Land Data FWI: Methods and Strategies

Abstract

Abstract There have long been attempts to apply Full Waveform Inversion (FWI) to land data sets because it has the potential to build higher-resolution velocity models compared to traditional ray-based model building techniches. However, several difficulties hinder the successful application of FWI to land data: cycle skipping, unreliable amplitudes, as well as source and receiver signature estimation. To address these issues, we have developed three new FWI methods: the Dual Frequency Phase Difference (DFPD) Laplace Fourier domain FWI, Surface Offset Gather Flattening (SOGF) Wave Equation Migration Velocity Analysis (WEMVA), and Seismic Reflection Slope FWI. The Dual Frequency Phase Difference (DFPD) Laplace-Fourier domain FWI greatly simplifies the input data and promises the global minimum, and thus, is ideal for initial velocity model building. The Surface Offset Gather Flattening (SOGF) WEMVA method aims at maximizing the flatness of surface offset gather, which is a better objective than maximizing the focus of subsurface time/spatial lagged gather in terms of handling models with large lateral change. The Seismic Reflection Slope FWI method is capable of identifying small velocity anomalies such as gas clouds and faults, thereby producing a higher-resolution velocity model. Three model data examples are used demonstrate the capability of our methods: first we use a simple two-layer model to demonstrate that the Dual-Frequency Phase Difference (DFPD) Laplace-Fourier domain FWI is indeed immune to cycle skipping; second we use a steep boundary model to demonstrate that our Surface Offset Gather Flattening WEMVA is efficient at handling models with large lateral velocity change; finally we use the BP 2004 benchmark model to demonstrate the Seismic Reflection Slope FWI method is capable of identifying small velocity anomalies. We also note that these three methods can be combined together to form a FWI based model building strategy that produces high-resolution velocity model for land data set. Two field data examples are used to illustrate the effectiveness of this model building strategy. First, we applied the strategy to a 2D field data set acquired in Inner Mongolia. Two starting models were used, one is a simple model converted from time domain RMS velocity, and the other is from the traditional ray based model building approach. For both cases, our strategy was able to greatly improve the velocity model and the migration image. The migration image starting from the traditional ray based model had a superior focus and better followed the geological setting, indicating that our FWI strategy is compatible with traditional method and that, in order to get the best possible result, one would use our strategy subsequent to utilizing the traditional method. We then demonstrated this by applying our refined workflow to a 3D field data set acquired in the Pre-Caspian Basin, Kazakhstan, producing a velocity model with an improved migration image, containing fine details such as faults and channels.