Improved NASVD smoothing of airborne gamma-ray spectra

Abstract

Noise-adjusted singular value decomposition (NASVD) is a spectral component analysis procedure for the removal of noise from gamma-ray spectra. The procedure transforms observed spectra into orthogonal spectral components. The lower-order components represent the signal in the original observed spectra, and the higher-order components represent uncorrelated noise. Noise is removed from the observed spectra by rejecting noise components and reconstructing the spectra from lower-order components. A synthetic dataset has been used to obtain new insights into the NASVD method. The dataset is based on an airborne survey over the Jemalong Plains area, NSW. The estimated ground concentrations of potassium (K), uranium (U) and thorium (Th) were spatially filtered and then used to synthesise airborne spectra using simulated Poisson noise. These spectra include a background component based on typical aircraft and cosmic background, and a simple model of the distribution of atmospheric radon. The application of NASVD smoothing to this dataset gives a much greater reduction in U concentration errors than that normally experienced. Careful investigation suggests that the reason for this is that the Jemalong Plains dataset exhibits high correlation between U and Th, and because the survey was flown on a day of low and near constant atmospheric radon concentration. If the dataset is modified by either adding spectra derived from an anomalous U/Th source, or by including atmospheric radon variations typical of most airborne surveys, then the large reductions in U concentration error are no longer achieved. Tests on the synthetic data suggest that the smaller the variation in spectral shape within the input signal, the greater the noise reduction. This is used to develop an improved implementation of the NASVD method that can be applied to large survey datasets. Instead of processing the survey data by flight, the entire survey database is sorted into clusters on the basis of similarity in spectral shape, and the NASVD method is applied to these clusters. This typically halves the K, U and Th fractional errors compared with those obtained when the data are processed according to flight. This increases the amount of geological information that can be extracted from enhanced images of the processed data.