Lithological mapping using Spatially Constrained Bayesian Network (SCB-Net): A deep learning model for generating field-data-constrained predictions with uncertainty evaluation using remote sensing data

Abstract

Geological maps are an important source of information for the Earth sciences. These maps are created using numerical or conceptual models that use geological observations to extrapolate data. Geostatistical techniques have traditionally been used to generate reliable predictions that take into account the spatial patterns inherent in the data. However, as the number of auxiliary variables increases, these methods become more labor-intensive. Additionally, traditional machine learning methods often struggle to capture spatial context and extract valuable non-linear information from geoscientific datasets. To address these challenges, we developed the Spatially Constrained Bayesian Network (SCB-Net)—an architecture designed to effectively integrate auxiliary variables while generating spatially constrained predictions. SCB-Net employs a late-fusion strategy, processing auxiliary data and ground-truth information through two parallel encoding paths. Additionally, it leverages Monte Carlo dropout as a Bayesian approximation to quantify model uncertainty. The SCB-Net has been tested on two real-world datasets from northern Quebec, Canada, demonstrating its effectiveness in generating field-data-constrained lithological maps while providing uncertainty estimates for unsampled locations. Our method outperformed the Attention U-Net – a widely used model in image segmentation – by at least 4.7% in accuracy across all tested datasets.