The near surface in the Middle East, particularly in the Sultanate of Oman, is characterized by very shallow high-velocity carbonates and anhydrites interleaved by slow-velocity clastic layers, resulting in sharp velocity inversions in the first few hundred meters below the surface. In addition, the surface is characterized by features such as unconsolidated materials within dry riverbeds (known as “wadis”), small jebels, and sand dunes, which cause distortions in the underlying shallow and deeper seismic images. This work presents the building of a near-surface model by using multiwave inversion that jointly inverts information from P-wave first breaks and surface-wave dispersion curves. The use of surface waves in combination with first breaks captures the lateral and vertical velocity variations, especially in the shallowest parts of the near surface. This paper focuses on the analysis of two drawbacks of this technology: the picking of the input data information, which can be cumbersome and time consuming, and the limited penetration depth of surface waves at the typical frequencies of active data. To overcome these issues, an innovative workflow is proposed that combines the use of an unsupervised machine learning technique to guide the pick extraction phase and the reconstruction of ultra-low-frequency surface waves (0.5 to 1.5 Hz) through an interferometry process using information from natural and ambient noise. Deeper near-surface P- and S-wave velocity models can be obtained with multiwave inversion using these ultra-low frequencies. The integration of a near-surface model into the velocity model building workflow brings a major improvement in depth imaging from shallow to deep structures, as demonstrated on two data sets from the Sultanate of Oman.
