Sub3DNet1.0: a deep-learning model for regional-scale 3D subsurface structure mapping

Zhenjiao Jiang,Lei Gao,Gregoire Mariethoz,Tim Munday,Luk Peeters,Dirk Mallants

doi:10.5194/gmd-14-3421-2021

Abstract

Abstract. This study introduces an efficient deep-learning model based on convolutional neural networks with joint autoencoder and adversarial structures for 3D subsurface mapping from 2D surface observations. The method was applied to delineate paleovalleys in an Australian desert landscape. The neural network was trained on a 6400 km2 domain by using a land surface topography as 2D input and an airborne electromagnetic (AEM)-derived probability map of paleovalley presence as 3D output. The trained neural network has a squared error <0.10 across 99 % of the training domain and produces a squared error <0.10 across 93 % of the validation domain, demonstrating that it is reliable in reconstructing 3D paleovalley patterns beyond the training area. Due to its generic structure, the neural network structure designed in this study and the training algorithm have broad application potential to construct 3D geological features (e.g., ore bodies, aquifer) from 2D land surface observations.

Highlights

Imaging the Earth’s subsurface is crucial for the exploration and management of mineral, energy, and groundwater resources, its reliability depending on the availability and quality of geological data
The effectiveness of our deep-learning model is tested on predicting 3D paleovalley patterns in the Anangu Pitjantjatjara Yankunytjatjara (APY) lands, part of the Musgrave Province of South Australia (Fig. 3a and b)
The 3D structure of a paleovalley was interpreted from an airborne electromagnetic (AEM) survey (Soerensen et al, 2016)

Summary

Introduction

Imaging the Earth’s subsurface is crucial for the exploration and management of mineral, energy, and groundwater resources, its reliability depending on the availability and quality of geological data. In Australia, as an example, the former are readily available at relatively low or no-cost, while the latter are often non-existent and expensive for remote desert areas, where a key challenge is to secure groundwater for town/community supply, primarily from shallow aquifers (Munday et al, 2020a, b). With a regional AEM line spacing of 2 km, smaller infill areas were defined close to remote isolated communities where line spacing was reduced to 250 and 500 m. This provided greater detail of the character of the subsurface electrical conductivity, enabling more accurate mapping of paleovalley aquifers to be achieved (Munday et al, 2020a). The application of such high-resolution data to much larger areas like the entire Victorian Desert would be cost prohibitive, so alternative approaches to the definition of paleovalley systems are required

Objectives

Results

Discussion

Conclusion