I would have a look at point cloud features in FME.

Aggregating the points using the Aggregator and then changing the multipoint to a point cloud using the GeometryCoercer will get you started.

I hope this helps.

I don't see an option to do IDW in any related transformer, and I know we don't have kriging support yet. Still, I wonder what is wrong with the RasterDEMGenerator output that it doesn't look satisfying? That's certainly the transformer we would expect you to use. Perhaps you can clarify? I would suggest the Planar interpolation method and try a tolerance of zero to give the best results.

Hi @willem85, Kriging is tough to implement, but IDW is relatively easy. The summary is:

1. Create grid points covering the entire country with preferable spacing.
2. Collect neighbor observation points for each grid point using the NeighborFinder. You have to define maximum distance and/or maximum number of neighbor observation points that will be used to calculate IDW.
3. Calculate inverse distance weighting for each grid point, and assign the value to z-coordinate.
4. Create a numeric raster based on the grid points with the NumericRasterizer.

Note that it could take a long time, if the number of grid points is very large. Depending on the actual condition, consider dividing the entire grid points into some groups (tiles) and processing for each group.

https://knowledge.safe.com/content/idea/19748/maxi...

I posted an idea to have image classification tools. A transformer that could create IDW, MLC, PCA and kriging outputs would be brilliant.

In FME 2016.1 we have an RCaller transformer that can call out to the R language to do operations. One of the demos I saw involved Kriging. I'll ask for the knowlege team to write something up about how to do this.

Here's a demo built by @DeWetAtSafe that shows how you can do kriging with the RCaller (New for FME 2016.1-available now in beta). Note this workspace requires users to install the 'geoR', and 'sp' R modules: kriging.fmwt

Just pointing out that this is very rough and I was basically just aiming for something that demonstrates that you can generate a raster from point data using the RCaller. I make no claims about any sort of statistical accuracy.