Discussion and Conclusion

We have proposed a framework for joint inversion for both passive seismic sources and velocities using diagonal preconditioning. The linear inversion of passive source functions is based on an iterative least-squares inversion using PCG. It is then incorporated into the FWI for velocity model using the method of variable projection. The framework can be efficiently parallelized on distributed computer clusters by separating passive seismic data into different stages and simultaneously inverting for the corresponding source functions. Synthetic examples demonstrate that the preconditioned LSTRI can dramatrically improve the convergence rate, and the inverted source functions can be further incorporated in FWI for velocity model updates.