Distributed Acoustic Sensing (DAS) has emerged as a promising observational tool for a variety of geophysical monitoring applications. Its cost-effectiveness and high spatial sensor density offer a compelling alternative to traditional seismic sensors, particularly in regions where conventional deployment is challenging. DAS inherently measures strain (or strain rate), whereas conventional seismic sensors record displacement (or velocity). However, most seismological algorithms are optimized for translational ground motion data, motivating robust methods for converting DAS data into equivalent ground motion. In this work, we present a novel deep learning model that accurately converts DAS strain into geophone particle velocity trained on co-located nodal seismometers for the PoroTomo data obtained at Brady geothermal field in 2016. The model combines Fourier Neural Operator (FNO) and Bidirectional Long Short-Term Memory (BiLSTM) with an attention mechanism (FNO-BiLSTM-Attention). The model is trained and evaluated using earthquake waveform data recorded simultaneously by co-located DAS channels and geophones. To validate the conversion process, we compared it with both geophone data and a physics-based conversion method. Then, a seismic beamforming analysis was performed using the deep learning-based converted DAS data, with results compared to those from the geophones. The results show an excellent match between both estimations and they are notably better than using DAS strain directly. The further improvement over using nodal data comes from improved signal coherency and density of spatial data.
Al-Qadasi et al. (2026) studied this question.