-
Korea Institute of
Atmospheric Prediction Systems (KIAPS)
(재)한국형수치예보모델개발사업단
LETKF Data Assimilation System for KIAPS AGCM: Progress and
Plan
Ji-Sun Kang, Jong-Im Park,
Hyo-Jong Song, Ji-Hye Kwun,
Seoleun Shin, and In-Sun Song
UMD Weather-Chaos Group Meeting June 17, 2013
-
Korea Institute of Atmospheric Prediction Systems
Goal – To develop a global NWP system optimized to the
topographic &
meteorological features of Korean peninsula
Period – 2011~2019 (total 9 years)
Total fund – About $ 100 million
-
Korea Institute of Atmospheric Prediction Systems
There are 14 researchers in a group for data assimilation
(DA)
– 7 for observation pre-processing (QC, bias correction for
observation data)
– 7 for DA system development
(12)
(1)
(31) (22)
-
Computing Facilities
Haenam (Cray XE6, AMD 2.1G, 16.9Tflops) – One of KMA’s
supercomputers – 2016 cores
Gaon1 (Dell, AMD Opteron 2.3G*4cpu, 2.9 Tflops) – Belongs to
KIAPS – 5 nodes with 320 cores (64 cores per node)
Gaon2 (IBM, Intel Xeon 2.9G 2cpu, 11.5 Tflops) – Belongs to
KIAPS – 36 nodes with 576 cores (16 cores per node)
Gaon3 – Will be purchased this year or early next year –
Expected a similar one with Gaon2 (576 cores)
Total 3,488 cores (5325 cores normalized by the performance of
Haenam)
Backup & Archiving system – 210 TB for backup & 330 TB
for archiving
-
KIAPS Data Assimilation Systems
Plans for KIAPS Data Assimilation System
Ensemble Data Assimilation ─ LETKF would be the first system to
be constructed as an operational system
Hybrid(3D-Var/LETKF) ─ 3D hybrid assimilation systems of which
components consist of the
ensemble assimilation system, descent algorithm for minimization
of 3D-Var cost function
Variational data assimilation (3D/4D-Var) ─ KIAPS also plan to
develop a 4-d variational data assimilation system
-
Framework of EnKF data assimilation system
LETKF – CAM_SE (Kang)
We are implementing LETKF to NCAR CAM_SE (Community Atmospheric
Model-Spectral Element) model, because it has the same
horizontal/vertical coordinates as HOMME/KIAPS that will be
released as the first version at the end of this year.
LETKF implemented to NCAR CAM-SE model can be immediately used
for the HOMME/KIAPS when released
LETKF – SPEEDY (Park)
While developing LETKF-CAM_SE(HOMME/KIAPS), we would develop
or/and test many essential methods to advance the current LETKF DA
system, using the simplified model as a testbed.
-
Model
HOMME/KIAPS & NCAR CAM-SE – Spectral Element dynamic
core
– The SE dycore uses accurate, high-order numerical methods on
rectangular elements in a cubed-sphere geometry (six faces)
Each face has Ne*Ne elements (Ne: # of elements in one side of a
face)
Each element has Np*Np grid points (Np: # of points in one side
of a element)
Horizontal resolution can be addressed by neXnpY (e.g.
ne16np4~2° resolution, ne120np4~0.25°, ne240np4~0.125°)
5
-
LETKF implemented to NCAR CAM-SE Model
– NCAR CAM 5.2 with Spectral Element dynamic core
– Horizontal resolution: ne16np4 (~ 2°)
– 30 vertical levels with hybrid sigma-pressure coordinate
Major modifications of LETKF
– I/O of the model
– Data search process
The original LETKF codes (from Dr. Miyoshi) compute (ri, rj) for
a location of each observation which is a relative position to the
model grid (i, j).
Since (ri, rj) requires too much consideration near the boundary
for cubed-sphere domain when LETKF searches for data within local
area, I modified this part so that just (lon, lat) is used instead
of (ri, rj).
vs.
-
Test of LETKF-CAM_SE
Observing System Simulation Experiments
– Simulated observations for U, V, T, q
Radiosonde distribution
– Simulated observations for Ps
Surface stations
Observation distribution has been determined by real observation
data (NCEP bufr)
Observation errors: 1m/s for (U, V), 1K for T, 1g/kg for q, 1hPa
for Ps
64 ensembles (cam, clm2, and cice)
Random initial condition
– States at 64 arbitrary timesteps from a nature run +
perturbations
-
Preliminary Result
Reduced RMS difference after the first analysis step
It seems working well.
– I’ll make an analysis cycle with ensemble forecast right after
getting back to Korea
RMS difference of background (red) and analysis (blue) to
observations in the observation space
-
Progress
Understanding the model with special horizontal grid of Spectral
Element dynamic core on rectangular elements in a cubed-sphere
geometry
– Installing and running CAM-SE (coupled with CLM2 & CICE
components) with every 6-hour restart
Modifying standard LETKF codes for the model HOMME/KIAPS (NCAR
CAM-SE)
– Analysis system of LETKF-CAM_SE assimilating radiosonde and
surface station data has been developed.
– Preliminary result looks good
-
Plans
We plan to include satellite data of AMSU-A & IASI, and GPS
radio occultation data into LETKF data assimilation system. –
Takemasa (radiance DA) & Shu-Chih (GPS RO DA) will visit
KIAPS
in August for giving us an advice.
AIRS retrieval data can be also assimilated and compared.
Target resolution of HOMME/KIAPS is ne240np4 (~0.125°, very
high) with 70 vertical layers – Ensemble forecast may have coarser
resolution than ne240np4.
If the resolution for ensemble forecast is too coarse, we may
not be able to get comparable results with others.
It would be good to incorporate a mixed resolution of background
(Rainwater and Hunt, 2013)? I can test it using CAM!
We plan to test many useful techniques in EnKF, especially
forecast sensitivity to observations (Kalnay et al. 2012) which KMA
is very interested in.
Carbon cycle data assimilation (LETKF-C) will be also tested
using CAM5.2 with SE, or CAM3.5 with FV
Thank you very much for your attention!