1. Regional re-analysis without regional data
VON STORCH Hans
Institute of Coastal Research, Helmholtz Zentrum Geesthacht, Germany
with material of 李德磊
IOCAS, Qingdao
17+3 minuten
29 September NUIST, 南京 (Nanjing)

2. VON STORCH Hans Climate researcher (in the field since 1971)
Coastal climate (storms, storm surges, waves; North and Baltic Sea, North Atlantic, Yellow Sea); statistical analysis
Director emeritus of the Institute of Coastal Research of the Helmholtz Zentrum Geesthacht, Germany
Professor at Universität Hamburg
Guest professor at 中国海洋大学

3. Overview
Spectral nudging approach for constraining regional modelling - Improved representation of medium scale variability - Constructing a homogeneous regional re-analysis.
Improved representation in coastal regions: - variability at medium scales. - sub-synoptic phenomena (e.g., polar lows) - forcing fields for impact models (ocean waves, storm surges)
The regional reanalysis CoastDat for governmental agencies, commercial companies and the public/media

4. Climate and downscaling
The genesis of „regional“ climate my be conceptualized (only at midlatitudes?) as R = f(L, Φs) with L = larger scale climate, R = smaller scale climate, Φs = physiographic detail at smaller scale (mountains, cities, coastlines etc.).
Climate models generate numbers for all scales, beginning at the grid resolution. The smallest of these scales are heavily disturbed because of an insufficient representation of the small-scale physiography but also because of the abrupt truncation at the grid resolution.
The largest scales in global re-analysis may be in many cases considered as well and homogeneously described. The smallest scales are considered less realistic and subject to inhomogeneities (sensitive to changes of observational local data availability).

5. Skillfully represented scales Insufficiently represented scales
Maximum
resolution
Grid point resolution
Skillfully represented scales
Insufficiently represented scales
Global Analysis
Maximum
resolution
Grid point resolution
Regional
Analysis
Skillfully represented scales
Insufficiently represented scales
Con-straining
Added value
Downscaling using R = f(L, Φs) processes the reliably and homogeneously described large scales L. The constraining may be implemented by spectral nudging into a regional or even global dynamical model. (von Storch, H., H. Langenberg and F. Feser, 2000: A spectral nudging technique for dynamical downscaling purposes. Mon. Wea. Rev. 128: )

6. Constraining of large scales
global
Pattern correlation coefficients for meridional wind at 500 hPa between the driving global reanalysis and the
RCM-output with standard forcing via the lateral boundaries
and the
RCM-output with spectral nudging
regional

7. Regional re-analysis without regional data.
exploiting the presence of a “downscaling” situation L > R
and the availability of long, homogeneous re-analysis of “large-scale” dynamics L.
Added value is in R, i.e. in the regional detail, which results from L and the regional physiographic detail (such as coastlines, mountains)
Done for - Europe (including hydrodynamic of marginal seas; COASTDAT) - China (Yellow Sea region, see 李德磊), - Central Siberia and Southern Atlantic, - nd recently the Globe.

8. A 5-year simulation 2001-2005 for the region of South China李德磊, pers
A 5-year simulation for the region of South China李德磊, pers. Comm.)
CCLM_JRA55: CCLM simulation driven by JRA55 (and, alternatviely ERA-I) reanalysis dataset (7km, )
is compared to APHRODITE daily precipitation and temperature datasets with high-resolution grids for Asia (0.25°)
Domain-averaged seasonal climatology and temporal evolution (annual cycle removed) of surface temperature and total precipitation:
PS: Reanalysis dataset: ERA-Interim 1979 – present, JRA55: 1958 – present

9. precip amounts
2m temperature
JJA %
DJF % K
CCLM_JRA55 bias relative to observation for daily precip amounts and mean daily mean temperature in summer and in winter

10. precip amounts
2m temperature
JJA %
DJF
Correlations between CCLM_JRA55 model data and observations for daily precip amounts and mean daily mean temperature in summer and in winter

11. precip amounts
2m temperature
JJA %
DJF mm K
Root mean square errors between simulations and observation dataset
for daily precip amounts and mean daily mean temperature in summer and in winter

12. Improved presentation of in coastal regions
ERA-I-driven multidecadal simulation with RCM CCLM over the Bohai Sea and the Yellow Sea region (李德磊, 2015)
Grid resolution: 0.06o
Material by 李德磊

13. Model setups and datasets
Long-term simulation
COSMO-CLM (CCLM): , spectral nudging
Forcing dataset
ERA-Interim (1979 – 2013,
~ 80km, 6 hourly)
Physical Para-meterization
Tiedtke convection scheme and
Charnock-formula for RL
Resolution
Spatial: ° (~7 km),
40 vertical layers, hourly output
Grid
Rotated coordinate system
168 x 190
Land Station Data
67 land surface stations, 3-hourly (HadISD)
Marine Station
Data
7 stations, including 3 offshore stations and 4 coastal stations (NCDC and KMA)
Satellite Data
QuikSCAT L2B in 12.5 km swath data (QSCAT, )
This is my model domain. This part is Bohai Sea, and this is Yellow Sea. A long-term simulation driven by ERA-Interim was run for 35 years from , with spectral nudging technique. Tiedtke convection scheme and land surface process TERRA-ML scheme were used in the physical parameterization. The resolution is 7km, and has hourly output and 40 vercial layers. To verify the model wind speed output, several different datasets were used, including 67 land surface station; 7 marine station with 3 offshore stations and 4 coastal stations. QuikSCAT swath data was used as reference as well.

