

MetOp-A AMSU/MHSMetOp-B AMSU/MHS

GCOM-W1 AMSR2

Megha-TropiquesSAPHIR/MADRAS

TRMM TMI

NOAA-18 AMSU/MHSNOAA-19 AMSU/MHS

Inversion Process Inversion/algorithm consistent across all sensors Uses CRTM for forward and Jacobian operators Use forecast, fast regression or climatology as first guess/background

Benefit of the 1DVAR preprocessor is to enhance QC, as well as increase the number and types of observations assimilated (e.g. imager data)

**CrIS** **IASI****MIIDAPS extended to the hyperspectral Infrared for IR only or IR+MW 1DVAR analysis

6MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Obs Error [E]

No

Convergence

1DVAR Retrieval/Assimilation Process

Init

ial S

tate

Vecto

r [X

]

Climatology

Forecast

Retrieval mode

Assimilation mode CRTM

Simulated TBs

Observed TBs

(processed)

Compare

ConvergenceSolution

[X]Reached

ComputeDX

K

Update State Vector

[X]

Iterative Processes

CovarianceMatrix [B]

Bias Correction

7MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS Outputs

Extended state vector with hyperspectral IR covers trace gas absorption

IR expansion

CO

CO2

O3

CH4

N2O

Extended state vector to hydrometeor effective size

General Expansion

Cloud radius

Rain radius

Graupel radius

8MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS IR Expansion

• MIIDAPS atmospheric state vector extended to trace gas profiles

• MIIDAPS emissivity state vector expanded to IR channels (AIRS, CrIS, IASI)

• ATMS/CrIS brightness temperatures simulated using ECMWF analysis (T, Q, CLW), trace gas climatologies, and various IR/MW emissivity models

• MIIDAPS applied to simulated data to retrieve state vector elements for MW only, IR only and combined IR+MW

9MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS IR Compared to ECMWF

MW Only94.5% Conv.

MW+IR94% Conv.

Temp WV Temp WV

MW Only IR Only ECMWFIR+MW

10MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Overview of the MIIDAPS 1DVAR

Integration Status

Preliminary Forecast Impacts

Next Steps

11MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS GSI InterfaceGSI

2nd loop1st loop

Setuprad Module PP1dvar Module

Initialize CRTM structuresCollocate guess to obsCall pp1dvarCall CRTM for background calcCall quality control subroutinesBias correctionGross error checkDiagnostic file output

MiRS Library

Obs Error [E]

CovarianceMatrix [B]

Bias correction

Guess fieldsT(p), q(p),

pSfc, Windsp

Brightness Temperatures/scan/geo info

QC fields(flags, geo)

1dvar fields(clw, emiss)

InputsOutputs

12MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS Integration Updates

• Integration updates– 1DVAR analysis and guess fields written to binaries

and archived (tar) (1st outer loop only)– 1DVAR controlled with GSI namelist variable• Initialized to FALSE by default (EnKF)

– Extended to SSMI/S (1DVAR, QC)

13MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS GSI Timing

CPU

Hours

56:24:00

56:38:24

56:52:48

57:07:12

57:21:36

57:36:00

57:50:24

58:04:48

Wall

Time

0:36:43

0:37:26

0:38:09

0:38:52

0:39:36

0:40:19No 1DVARWith 1DVAR

• 1 GDAS Cycle• T670 Resolution• 200 CPU• 1DVAR run on ATMS and F18 SSMI/S only (22,000 Observations)

14MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Overview of the MIIDAPS 1DVAR

Integration Status

Preliminary Forecast Impacts

Next Steps

15MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Experiment Setup

• GSI r46725, T670/254 (GFS/GSI)– PRCN: GDAS/GFS Operational configuration– PR1D: PRCN + MIIDAPS applied to ATMS only

• Summer season– August 1, 2014-September 10, 2014

• MIIDAPS Applied to ATMS only– ATMS QC based on MIIDAPS output– SSMI/S QC still being tuned– Use of 1DVAR Geophysical outputs still being

explored

16MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

CLW > 0.2 Remove:19-51 GHz 89-190 GHz

ChiSq > 5Remove: All Channels

GWP/RWP > 0.05Remove :19-53 Ghz, 89-190 GHz

MIIDAPS-based Quality Control

Heritage GSI QC Flags for LEFT: SNPP ATMS Channel 5 (52 GHz) and RIGHT: Channel 7 (54 GHz)

MIIDAPS-based GSI QC Flags for LEFT: SNPP ATMS Channel 5 (52 GHz) and RIGHT: Channel 7 (54 GHz)

MIIDAPS-based QC Scheme: Sounders

With 1DVAR QC With 1DVAR QC

Operational QC Operational QC

Points Passing MIIDAPS QC but Failing Operational QC

ATMS 52.8 GHz ATMS 54.5 GHz

O-B (all observations) for LEFT: SNPP ATMS Channel 5 (52 GHz) and RIGHT: Channel 7 (54 GHz)

17MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS Forecast Impact

18MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS Forecast Impact

19MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS Forecast Impact

1. PR1D 2. PRCN

FCST: 00Z 8/7/2014

1. PR1D 2. PRCN

1. PR1D 2. PRCN 1. PR1D 2. PRCN1. PR1D 2. PRCN

1. PR1D 2. PRCN1. PR1D 2. PRCN

FCST: 00Z 8/8/2014 FCST: 00Z 8/9/2014

FCST: 00Z 8/4/2014 FCST: 00Z 8/5/2014 FCST: 00Z 8/6/2014

20MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS SSMI/S Quality Control

ChiSq > 5Remove: All Channels

GWP/RWP > 0.05Remove :19-53 Ghz, 89-190 GHz

CLW > 0.15 Remove:19-51 GHz 89-190 GHz

About 1000 more obs per channel passing quality control

21MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

Overview of the MIIDAPS 1DVAR

Integration Status

Preliminary Forecast Impacts

Next Steps

22MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

• Finalize QC implementation for current sensors and extend to new sensors– ATMS, SSMI/S, AMSU/MHS, GMI, AMSR2, SAPHIR,

IR sensors• Use of 1DVAR geophysical fields– Surface emissivity, hydrometeors

• Explore use of 1DVAR analysis as background and all-sky radiance assimilation– Resolve displacement errors– Linearization

Next Steps

Figure. MIIDAPS retrieved liquid water path and GFS 6hr forecast valid 12Z Jul 3, 2014, for Hurricane Arthur event off the U. S. Southeast coast. a) Displacement of MIIDAPS 1DVAR analysis and GFS forecast; b) 1DVAR liquid water path; c) GFS 6hr liquid water path; and d) MIIDAPS-GFS liquid water path.GFS forecast is collocated in space/time to GPM GMI observation points.

b) c) d)a)

23MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

BACKUP

24MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

MIIDAPS Example OutputNumber of Attempts Liquid Water Path

TPW

Chisq

Graupel Water PathNumber of Iter

MIIDAPS Analsyis for 2014-08-01 00Z GDAS Cycle

MIIDAPS Status – 13th JCSDA Technical Review and Science Workshop, College Park, MD

• Cost Function to Minimize

• To find the optimal solution, solve for:

• Assuming Linearity

• This leads to iterative solution:

25

Mathematical Basis

0(X)'JX

J(X)

nΔXnK)nY(XmY

1ET

nBKnKTnBK

1nΔX

0

xxK)0

y(xy(x)

Y(X)YEY(X)Y

21

XXBXX21

J(X) m1Tm0

1T0

Jacobians & Radiance Simulation from Forward Operator: CRTM