Summary
The aim of this diploma thesis was to produce the most automated extraction of selected elements of ZABAGED® from DTM of airborne laser scanning data. The theme arose in National park Czech Switzerland, where ZABAGED® vector data (1:10 000) are used as representation of geographic elements. NP has also the data of DTM from ALS so it occurred if it was possible to create new vector layers by extraction from DTM.
The first step was to select elements suitable for the extraction. The final list includes quoted points, points of positioning and elevation fields, contours, watercourses, watershed, terrain and rock stages a paths. Each selected element was treated separately in program ArcGIS 9.3 and 10.
The main aspect that affected the entire work was the presence of DTM data only. For element extraction is generally used combination of LIDAR data, IfSAR and photogrammetric data. But for this work the additional data wasn´t provided.
ALS data high accuracy is guaranteed by the density of several points per m2. Therefore the interest of creating more accurate vector data is a logical consequence of availability of accurate DTM data.
During the detection of all types of points has been solved one problem – it is not in fact the detection, but the extraction of new height information in its original location (coordinates x, y). It was obvious that also the x, y coordinates are probably incorrect in some locations. If the altitude of original points differed significantly from the new value and exactly the same old value is placed in some very near pixel, then new location x, y could be found. The problem is that it can be done on one condition – the point has to be local minimum or maximum. But the points appear also on some significant objects in terrain (e.g. building) which aren´t local minimum or maximum.
Contour extraction has avoided grater complications, since the algorithm for contours generation is embedded the ArcGIS program. The contours was designed to be printed in a map 1:10 000, so the most important question was how much generalize the line. There is no problem of creating very accurate contours, but it is inappropriate for printing in specified scale. On the other hand it can be created smoothed contours which are well readable in a map, but it loses the information of terrain characteristics. Therefore, it was necessary to find a way of extraction and proper degree of generalization so that the output was as most accurate as possible but also readable. Thus, there were generated slightly smoothed contours and then erased polygons with slope higher than 45°. In locations with erased contours were left main contours (e=25 m) in order to maintain the true character of relief for the map reader.
Detection of water flow is a complex issue and thus brings more questions. For river start ("Stream Definition" from the Flow Accumulation ") were selected 20 000 pixels (i.e. m2). This approach generated much more water flows than exists in reality, but it ensured that all real streams were actually modelled. The value for river start depends on type of area, so his concrete value can´t be applied to any type of terrain. Another important fact is that witch this chosen method the line of maximum slope of the watercourse is modelled, which does not necessarily mean the actual course of flow. In places where the location of a channel can´t be verified, such as on the orthophoto it has to be verified in terrain.
Generating of watercourses is closely related to the resulting layer of watershed. The line of generated watershed is precise on ridges, because in these location plays role only the elevation of pixels. The line in valleys is very dependent on watercourse. If there is for instance the confluence of two streams modelled on the spot a few meters shifted from the original junction it is changing the line of watershed. Therefore, it is necessary to check the confluences on aerial images. If it can´t be seen on image then it can be checked by field investigation.
The output of terrain and rock stages is lines separating their upper and lower edges. It is necessary to say that this output is not for printing. From the output were removed all edges shorter than 5 meters, because of excessive file size and also because of the low explanatory power of lines so short.
Automatic extraction of paths was resolved only to certain extent. There were only elevation data available, so it was possible to extract only the paths which differed from form the surrounding terrain by its height. Detection of paths was done by the same procedure as the extraction of terrain and rock stages. There were found the most paths in the interval of 5- 10 % of slope change after visual examination of the output and the original data ZABAGED ®. Therefore, the layer was classified, so that only these values can be extracted. This output can´t be used as a final layer of paths due to its incompleteness. However, it can serve as an input for manual digitization.
The work can be summarized positively. Extraction of elements from the DTM ALS data is suitable for creating layers more precise than ZABAGED ® layers. However it is never a fully automated process. There is always needed more or less demanding post editing of the extracted data. In the case of paths the author considers the least expensive method is to digitize the elements manually on a base of orthophoto, shaded relief using extracted features.