Depth estimation has been used widely as a tool for rapid interpretion of large-scale potential-field data in applications such as mapping basement relief. Nearly all of these techniques rely on the analysis of the local shape of the anomalous field in determining the depth and location of the subsurface sources. These methods focus on the phase information at the expense of the amplitude of the data. Consequently, these methods often produce a large number of solutions and interpretation of the result is difficult. We develop a method for enhancing these techniques by incorporating amplitude information back into the depth estimation process. The method statistically identifies significant source solutions from the estimation based on their relative source strengths, and discards false solutions due to noise and spray effects. The result is a subset of solutions that is more amenable for direct interpretation. We illustrate this new approach by applying it to the solution of the Euler and extended Euler deconvolutions. We demonstrate the improvement using magnetic data from the Bishop model, and present a field dataset from petroleum exploration.
Depth estimation is a class of well-known techniques for finding the location and depth of potential-field sources. Traditionally, it has primarily been used to find depth to magnetic basement for petroleum exploration problems. There has been much published on the different techniques for depth estimation. These include the Naudy method (Naudy, 1971), Werner deconvolution (Werner, 1953), Euler deconvolution methods (Reid et al., 1990), the continuous wavelet method (Moreau et al., 1999), and the source parameter imaging methods (Thurston and Smith, 1997). A good summary of these methods is given by Li (2003). The methods can work well when the basic assumptions are met; however, they can be severely affected by any noise in the dataset and the choice of window size that may capture a partial anomaly or multiple anomalies.
To deal with these problems, much work has been done in several different aspects of data preparation such as enhancing the signal-to-noise ratio (SNR) by examining the derivatives of the field (e.g. Florio et al., 2006; Silva and Barbosa, 2003) or by combining different depth estimation methods (Salem and Ravat, 2003). These approaches focused upon the ability to more accurately calculate source location from the observed decay of the field from a single window. Although the solutions in general are improved, there may still be many that are manifestations of the noise within the data set or partial anomalies. Others focused on post-processing to improve the interpretability of the solutions. For example, Mikhailov et al. (2003) used artificial intelligence to better cluster the vast amount of solutions. Despite these efforts, depth estimation methods have had limited success. Part of the reason is that these methods still produce large numbers of depth solutions and, collectively, they are difficult to interpret.
To illustrate our methodology, we will use the popular approach of Euler deconvolution (Thompson, 1982) and extended Euler deconvolution (Mushayandebvu et al., 1999; Nabighian and Hansen, 2001).