2 replies to this topic

[Rhienne]

Hi,

I have some elevation data which I interpolated using ordinary kriging, and didn't get great error results from it:

mean error: -0.02
rms: 0.31
average SE: 0.46
mean standardized: -0.03
rms standardized: 0.66
average standard error/rms error: 1.48

I applied a log transform to the data and got much better results. However, the DTM is kind of meaningless when it displays the log values. Is there any way to remove the log transform and display it as normal elevations once it has been interpolated?

Thanks.

[NicKV]

Hi,

I have some elevation data which I interpolated using ordinary kriging, and didn't get great error results from it:

mean error: -0.02
rms: 0.31
average SE: 0.46
mean standardized: -0.03
rms standardized: 0.66
average standard error/rms error: 1.48

I applied a log transform to the data and got much better results. However, the DTM is kind of meaningless when it displays the log values. Is there any way to remove the log transform and display it as normal elevations once it has been interpolated?

Thanks.

The log values are log10 values (presumably). So you need to convert to values to 10^power.

However, I would not recommend kriging for DEM creation, it is not an interpolator but a geostatistical estimator so smoothes and does not honour the field variable value (height in this case) at data points even if they are exactly on a grid point. It cannot handle abrupt changes in data point value.

Recommend you use a thin plate spline base method such as ANUDEM or a triangulation based method such as DEST as provided by Manifold (but beware Manifold does not recognise nor can it create point-type arrayed information, only cell-type).

Thin-plate spline methods are better for smooth rolling terrain without abrupt changes in elevation, cliffs etc. Triangulation is better for rugged mountainous terrain where there are abrupt changes in elevation and where there is a high number of data points per unit area and these are more or less evenly distributed.

[Rhienne]

Thanks very much for your help. The reason I am using kriging is that it is important for my data that I have an indication of the error associated with my final surface, and (as far as a I'm aware), kriging is in the only method that produces this?