Results - Response of Flood Events to Land Use and Climate Change - page 126

6.2 DEM Quality Control Modeling

6.2.1 Interpolation Techniques

In this section, eight digital elevation models (DEMs) were generated using dif-
ferent interpolation techniques and different parameter settings. The spatial reso-
lution of all surfaces was set to 5 m. Three DEMs were generated based on
deterministic techniques including IDW, Spline, and Topo to Raster (using the
ANUDEM program developed by Michael Hutchinson), and five DEMs were
generated using ordinary Kriging methods including Circular, Exponential,

Gaussian, Stable, and Spherical.
To select the best DEMs generated with different interpolation techniques, a set
of critical measurements were used to compare the results. In the case of the
Geostatistic model—Kriging-based method—the selection of an optimal set of
parameters to be used in ordinary Kriging required an exploration of different
combinations of the values. Different combinations of the lag size and the number
of lags were used in the generation of empirical semi-variograms for different
surfaces, interpolated as well. Even though the selection of an appropriate lag size
depends on the particular phenomenon being modeled, too large of lag sizes may
mask short distance autocorrelation, and too short of lags can create in the pre-
dictions. In this study a variety of lag sizes were tested, and finally the best results
were obtained using a relatively small lag size of 100 and 25 m. One aspect of the
interpolation methods that was qualitatively quantified was time efficiency. For the
training dataset used in this study, the surface interpolation, with a final resolution
of 5 m using Kriging took an average of 1 h for each test. IDW, Topo to Raster
and Spline methods were the fastest interpolation techniques (less than 10 min) to
interpolate the surfaces to the final resolution. The defined lag size suggests that
given the complex topographic conditions of the study area, the autocorrelation
between locations is significant only in reduced neighborhoods beyond which
conditions change abruptly. Some important conditions were maintained constant
in all deterministic and geostatistic models. First, no anisotropy was considered in
any of the models. Second, four-sector circular neighborhoods were used in all
models and four points per sector were used to make predicted points at unknown
locations. Therefore, in this section, two assumptions are considered and the
related results are discussed below:
As for the first assumption, interpolation methods that often assume data
points in original DEM are correct (exact), but it can be assumed that they are
subject to error (generally of a known or estimated extent). This means the
interpolation technique is not only for creating DEM from contour lines or
observed data points, but also for when we have a DEM that we could correct or
find error with, such as sink errors. Therefore, the main purpose was to correct
the original DEM and find errors and to address the best interpolation techniques
in order to erase errors in DEM. Therefore, 70 % of the observed data points