oa Supervised Deep Learning for DAS Denoising: Impact of the Training Dataset

Distributed acoustic sensing (DAS) has become a fast-emerging tool for seismic exploration and monitoring notably due to its numerous advantages such as high resolution, high coverage and low cost. However, DAS typically exhibits high levels of various types of noise, which may deteriorate the signal-to-noise ratio (SNR) in seismic acquisition acquired with low-power active sources or impede the detection of microseismic events in passive data. To address this noise, several studies have employed deep supervised learning approaches. Though promising, these methods are highly sensitive to the dataset that is used to train the neural network. In this work, we attempt to better understand the impact of the training dataset on the denoising results and eventually provide guidelines on how to optimize this dataset for improved results.