pointkriging: interpolates regularly spaced data on a grid.
In geo: Draw and Annotate Maps, Especially Charts of the North Atlantic

Description Usage Arguments Value Side Effects Details See Also Examples

The function interpolates regularly spaced data on a grid. The program uses inverse distance method that takes clustering into account. Under certain conditions this method can be called kriging. For each gridpoint the program looks for neighbourhood points according to certain criteria. The program has a possibility to prevent smearing data out from one area to another where it is not wanted like between two fjords. The program has the possibility of universal kriging with the drift in lat,lon or due to an external variable.

pointkriging(lat, lon, z, xgr, vagram, maxnumber=16, scale="km",
             option=1, maxdist=0, rat=3, nb=8, set=0, areas=0,
             varcalc=F, sill=0, minnumber=2, suboption=1, outside=T,
             degree=0, lognormal=F, zeroset=F)

`lat`	Latitude of datapoints.
`lon`	Longitude of datapoints.
`z`	Values at datapoints.
`xgr`	Description of the grid. Can be output from program grid or just a list with components \$lat and \$lon.
`vagram`	Components of the variogram, a list with a least 3 components, \$sill, \$nugget & \$rang1.
`maxnumber`	Number of neighbourhood points used , default is 16. In some cases 16 points are not found.
`scale`	Scale "km" or "miles", default is "miles".
`option`	Option used for selecting neighbourhood points. Allowed values 1,2,3 and 4. Default value is one. For further information see below.
`maxdist`	If option = 4 all points within maxdist are used. If option = 4 the default value of maxdist is the range of the variogram. maxdist has also meaning when option = 1,2 or 3. In those cases points further away from any datapoint than maxdist are set to zero, mean z or NA.
`rat`	The number of points in the first step of the search is rat*maxnumber.
`nb`	Parameter describing the extent of the area where neighbourhood points are searched in the first round. Default value is 8 which means that and area of 16x16 gridpoints is searched.
`set`	Points outside region or further away than maxdist from any datapoint are set to either zero (set=0) or mean(z) set=1 or NA set =-1.
`areas`	A list defining a number of areas. NA is between areas. Points in different areas are treated as independed. Two neighbourhood fjords could be defined as different areas so the program does not interpolate between them.
`varcalc`	If varcalc is true the estimation variance at each datapoint is calculated. Default value is F.
`sill`	Sill in variance calculations. Default value is the sill of the variogram.
`minnumber`	If number of neighbourhood points found is less than or equal to minnumber the point is considered outside the areas covered by the datapoints and set to NA,0 or mean(z)
`suboption`	If option = 4 and more than maxnumber points are found in the area the program switches to option 1, 2 or 3 in the point search. Default value is 1.
`outside`	If outside is T points further away from the area than nb*dx are excluded. dx is the grid interval. Too many points outside of the area can problems in the search because the grid is used to divide the area in squares and everything left and below the first gridpoint is for example one square.
`degree`	Degree of drift polynomial for universal kriging. 0, 1 or 2. Default 0.
`lognormal`
`zeroset`

A vector with the calculated values at the gridpoints.

The program is partly written in C so if it crashes Splus is exited.

The area covered by the gridpoints is divided into rectangular elements with the gridpoints at the corners of the elements. For each gridpoints the search for neighbourhood points is done in two steps. <s-example>

1. Neighbourhood elements searched for points. How far away the element searched are is defined by the parameter nb. This step returns in most cases more points than will be used.

2. Select the final list of neighbourhood points. 4 different algorithms are available. 1. Select the nearest points. 2. 4 directions are selected 0 - 90, 90 - 180, 180 - 270 and 270 - 360 degrees. The program selects the nearest points but weighs more points in direction with few points. 3. 8 direction selected. 4. All points within maxdist selected. </s-example> If too many points are found in option 4 the program switches to some other options.

When the neighbourhood points have been selected the value is calculated using kriging. The user can either use variogram found from the data or some other variogram. By changing nugget/sill and range the degree of smoothing can be controlled. High nugget/sill generally means much smoothing. It is not desirable to use low nugget/sill (lower than 0.01 - 0.05) if any two datapoints are near each other. If nugget/sill >= 0.01 - 0.05 two datapoints can be in the same location without problems. Two too much correlated datapoints cause numerical problems.

variogram, variofit, grid, geocontour.fill.