14. CCLM compared with QSCAT grid data (7 km)
Mean wind speed:
6.5 – 9 m/s
Negative Bias:
around -1 m/s
Correlation coeff. :
RMSE: 1.5 – 3.0 m/s
(m/s)
(m/s)
In some coastal areas : larger bias, lower correlation coeff., larger RMSE
This one shows QSCAT mean wind speed, which varies from 6.5 m/s to 9 m/s, and increase from northwest to southeast direction. The bias between CCLM and QSCAT are mostly negative, around -1 m/s. The Correlation coefficient are in the range of Root mean square error is generally in the range of 1.5 – 3 m/s. We see that in the coastal region, we have larger bias, lower correlation coefficient and larger RMSE than over the open see areas. Comparing with some other studies on modeled wind speeds assessment, we can say the CCLM wind speeds are reliable. based on the statistical measures.
(m/s)

15. Is there an added value? SEO IEO
BSS (CCLM, ref: ERA-I, obs:QSCAT > 3 m/s)
BSS (CCLM, ref: ERA-I, obs:QSCAT > 10.8 m/s)
Modified Brier Skill Score: BSS (Winterfeldt et al. 2011)
𝐵𝑆𝑆= 1− 𝜎 2 𝑥 𝐶 , 𝑥 𝑂 𝜎 2 𝑥 𝐸 , 𝑥 𝑂 𝑖𝑓 𝜎 2 𝑥 𝐶 , 𝑥 𝑂 < 𝜎 2 𝑥 𝐸 , 𝑥 𝑂 𝜎 2 𝑥 𝐸 , 𝑥 𝑂 𝜎 2 𝑥 𝐶 , 𝑥 𝑂 −1 𝑖𝑓 𝜎 2 𝑥 𝐶 , 𝑥 𝑂 > 𝜎 2 𝑥 𝐸 , 𝑥 𝑂
𝜎 2 𝑥 𝐶 , 𝑥 𝑂 is error variance between the regional model data and QSCAT;
𝜎 2 𝑥 𝐸 , 𝑥 𝑂 is error variance between the reference data ERA-I and QSCAT.
SEO
BSS: [-1, 1]
BSS =1, Perfect model
BSS > 0, added value (improvement)
BSS < 0, no added value
IEO
The next question we want to answer is: is there an added value? We used a metrics named modified Brier Skill Score to assess the added value issue. The equation here is to calculate modified BSS. Sigma square Xc is error variance between the regional model data and QSCAT; sigma square Xe is error variance between the reference data ERA-I and QSCAT. BSS is in the range of negative 1 to positive one. In case of perfect model, Sigma square Xc equals to zero, then BSS=1, if BSS > 0, there is an added value, and vise verse.
For wind speed larger than 3 m/s, we see that positive BSS are mainly along coastal region, while negative in the offshore region. For strong wind speeds > 10.8 m/s, there are more areas with positive BSS, not only coastal areas but also some offshore areas. The results shown that added value (positive BSS) along coastal areas for winds > 3 m/s; while areas with added value is larger for strong winds (right) than for winds > 3 m/s (left).
To verify the finding based on QuikSCAT observation data, we further compared the CCLM wind speeds with station observations: one in the offshore area and another in the coastal area.
Brier Skill Score is a parameter reflecting to what extent the CCLM wind gives a better reproduction (added value) of the observation than ERA-Interim
when we analysis the added value of strong winds (> 10.8), we can see larger area with positive BSS than for all winds.
Added value (positive BSS) in coastal areas for wind speed > 3 m/s
Areas with added value are larger for strong wind speed (right) than for wind speed > 3 m/s (left)

16. Comparison of CCLM/ERA-I and in-situ data: offshore station (IEO)
qq-plot
The left one shows the comparison between CCLM and observation wind speeds, the right one is ERA-interim and observation. At offshore station IEO, we sWe can see that the Q-Q plot, scatter distribution and wind density distribution are similar with each other. About the statistical measures, they are also very similar with each other. There is no improvement by CCLM wind speeds to ERA-Interim wind speeds.
scatter plots, qq-plots as well as Kernel density estimation contours
No improvement of CCLM wind speed (left) relative to ERA-I wind speed (right) at offshore station IEO

