Skip navigation.
Home
INCUSAR
Home

3rd bimonthly progress report (October-November 2009)

ASAR Doppler Centroid calibration

In this period, more work has been done together with Morten Wergeland Hansen at NERSC on the calibration of the Doppler Anomaly from Envisat Wide Swath images. Various methods have been tested to remove the so called "range bias". The range bias is of instrumental rather than geophysical origin, and leads to a strong variation of the Doppler anomaly along the range direction. The effect is due to misfunctioning tiles of the ASAR antenna, which modifies the antenna pattern, and has been qualitatively and qualitatively explained by numerical simulations by Davide D'aria at Aresys, in cooperation with Fabrice Collard and Pierre Fabry at CLS. Although it can be modelled, the most practical solution is still to perform empirical corrections on the data. Previously for the proecssing at NERSC, the mean Doppler anomaly of each range line of any SAR scene was subtracted, but this had the unwanted side effect that also any mean geophysical Doppler was removed. In the updated processing scheme, the mean geophysical Doppler along each range line is now re-added with the CDOP function, after the subtraction of the overall mean Doppler. Apriori wind information input to CDOP is taken from the NCEP forecast model. Since the instrumental bias has been found to vary slowly, future enhancements can be to improve the accuracy by estimating it over several scenes of the same or preceding orbits, and to use blended (Bayesian) SAR/model wind for the correction. Hence, a sprial structure is seen in this work: an improved (Bayesian) SAR wind retrieval can improve the Doppler calibration, which again can improve the wind retrieval; and likewise for the subsequent current retrieval.

A presentation of this work will be given by Morten Wergeland Hansen at the SeaSAR workshop in ESRIN in January 2010. A paper is also in preparation for the upcoming IEEE Special Issue on Advances in Remote Sensing Image Processing in 2010.

Bayesian wind retrieval - grid size

Whereas the principle of Bayesian Wind retrival is straightforward (see e.g. 1st progress report), an important question is at which spatial resolution to evaluate and minimize the cost function. In a typical Wide Swath scene there are on the order of 100 million pixels of 75 m. The variation of sigma0 at this scale can not be all attributed to real changes in wind speed and/or direction, so for traditional wind estimation from SAR with the CMOD-function, pixels are normally averaged to blocks of between 500 m and 2 kilometres. Even when averaged to 500 m, the number of pixels is too large for practical evaluation and minimization of the cost function. This problem was solved in two ways in the algorithms described in previous works:

  • In Portabella et al. (2002) the cost function was only evaluated in vicinity of the a priori (model) wind vector. The drawback of this solution is that the optimum solution may not be found in vicinity of sharp and large changes in wind direction, such as close to fronts. Note that for this work the Doppler anomaly was not an available resource.
  • In Kerbaol et al. (2007) Mouche et al. (2009) the cost function was only evaluated for wind speeds and directions which satisfied perfectly the CMOD relationship with roughness. In other words, both the observed sigma0 and the CMOD function itself was assumed free of errors.

In the latter work, a problem was found that in some cases no solutions were found which could also satisfy the CDOP relationship with the observed Doppler, and the a priori wind, within the bounds set by their respective error estimates. In the Incusar project, an error estimate will be provided also for sigma0, like in Portabella et al. (2002), but the cost function will still be evaluated over the full solution space (i.e. for U and V having any value between -25 and +25 m/s). Instead the computational challenge will be solved by increasing the pixel size for the computational grid. Decreasing the resolution may in fact be an advantage, since validation of (traditional) SAR wind versus buoys showed improved agreement for increasing pixels sizes up to 50 km (1st progress report), and secondly because the Doppler and apriori wind is also given a coarser scales. Hence a tradeoff must be found between resolving fine resolution wind variation and improving accuracy by decreasing resolution.

For development and testing of the approach outlined above, the ASAR image shown below has been selected. In this case the model wind field does not show a wind front which is clearly visible in the Doppler image, and also observable in the roughness image.

ASAR scene off the coast of Norway 4 May 2009 20:56 UTC. Left is roughness image, middle is with Doppler anomaly overlaid in colour. Right is NCEP wind speed and direction from 21 UTC. Manually estimated wind direction based on Doppler and roughness variations is drawn by hand over the roughness image. The front is drawn with a dashed line.
ASAR scene off the coast of Norway 4 May 2009 20:56 UTC. Left is roughness image, middle is with Doppler anomaly overlaid in colour. Right is NCEP wind speed and direction from 21 UTC. Manually estimated wind direction based on Doppler and roughness variations is drawn by hand over the roughness image. The front is drawn with a dashed line.

 

The development of the algorithm with be presented on the SeaSAR workshop in January 2010, where the scene above, an potentially others, will be used to illustrate the concepts.