Simplified Direct Sampling Method for Geostatistical Multiple-point Simulations
P. Juda, J. Straubhaar and P. Renard
Event name: Petroleum Geostatistics 2019
Session: Poster Session 1
Publication date: 02 September 2019
Info: Extended abstract, PDF ( 4.85Mb )
Price: € 20
The Direct Sampling (DS) algorithm is a statistical multiple-point simulation technique based on training images. It allows modeling spatial fields that contain a wide range of complex structures and has applications in reservoir characterization (in hydrology and petroleum engineering), mining (ore reserve estimation), or climate modeling. The DS simulation quality depends in a complex manner on the choice of three main parameters (threshold, number of neighboring nodes and scan fraction), whose selection can be tedious and computationally expensive. To reduce the parameter space, we propose a modified version of the DS algorithm without the distance threshold parameter. While this version of the algorithm produces simulations of comparable quality, it has only two main parameters, and thus it is easier to tune and understand for users. It also requires comparable CPU time and can be applied to the same class of problems as the original Direct Sampling algorithm.