Solved

FME Raster Convolver compared to Focal Statistics in ArcGIS

  • 9 March 2020
  • 4 replies
  • 12 views

I'm trying to generate a cell averaged DEM to generate smoothed contours within FME. After converting bare earth las points to a DEM using the Raster DEM Generator (1' cell), I want to average the values of those cells for smoother contours and it seems that the Raster Convolver is my best bet. Within the Raster Convolver I'm using a 5x5, weights all set to 1, Divisor set to 1 and Operation set to Mean but my results arent really matching what I can create in ArcGIS, they appear to be shifted to the side, about 2.5'. The Focal Statistics tool in Arc can be used to average the cell values of my initial DEM in a 3 cell radius with proper results.

 

Are there different settings I should be using in the Raster Convolver or is there a better way to go about this?

icon

Best answer by alyssaatsafe 14 March 2020, 01:02

View original

4 replies

Hi @kreid, by using 3 cell radius in ArcGIS, I am assuming you were working with a circle? Have you tried with the set of parameters as in the screenshot below? I am creating a 7x7 window (three additional cells either side of the central cell) and assign the weight of the corners to zeros as an approximation of the shape.

You are very welcome to share a small set of sample data, if this still doesn't work for you.

Hi @alyssaatsafe. Yes, I was working with a circle.

 

I just tried the parameters you suggested and the results aren't correct when comparing to my dataset generated in ArcGIS. Red contours were generated in Arc, the purple in FME with the suggested settings above.

The contours below might be a bit difficult to look at but the red contours were generated in Arc. The blue, generated in FME with the settings you suggested above except that I set the divisor to 1 and the green, FME with a similar 5x5 grid (divisor set to 1).

My bare earth las file's attached. I'm running a Raster DEM Generator with the settings below, then the Raster Convolver and finally the Contour Generator with an interval of 1.

 

be.zip

 

Your help is much appreciated!

Hi @kreid, thank you for sharing you data and screenshots. They are very helpful.

I realized I made a mistake in the original parameter settings. Size of kernel is 7x7 still and weights remain the same. Would you like to try with Divisor as Sum of Kernel Weights and Operation as Sum (instead of Mean)? This way it takes the sum of all cell values with weight of 1, then it is divided by the total number of 1's. Sorry for the confusion.

I've tested the above and compared to the contour lines generated from raster created with ArcGIS Focal Statistics. They appear much closer. Some differences still exist (mostly along the edges of the raster and some curves), but I've confirmed with the differences in the raster values, which are around +/- 0.005 in elevation.

Hope this works for you!

Hi @kreid, thank you for sharing you data and screenshots. They are very helpful.

I realized I made a mistake in the original parameter settings. Size of kernel is 7x7 still and weights remain the same. Would you like to try with Divisor as Sum of Kernel Weights and Operation as Sum (instead of Mean)? This way it takes the sum of all cell values with weight of 1, then it is divided by the total number of 1's. Sorry for the confusion.

I've tested the above and compared to the contour lines generated from raster created with ArcGIS Focal Statistics. They appear much closer. Some differences still exist (mostly along the edges of the raster and some curves), but I've confirmed with the differences in the raster values, which are around +/- 0.005 in elevation.

Hope this works for you!

This method worked great. I really appreciate your help, @alyssaatsafe and thank you!!

Reply