17. Comparison of CCLM/ERA-I and in-situ data : coastal station (SEO)
qq-plot
While for coastal station SEO, we see much improvement by CCLM wind speeds to ERA-I based on the Q-Q distribution, scatter plot. The metrics also show obvious improvement by CCLM wind speeds relative to ERA-Interim wind speeds.
Therefore, based on satellite data and station data, we get the conclusions that CCLM can add value to ERA-I in the coastal areas but not in the offshore and open sea areas.
CCLM wind speed (left) is in better agreement with observations than ERA-I wind speed (right) at coastal station SEO

18. Seasonality of water-region-averaged wind speed
This bar plots show the seasonality of water-area-averaged wind speed for three products.
As we can see, CCLM and ERA-I wind speeds underestimate QSCAT observation in all seasons. In winter, CCLM is consistent with ERA-I, and in the other three seasons, we see improvement by CCLM, especially in summer season.
CCLM and ERA-I wind speeds underestimate QSCAT observations in all seasons
More improvement of CCLM wind speed relative to ERA-I wind speed in summer than in other seasons especially in winter

19. Climatological mean of annual extreme surface wind speed (99%-iles)
About the climatological mean of annual extreme wind speed, we can see that the Similar distribution patterns are similar between ERA-I and CCLM wind fields
However, over the marine area, CCLM wind speeds tend to be stronger. while over the mountainous areas, CCLM wind speeds are lower than ERA-Interim wind speed. The wind speed of CCLM has stronger spatial variability as well.
When comparing with station observation, we see that CCLM is better in agreement with observation in many areas. These above are quantitative assessment of CCLM surface wind speeds. Next we want to know whether there is any physical pheonomena that underlying the added value revealed based on statistical scores.
(m/s)
Similar distribution pattern, more spatial variability and detail in CCLM wind field
CCLM: lower values over mountainous areas and larger values over some water areas
CCLM: better in agreement with observations

20. Generating additional regional dynamical detail
NCEP-driven multidecadal simulation with RCM CLM over North Pacific (陈飞 et al., 2012, 2013, 2014)
Grid resolution: about 0.4o
Employing spectral nudging (wind above 850 hPa, for scales > 800 km)
Simulation of sub-synoptic phenomena
Polar lows in the Northern North Pacific

21. North Pacific Polar Lows
(陈飞 et al., 2012, 2013 and 2014)
North Pacific Polar Low
on 7 March 1977
NOAA-5 infrared satellite image at 09:58UTC 7th March 1977

22. Annual frequency of past polar lows in the North Pacific
Number of detected Polar Lows in the North Pacific per Polar Low season (PLS; October to April). The trend from 62 PLSs, from 1948/1949 to 2009/2010, amounts to 0.17 cases/year.
陈飞 et al., 2013

23. Improved representation of forcing fields for impact models
NCEP-driven multidecadal simulation with RCM REMO in Europe
Grid resolution: 0.5 o
Employing spectral nudging (wind above 850 hPa, for scales > 800 km)
simulation
Wind and air pressure used to drive hydrodynamical models for describing currents and sea level
Wind used to drive models of the statistics of surface waves (ocean waves) in coastal seas (North Sea).

24. Red: buoy, yellow: radar, blue: wave model run with REMO winds
significant wave height
[days]
wave direction
[days]
Red: buoy, yellow: radar, blue: wave model run with REMO winds
Gerd Gayer, pers. comm., 2001

25. Interannual variability of mean water levels
(Weisse and Plüß 2006)
Annual mean winter high waters Cuxhaven red – reconstruction, black – observations

26. The CoastDat-effort at the Institute for Coastal Research@HZG
Long-term, high-resolution reconstructions (60 years) of present and recent developments of weather related phenomena in coastal regions as well as scenarios of future developments (100 years)
Northeast Atlantic and northern Europe.
Assessment of changes in storms, ocean waves, storm surges, currents and regional transport of anthropogenic substances.
Applications
many authorities with responsibilities for different aspects of the German coasts
economic applications by engineering companies (off-shore wind potentials and risks) and shipbuilding company
Public information
Integration area used in HZG reconstruction and regional scenarios 26

27. Changing significant wave height, 1958-2002
waves
wind
waves
Weisse, pers. comm.
Yantai, 18 June 2007

28. Some applications of Navigational safety Offshore wind
Ship design
Navigational safety
Offshore wind
Interpretation of measurements
Oils spill risk and chronic oil pollution
Ocean energy
Scenarios of storm surge conditions
Scenarios of future wave conditions
Wave Energy Flux [kW/m]
Currents Power [W/m2]
Weisse, pers. comm. 28

29. Conclusion …
Dynamical downscaling (R = f(L,Φs)) works … - Large scales are hardly affected but smaller scales are more realistically described in heterogeneous regions (coasts, mountains)
Downscaling allows the generation of homogeneous data sets i.e., a regional re-analysis with uniform quality (across time).
Added value in describing medium scale phenomena – such as wind storms
Added value in generating regional impact variables, such as wind for storm surges and ocean waves.
Several such regional re-analysis has been done, among them for the region of the Bo Hai and the Huang Hai on a 7-km grid by 李德磊.