I am introduced with methods for determining the right cell size according to sample characteristics and with visual validation of DEM resolution suitability, but is there any statistical indicator which is calculated after DEM generation (e.g. descriptive statistics), which is relevant ?
Not sure I completely understand the question. Cell size is usually equal resolution, thus a resolution of .3m equals a .3x.3m size pixel, 1m equals a 1x1m pixel, a 5m resolution equals a 5x5m cell size, etc. Suitability is determinate on the application of the DEM. Thus to located man-made features 1m resolution or less is required, to assess the drainage or topographic features of an area can be performed using 5m to 10m resolution, while characterizing drainage patterns or topography for a large region can best be accomplished using 30m cell size.
Thanks. Actually, I was wondering is there some statistical indicator related to DEM resolution suitablity after generating DEM, like RMSE in the case of interpolation methods assesment ... How to evaluate chosen DEM resolution and its terrain representation accuracy, apart from visual validation ? I found some papers where is mentioned a comparison of DEM cell size value with some relevant control points from other dataset ...
That is a good question. Have you checked the algorythm by which certain software like ArcGIS automatically provides a cell size when ellaborating a DEM?
More-less yes. I thought of something more like on this link ...
https://blogs.esri.com/esri/arcgis/2007/07/03/choosing-an-appropriate-cell-size-when-interpolating-raster-data/
This is far from straightforward. Firstly, how has the original data been collected? Are the sampling points distributed evenly or are they clustered? Is the clustering due to access/line of sight or due to intelligent supervision (higher data density in areas of greater variation)?
Secondly, has the data been interpolated? Was a TIN created from clustered points or was bilineal interpolation used. Or perhaps kriging or simply nearest neighbour assignment.
Thirdly, what is "good enough"? What is it that you need to be able to detect in the DEM? Interpolation methods may provide you with a continuous surface but this may simply mean that features are missing due to errors in the data collection. Small pixel size does not necessarily mean high spatial resolution.
Mirko,
Your link basically says what Andrew and I noted, that it depends on the size and complexity of the area and what you want to accomplish. While the suggestions in link tell you whether you have enough data points to interpolate across the landscape, they can not validate the data. There is no statistical program or method that I know of that can confirm the accuracy or appropriateness of cell size. Nearest neighbor and other statistical interpolation schemes can only calculate assumed values across a data void based on the sample set. I have two suggestions if you are working with elevation data. First, project your points on a topographic map of the area and examine where they fall in relation to topographic features. A good example is in a karst environment that has 40 sinkholes, if your sample set does not have points which fall in a majority of the sinkholes, then the sample set is not adequate. The most reliable method is to overlay your DEM onto topographic data to see if the cell size picks up most of the relevant features.
This can also be done for forest canopy and vegetation cover analysis by comparing the DEM results to false-color aerial/satellite images.
Hi Mirko,
Have a look on this nice paper by Tomislav Hengl:
Hengl, T. (2006) Finding the right pixel size. Computers & Geosciences, 32, 1283–1298.
Best,
martin
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics
- Geographic Information System