Predicting DAS channel depths in Vertical Seismic Profile (VSP) using a Long Short-Term Memory (LSTM) model

Arvin Karpiah, Daniel O. Trad, Ivan Sanchez

Depth registration is a persistent issue with Distributed Acoustic Sensing (DAS) data. Although the positions of DAS channels along the optical fiber are known, their true depths relative to the formation remain uncertain due to installation conditions, cable slack, and coupling effects. This depth ambiguity complicates the calibration, interpretation, and integration of DAS measurements with other subsurface datasets. To address this challenge, we focus on a Long Short-Term Memory (LSTM)–based deep learning approach. LSTMs are a class of recurrent neural networks specifically designed to capture long-range temporal dependencies and patterns within sequential data, making them particularly well suited for analyzing geophysical time series. Their ability to retain information across long time windows allows them to recognize subtle waveform signatures that correlate with formation properties at specific depths. In this study, we construct a synthetic model to systematically test the LSTM model’s capability to learn depth-dependent patterns from geophone data and use those learnings to predict channel depths by identifying similar waveform patterns within the DAS time series.