The min options specifies the minimum number of points to be used for interpolation. r.resamp.tps always performs tiled local TPS interpolation. Tile sizes are variable and dependent on the extents of the min nearest neighbors when a new tile is generated.
The smooth option can be used to reduce the influence of the splines and increase the influence of the covariables. Without covariables, the resulting surface will be smoother. With covariables and a large smooting value, the resulting surface will be mainly determined by the multiple regression component.
The overlap option controls how much tiles are overlapping when the min option is smaller than the numer of input points. Tiling artefacts occur with low values for the min option and the overlap option. Increasing both options will reduce tiling artefacts but processing will take more time.
The module works best with evenly spaced points. In case of highly unevenly spaced points, e.g. remote sensing data with gaps due to cloud cover, the module will take a long time to finish. For data with large gaps, it is recommended to use first a different interpolation method and then optionally use r.resamp.tps with the smooth option to identify outliers (difference between the output of r.resamp.tps and the data interpolated with a different method).
When using covariables, outliers might be created if the values of the covariables of the current output cell are far outside the observed range of covariables, or if the linear regression component of the TPS interpolation for the covariables does not provide a good solution. Two methods are provided to avoid outliers caused by covariables. The first method (lmfilter) will discard covariables if R squared is larger than the value provided with the lmfilter option. The second method (epfilter) will discard covariables if the current value of a covariable is outside the observed range of covariables by a factor of (epfilter). The epfilter option typically results in more interpolations using the supplied covariables than the lmfilter option when both are adjusted to reject the same outliers.
The memory option controls only how much memory should be used for the covariables and the intermediate output. The data needed for TPS interpolation are always completely loaded to memory.
Last changed: $Date$