In EFWI, the seismic source and receiver geometries remain aligned with the OPENFWI dataset [1], except that the grid spacing is reduced to 5m in order to preserve the stability of elastic wave propagation. Each velocity map is associated with a total of 5 seismic sources and 70 receivers, which are evenly distributed on the upper surface. This configuration ensures an abundance of source-receiver pairs for the purpose of seismic data generation.

Our Python forward modeling algorthm follows the Matlab code at https://csim. kaust. edu. sa/files/ErSE328. 2013/LAB/Chapter. FD/lab. fdpsv/lab2. html. The seismic data is simulated from the velocity maps using finite-difference solver [2] with the elastic equations [3] with a 2nd-order accuracy in time and a 4th-order accuracy in space. A 350 grids absorbing boundary [4] is adopted to avoid the reflections from the model boundaries. This method provides a robust and computationally efficient mean of generating accurate seismic data that align with the VP and VS models. The point source function utilized in this study is a Ricker wavelet with a central frequency of 15 Hz. This particular wavelet is applied to the vertical component of particle displacement.