I am trying to create a raster dataset from a vast number of bathymetric data points. I want to use the kriging method with a cell size of ~20m but when I create the raster surface it is interpolated beyond/outside the zone where the points are located resulting in a surface that is fine within the point area but is incorrect outside. I know that a barrier polyline can be used for the IDW interpolation method to prevent this from happening but does anyone know a way of doing this in the kriging method as the barrier polyline option is not available?
You can limit the area where the interpolation will be done by setting a mask in the environments. To do this you will need a polygon feature class/shape file or a raster which delimits the area of your interpolation. When you open the kriging tool, click on the Environments button. Under Raster Analysis set the mask to the polygon or raster. This will limit the interpolation to this area.
Thanks for your responses. So I am guessing the MASK function isn't as effective as the barrier option as the MASK is undertaken after the rater interpolation is complete, i.e. the interpolation process outside the point area has been undertaken and the MASK tool is just cropping the raster back to the specificed polygon? Is this the same as using the extract by polygon tool once the raster has been created?
From experience, which interpolation methods would you recommend in order to create a high resolution raster surface (mainly for visualisation purposes only) from dense point data using a polyline as a barrier to prevent the interpolation from going outside the area of interest? The IDW method seems to crsh when I try and create a raster using a polyline barrier shapefile?
Bill is right. Interpolators are good at, well, interpolating. Unless you have an expert understanding of the physics involved with the process (and build your model around the physics), extrapolated predictions will be unreliable.
You can force the interpolators to make predictions as far outside the data extent as you want, but this is generally a very bad idea. If you build the model correctly, you can trust the predictions within the data extent, but you should be very careful about any conclusions you draw outside that extent.