Mean traveltime curves analysis: a method to improve understanding of data behaviour in 2-D transmission tomography at the pre-inversion stage (Q1001702)

Transmission tomography methods show a great sensibility to data variability, which eventually includes data errors, often present in field experiments. Local optimization methods, traditionally used to solve this inverse problem, are very sensitive to these difficulties, failing to converge properly in the presence of spurious data. Regularization methods partially cope with these weaknesses, damping the instabilities. A complementary approach, adopted here, is to perform a structured analysis of data variability before the inversion, oriented to discriminate the contribution of errors from that of true geological heterogeneities. The key concept of mean traveltime curves (\(\overline t\)- curves and \(\sigma\)- curves) is introduced and described. Their analytical equations are deduced for isotropic homogeneous media and any recording geometry. Empirical mean traveltime curves can be inferred based solely on traveltime data, using the corresponding discrete estimators. The methodology proposed here proceeds through a user-defined subdivision of the domain of interest into isotropic homogeneous areas. Least squares velocity estimations and associated data misfits are used to scrutinize the behaviour of the implied source-receiver sets and of the ray-swept part of the geologic medium. Data are considered suspicious if zonal estimated velocities are non-consistent with a priori information. Also, independent fitting of both empirical curves helps to classify the genesis of the residuals: some situations are illustrated. Finally, we show the application of this technique to a data set from the Grimsel test site in Switzerland. Using this methodology, we detect some anomalous gathers, which may be responsible for the large range of velocities found in the initial imaging with this data set. Also, we give some indications of the possible sources of these anomalies. This approach offers a quick data variability analysis in the pre-processing stage, which, even if no data editing algorithms are finally used, always improves the understanding of the data structure.