A Case Study of Effective Metrics for History Matching the 4D Seismic Data

Challenges specific to the utilisation of the 4D seismic data in history matching workflow are discussed. In particular, we focus on selection of an appropriate metric for the seismic objective function where the observed and synthetic 4D seismic attribute maps are compared. Synthetic 4D seismic maps for seven seismic surveys are created for three realisations of the simulation model for Segment 1 of the Schiehallion Field using simulator-to-seismic modelling (sim2seis). Subject matter experts (SMEs) ranked the models according to the consistency between the sim2seis and the observed 4D attribute maps. We then benchmark some popular numerical metrics for the seismic objective function against the scores provided by SMEs. These include L2-norm, cross-correlation, binary image comparison using manual thresholding, and comparison of the binary images after segmentation using Gaussian Mixture Model (GMM) clustering. We find that the binary image comparison using manual thresholding best agrees with the SMEs assessments. The L2-norm metric and binary image comparison using GMM segmentation both fail to tell the models apart. While the cross-correlation metric is able to differentiate the models, the variation of the cross-correlation indices across different monitor surveys are different from to those given by SMEs.