Abstract

Seismic acquisitions in urban areas represent a challenge due to the noisy environment during the recording of the data. This often prevents the obtainment of high-resolution images of the subsurface in the interest area. One way to achieve better results, despite the not ideal circumstances, is the full-waveform inversion (FWI). This method iteratively updates a starting model until the simulated data and the observed field data matches to a satisfactory level. This eventually yields the desired underground model. However, FWI comes with some major limitations. Because each iteration requires a calculation of the wave propagation several times for each shot, it is highly computationally expensive. Furthermore, due to storage limitations, especially if high performance computing (HPC) clusters are not available, the model size is restricted as well.

This work aims to find a solution for these limitations in an elastic 2D case by implementing the random shot workflow (RSW). In a conventional FWI workflow which is called full shot workflow (FSW) in this work, each iteration uses all shots of the recorded data before updating the whole model at once which leads to the high computational costs and storage usage. The RSW randomly picks only one shot, applies the FWI within the area of the shot and its corresponding receivers and then updates the model only within this subset. This should, in my expectation, divide the workload to several smaller ones without sacrificing the effectiveness of FWI.

After the implementation, I compare the RSW to the FSW in a synthetic benchmark where I verify the functionality. Afterwards, the workflows are compared to each other by using a fraction of the field data that was acquired in Salt Lake City. This data represents a large urban seismic land data set with a low signal-to-noise ratio. It also has suboptimal conditions such as the end-on spread geometry due to the land streamer setup and occasionally bad coupling of the geophones. This turned out to be a big challenge for the FWI algorithm.
The starting model used for the field data inversion is based on the multichannel analysis of surface waves which deals with the inversion of surface-wave dispersion curves. With a subset of this model and the data, I investigate the influence of several different configurations, such as different multi-stage configurations, large or small number of iterations and whether or not killing traces near the source, which suffer heavily from the suboptimal acquisition conditions, lowers the misfit.
The comparisons and investigations show that while RSW has the computational advantage in storage usage and per-shot performance, it falls behind the FSW if a great shot parallelisation is used for FSW. It's also too sensitive towards bad traces which disrupt the model heavily.

At the end I do a final comparison of both workflows by applying the inversion to the complete Salt Lake City field data set. An application of FWI on such a large scale land streamer acquisition with such a large number of shots hasn't been done yet to my knowledge, most likely because of the not easily feasible computational costs, even in 2D. 
While the computational advantages of RSW still hold true in the per-shot performance and storage usage, it also still lacks behind if shot parallelisation can be abundantly used. Together with the strong disruptions of bad traces in the final model, the RSW is assessed as not applicable for the Salt Lake City data set. However, even FSW barely manages to converge here which supports my assumption that the data is generally very challenging for the FWI algorithm.