This article was downloaded by:[National Taiwan University] On: 28 December 2007 Access Details: [subscription number 769798964] Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK International Journal of Remote Sensing Publication details, including instructions for authors and subscription information: http://www.informaworld.com/smpp/title~content=t713722504 Comparison between automated and manual mapping of typhoon-triggered landslides from SPOT-5 imagery A. M. Borghuis a ; K. Chang b ; H. Y. Lee c a Utrecht University, Department of Physical Geography, The Netherlands b Department of Geography, National Taiwan University, Taipei, Taiwan 106 c Department of Civil Engineering, National Taiwan University, Taipei, Taiwan 106 Online Publication Date: 01 January 2007 To cite this Article: Borghuis, A. M., Chang, K. and Lee, H. Y. (2007) 'Comparison between automated and manual mapping of typhoon-triggered landslides from SPOT-5 imagery', International Journal of Remote Sensing, 28:8, 1843 - 1856 To link to this article: DOI: 10.1080/01431160600935638 URL: http://dx.doi.org/10.1080/01431160600935638 PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.
15
Embed
International Journal of Remote Sensing - UN-SPIDER ... between automated... · applied to mask roads, ... forested watersheds where soil erosion rates are relatively low ... satellite
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
This article was downloaded by:[National Taiwan University]On: 28 December 2007Access Details: [subscription number 769798964]Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
International Journal of RemoteSensingPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713722504
Comparison between automated and manual mappingof typhoon-triggered landslides from SPOT-5 imageryA. M. Borghuis a; K. Chang b; H. Y. Lee ca Utrecht University, Department of Physical Geography, The Netherlandsb Department of Geography, National Taiwan University, Taipei, Taiwan 106c Department of Civil Engineering, National Taiwan University, Taipei, Taiwan 106
Online Publication Date: 01 January 2007To cite this Article: Borghuis, A. M., Chang, K. and Lee, H. Y. (2007) 'Comparisonbetween automated and manual mapping of typhoon-triggered landslides fromSPOT-5 imagery', International Journal of Remote Sensing, 28:8, 1843 - 1856
To link to this article: DOI: 10.1080/01431160600935638URL: http://dx.doi.org/10.1080/01431160600935638
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expresslyforbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will becomplete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should beindependently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with orarising out of the use of this material.
of concordance between the automated method (black areas) and the manual method
(white outlines). The landslide at the bottom of figure 4(d) also exhibits an important
difference between the methods. The manual method produced a single, enclosed
landslide whereas the automated method resulted in a fragmented landslide.
6. Discussion
An important finding of the study is that both the supervised and unsupervised
classification produced many more small landslides than the manual delineation.This was further confirmed by the t-tests for all landslide areas. The differences in
landslide areas and numbers can be explained as follows. First, in the manual
delineation we connected all areas that appeared to belong to one single landslide,
leading to fewer and larger landslides. In contrast, the raster-based automated
classification methods identified landslides on a cell-by-cell basis. Large landslides
were therefore split into smaller fractions when, for example, patches of vegetation
were present on the landslide surface. Second, despite the use of a slope filter, parts
of farmland, stretches of roads or parts of riverbeds were still erroneouslycommissioned as landslides, leading to a high number of small landslides (figure 5).
Because of the incorrect classification of roads and riverbeds, the automated
Table 2. Unsupervised classification (MLC) using all bands and slope and noise filters
Overlay 12 April 2004 10 July 2004 12 October 2004
All three methods 14.3 20.2 37.6Supervised/Manual 15.7 37.6 39.4Unsupervised/Manual 58.6 53.3 63.1
1852 A. M. Borghuis et al.
Dow
nloa
ded
By:
[Nat
iona
l Tai
wan
Uni
vers
ity] A
t: 02
:31
28 D
ecem
ber 2
007
classification methods likely overestimated the number and areal extent of
landslides. Third, we found that the manual delineation of supermode 2.5 m
imagery proved not suitable for the identification of small landslides. As a rule,
landslides smaller than 400 m2 (868 or 64 pixels) could not be distinguished by eye
in the supermode colour images. However, towns, farms and clouds could be easily
identified.
The area concordance statistics show large differences between the supervised and
unsupervised classifications compared to the manual delineation. The supervised
classification result was especially poor (an area concordance of 16%) for the April
scene because of the absence of fresh landslides. We used two existing landslides,
which we visited in the field, as training areas. The first was a partly re-vegetated
landslide appearing very similar to areas where vegetation was still sparse in April.
