Due to SPAM attacks, new members must be approved before posting.  Please email when registering and your account will be approved.

Main Menu

Low resolution DEM

Started by j_silver, November 02, 2012, 02:43:12 PM

Previous topic - Next topic


We would like to run SLAMM at a site where we only have a 30m DEM. 
Before we do, however, we would like to know how coarse is 'too coarse' a DEM to run SLAMM successfully. 
We are aware of the elevation pre-processor for poor quality DEM.



Jonathan S. Clough

I would pay attention to the important difference between vertical and horizontal resolution.

A LiDAR DEM with a 30M cell size is likely to be very accurate indeed vertically.  This is because so many LiDAR returns will be included in the average for each cell.  High and low errors should balance out.  Unless the data collection is vertically biased, increasing cell sizes tends to increase vertical resolution.

Some analysis done with SLAMM indicated that final model results are considerably more affected by vertical accuracy than horizontal accuracy (or cell size).

With respect to how vertically accurate elevation data need to be to enable valid predictions, there is some disagreement about that.  Some folks look at the vertical accuracy of the data set (The 95% confidence interval is often estimated as 1.96*RMSE) and state that this error must be less than the RSLR being simulated to have a valid prediction.  While this seems like a reasonable and scientific assertion, it ignores the fact that error statistics represent the error for each cell of the data set, not the bias of the data set as a whole.  There is essentially no chance that all of the cells will be off by 1.96*RMSE.

Because the 95% confidence interval represents an individual cell, to have each cell in error to this degree would have a likelihood of 5% raised to the number of cells (or less than 1E-130 for a map with 100 cells).

For some  projects, we have run SLAMM "elevation-uncertainty analyses" by creating a number of equally-likely DEMs using error statistics and spatial autocorrelation to create random error fields that are added to the DEMs (in the method of Hunter & Goodchild, 1997 and Heuvelink,1998, e.g .  Our findings have been that elevation error from LiDAR can be significantly overstated as a component of model error, especially when using the 1.96*RMSE metric for acceptability.

If you are concerned about the vertical error of your data set you should consider running an elevation data uncertainty analysis, which is available in SLAMM 6.1 and we can make available to you upon request. 

That all being said, using non LiDAR (USGS-contour-derived) DEMS and the pre-processor increases model uncertainty dramatically.  It should be considered a "best estimate in the absence of better data."

64-bit SLAMM should be available within this calendar year and will run 2x faster and include all uncertainty analyses, etc.

Best Regards