Young-Kwon Lim
Collaborators: Climate modeling group (Drs. L. Stefanova, D.W. Shin, S. Cocke, and T. E. LaRow) at
FSU/COAPS, Dr. G. Baigorria (Univ. of Florida),
Dr. K. H. Seo (Pusan Nat’l Univ., Korea), Dr. S. Schubert (NASA/GSFC), Dr. H. Juang (NOAA)
Center for Ocean-Atmospheric Prediction Studies, Florida State University, Tallahassee, FL, USA
Development of Multi-model High Resolution Seasonal Forecasting System:
An Application to SE US
Current high-resolution seasonal forecast in FSU/COAPS
Current high-resolution seasonal forecast in FSU/COAPS
One dynamical model (FSU/COAPS RSM) and one statistical downscaling model at 20km resolution (Cocke et al. 2007; Lim et al. 2007, 2009).
Real-time seasonal forecasts (up to 6 month ahead) are updated four times a year through CMO web (http://coaps.fsu.edu/cmo).
Realtime high-resolution forecast by FSU/COAPS
(example: Nov. 2009, 20km resolution) Realtime high-resolution forecast by FSU/COAPS
(example: Nov. 2009, 20km resolution)
T mean
T ano.
Rainfallmean
Rainfallanomaly
Corr. skill of the current FSU/COAPS downscaling system:
1) Winter rainfall: Corr. > 0.5 (FL,GA) (Cocke et al. 2007)
2) Crop growing seasons (spring and summer) :
Sfc. air T. (Lim et al. 2007): Corr.=0.3~0.8,
Rainfall: Improvement of correlation over the large scale CFS. Statistical significance problem (Lim et al. 2009).
Skill (correlation and categorical predictability) tends to be model dependent (e.g., summer rainfall: higher skill over inland by FSU model, while higher skill over Florida peninsula by NCEP model)
Question: Can we improve the skill over the entire SE US with statistical signifcance via MM downscaling system?
Capability of the current FSU/COAPS downscaling system, and MotivationCapability of the current FSU/COAPS downscaling system, and Motivation
Error variance and Seasonal Anomaly Correlation (current downscaling system) (Lim et al. 2009)
Downscaled seasonal forecast with an improvement of Corr.
Reduction in Relative error variance (REV) (≈ 2 0.6~1.4)
Downscaled seasonal forecast with an improvement of Corr.
Reduction in Relative error variance (REV) (≈ 2 0.6~1.4)
REV Corr.
Corr. (0.~0.2)
Corr. (0.4~0.6)
REV > 6.0
REV < 1
Down. from FSU model
Categorical predictability (HSS) for the frequency of rainfall extremes (Lim et al. 2009)
Downscaling
Difference (Down. - Rescaling)
Rescaling (OA) from the CFS
Downscaling:
Florida and S. Georgia : > 0.1, Alabama and C. Georgia : -0.1 ~ 0.2,
Rescaling: -0.2 ~ 0.2
1 std. + climatology
0.1~0.5
-0.2 ~ 0.1
≥0.1
Dynamical models
1. FSU/COAPS NRSM
2. RSM (NCEP, ECPC)
3. RegCM3 (ICTP)
Statistical models
1. CRT (CSEOF + Regression + Time series generation)
2. NLCCA (Neural network based CCA) (Hsieh et al. 2006)
3. Geo-spatial weather generator (Baigorria et al. 2007)
Dynamical and statistical models involved in the multi-model downscaling study
Dynamical and statistical models involved in the multi-model downscaling study
Difference between FSU/COAPS downscaling works and other downscaling projects
Difference between FSU/COAPS downscaling works and other downscaling projects
NARCAPP MRED CORDEX FSU/COAPS
AimClimate change
projection
Seasonal forecast (winter)
Climate change
projection
All seasonal forecast, Climate
change proj.
DownscalingModels
RegCM3, CRCM, RSM,
HadRM3, WRF, MM5
MM5, WRF, RSM
RCMs……
FSU,RegCM3,RSM,
+ 3 Statistical models
Global Models
HadCM3,CGCM3,CCSM,
GFDL
CFS, GEOS-5
CMIP5 models
FSU, CCSM,GEOS-5,CFS
Resolution 50km 32km 50km 20km
Domain N. America US World SE US
ProceduresProcedures
Downscaling large-scale reanalysis using dynamical
models for model validation (bias, reliable distribution)
Downscaling large-scale retrospective forecasts
High-resolution seasonal forecasts on real-time basis
Probabilistic forecasts and application of the MME for
the improved deterministic forecasts
Expansion to high-resolution climate change
projection
Downscaling large-scale reanalysis using dynamical
models for model validation (bias, reliable distribution)
Downscaling large-scale retrospective forecasts
High-resolution seasonal forecasts on real-time basis
Probabilistic forecasts and application of the MME for
the improved deterministic forecasts
Expansion to high-resolution climate change
projection
Preliminary result: RCM response to
downscaling (2.5˚ → 20km) Preliminary result: RCM response to
downscaling (2.5˚ → 20km) 2m T. (JJA/2004) 2m T. (JJA/2005)
Preliminary result: RCM response to
downscaling (2.5˚ → 20km) Preliminary result: RCM response to
downscaling (2.5˚ → 20km) Prcp. (JJA/2004) Prcp. (JJA/2005)
Preliminary result:
Statistical downscaling models Preliminary result:
Statistical downscaling models 2m T. (JJA/2004) 2m T. (JJA/2005)
Preliminary result:
Statistical downscaling models Preliminary result:
Statistical downscaling models Prcp. (JJA/2004) Prcp. (JJA/2005)
SummarySummary
Multi-model high-resolution seasonal forecasting system study at FSU/COAPS was begun in September.
Three dynamical and three statistical models have been involved in this study.
We aim at the spatial resolution as fine as 20km for the southeastern US (FL, GA, AL, NC, and SC).
Improvement of the skill over our existing downscaling system (one dynamical and one statistical model) is expected.