Model based seismic inversion using non-Gaussian autoregressive moving average initial model

Often exploration seismic data lacks low and high frequency band signals. The low frequency information provides crucial information about the mean model. Thus, estimation of absolute models using inversion schemes is difficult in case of band limited seismic data. We present a new method to synthesize initial model for inversion of seismic data using autoregressive and moving average modeling. The technique uses available well logs to estimate ARMA model. The estimated ARMA models are used to synthesize initial model as an input to the inversion algorithm. The use of ARMA based modeling ensures realistic frequency band in the initial model which are close to the frequency band available in the well log. The inversion is carried out using very fast simulated annealing (VFSA) technique. This technique offers directed Monte Carlo search of the model space. ARMA based initial model facilitates better search of the models residing in the null space. The method has been tested on synthetic and real data.