The bottom sounding survey data set collected by Zvi Ben-Avraham in 1986-87 on the Sea of Galilee in Israel Ben-Avraham et al. (1990) is often used at SEP to test new algorithms for data interpolation. The main difficulties associated with interpolating the Galilee data set onto a regular grid are inconsistencies in the data due to random spikes, and an acquisition footprint (ship tracks) left in the final image caused by systematic error in the data.
There are different approaches developed at SEP to confront these difficulties Brown (2001); Claerbout (1999); Fomel and Claerbout (1995); Fomel (2001). Fomel and Claerbout (1995) suggested to use a derivative operator to filter out low-frequency components of the residual, and the Iteratively Reweighted Least Squares (IRLS) technique to suppress non-Gaussian spikes in the data. However, they showed that while the acquisition footprint and the noisy portion of the model disappear, the price of the improvement is a loss of image resolution.
In this paper we implement an idea from Claerbout (1999), and use a bank of a Prediction Error Filters instead of the derivative operator to whiten the residual on individual data tracks before implementing IRLS. We show that our algorithm leads to reduced artifacts in the final map without an apparent decrease in map resolution.