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

Main Menu

DEM in uncertainty analysis

Started by Nava, March 24, 2015, 02:05:25 PM

Previous topic - Next topic


When running an uncertainty analysis, are there any down sides to including the DEM uncertainty as a parameter? Is there strong criticism of uncertainty model results that don't include it?

Jonathan S. Clough

The DEM uncertainty module tries to create a number of equally likely maps that would have the same error statistics as the (hopefully LiDAR-derived) elevation layer.  The fact that elevation errors are spatially auto correlated complicates the matter to some degree.

The only down-side to including DEM uncertainty in an uncertainty analysis would be that the default random spatial auto-correlation is not good enough.  LiDAR error is known to be greater in marshes for example, and possibly biased high (due to difficulty in lasers penetrating the marsh surface).   DEM error is also known to be greater in higher sloped areas.  There may be a known spatial component to error in ground-truthed LiDAR and this cannot yet be brought into the existing SLAMM analysis.  However, if you only know the RMSE of the data set, this is probably the best that can be done for the moment.

The lack of accounting for elevation data uncertainty was a criticism of early SLAMM work.  Some researchers will take the 95th percent confidence interval for the entire data set (based on RMSE) and apply that uniformly across the data set.  I don't believe that this is an appropriate characterization of elevation data uncertainty as you are applying individual cell statistics as though they apply to model bias.  If you assume a data set bias of zero but apply error fields that reflect the RMSE of the LiDAR data that is a much better way of characterizing additional uncertainty.  The effects on model results have been fairly modest when evaluating effects of state-of-the-practice LiDAR data error.


This is helpful. We are working with recent LiDAR, so that should help, but we do also have a lot of the types of places where error may be higher than average (e.g. marsh and steep slopes).

In the SLAMM interface, when you go to specify the DEM uncertainty parameter, it seems to be briefly running or calculating something before it shows the "map". But I don't think it is anything specific to my data. If I understand correctly I would need to enter my own RMSE from LiDAR metadata. Is that right?


In the uncertainty analysis set up for the DEM you can define the RMSE and the Autocorrelation Coefficient. SLAMM calculates and shows the uncertainty field (more random when less correlated). This is applied to the entire area and there is no option to input different elevation errors depending on the land cover, slopes or subsite. To be safe, I would use the highest estimated RMSE.


Do you also recommend using a high autocorrelation coefficient? What guidlines can I use to choose this value?


Unfortunately this information is rarely (if ever) provided with the elevation data. Therefore, using professional judgement, we recently select 0.2495, which is relatively high. This values provides clustered areas of uncertainty, meaning that when there is an elevation point with an error it is also likely that points around have errors with similar amplitude and sign (e.g. all elevations are higher or lower in the area).