![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Testing Quantification Methods with Synthetic Drumlins in a real DEM |
VerfasserIn |
John Hillier, Mike Smith |
Konferenz |
EGU General Assembly 2011
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 13 (2011) |
Datensatznummer |
250052259
|
|
|
|
Zusammenfassung |
Poorly understood errors are a key problem for proposed computational methods that aim to
extract and quantify the morphology of landforms (e.g. drumlins, craters, seamounts). A
novel way of quantitatively investigating such errors is presented, using a case-study to i)
determine which parameters are recovered accurately ii) test claims about apparently
morphologically distinct sub-populations and iii) compare extraction methods’ accuracy in
the presence of topographic ‘noise’ e.g. surface clutter. The approach could be readily
adapted to assess a variety of landforms.
Drumlins’ attributes, such as height (h) and volume (V ), may preserve information about
the dynamics of former ice sheets. A computational regional-residual separation method, the
‘Cookie-Cutter’, has recently been proposed to reproducibly quantify h and V
from DEMs. However, error in the ‘regional’ basal surfaces passing underneath
the drumlins is significant and poorly understood. For instance, it is not known if
important parameters, e.g. recovered mean volume (V r), reflect the actual population at
all. A new way to quantitatively investigate errors in such methods is presented.
Its use of idealised drumlins located randomly with respect to noise (e.g., trees)
and larger features in a real landscape is novel, as is the drumlins’ 2D Gaussian
shape.
184 drumlins with digitized outlines in western Central Scotland are used as a case
study. The normalised and stacked profiles of these drumlins are demonstrably
Gaussian. Length (lin), width (win), hin and V in determined from these outlines and
profiles provide an initial description of 178 drumlins with V in > 0. An initial
estimate of drumlins’ 3D form is then removed from within the outlines to leave
only height associated with surface clutter and larger features. Finally, idealised
‘synthetic’ drumlins are inserted with random locations and orientations to create a
synthetic DEM. Critically, because the synthetic DEM is almost entirely still the
original landscape, usual concerns when generating synthetic landscapes such as
replicating the statistical properties of the original are avoided. 10 DEMs were
created (1780 drumlins), and then the Cookie-Cutter used to retrieve the drumlins’
parameters.
For groups of 178 drumlins, 14.3 ± 6.4 (2Ïă) drumlins with negative (i.e., incorrect)
volumes suggests that 9 found for the digitized landforms is typical for the Cookie-Cutter,
and that the synthetic DEMs are acceptably realistic. Individual volumes are recovered
poorly, with 39.2% being within Ã0.75 - 1.25 of V in. In contrast, mean volume (V in) of 1.59
à 105 m3 is recovered well as 1.56 ± 0.16 à 105 m3 (2Ïă), implying that individual errors are
randomly distributed. Mean height (hin) of 6.8 m is recovered poorly at 12.5 ±
0.6 (2Ïă) m, demonstrating that consideration of errors is important before stating
the value of parameters. Variants on the Cookie-Cutter may also be assessed. For
the 1780 individual synthetic drumlins, the tensioned spline used induces about
half as much error as an un-tensioned spline, with standard deviations of the ratio
V recovered/V in being 1.28 and 2.29 respectively. Finally, by linking input and recovered
values for synthetic drumlins, it is possible to deduce that the statistically significant
(p = 0.007) difference in recovered mean volumes between Younger Dryas (YD)
and Last Glacial Maximum (LGM) age sub-populations observed for the digitized
landforms is only 40-45% likely to exist in reality. Testing variants demonstrates
that results are insensitive to the exact method used to create the synthetic DEMs |
|
|
|
|
|