Decomposing and recovering airborne radiometric data through principal component analysis applied on flight-line data: An alternative to reduce noise

Abstract

Airborne gamma-ray spectrometry (AGRS) data provide valuable information about the distribution of radiometric elements on Earth’s surface. However, the presence of noise can hinder the interpretation or the identification of subtle variations of radioelement concentrations that can be economically attractive. Previous research demonstrated that techniques based on matrix factorization, such as Noise Adjusted Singular Value Decomposition (NASDV) and Minima Noise Fraction (MNF), can reduce noise when applied to AGRS raw spectra. Nevertheless, the raw spectra are often unavailable for end-users, limiting the widespread adoption of such techniques. In this context, we propose using Principal Component Analysis (PCA) with the flight-line data before interpolating the data onto a regular grid as a means to reduce noise when the raw spectra are no longer available. We used our approach on two AGRS datasets located in Brazil and one in the United States of America (U.S.A). For Brazil’s AGRS data, results show that noise can be attenuated through eigendecomposition projection and recovery of the radiometric variables. Furthermore, the technique we propose can highlight some geological features dependent on the number of eigenvectors used to reconstruct the database. For the U.S.A's AGRS dataset previously filtered with NASDV, the proposed methodology seems to produce only marginal improvement. Therefore, our methodology might be particularly successful for AGRS data whose acquisitions were conducted before NASDV and MNF were proposed as radiometric data processing techniques.#xD;