If there is a current working mask, it applies to the output raster file. Only those cells falling within the mask will be assigned interpolated values. The search procedure for the selection of nearest neighboring points will consider all input data, without regard to the mask.
Unlike r.surf.idw2, which processes all input data points in each interpolation cycle, r.surf.idw attempts to minimize the number of input data for which distances must be calculated. Execution speed is therefore a function of the search effort, and does not increase appreciably with the number of input data points.
r.surf.idw will generally outperform r.surf.idw2 except when the input data layer contains few non-zero data, i.e. when the cost of the search exceeds the cost of the additional distance calculations performed by r.surf.idw2. The relative performance of these utilities will depend on the comparative speed of boolean, integer and floating point operations on a particular platform.
Worst case search performance by r.surf.idw occurs when the interpolated cell is located outside of the region in which input data are distributed. It therefore behooves the user to employ a mask when geographic region boundaries include large areas outside the general extent of the input data.
The degree of smoothing produced by the interpolation will increase relative to the number of nearest data points considered. The utility may be used with regularly or irregularly spaced input data. However, the output result for the former may include unacceptable nonconformities in the surface pattern.
The -e flag option provides a standard surface-generation error analysis facility. It produces an output raster map of the difference of interpolated values minus input values for those cells whose input data are non-zero. For each interpolation cycle, the known value of the cell under consideration is ignored, and the remaining input values are used to interpolate a result. The output raster map may be compared to the input raster map to analyze the distribution of interpolation error. This procedure may be helpful in choosing the number of nearest neighbors considered for surface generation.