Monte-Carlo Statics: A Stochastic Approach on Wide-Azimuth Middle-East Data

Poulain, Guillaume ^{*1}; Le Meur, David ^{2}

(1) CGGVeritas, Massy, France. (2) CGGVeritas, Massy, France.

Estimation of surface-consistent residual statics on large wide-azimuth Middle-East data using a Monte-Carlo approach is a challenge. This non-linear method that uses Simulated Annealing to compute large magnitude statics is characterized by its efficiency. However, by using advanced methods like high performance computing, the computation cost of this algorithm can be drastically reduced. This paper demonstrates that a non-linear approach to estimate large magnitude statics is now possible with a reasonable turn-around without chunking the statics computation into several swaths. Most of current methods are based on the use of the cross-correlation functions and solution of linear system of equations at a local minimum. A non-linear approach, however, that uses the Simulated Annealing concept coupled with a Monte-Carlo technique allows computing surface-consistent residual statics at the global minimum. The Monte-Carlo approach we chose to implement uses a cost function that is based on the coherence of the stack combined with some robust criteria to stabilize the results. Data access is the main bottleneck when non-linear methods based on Simulated Annealing are used on large Wide-Azimuth data. At each simulation step, stations must be visited in a random order. For each station, several associated collections shots (or receivers) and CMPs are used to compute the cost function. This means that the input pre-stack dataset has to be read several hundred times in a random order! Indeed, for a given simulation step, residual static corrections cannot be computed independently because stations overlap each other in the CMP domain, i.e., the algorithm cannot be massively parallelized by using stations as described. Processing Wide-Azimuth surveys in a reasonable turn-around time was a real challenge for that method. This extremely high computation cost could be decreased only by using an advanced high performance computing solution. We made it possible to perform the method efficiently and without chunking the input data by swaths by implementing a careful fine-grained multi-core parallelism as well as some software optimization. These high performance computing techniques allowed us to optimize and minimize the data access time and therefore drastically improve the efficiency of the non-linear inversion. In this paper, we illustrate the efficiency of this stochastic approach on different 3D Wide-Azimuth surveys from the Middle-East.

AAPG Search and Discovery Article #90141©2012, GEO-2012, 10th Middle East Geosciences Conference and Exhibition, 4-7 March 2012, Manama, Bahrain