Young-Kwon Lim, D.W. Shin, S. Cocke, T. E. LaRow, J. J. O - PowerPoint PPT Presentation

Loading...

PPT – Young-Kwon Lim, D.W. Shin, S. Cocke, T. E. LaRow, J. J. O PowerPoint presentation | free to download - id: ba481-ZDc1Z



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Young-Kwon Lim, D.W. Shin, S. Cocke, T. E. LaRow, J. J. O

Description:

YoungKwon Lim, D'W' Shin, S' Cocke, T' E' LaRow, J' J' OBrien, and E' P' Chassignet Center for Ocean – PowerPoint PPT presentation

Number of Views:11
Avg rating:3.0/5.0
Slides: 29
Provided by: cpcNce
Category:
Tags: larow | cocke | kwon | lim | nssdc | shin | young

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Young-Kwon Lim, D.W. Shin, S. Cocke, T. E. LaRow, J. J. O


1
Young-Kwon Lim, D.W. Shin, S. Cocke, T. E. LaRow,
J. J. OBrien, and E. P. Chassignet Center
for Ocean-Atmospheric Prediction Studies,
Florida State University, Tallahassee, FL, USA
Regional Climate Simulation of Surface Air
Temperature (Tmax) and Precipitation by
Downscaling over the Southeast US
2
Why downscaling over the SE USA?
  • Extremely high temperature and heavy rainfall
    with severe storms during summer, resulting in
    potential property damage and injuries.
  • The largest areas of agricultural farms in the
    nation.
  • An accurate forecast with higher spatial
    resolution is essential to adapt management,
    increase profits, reduce production risks, and
    mitigate damages.

3
Simulation of regional climate by FSU
  • FSU/COAPS Global Spectral Model (FSU/COAPS GSM)
    has been downscaled to the 20km grid resolution
    by FSU/COAPS nested regional spectral model
    (FSU/COAPS NRSM) over the southeast US. ?
    Dynamical Downscaling
  • FSU/COAPS NRSM 1) Same physics as the GSM, 2) 3
    or 6 hr nesting interval, and 3) Output Surface
    T, prcp., and radiative variables.
  • Statistical downscaling model has been also
    developed. (CSEOF, multiple regression, and
    stochastic PC generation are used.)

4
Observation
FSU GSM
0.2? 0.2 (20km res.)
1.8? 1.8 (180km res.)
CSEOF analysis
Training

Predictor model output
Regression
Predictand observation
Regressed eigenfunctions of GSM runs over
training period used
CSEOF PC generation
(for the prediction period)
Prediction period
Eigenfunctions of the Obs. over training period
and the Generated CSEOF PC used
Downscaled data construction
Withholding different year for Cross-validation
Output
5
Data (Obs. Model) and period
  • Variables Daily Tmax, Tmin, and precipitation
  • Period 1994 2002 (March September each year
    (daily))
  • Observed data source
  • National Weather Service Cooperative
    Observing Program surface data over the southeast
    US 20km20km
  • Large-scale model data
  • FSU/COAPS GSM 1.81.8 (T63), initial
    condition centered on Mar. 1 each year,
    seasonally integrated.

6
Results
  • 2-d monthly mean field (Obs. GSM, NRSM, and
    Statistical Down.)
  • Time series of monthly Tmax anomaly over the
    selected local grids (Tallahassee, Jacksonville,
    Orlando, Miami, Atlanta, Tifton, Birmingham, and
    Huntsville)
  • Time series of seasonal T anomaly and
    correlations
  • Categorical Predictability () for above/below
    seasonal T climatology
  • Predictability (e.g., rainy/dry, false alarm,
    HSS) for precipitation
  • Correlation and 3-category predictability for
    summer monthly prcp.

7
Monthly mean field (1994)
Spring
Summer
8
Monthly anomaly time series
Black solid ObservationRed solid statistical
downscalingBlue solid FSU/COAPS NRSM
  • Peaks seen in the observation are reasonably
    captured by both downscaling methods.
  • Both methods appear to have comparable skill in
    reproducing the observed fluctuations.
  • Poor coincidence in peaks between the downscaled
    and the observed time series are found at a few
    time steps (e.g., e, g, and h in 96 and 97).

9
Seasonal anomaly Time series
Black solid ObservationRed solid statistical
downscalingBlue solid FSU/COAPS NRSM Green
dashed GSM
  • Both downscaled time series tend to undulate in
    accordance with the observed time series
  • Incorrect predictions 94 summer, 95 spring, and
    97 spring
  • The relatively poor downscaling at these periods
    arises from poor simulation of the GSM anomaly.

10
Anomaly Correlation Top Statistical
downscalingMiddle FSU/COAPS NRSMBottom
Difference
seasonal, monthly
  • Correlation ranges from 0.3 to 0.8 over most of
    grids (seasonal).
  • Florida region tends to be highly correlated with
    observation.
  • Differences do not exceed the magnitude 0.1,
    indicating any of these methods is not
    significantly better than the other.

11
Categorical evaluation Left Correct forecast
(), second column () forecast but (-)
obs.(), third (-) forecast but () obs. (),
right Heidke skill score
SD
NRSM
12
MAE and Correlation for frequency of daily
extreme event
Top Statistical downscalingMiddle FSU/COAPS
NRSMBottom Difference
  • Correlations exceed 0.4 except for N. Georgia and
    Alabama, and SW tip of Florida.
  • Corr. Statistical downscaling shows higher
    correlations.
  • MAE Statistical downscaling shows greater MAE
    than dynamical downscaling. (significant
    overestimation / underestimation should be
    improved specifically in the statistical
    downscaling method.)

