What would be the best statistical approach to compare two DEMs from diffrent sources?

Dear Emmanuel Kwadzo Brempong ,

If you want something straightforward to use, I would suggest the tool DEMcompare by CNES (https://github.com/CNES/demcompare)

To a statistical approach I would prefer one based on fuzzy inference systems (FIS). DEMS are collections of geomorphometric entities: pit, peak, pass, channel, ridge and plane. The two DEMS could then be compared on the basis of these meaningful geomorphometric components instead of some abstract statistics.

Although I have not had the pleasure to work in this domain since a finished my PhD, I consider the fuzzy approach more promising.

Dear Alessandro Ferrarini, it would be a great help if you please suggest some published papers for the absolute and relative DEM difference measures you suggested.

Thank you

It depends on your aim(s). If the two DEMs represent the surface at +/- the same point in time, you're likely interested in the deviation between the two (=> uncertainty analysis). Maybe one of the DEMs is a "reference", and can be used to check the uncertainty of the other. If this is not the case, the difference between the two only yields the "combined" uncertainty, which is the uncertainty of the difference (DEM1-DEM2), not of the DEMs themselves. The latter can be investigated only using a "correct" DEM for reference or independently measured ground control points (e.g. using dGNSS).

If the two DEMs represent the surface at different points in time, you're heading for a "DEM of difference" to detect and quantify surface changes. In any case an uncertainty assessment is necessary, e.g. by computing statistics of the deviations (DEM1-DEM2) in a "stable area" where no surface change is plausibly expected.

I'd suggest to analyse the deviations (i) statistically, i.e. by looking at parameters like mean (=>systematic error?) and standard deviation (=>random error?), but also using (ii) spatial statistics, for example by looking at the variogram of a sample of point differences: Does the deviation show spatial autocorrelation (i.e. are deviations very similar on neighbouring raster cells), and if so, how large is the range (beyond which differences in deviations are not depending on distance any more) ? Does the spatial pattern of the deviations indicate systematics such as a "doming effect" ?

Quite a lot of references should appear when you search for "dem of difference" or "digital elevation uncertainty" in google scholar. Ask back if you have trouble finding a reference that suits your purpose.

Cheers

Tobias

Brand new open access references:

* Mesa-Mingorance JL, Ariza-López FJ. 2020. Accuracy Assessment of Digital Elevation Models (DEMs): A Critical Review of Practices of the Past Three Decades. Remote Sensing 12(16): 2630.

* Polidori L, El Hage M. 2020. Digital Elevation Model Quality Assessment Methods: A Critical Review. Remote Sensing 12(21): 3522.

Hope this helps

Cheers

Tobias

Taking a look at the model of differences between models (DEM1-DEM2) is interesting. Probably you have to resample one of the models to match the grid of the other and in this process you will be adding an interpolation error. In the model of differences will find the areas with larger differences and then you will need to choose which model is better. Probably the differences concentrate in areas with high vegetation, low texture, steep slopes o high curvature. Slope and shadow models from each of the original DEMs are also very important to do a qualitative assesment of quality. For a quantitative assesment you need some very precise data as lidar, some GNSS points with RTK or something like that.

Try to coregister both the dems to a common reference DEM using ENVI+IDL and then the coregistered DEM should be applied with mathematical operation like minus in GIS and then go in properties and try to see their std. And mean error before and after coregistration

I strongly recommend the commonly applied metrics such as RMSE and StD.

Simply, you can try to finite difference methods in different orders

Dear Emmanuel Kwadzo Brempong

You can statistically compare two DEMs using bench mark heights by means of RMSE, ME and Std. dev etc. Reliability of this accuracy measures can also be estimated. This will give you absolute accuracy of DEMs. However, you can also compare relative accuracy of two DEMs by comparing derivatives such as drainages. In my opinion, relative accuracy more useful than absolute accuracy. You can refer our paper (attached here) for absolute and relative accuracy measurement.