No abstract

A new inversion scheme for common-offset ground-penetrating radar measurements at multiple antenna separations was proposed, which is intermediate between inverting of picked reflectors using ray-tracing and full-waveform inversion. The measurements are modeled similarly to the real data using 2D finite-difference time-domain simulations. These simulations are obtained with a parameterized model of the subsurface that consists of several layers with constant dielectric permittivity and an explicit representation of the layers’ interfaces. Then, reflections in the modeled and in the real data are detected automatically, and the reflections of interest of the real data are selected manually. The sum of squared residuals of the reflections’ traveltime and amplitude is iteratively minimized to estimate subsurface water content and geometry, i.e., the position and shape of the layer interfaces. The method was first tested with a synthetic data set and then applied to a real data set. The comparison of the method’s result with ground-truth data showed an agreement with the subsurface geometry within [Formula: see text] and with the water content, a difference less than [Formula: see text] volume.

No abstract

Abstract.We show the potential of on-ground GroundPenetrating Radar (GPR) to identify the parameterisation of the soil water retention curve, i.e. its functional form, with a semi-quantitative analysis based on numerical simulations of the radar signal. An imbibition and drainage experiment has been conducted at the ASSESS-GPR site to establish a fluctuating water table, while an on-ground GPR antenna recorded traces over time at a fixed location. These measurements allow to identify and track the capillary fringe in the soil. The typical dynamics of soil water content with a transient water table can be deduced from the recorded radargrams. The characteristic reflections from the capillary fringes in model soils that are described by commonly used hydraulic parameterisations are investigated by numerical simulations. The parameterisations used are (i) full van Genuchten, (ii) simplified van Genuchten with m = 1 − 1 n and (iii) Brooks-Corey. All three yield characteristically different reflections, which allows the identification of an appropriate parameterisation by comparing to the measured signals. We show that for the sand used here, these signals are not consistent with the commonly used simplified van Genuchten parameterisation with m = 1 − 1 n .

Neural networks have proven their capabilities by outperforming many other approaches on regression or classification tasks on various kinds of data. Other astonishing results have been achieved using neural nets as data generators, especially in settings of generative adversarial networks (GANs). One special application is the field of image domain translations. Here, the goal is to take an image with a certain style (e. g. a photography) and transform it into another one (e. g. a painting). If such a task is performed for unpaired training examples, the corresponding GAN setting is complex, the neural networks are large, and this leads to a high peak memory consumption during, both, training and evaluation phase. This sets a limit to the highest processable image size. We address this issue by the idea of not processing the whole image at once, but to train and evaluate the domain translation on the level of overlapping image subsamples. This new approach not only enables us to translate high-resolution images that otherwise cannot be processed by the neural network at once, but also allows us to work with comparably small neural networks and with limited hardware resources. Additionally, the number of images required for the training process is significantly reduced. We present high-quality results on images with a total resolution of up to over 50 megapixels and demonstrate that our method helps to preserve local image details while it also keeps global consistency.