Pushing sensors and algorithms to the limit to minimize the chance of overlooking unexploded ordnance (UXO) increases the chance that noise will be misidentified as signal and resources used to excavate scrap metal or chunks of magnetic rock. Approaches to improve the signal-to-noise ratio typically follow one of three tracks: (1) development of new sensors that are either more sensitive or less noisy; (2) fusion of data from multiple sensors so that the chances of confounding all of the detectors simultaneously is reduced; or (3) development of computer algorithms to extract the signal(s) from the noise. In the case of wide-area surveys that use methods such as helicopter-borne magnetometry, the battle against noise has focused primarily on instrument noise (e.g., thermal noise), platform noise (e.g., magnetic noise created by the helicopter), or interference created by multiple UXO targets in cluttered settings. What has been largely ignored is the "noise" created by the rock and soil that surrounds the buried ordnance.

The objective of this project was to investigate a new approach to the characterization and simulation of geologic noise using multifractal analysis that captures the scale-dependent variability arising from geologic heterogeneity in different environments.

By combining geologic noise simulations with models for the geophysical signatures of UXO, the researchers aimed to create synthetic data sets that can be used both to test and improve UXO discrimination algorithms developed by other researchers and to develop more reliable estimates of the ratio of false positives to false negatives for a particular geologic environment. This project addressed the multifractal methodology and its application to three data sets: Sierra Army Depot, California; Fort Ord, California; and the Isleta Pueblo, New Mexico. The three data sets encompass a wide range of geologic conditions likely to be encountered at UXO sites.

The geologic noise at Sierra Army Depot was extremely low, less than a nanoTesla (nT), well below the magnetic noise level produced by the helicopter even after compensation. Improved UXO detection at such low noise sites would require improved data acquisition before multifractal modeling would be useful.

Fort Ord magnetic data showed higher levels of magnetic noise, up to ± 20 nT or more. Analysis showed that the data could be fit with a multifractal model, but that the magnetic background at this site is anisotropic. The researchers produced reasonable anisotropic simulations with a fairly simple modification to their methodology. A more generalized approach would be required to incorporate, for example, a rotational anisotropy where the direction of anisotropy changes with scale.

The magnetic data collected at Isleta Pueblo had the highest levels of geologic background noise of the three data sets investigated, with levels reaching more than ± 40 nT for the test subset. Problems with magnetic background were noted by the teams that flew surveys of this area. Again, the data proved to be multifractal, although in this case an isotropic simulation sufficed.

Tests showed the data to be multifractal, and the simulation results demonstrate that the multifractal methodology provides a versatile tool for researchers to experiment with new detection and discrimination algorithms that could potentially be used for quality assurance assessments at UXO remediation sites.

Future work might focus on improved methods for characterizing and modeling anisotropy, on incorporating remnant magnetization, and on joint multifractal modeling of other geophysical properties such as electrical conductivity and dielectric permeability