The second was a large and scarcely vegetated landslide with exposed rock materials
that appeared to be spectrally similar to the sediment in the dry river floodplain. The
result was that many areas were wrongly commissioned as landslides in the April
scene.
The difference in the area concordance results between the two methods can be
further explained by examining maximum landslide areas in tables 1–3. Landslides
generated by the supervised classification are all relatively small and fragmented,
with maximum areas ranging from 3.48 ha to 7.67 ha, whereas landslides generated
by the unsupervised classification are much larger, with maximum areas ranging
from 14.37 ha to 24.61 ha, a range similar to that obtained by the manual method,
Figure 4. (a) Small landslides along a stream. (b) The landslides in (a) correctly identified bythe automated method. (c) Errors of commission by the automated method caused by roads.(d) Area concordance between the manual and automated methods.
Landslide mapping 1853
Dow
nloa
ded
By:
[Nat
iona
l Tai
wan
Uni
vers
ity] A
t: 02
:31
28 D
ecem
ber 2
007
18.70 ha to 24.89 ha. The overlap of large landslide areas with the manual method
probably accounts for the higher area concordance for the unsupervised
classification.
The validation by 0.35 m resolution orthophotographs confirmed the errors of
omission and commission that were made by all three methods. Counter intuitively,
we found that the automated method using 10 m resolution multi-spectral images
was able to correctly identify many small landslides around 10 m wide that could not
be positively identified in the 2.5 m supermode imagery.
The application of a 28u slope mask to filter out non-vegetated features with
relatively low slopes that were not landslides was not fully successful. This can be
explained as follows. The slope mask had a resolution of 40 m; therefore, 10 m to
15 m wide roads, parts of riverbeds and parts of farm fields close to steep hill slopes
would still remain in the scene and be wrongly classified as landslides.
7. Conclusion
In this study we tested automated classification methods and manual delineation of
SPOT-5 products for mapping typhoon-triggered landslides. Unsupervised classi-
fication using all SPOT-5 wavelength bands combined with a slope mask produced a
63% area concordance with manual mapping results using 2.5 m supermode
imagery. Statistical tests revealed that the automated methods tended to produce
significantly smaller landslide areas when compared to the manual delineation
Figure 5. SPOT-5 supermode imagery and classification results from the automated andmanual methods for a select area for three different acquisition dates.
1854 A. M. Borghuis et al.
Dow
nloa
ded
By:
[Nat
iona
l Tai
wan
Uni
vers
ity] A
t: 02
:31
28 D
ecem
ber 2
007 results. Classification results were validated by 0.35 m orthophotographs and
showed many errors of omission by the manual delineation method for small
landslides. In contrast, the majority of errors made by the automated methods were
errors of commission that could be attributed to the presence of roads, riverbeds and
bare farm fields in the scene. When a higher resolution DEM becomes available, we
believe that these errors can be reduced to a minimum, thus enabling a cost and time
effective classification of landslides based on medium- and high-resolution multi-
spectral satellite imagery.
Acknowledgments
We thank the Water Resource Agency Southern Bureau for funding the 2004
Tsengwen Project. We thank Shou-Hao Chiang for assisting in the validation of
landslides using orthophotographs. Thanks also to Simon Dadson, Colin Stark and
the anonymous reviewers for their useful comments.
ReferencesALEOTTI, P. and CHOWDHURY, R., 1999, Landslide hazard assessment: summary review and
new perspectives. Bulletin of Engineering Geology and the Environment, 58, pp. 21–44.
BORGHUIS, A.M. and CHIU, Y.J., 2005, Reconstructing more than 40 years of soil
redistribution in two Taiwanese mountainous watersheds. Proceedings of the 2005
European Regional Science Association (ERSA) Conference, Vrije Universiteit,
Amsterdam, The Netherlands.
CHADWICK, J., DORSCH, S., GLENN, N., THACKRAY, G. and SHILLING, K., 2005, Application
of multi-temporal high-resolution imagery and GPS in a study of the motion of a
canyon rim landslide. ISPRS Journal of Photogrammetry & Remote Sensing, 59, pp.
212–221.
CHANG, K., 2006, Introduction to Geographic Information Systems, 3rd edn (New York:
McGraw-Hill).
CHANG, J.C. and SLAYMAKER, O., 2002, Frequency and spatial distribution of landslides in a
mountainous drainage basin: Western Foothills, Taiwan. Catena, 46, pp. 285–307.
CRUDEN, D.M., 1991, A simple definition of a landslide. Bulletin of the International
Association of Engineering Geology, 43, pp. 27–29.