Data Driven Initial Model Estimation with Application to Marine CSEM Data

Initial model estimation can be fully data driven. Normally only one inversion run is required to achieve a good estimate for a smooth initial model with this approach. The scheme saves the operator in charge of the processing from repeated tests to find a proper initial model. It is very common in a production setting that several initial models are tested by the operator for each survey dataset. Thus there is a search not only for the final model but also for the initial model. These initial-model tests have a fairly high cost since they are full-scale inversions. A data driven start model generation approach has the potential to improve marine controlled-source electromagnetic imaging, especially in the deeper sections. A real data test from Brazil confirms the potential, the approach, giving a more focused inversion that has a better match with a priori information regarding reservoir placement, both laterally and in depth.