| Thèse soutenue
le 10 juin 2019
| Dong Eon Kim
|
| Directeur
Thèse Polytech'Lab |
Philippe Gourbesville -
Shie-Yui Liong (NUS) |
| Titre |
Simple-and-yet-novel approach
in flood assessment to overcome data scarcity: High quality
DEM and Rainfall proxies
|
|
Résumé
|
Many urban cities in Southeast Asia
witness severe flooding associated to increasing rainfall
intensity and rapid urbanization often due to poor urban planning. Two
important inputs required in flood hazard assessment are: (1) high
accuracy
Digital Elevation Model (DEM), and (2) long rainfall record. High
accuracy DEM
is both expensive and time consuming to acquire. Long rainfall records
for
areas of interest are often not available or not sufficiently long to
determine
the probable extremes. This thesis presents a notably cost-effective
and
efficient approach to estimate high-resolution and accuracy DEM, and
suggests
proxies for long rainfall data.DEM data from a publicly accessible
satellite,
Shuttle Radar Topography Mission (SRTM), and Sentinel 2 multispectral
imagery
are selected and used to train the Artificial Neural Network (ANN) to
improve
the quality of the DEM. In the training of ANN, high quality observed
DEM is
the key leading to a well-trained ANN. The trained ANN will then be
ready to
efficiently and effectively generate high quality DEM, at low cost, for
places
where DEM data is not available.
The performance
of the DEM improvement scheme is evaluated in places of various landuse
types
(e.g. dense urban areas, forested areas), and in different countries
(Nice,
France; Singapore; Jakarta, Indonesia) through various criteria, e.g.
whenever
possible visual clarity, scatter plots, Root Mean Square Error (RMSE)
and
drainage networks. The DEM resulting from the latest version of
improved SRTM
(iSRTM_v2 DEM) performs (1) significantly better than the original SRTM
DEM, a
34 % to 57 % RMSE reduction; (2) the visual clarity is so much better;
and (3)
much closer drainage network with the actual. The much improved DEM
allows
flood modelling to proceed with high confidence.
Rainfall
data
resulting from a high spatial resolution Regional Climate Model (RCM),
Weather
Research and Forecasting driven by ERA-Interim (WRF/ERAI) dataset, is
extracted, analyzed, and compared with high quality observed rainfall
data of
Singapore with regard to accuracy. The comparisons are performed, among
others,
on their Intensity-Duration-Frequency (IDF) curves, the essential
design curves
for flood risk assessment; they matched quite well. The rainfall data
(from the
RCM) are then used as proxies for Greater Jakarta
(Indonesia),
where no rainfall data were available, to derive the IDF curves
required for
the flood analysis. MIKE 21 Flow Model Flexible Mesh
(MIKE 21 FM) is applied to Greater Jakarta, with input data from the
above
mentioned much improved DEM and precipitation proxy data, for flood
simulations
of 2 return periods (50- and 100-years). Qualitative agreement of model
results
and observation of the 2013 Jakarta
flood were obtained. This demonstrates the applications of the
approaches/methodologies, proposed in this thesis, on catchments where
most
essential data for flood risk assessment (high resolution and high
accuracy DEM
and long and high accuracy rainfall data) are not available.
This thesis should be of interest to readers of the
areas of remote sensing, artificial intelligence and flood management,
possibly
also for the policy makers in proposing relevant flood mitigation
measures
under climate change with increasing devastating flood damages and
casualtiese.
|
|
|
