offline assessment of nesdis oscat data li bi 1 sid boukabara 2 1 rti/star/jcsda 2 star/jcsda...
TRANSCRIPT
Offline Assessment of NESDIS OSCAT data
Li Bi1
Sid Boukabara2
1RTI/STAR/JCSDA2STAR/JCSDA
American Meteorological Society 94th Annual Meeting02/06/2014
Outline
• Motivation• Introduction of statistics used in this study• Preliminary results– Raw data– Filtered data
• Conclusion • Future work
Motivation
• Optimize the usage of OSCAT data by performing offline assessment before the assimilation.
• Based on the offline assessment statistics of wind speed, direction, u/v component bias and STDV comparing with GDAS analysis, suggest optimal QC methods.
• Data usage:– NESDIS OSCAT 50km data June 2012
HDF5/BUFR data.• Selected wind vector cell quality flags:
– Land-sea boundary flag– Land flag– Ice flag– Rain impact flag– Wind retrieval flag
Fig.1. Selected wind vector cell quality flags
Statistics for raw data
• Calculated O-B bias and STDV w.r.t. GDAS analysis– Wind speed, direction, U/V components
• Geographic stats• By observation bins• By SST range• By cell index
– Ascending, descending separated.– Histogram of counts in each bins– Latitude band profile
20120601-20120630 speed bias raw data 20120601-20120630 speed bias filtered data
20120601-20120630 speed STDV raw data 20120601-20120630 speed STDV filtered data
20120601-20120630 direction bias raw data 20120601-20120630 direction bias filtered data
20120601-20120630 direction STDV raw data 20120601-20120630 direction STDV filtered data
Raw data ascending Raw data descending Raw data all
Filtered data ascending Filtered data descending Filtered data all
Raw data ascending Raw data descending Raw data all
Filtered data ascending Filtered data descending Filtered data all
Summary finding after applying basic retrieval filtering
• The basic retrieval filtering effectively remove land/sea boundary flag, ice flag, etc.
• Slightly reduce latitude band wind speed STDV. Direction latitude profile remains similar except for high latitudes.
• Latitude band cut off (60oS-60oN) is suggested.• Further assessment for optimal filtering is very
necessary.
Raw data ascending (speed, dir, u/v)
Raw data descending (speed, dir, u/v)
Raw data all (speed, dir, u/v)
Filtered data ascending (speed, dir, u/v)
Filtered data ascending (speed, dir, u/v)
Filtered data all (speed, dir, u/v)
Layout of OSCAT vs. GDAS stats bias and STDV
Wind Speed - Raw data
Wind Speed Filtered data
Speed Relative Bias - Raw data
Speed Relative Bias - Filtered data
5m/s cut off for wind speed is suggested
Wind Direction - Raw data
Wind Direction - Filtered data
Count Histogram - Raw data
Count Histogram - Filtered data
Basic filter + 5m/s cut off for wind speed + high wind speed cut off for U/V components
Raw data
Filtered data
15m/s cut off for U-Comp is suggested
Raw data
Filtered data
15m/s cut off for V-Comp is suggested
Raw data ascending Raw data descending Raw data all
Filtered data ascending Filtered data descending Filtered data all
Wind Speed - Raw data
Wind Speed - Filtered data
Wind Direction - Raw data
Wind Direction - Filtered data
Raw data all
Filtered data all
Wind Speed - Raw data
Wind Speed - Filtered data
Wind Direction - Raw data
Wind Direction - Filtered data
Count Histogram - Raw data
Count Histogram - Filtered data
Conclusion
• The optimal filtering effectively reduced large STDV for wind speed, as well as wind direction in the observation bins.
• Suggested new filtering includes:– Old filtering provided during the retrieval to remove ice/rain flag
etc. – 5m/s cut off for wind speed– Latitude band cut off (60oS-60oN only)– 15m/s cut off for high u and v components of the winds
• Wind speed biases generated in the observation bins.• Largely reduce wind direction STDV in scan position which is
mainly due to low winds.
Future Work
• Test the optimal observation error • Possible bias correction based on the preliminary results• Revisit thinning methods• Test with high resolution OSCAT data (25km)• Exploring assessing and assimilating the scatterometer
data in rainy conditions (pending maturity of the products)
• Perform the data assimilation experiment with the optimal filtering, bias correction and observation error.