New features in gstat

2.0f:

• ER-Mapper grid map support (read/write; contrib. by Steve Joyce)
• GMT/NetCDF grid map support (read/write; contrib. by Konstantin Malakhanov)
• removed cross validation bug  in 2.0
• previously calculated sample variograms can be loaded for model fitting, for zinc.sem e.g. by variogram(zinc): 'zinc.sem' : 0.5 nug() + 1 sph(600);
• specifying `dummy' is now implicit when no data file name is set
• removed the percell variable
• added  action -e mapdiff

2.0:

• online manual
• simulation locations may now coincide with data locations
• easier variogram model editing from within user interface
• allow control over multiple gnuplot windows from user interface (X11)
• optional neighbourhood search maxima per quadrant/octant
• default random number seed now uses microseconds, if available (non-ansi)
• file names for filtering input or output (using pipes) or append-to-files

added several executable ``actions'' (command-line driven programs):
• map2gif [create gif from map] (needs gd gif library from http://www.boutell.com/gd/)
• convert [convert map formats]
• cover [combine non-missing valued areas from maps]
• cut [cut square area from map]
• nominal [convert n 0-1 maps into one 0...n-1 map]
• semivariance [semivariance table for variogram model]
• covariance [covariance table for variogram model]
• semivariogram [sample variogram from data]
• covariogram [sample covariogram from data]
• statistics [report basic statistics]

Speed issues

2.0f:

• Faster neighbourhood selection routine, using quad-trees and a priority queue (check the Bucket PR-Quadtree in  this demo; bucket size is set by the split variable)
• Faster sample variogram calculation (contrib. by Konstantin Malakhanov)
• now use custom 1D/2D/3D point-point and point-block distance functions

2.0:

• improved neighbourhood selection routine, contributed by Steve Joyce. Steve implemented a spatial tree based, recursive indexing structure to speed up searching. Depending on the application, speed improvements may be a factor 5-10! (see search.c)
• simulation of multiple realizations is now much faster
• faster filling of point-to-point generalized covariance matrix
• fast sample semivariogram calculation for gridded data
• gstat variogram model fitting code now uses Gauss-Newton non-linear fitting

Technical improvements

2.0f:

• improved portability of lex/yacc files and makefile

2.0:

• use configure script for platform configuration (see configure.in)
• command parser completely rewritten in yacc (see parse.y)
• use QR decomposition for all uncorrelated linear models (more stable)
• byte map support for indicator simulation (PCRaster and Idrisii format)
• secure mode prevents calls to popen(), system() and remove().
• allow perfect fit of variogram model