Although least–squares inversion is a useful tool in data analysis, nonuniqueness is an inevitable problem. This problem can be analyzed by considering the sensitivity of a model response to the parameter estimates. Such sensitivity methods produce extremal solutions which barely satisfy some resolution (or stability) criteria. Two closely related methods for producing such solutions are the “Edgehog” and “Most Squares” methods due to Jackson (1973, 1976).
These techniques, which evaluate the “degree of nonuniqueness” in a least–squares inversion, require only the information computed in a singular value decomposition (SVD) solution. While the “edgehog” and “most squares” techniques are mathematically similar, the “most squares” estimate is the simpler to compute. Both methods show that the credibility of an inversion depends on both the specified error criterion as well as on the properties of the Jacobian matrix associated with the least–squares solution. The similar performance of these two closely related methods is demonstrated with the traveltime inversion of both synthetic and real vertical seismic profiles (VSPs). The sensitivity analysis of this inversion problem provides a quantitative measure of solution reliability.