13
Monthly anomaly time series (Prcp.)
14
Categorical evaluation for rainfall event Left
Correct forecast (), second column False
alarm ratio (), third Prcp. missed (), right
Heidke skill score
SD
NRSM
15
Monthly anomaly correlation Categorical
predictability (summer)
16
Concluding remarks
  • Daily Tmax and Prcp. obtained from FSU/COAPS GSM
    (1.8lon.-lat., T63, seasonal integration) run
    have been downscaled to local spatial scale of
    20km for the southeast US region, covering
    Florida, Georgia, and Alabama.
  • Both downscalings better reproduce the
    regional-scale features of temperature and
    precipitation than the GSM.
  • A series of evaluations reveal that both
    downscaling methods reasonably produces the local
    climate scenario from large-scale simulations.
    Skills for T is greater than precipitation.
    Skills of both methods are comparable to each
    other.
  • FSU COAPS is the leading institution for regional
    climate simulation (downscaling) for seasonal
    forecast and crop model application over the
    southeast US.
  • Still remains a room for the improvement in
    predictive skill.

17
Statistical downscaling procedure (1)
  • 1. Cyclostationary EOF analysis for the model
    output and the observation
  • CSEOF (Kim and North 1997) analysis
    technique for extracting the spatio-temporal
    evolution of physical modes (e.g., seasonal
    cycle, ENSO, ISOs, etc.) and their long-term
    amplitude variations.
  • P(r,t)?n Sn(t)
    Bn(r,t)
  • Bn(r,t) time-dependent eigenfunctions,
    Sn(t) PC time series.
  • In this study, CSEOF is conducted on both
    observation and FSUGSM runs over the training
    period.

18
Statistical downscaling procedure (2)
  • 2. Multiple regression between the model output
    and the observation
  • CSFOF PC time series of the first significant
    modes of a predictor variable (FSUGSM data) are
    regressed onto a certain PC time series of the
    target variable (observation) in the training
    period.

  • PCTn(t)?ianiPCPi(t)e(t) i1,2,10
  • PCTn(t) target PC time series, ani
    regression coefficient
  • PCPi(t) predictor PC time series
  • Relationship between model output and the
    observation is extracted from CSEOF and multiple
    regression.

19
Result of multiple regression
  • PC time series

?
(training period)
forecast period
Eigenfunction (from Observation)
Regressed Eigenfunction (model)
Both are physically consistent.
20
Result of multiple regression
Eigenfunction (from Observation)
Regressed Eigenfunction (model)
21
Statistical downscaling procedure (3)
  • 3. Generating CSEOF PC of the model data over the
    forecast period from the regressed fields in the
    training
  • CSFOF PC time series of the model data are
    generated for the prediction period. Modeled data
    and the regressed eigenfunctions identified from
    training are used.

  • PCn(t)?gP(g,t)Bn(g,t)
  • PCn(t) the nth mode PC time series for the
    prediction period
  • g large-scale grid point
  • Bn(g,t) regressed CSEOF eigenfunctions
  • P(g,t) global model anomaly over the
    prediction period

22
Statistical downscaling procedure (4)
  • 4. Downscaled data construction from the
    eigenfunctions of the observation and the
    generated CSEOF PC time series

  • D(s,t)?nPCn(t)Bno(s,t)
  • PCn(t) generated PC time series from the
    previous step
  • Bno(s,t) CSEOF eigenfunctions of the
    observation (training period)
  • D(s,t) downscaled output
  • 5. Generating downscaled output for the entire
    period (9yrs) by cross-validation framework

23
Observation
FSU GSM
0.2? 0.2 (20km res.)
1.8? 1.8 (180km res.)
CSEOF analysis
Training

Predictor model output
Regression
Predictand observation
Regressed eigenfunctions of GSM runs over
training period used
CSEOF PC generation
(for the prediction period)
Prediction period
Eigenfunctions of the Obs. over training period
and the Generated CSEOF PC used
Downscaled data construction
Withholding different year for Cross-validation
Output
24
Monthly time series(Tmax) Black solid
ObservationRed solid statistical
downscalingBlue solid FSU/COAPS NRSMGreen
dashed FSU/COAPS GSM
  • Downscaled results are closer to observation than
    FSU/COAPS GSM.
  • Warm or cold biases unveiled from GSM have been
    corrected by downscaling.

25
Seasonal mean field (example 97-98 summer)
  • Interannual temperature difference between the
    two years.
  • Higher (lower) T in 98 (97) with detailed spatial
    structure is simulated by the two downscaling
    methods.
  • The GSM fields have limited capability to realize
    the regional temperature fields over the domain.

26
The number of extreme Tmax events
Black solid ObservationRed solid statistical
downscalingBlue solid FSU/COAPS NRSM
  • Extreme T events exceed the one standard
    deviation plus climatology.
  • Interannual change in the occurrences of extreme
    Tmax (warmer T) events are fairly captured at
    individual grids by both downscalings.

27
Mean absolute error
Top Statistical downscalingMiddle FSU/COAPS
NRSMBottom FSU/COAPS GSM (interpolated)
  • MAE 0.8 2C (GA, AL).
  • MAE 0.4 1.5C (FL).
  • FSU/COAPS NRSM (dynamical downscaling) has the
    smallest biases.

28
Categorical evaluation
  • Two categories above average and below average
  • Correct forecast the same sign of anomalies
    between observation and the downscaled forecast
    (Paa, Pbb) ?
  • Incorrect forecast opposite anomalies between
    observation and downscaled forecast (Pab, Pba) ?
    ,
  • Heidke skill score
  • PE probability of a
  • random forecast
  • (F and P are independent)

Verifying analysis Verifying analysis Forecast Forecast
Verifying analysis Verifying analysis above normal below normal
Obs. above Paa Pba PaP
Obs. below Pab Pbb PbP
PaF PbF 1
About PowerShow.com