High-resolution paleovalley classification from airborne electromagnetic imaging and deep neural network training using digital elevation model data

Abstract

Abstract. Paleovalleys are buried ancient river valleys that often form productive aquifers, especially in the semiarid and arid areas of Australia. Delineating their extent and hydrostratigraphy is however a challenging task in groundwater system characterization. This study developed a methodology based on the deep learning super-resolution convolutional neural network (SRCNN) approach, to convert electrical conductivity (EC) estimates from an airborne electromagnetic (AEM) survey in South Australia to a high-resolution binary paleovalley map. The SRCNN was trained and tested with a synthetic training dataset, where valleys were generated from readily available digital elevation model (DEM) data from the AEM survey area. Electrical conductivities typical of valley sediments were generated by Archie's law, and subsequently blurred by down-sampling and bicubic interpolation to represent noise from the AEM survey, inversion and interpolation. After a model training step, the SRCNN successfully removed such noise, and reclassified the low-resolution, converted unimodal but skewed EC values into a high-resolution paleovalley index following a bimodal distribution. The latter allows us to distinguish valley from non-valley pixels. Furthermore, a realistic spatial connectivity structure of the paleovalley was predicted when compared with borehole lithology logs and a valley bottom flatness indicator. Overall the methodology permitted us to better constrain the three-dimensional paleovalley geometry from AEM images that are becoming more widely available for groundwater prospecting.

Nonparametric machine learning for mapping forest cover and exploring influential factors

“…The computational expenses for extending the ML models to a larger scale (e.g., the national scale) will considerably increase, and this computational requirement may demands high-performance computing and advanced machine learning methods. Recent deep learning technologies (Jiang et al 2019;LeCun et al 2015;Wei et al 2020) that have been making ground-breaking advances in many fields are suitable for big datasets and are expected to bring new insights into the projection of forest cover changes. In addition, the trained models are likely to underperform for projecting forest cover in other areas.…”

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

“…AEM data of sufficient spatial granularity (line spacing of < 400 m) to effectively define the spatial extent of near-surface aquifer systems only exists in a limited number of prospective areas within close proximity to isolated townships. Our previous work evidenced that the paleovalley geometry is correlated to the contemporary valley pattern in this region (Jiang et al, 2019) (compare Fig. 4a and b).…”

“…defining the neural network structure are determined by trialand-error tests (Supplement). Weight and bias in the generator and discriminator are trained to minimize L g and L4 using the stochastic gradient descent algorithm, referred to as adaptive moment estimation (ADAM) (Kingma and Ba, 2014). We implemented the above convolution neural network using the TensorFlow Python library (Abadi et al, 2016).…”

Surf3DNet1.0: A deep learning model for regional-scale 3D subsurface structure mapping

“…We use a CSIRO dataset to test the effectiveness of our deep-learning approach in predicting 3D palaeovalley patterns in the Anangu Pitjantjatjara Yankunytjatjara (APY) lands of South Australia (Fig. 2a are a proxy for palaeovalley presence (Jiang et al, 2019;Munday et al, 2013;Taylor et al, 2015). Thus, a palaeovalley aquifer index (PAI) is defined as:…”