Abstract
Earth scientists increasingly deal with ‘big data’. For spatial interpolation tasks, variants of kriging have long been regarded as the established geostatistical methods. However, kriging and its variants (such as regression kriging, in which auxiliary variables or derivatives of these are included as covariates) are relatively restrictive models and lack capabilities provided by deep neural networks. Principal among these is feature learning: the ability to learn filters to recognise task-relevant patterns in gridded data such as images. Here, we demonstrate the power of feature learning in a geostatistical context by showing how deep neural networks can automatically learn the complex high-order patterns by which point-sampled target variables relate to gridded auxiliary variables (such as those provided by remote sensing) and in doing so produce detailed maps. In order to cater for the needs of decision makers who require well-calibrated probabilities, we also demonstrate how both aleatoric and epistemic uncertainty can be quantified in our deep learning approach via a Bayesian approximation known as Monte Carlo dropout. In our example, we produce a national-scale probabilistic geochemical map from point-sampled observations with auxiliary data provided by a terrain elevation grid. By combining location information with automatically learned terrain derivatives, our deep learning approach achieves an excellent coefficient of determination (R^{2} = 0.74) and near-perfect probabilistic calibration on held-out test data. Our results indicate the suitability of Bayesian deep learning and its feature-learning capabilities for large-scale geostatistical applications where uncertainty matters.Graphic
Highlights
Maps are important for our understanding of Earth and its processes, but it is generally the case that we are unable to directly observe the variables we are interested in at every point in space
We demonstrate the power of feature learning in a geostatistical context by showing how deep neural networks can automatically learn the complex high-order patterns by which point-sampled target variables relate to gridded auxiliary variables and in doing so produce detailed maps
We present a two-branch deep neural network architecture—convolutional layers for feature learning combined with fully connected layers for smooth interpolation—that brings the benefits of deep learning to geostatistical applications, and we do so without sacrificing uncertainty estimation: Our approach estimates both aleatoric and epistemic uncertainties in order to provide a theoretically grounded predictive distribution as output, which is composed of spatially coherent realisations
Summary
Maps are important for our understanding of Earth and its processes, but it is generally the case that we are unable to directly observe the variables we are interested in at every point in space. For this reason we must use models to fill in the gaps. Additional sources of information are often available, thanks in part to the rise of remote sensing (Mulder et al 2011; Colomina and Molina 2014), which provides grids of what we consider here to be auxiliary variables (e.g., terrain elevation, spectral imagery, subsurface geophysics). These are complete maps of variables that we are not directly interested in but which are likely to contain information relating to our variables of interest
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.