|
Titel |
Flood damage assessment performed based on Support Vector Machines combined with Landsat TM imagery and GIS |
VerfasserIn |
Y. Alouene, G. P. Petropoulos, A. Kalogrias, F. Papanikolaou |
Konferenz |
EGU General Assembly 2012
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 14 (2012) |
Datensatznummer |
250064157
|
|
|
|
Zusammenfassung |
Floods are a water-related natural disaster affecting and often threatening different aspects of
human life, such as property damage, economic degradation, and in some instances even loss
of precious human lives. Being able to provide accurately and cost-effectively assessment of
damage from floods is essential to both scientists and policy makers in many aspects ranging
from mitigating to assessing damage extent as well as in rehabilitation of affected
areas. Remote Sensing often combined with Geographical Information Systems
(GIS) has generally shown a very promising potential in performing rapidly and
cost-effectively flooding damage assessment, particularly so in remote, otherwise
inaccessible locations.
The progress in remote sensing during the last twenty years or so has resulted to the
development of a large number of image processing techniques suitable for use with a range
of remote sensing data in performing flooding damage assessment. Supervised image
classification is regarded as one of the most widely used approaches employed for this
purpose. Yet, the use of recently developed image classification algorithms such as of
machine learning-based Support Vector Machines (SVMs) classifier has not been adequately
investigated for this purpose.
The objective of our work had been to quantitatively evaluate the ability of SVMs
combined with Landsat TM multispectral imagery in performing a damage assessment of a
flood occurred in a Mediterranean region. A further objective has been to examine if the
inclusion of additional spectral information apart from the original TM bands in
SVMs can improve flooded area extraction accuracy. As a case study is used the
case of a river Evros flooding of 2010 located in the north of Greece, in which TM
imagery before and shortly after the flooding was available. Assessment of the
flooded area is performed in a GIS environment on the basis of classification accuracy
assessment metrics as well as comparisons versus a vector layer of flooded area
obtained from the local authorities based on image photo-interpretation. Damage
assessment from the flood is performed on the basis of land use/cover information
derived from the GlobeCover2009 dataset freely-distributed from the European Space
Agency.
Results from our study indicated the ability of SVMs in extracting the flooded area as
well as in performing a flooding damage assessment. The use of additional spectral
information layers in SVMs showed an improvement in the flooded area extraction from the
surrounding environment, in comparison to when only the original spectral bands of the
sensor were used. All in all, our proposed SVMs scheme proved quite capable in performing
flooding damage assessment in the complex and highly fragmented case of our
Mediterranean study site.
Keywords: flooded area mapping, flood damage assessment, remote sensing,
Geographical Information Systems, Support Vector Machines, Landsat TM, Evros river,
Greece. |
|
|
|
|
|