ELECTROMAGNETIC SURVEY DESIGN USING ARTIFICAL NEURAL NETWORKS

Acquiring geophysical information requires selection of the geophysical method based upon the defining physical property and a survey design with adequate resolution but cost effective based upon the size of the area to be surveyed. The objective of this study is to use artificial neural nets (ANNs) to design an optimal survey. The developed approach is tested for the case of soil pipe surveying using electromagnetics. Soil pipes are tortuous voids located within 1.5 m depth of the ground surface. They trend perpendicular to the slope and have cross-sectional dimensions on the order of millimeters to tens of centimeters. The contrast in electrical conductivity (EC) is significant especially if the soil pipe is filled with air. Based upon these characteristics an EM38B is chosen to survey the area. The EM38B is relatively fast and its maximum exploration depth is approximately 1.5m. The measured apparent electrical conductivity (ECa) is a dipole dependent weighted average over a soil volume of about 1m3. A benchmark high resolution survey was conducted having a 2D cross grid pattern with a 0.5m line spacing to ensure an overlap of soil volume being interrogated. The benchmark data set is then decimated (7 options) based upon orientation and line spacing options to simulate various surveying patterns. ANN models are developed using the various reduced datasets. The quantile method is used to generate a table to guide the choice of survey for a given ECa range. To validate the concept, an exercise is conducted starting with a reconnaissance survey consisting of a few lines based on surface features of soil pipes. Using the table as a guide, a survey plan is proposed and the ANN models are created using this data set. The measured and model generated data are used to create the 2D ECa map using kriging interpolation. This map is in good agreement with the benchmark ECa map, although the second map required 60% less data