Paper submitted to EAGE Annual 2026
Summary
The Krishna Godavari (KG) Basin mega merge survey involved the reprocessing of several vintage surveys through a modern processing sequence. The basin is located in the Bay of Bengal, offshore India. The survey area is shown in Figure 1a. The aim of the reprocessing was to create a continuous broadband seismic volume which maximised the value of the vintage surveys and allowed for existing fields and prospects to be tied on a basin wide scale.
A key aspect of the processing was deghosting vintage datasets acquired with varying and relatively shallow tow configurations. Deghosting is a key step in a modern processing sequence where source and receiver side down going reflections from the sea surface, which distort the image and decrease the bandwidth, are removed. Traditional methods use multidimensional transforms such as sparse Tau-p (Seher et al., 2021) and are largely deterministic requiring reasonably accurate source and receiver depth information. Such information can be challenging to acquire or extract from vintage surveys and inaccuracies can lead to artefacts such as ringing.
We describe an augmented machine learning approach to deghosting where a generalised model is updated for each survey from a deterministic sparse tau-p result. Such a method allows for a faster turnaround compared to deterministic approaches and facilitates consistent results across multi-survey projects. This consistency is demonstrated using an RGB QC attribute. The method is illustrated for a sample of four of the surveys, with varying acquisition parameters, from the KG Basin merge project as shown in Figure 1b. These will be referred to as surveys 1,2,3, and 4.