Visualizing DEM Error

Project Overview

Learn how to resample DEMs and why resampling may be needed for raster analysis
Visualize elevation differences near Signal Mountain, UT between DEMs of varying resolution and type

Tutorial for this exercise by the Utah Geospatial Consortium can be found here. Data was sourced from UT AGRC and the USGS National Map. All maps are displayed in NAD 1983 UTM Zone 12N.

DEM Comparison

Figure 1 - Elevation differences between 5-m Autocorrelated DEM/10-m 3DEP DEM and 0.5-m LiDAR DEM overlaying 0.5-m hillshade. Differences in elevation are represented on a continuous color ramp from blue (+120 m from 0.5-m DEM elevation) to orange (-120 m from 0.5-m DEM elevation), where no color indicates 0 difference from the 0.5-m DEM elevation. Both the 5-m DEM and 10-m DEM overestimate elevation more often than underestimate elevation in this location. However, the 5-m autocorrelated DEM has a much larger area of significant differences from the 0.5-m DEM than the 10-m DEM. These results suggest that the 5-m DEM is a less accurate representation of elevation in this area, despite the "higher" resolutions than the 10-m DEM. Additionally, the underlying topography (represented by the 0.5-m hillshade) suggests that the 5-m autocorrelated DEM underestimates elevations along valley bottoms, whereas the 10-m 3DEP DEM overestimates elevations along ridges.

A higher resolution version of this map can be found here.

Figure 2 - Elevation differences between 5-m Autocorrelated DEM/10-m 3DEP DEM and 0.5-m LiDAR DEM overlaying aerial imagery. This figure is in the same location/extent as Figure 1. However, patterns in elevation overestimation in the 5-m DEM were unclear using hillshade alone. The addition of aerial imagery shows that many areas of overestimated elevation in the 5-m DEM correspond with some densely vegetated areas. That being said, 120 m is quite tall for trees in this region, so there may have been other underlying issues with the 5-m DEM. Aerial imagery in this figure was provided by ESRI.

Mastery Observations

The hardest part of this lab for me was visualizing the difference rasters in a meaningful way. Max/min values for the 10-m DEM were much smaller than the values for the 5-m DEM, which made it hard to see the 10-m difference raster on the same color scale. To make the differences more visible, I did three things: 1) I zoomed in to an area where 10-m differences were visible, 2) I created a custom color ramp that had darker colors closer to the "0" value, and 3) I increased the transparency of the underlying hillshade so that any color really stood out.
The low max/min values for the 10-m difference raster indicated that the 10-m DEM was much closer to the 0.5-m LiDAR DEM than the 5-m autocorrelated DEM. I expected this result based on the results of the DEM Mastery lab, where we used contour lines and hillshades to show that the 5-m DEM was less accurate than the 10-m 3DEP DEM.
I liked learning about using resampling to perform calcuations on rasters with different resolutions. I'm glad we covered this in class because it is one of those concepts that makes a lot of sense, but I would not have thought about it much on my own. Also, learning to clip the rasters to each other during resampling was very important! I initially missed that step and ended up re-running it.
I'm glad I checked with the aerial imagery in the end. It was crazy how exactly the elevation differences lined up with forested areas! I would have never know that information if I had only looked at the hillshade for trends.

Page updated

Google Sites

Report abuse