| Recently geostatistics has been actively developed and applied to a wide range of fields. Researchers and practitioners face the challenge to rapidly implement new algorithms, or to customize existing algorithms to address new problems. Currently no geostatistical software provides an environment favorable to modifications and new implementations.; A new geostatistical library and software system S4 (S&barbelow;tanford S&barbelow;patial S&barbelow;tatistical S&barbelow;oftware) is proposed. Modularity, generality, efficiency and portability are among the most important design goals. The system is implemented in C++, combining object-oriented programming and generic programming paradigms.; The library consists of a set of template classes: (1) fundamental classes that represent spatial data, Cartesian grid, unstructured grid, data search and neighborhood, visiting path; (2) classes representing probability distribution functions; (3) covariance classes; (4) kriging classes that estimate point or block Z-attributes, the trend, trend coefficients, or perform factorial kriging/filtering. Simple kriging, ordinary kriging, kriging with external drifts, and universal kriging algorithms are implemented; (5) cokriging classes that implement simple cokriging, ordinary cokriging, standardized ordinary cokriging, and cokriging with external drift algorithms.; The software includes twelve highly configurable programs that implement: (1) data transformation from one probability distribution function to another; (2) estimation by kriging, cokriging, or indicator (co)kriging; (3) stochastic simulation algorithms: sequential Gaussian cosimulation (SGSIM), sequential indicator cosimulation, direct sequential cosimulation, Gaussian truncated simulation (GTSIM), p-field simulation (PFSIM), and conditional spectral simulation.; Major characteristics of this system are: (1) Generality. The library does not assume the spatial dimension or a Cartesian grid. (2) Modularity. Library components are loosely coupled, and the software is highly configurable by data types. (3) Efficiency. The software speed of is on par with their GSLIB counterparts written in Fortran77. (4) Versatility. Many functionalities are not available in other geostatistical softwares.; This dissertation presents the interfaces of the library components; the software functionalities and parameter files. Relevant concepts and theories are reclassified from a broad perspective.; The theoretical links between the three simulation algorithms GTSIM, PFSIM and SGSIM are established. The links allow the implementation of GTSIM and PFSIM through a single program. One corollary is that SGSIM can be applied to simulate categorical variables. |
