Jens Tronicke, Urs Boeniger

• Ground-based magnetic surveying is a common geophysical method to explore near-surface environments in a non-destructive manner. In many typical applications (such as archaeological prospection), the resulting anomaly maps are often characterized by low signal-to-noise ratios and, thus, the suppression of noise is a key step in data processing. Here, we propose the steering kernel regression (SKR) method to denoise magnetic data sets. SKR has been recently developed to suppress random noise in images and video sequences. The core of the method is the steering kernel function which represents a robust estimate of local image structure. Using such a kernel within an iterative regression based denoising framework, helps to minimize image blurring and to preserve the underlying structures such as edges and corners. Because such filter characteristics are desirable for random noise attenuation in potential field data sets, we apply the SKR method for processing high-resolution ground-based magnetic data as they are typically collected inGround-based magnetic surveying is a common geophysical method to explore near-surface environments in a non-destructive manner. In many typical applications (such as archaeological prospection), the resulting anomaly maps are often characterized by low signal-to-noise ratios and, thus, the suppression of noise is a key step in data processing. Here, we propose the steering kernel regression (SKR) method to denoise magnetic data sets. SKR has been recently developed to suppress random noise in images and video sequences. The core of the method is the steering kernel function which represents a robust estimate of local image structure. Using such a kernel within an iterative regression based denoising framework, helps to minimize image blurring and to preserve the underlying structures such as edges and corners. Because such filter characteristics are desirable for random noise attenuation in potential field data sets, we apply the SKR method for processing high-resolution ground-based magnetic data as they are typically collected in archaeological applications. We test and evaluate the SKR method using synthetic and field data examples and also compare it to more commonly employed denoising strategies relying, for example, on fixed filter masks (e.g., Gaussian filters). Our results show that the SKR method is successful in removing random and acquisition related noise present in our data. Concurrently, it preserves the local image structure including the amplitudes of anomalies. As demonstrated by derivative based transformations, the mentioned filter characteristics significantly impact subsequent processing steps and, therefore, result in an improved analysis and interpretation of magnetic data. Thus, the method can be considered as a promising and novel approach for denoising ground-based magnetic data.…