12 th JCSDA Workshop Ocean Data Assimilation Development of a GSI-based DA interface for operational wave forecasting systems at NOAA/NCEP Vladimir Osychny, Hendrik Tolman, Henrique Alves, Arun Chawla
The main objective of the project: to develop a GSI-based module in WAVEWATCH III for assimilation of total significant wave height (Hs) from altimeter missions Completed work: - developed a quality-control (QC) module for Near-Real-Time (NRT) Hs data from satellites - developed a strategy to adapt the GSI for Hs assimilation using RTMA 2D approach (in collaboration with RTMA team: Manuel Pondeca, Steven Levine ) - modified the GSI code (RTMA 2DVAR) to include the new variable - significant wave height - determined that current RTMA prepbufr has enough wave-height data to start preliminary tests
Development of the QC procedure was based on Jason-1 NRT Hs data for 2011 obtained via GTS In principle: 254 passes ~10 days exact repeat cycle ~ 6 km (1 sec) sampling rate 3-10 km Hs footprint In NRT GTS reality: Not quite exact repeat passes Not quite regular alongtrack sampling
Example of raw SWH Jason-1 data: Dec. 6, 2011
Developed QC procedure includes: 1. Valid value (range) test 2. Proximity to land test 3. Proximity to ice test 4. De-spiking (statistical outliers)
Data rejected based on proximity to land test For each data location: a data is flagged as being likely “bad”, if a land point is found within the area with radius approx 20 km Test is based on ETOPO-1 data set, which is also used in operational wave model
Data that are likely affected by proximity to floating ice
Ice Concentration NCEP operational (5’ grid)
For each data location: → a data is rejected, if ice is found within the area with radius approx. 20 km -- same as for the “land” test; → this search radius seems to be too small in the case of ice Details of the Proximity to Ice Test
Results of the proximity to ice test are more accurate with a larger search radius – 40 km Also shown are “spikes” identified by the de-spiking procedure
An iterative de-spiking procedure Iterative core: 1.A low-pass signal is obtained by using an order-statistic filter: Approx. 10 sec. (~60 km, ~11 points; 5 minimum) data window Mean is calculated for values between 20 th and 80 th percentile 2.Estimate STD based on the high-pass residue for the same data window and the same data selection 3.Flag outliers (> 3STD) 4.Additional constraints at each iteration: - test differences between neighboring data values for original data and for high-pass portion - introduce lower limit on RMS
The next step: use the Real-Time Mesoscale Analysis (RTMA) 2DVAR approach to adapt the GSI for Hs assimilation What is RTMA? operational hourly analysis of atmospheric surface data based on GSI- and an atmospheric forecast model RTMA is the best choice of development framework for our purposes because: similar set up (although different models, grids, etc.) relatively simpler case to start with substantial existing expertise opportunity to add a valuable (for forecasters) new analysis variable to an existing operational system (RTMA) while developing a data assimilation module for a wave model
Summary: Concluded development of a working version of the QC procedure In progress: -Transfer the QC procedure to FORTRAN or Python (currently in Matlab) -Test the QC procedure on real time GTS data (Jason-2) -Start pre-operational cycling on WCOSS In progress: -modify RTMA to include analysis of Significant Wave Height -work with EMC obsproc group to include altimeter wave height data into RTMA prepbufr -further modify RTMA code to build the GSI-based data assimilation module for the operational wave model