prior: The a priori model

A priori information is defined by the prior Matlab structure. Any number of different types of a priori models can be defined. For example a 1D uniform prior can be defined in prior{1}, and 2D Gaussian prior can be defined in prior{2}.

Once a prior data structure has been defined (see examples below), a realization from the prior model can be generated using

m=sippi_prior(prior);

The realization from the prior can be visualized using

sippi_plot_prior(prior,m);

A sample (many realizations) from the prior can be visualized using

m=sippi_plot_prior_sample(prior);

All a priori model types in SIPPI allow to generate a new model in the vicinity of a current model using

[m_new,prior]=sippi_prior(prior,m);

in such a way that the prior model will be sampled if the process is repeated (see Sequential Gibbs Sampling).

Types of a priori models

Six types of a priori models are available, and can be selected by setting the type in the prior structure using e.q. prior{1}.type='gaussian'.

The UNIFORM type prior specifies an uncorrelated ND uniform model.

The GAUSSIAN type prior specifies a 1D generalized Gaussian model.

The FFTMA type prior specifies a 1D-3D Gaussian type a priori model based on the FFT Moving Average method, which is very efficient for unconditional sampling, and for defining a prior Gaussian model with variable/uncertain mean, variance, ranges, and rotation.

The CHOLESKY type prior specifies a 1D-3D Gaussian type a priori model based on Cholesky decomposition of the covariance model.

The VISIM type prior model specifies 1D-3D Gaussian models, utilizing both sequential Gaussian simulation (SGSIM) and direct sequential simulation (DSSIM) that can be conditioned to data of both point- and volume support and linear average data.

The PLURIGAUSSIAN type prior model specifies 1D-3D pluriGaussian. It is a type if truncated Gaussian model that can be used for efficient simulation of categorical values.

The VORONOI type prior defines a number of Voronois cells in a 1D to 3D grid.

The MPS type prior model specifies a 1D-3D multiple-point-based statistical prior model, based on the MPS C++ library. Simulation types includes SNESIM (based on a search tree or list), ENESIM, and GENESIM (generalized ENESIM).

The SNESIM type prior model specifies a 1D-3D multiple-point-based statistical prior model based on the SNESIM code from Stanford/SCRF.

The SNESIM_STD is similar to the 'SNESIM' type prior, but is based on SGEMS.

The following sectionsdocuments the properties of each type of prior model.

Examples of using different types of prior models or combining prior models can be found in the examples section.