Dear Mohamed Benhsinat,

Firstly, based on things you said, what I think is that the accuracy is basically based on your data-set and calculation model. However, in my opinion,

for creating a good DEM model, ArcGIS or QGIS may be more powerful. And I think there are some way for judging your DEM model is good or bad.

Regard.

To my experience the ALOS World DEM (AW3D30) is currently the best of all freely available digital elevation models. It has a spatial resolution of 30 meters but in contrast to ASTER or STRM it was acquired at a resolution of 5 meters (commercially available) and then down-sampled to 30m as a free version.

I found that this preserved many more details which are not present in the other DEMs at the same resolution.

http://www.eorc.jaxa.jp/ALOS/en/aw3d30/index.htm

Your purchased DEM involved heights measured by which technology.? We have done several studies on the vertical accuracy of the traditional 90m 3 arc SRTM and recently released 1arc 30m SRTM. The crux of the matter is that the SRTM data contains many really bad outliers. These have to go first if you want to have any sense of accuracy of your DEM model. We have used IGS GPS stations, Static GPS stations as well as RTK GPS measurements to explore this (please check on my project on vertical accuracy of SRTM data for relevant publications). Once you have the outliers filtered, the error in the SRTM seems to include a systematic bias (depending on the region of interest) and a random error. You can remove the systematic bias to improve on the accuracy of your DEM model. The random error would remain and would be the eventual uncertainty associated with your DEM model. Hope this helps.

https://www.researchgate.net/project/Vertical-Accuracy-of-SRTM-datasets

Dear Mohamed Benhsinat,

The question is whether it is an accuracy problem, a measurement technique difference or it is caused by a reference grid shift. Furthermore, it is important not to forget the issue with the tropical forest.

(As DEM can mean anything (see Walker & al. 2007 Rem. Sens. Environ.), in the following I assume you want to have a DTM(sensu Miller & LaFlamme (1958)) (implied from the need for a run-off model) .)

As Andreas and Malay pointed out above, the result of a DTM creation  is very much dependent on the technology. (Lidar, radar, stereophotogrammetry, levelling, GPS measurements will result in quite different values, and that is perfectly okay.)

The other "inaccuracy" can be caused by the unaligned grids as well. If there is a shift in the grid and the data has a low resolution like in the case of SRTM, it is no wonder that you will get differing values if the grid points are shifted some 5-10 m away (horizontally).

By all this I did not mean that your data are completely accurate, I just wanted to draw your kind attention that a certain DIFFERENCE must be there, because of the different sampling of the same topography.

And last, but not least, in the case of a tropical forest, you have a hard time to generate a somewhat accurate DTM anyhow (unless you have ground-based measurements). It is probably the most difficult terrain modelling problem after karst surfaces.

Kind regards,

Balázs

@ Balázs Székely :How can a ground measurement points improve my DEM.I want to use DEM with Ground point in order to generate DTM of the tropical forest.

Regrads

Dear Mohamed Benhsinat,

You asked: "How can a ground measurement points improve my DEM"?

I assume that you have some remotely sensed data, i.e. a LiDAR point cloud. In case of tropical forests the problem is to figure out which points (echoes) are lying on the ground surface. Depending of the leaf density (or LAI) and the structure of forest it is possible that

(a) you have many echoes from the ground surface (untypical in tropical forests) or

(b) there are a few points on the ground surface (that is what one can expect) or

(c) you don't have any points, because all echoes are coming from the vegetation, no pulse reaches the ground.

In case (a), in general, you can create a proper DTM (please note the distinction) using robust filtering (see e.g., Kraus & Pfeifer 1998) Concerning cases (b) and (c) you can use ground based measurements to determine whether you have case (b) (i.e., there are at least a few ground points) or it is case (c) (there are no ground points). If the ground measurements correlate or even overlap with the lowermost points of the LiDAR point cloud then

(1) you can be sure that there are ground points

(2) you can design a filter that is able to choose ground points.

If there isn't any correlation between the ground measurements and the point cloud, you will not be able to generate a valid DTM for this area. In this case the ground measurements are the only data that you can use for the DTM generation. (In most of the cases, that would not satisfy your needs.)

Kind regards,

Balázs