6th bimonthly progress report

Sensitivity testing of Bayesian wind algorithm

An Envisat ASAR Wide Swath scene from the North- and Norwegian seas from 23 August 2010 is here used to exemplify the performance of the implemented Bayesian wind algorithm.

The NRCS image, the Doppler surface velocity, and the corresponding NCEP wind is shown below (links only, as the website editor is currently not accepting images). This case is very interesting, because it contains a "normal" wind front in the Northern part, and a sharp cyclonic wind field east of England.

The Doppler surface field agrees well with the NCEP wind field, but contains more sharp features. Since these sharp features agrees well with features of the NRCS image, the Doppler velocity seems here to give a good representation of the wind field, at least qualitatively.

Wind calculated with the Bayesian algorithm, with two different values used for the error of the Doppler (5 and 10 Hz), and three different values used for the error of the model (0.5, 1 and 2 m/s). Ref to equation from the 1st progress report. The NRCS error is kept constant at a value of 0.078*NRCS, as suggested by Portabella et al. (2002).

Modifying these error-parameters, changes the weighting of the Doppler velocity and the model information. At the one extreme, by taking model error of 0.5 m/s and Doppler error of 10 Hz, the Bayesian wind field is very smooth, like the original model field. At the other extreme, by taking a large model error of 2 m/s and a small Doppler error of 5 Hz, more features are seen, but unfortunately also some artifacts. A trade-off, which also corresponds to reasonable errors, model error of 1 m/s and Doppler error of 10 Hz, gives a compromise with no artifacts, but still some details provided by the SAR image.

For the case where artifacts are seen, it is interesting to note that much of this occurs south-west of Norway over the Norwegian coastal current which is known to flow with a speed of about 0.5 m/s. This is a reminder that the Doppler velocity is due to both wind-waves and current, and that a combined consistent retrieval should be performed. One method which will be tested, is based on the assumption/experience that the coastal current is very regular in time, whereas the wind is situation dependent. Hence a mean (climatology) current field can be subtracted from the Doppler velocity before wind inversion. This is a "poor-man's version" of the full 2d-surface current inversion scheme suggested in WP 2 of this project. This will soon be tested.

In parallel to development, testing and preparation for 2D current inversion, two papers are in preparation and soon ready for submission:

• A paper about the principles of the Bayesian SAR wind retrieval in cooperation with Alexis Mouche, CLS. This will be submitted to GRL. A planned longer follow-up paper to be submitted to JGR will contain more sensitivity tests and validation.
• A paper on the calibration of Doppler centroid values from ASAR is in preparation together with Morten W. Hansen at NERSC. This will be submitted to a special issue of IEEE TGARS. 

In addition, SAR-based and WRF-modeled Bora-winds in the Adriatic Sea has been compared and assessed together with Dr. Milivoj Kuzmic from Ruđer Bošković Institute of Croatia. A draft of a paper on this topic is being discussed.