Geoscience Reference
In-Depth Information
2.3 Algorithm for DEM generalization
The task of choosing the most appropriate algorithm for multiscale DEM
generalization entails running into some drawbacks. The initial one is that
most of existing algorithms need comparative testing on terrains of differ-
ent morphology and in different scales. The second issue is that most of
advanced generalization techniques are implemented by authors as inac-
cessible customized software.
To overcome these difficulties, a compromising strategy was taken. First,
careful analysis of existing algorithms was made and some useful ideas
that can be implemented using standard GIS functions were chosen to
combine. Then a new algorithm partially adopting selected ideas was
developed and tested on various DEMs and in several scales for multiscale
hypsometric mapping purposes.
The concise scheme of the algorithm is presented in
Figure 5
. It consists of
the following steps:
1.
Streams generation from flow accumulation (O'Callaghan and Mark,
1984) with density corresponding to source DEM detailization (S1).
2.
Streams generation with density corresponding to target DEM detaili-
sation needed (S2). Here we implemented algorithm by Leonowicz
et al. (2009) based upon least negative flow accumulation difference as
Python script for ArcGIS.
3.
Selection of the streams from S1 which are the direct tributaries of S2
streams (S3). Valleys of these streams will be filled by triangulation.
4.
Watersheds generation for S2 (W2).
5.
Watersheds generation for S3 (W3).
7.
Conversion from TIN to raster.
8.
Raster postprocessing by low and high quartile filters for slight exten-
sion of resulting valleys and watersheds. Here we adopted method by
Leonowicz et al. (2009).
We should note similar algorithms, which were developed previously by
Jordan (2007) and Li and Ai (2010). The main difference of our methodology
is that it does not rely upon stream ordering but uses stream length for
generalization instead, as proposed by Leonowicz et al. (2009). We believe
