Flood Modelling for Cities using Cloud Computing FINAL REPORT Vassilis Glenis, Vedrana Kutija, Stephen McGough, Simon Woodman, Chris Kilsby Summary Assessment of pluvial flood risk is particularly difficult because it is sensitive to the spatial- temporal characteristics of rainfall, topography of the terrain and surface flow processes influenced by buildings and other man-made features. “CityCat” is an urban flood modelling, analysis and visualisation tool which uses very accurate and computationally efficient solutions for free surface flow equations. In this pilot project a Cloud Computing compatible version of “CityCat” was developed and for the first time applied for estimating spatial and temporal flood risk at a city-scale. The use of Cloud Computing enabled modelling of flooding of larger domains (up to 1100km 2 ) with much higher resolution (up to 16.000.000 computational cells) than it has been done previously. This project has demonstrated that the use of Cloud Computing can enable efficient and detailed modelling of flooding at a city or even regional scale, with high resolution, using standard terrain and rainfall data and powerful Cloud enabled software which does not impose limitations on the computational domain size. Background The risk from pluvial flooding, where intense direct rainfall overwhelms urban drainage systems, in the UK cities is considerable and as a result of 2005 and 2007 floods, local authorities are now required to develop Surface Water Management Plans (SWMPs) However, assessment of pluvial flood risk is particularly difficult because it is sensitive to the spatial-temporal characteristics of rainfall, topography of the terrain, local runoff and surface flow processes influenced by buildings and other man-made features and the performance of urban drainage systems. Conventional assessments of urban flood risk are generally carried out at relatively small scales using commercial software resulting in very restricted coverage in space and number of design storms (Hunter et al., 2008). Alternatively, simplified codes may be used for city scales (Neal et al., 2009) .The use of fully detailed numerical codes for larger areas is in its infancy and the requirement for HPC or cluster facilities means it is limited to institutional platforms such as Condor grids which are restrictive in terms of power as well as, crucially, access for non-institutional users in industry. On the other hand, Cloud Computing (CC) offers (apparently) huge amount of processing power to anyone. Also, Cloud Computing has been successfully used within the business community allowing on-demand access to this power at a (relatively) cheap price. The motivation for this project was exploration of suitability of Cloud Computing for the assessment of city-scale flood risk
10
Embed
Flood Modelling for Cities using Cloud Computing FINAL ... · Vassilis Glenis, Vedrana Kutija, Stephen McGough, Simon Woodman, Chris Kilsby Summary Assessment of pluvial flood risk
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
Flood Modelling for Cities using Cloud Computing
FINAL REPORT
Vassilis Glenis, Vedrana Kutija, Stephen McGough, Simon Woodman, Chris Kilsby
Summary
Assessment of pluvial flood risk is particularly difficult because it is sensitive to the spatial-
temporal characteristics of rainfall, topography of the terrain and surface flow processes
influenced by buildings and other man-made features. “CityCat” is an urban flood modelling,
analysis and visualisation tool which uses very accurate and computationally efficient
solutions for free surface flow equations. In this pilot project a Cloud Computing compatible
version of “CityCat” was developed and for the first time applied for estimating spatial and
temporal flood risk at a city-scale. The use of Cloud Computing enabled modelling of
flooding of larger domains (up to 1100km2) with much higher resolution (up to 16.000.000
computational cells) than it has been done previously. This project has demonstrated that
the use of Cloud Computing can enable efficient and detailed modelling of flooding at a city
or even regional scale, with high resolution, using standard terrain and rainfall data and
powerful Cloud enabled software which does not impose limitations on the computational
domain size.
Background
The risk from pluvial flooding, where intense direct rainfall overwhelms urban drainage
systems, in the UK cities is considerable and as a result of 2005 and 2007 floods, local
authorities are now required to develop Surface Water Management Plans (SWMPs)
However, assessment of pluvial flood risk is particularly difficult because it is sensitive to the
spatial-temporal characteristics of rainfall, topography of the terrain, local runoff and
surface flow processes influenced by buildings and other man-made features and the
performance of urban drainage systems.
Conventional assessments of urban flood risk are generally carried out at relatively small
scales using commercial software resulting in very restricted coverage in space and number
of design storms (Hunter et al., 2008). Alternatively, simplified codes may be used for city
scales (Neal et al., 2009) .The use of fully detailed numerical codes for larger areas is in its
infancy and the requirement for HPC or cluster facilities means it is limited to institutional
platforms such as Condor grids which are restrictive in terms of power as well as, crucially,
access for non-institutional users in industry. On the other hand, Cloud Computing (CC)
offers (apparently) huge amount of processing power to anyone. Also, Cloud Computing has
been successfully used within the business community allowing on-demand access to this
power at a (relatively) cheap price. The motivation for this project was exploration of
suitability of Cloud Computing for the assessment of city-scale flood risk
“CityCat” is an urban flood modelling, analysis and visualisation tool which uses state of the
art numerical solutions for flow equations in a user-friendly visual environment. Its
numerical solutions of the 2D free surface flow equations are very accurate and
computationally efficient but it has been recognised that users would benefit from further
enhancement of the computational power brought in by Cloud Computing as that will
enable modelling of larger domains, extended simulation periods and/or different future
climate scenarios.
Objectives
A meaningful assessment of city-scale flood risk requires a very large number of simulations
on large domains that cannot be achieved by existing institutional HPC or cluster facilities.
Therefore in this pilot study we developed for the first time a truly city-scale application of
the hydrodynamic model “CityCat” for estimating spatial and temporal flood risk using
Cloud Computing.
This has been achieved through the following objectives:
1. Porting the existing state-of-the-art hydrodynamic model ”CityCat” from desktop to
Cloud
2. Selecting and using only readily available data sets for “CityCat” simulations so that
this kind of flood risk assessment can be easily applied nationwide
3. Generating a large number of design storms with different return periods and
durations as rainfall input for simulations
4. Applying “CityCat” to different large domains (4km2 to 1100 km2) for a variety of
extreme storm events.
Methodology and results
1. Porting ”CityCat” from desktop to Cloud
Prior to this project, “CityCat” was written in Delphi and compiled only under Windows.
Also, CityCat’s Graphical User Interface (GUI) was used for preparation and visualisation of
results. Therefore in order to run “CityCat” on the Cloud the following changes were made:
a. The numerical engine was separated from the GUI and input files were used for setting up the model. Also, the variables: water depth, velocity in the x direction and velocity in the y direction were saved at predetermined time step intervals.
b. A new version of “CityCat” was developed and compiled under Linux because if the Windows version was used then the Windows OS would need to be installed at each PC. That would have increased the cost and due to our limited resources it would in fact have reduced the number of runs we could do.
In this project we adopted a high throughput model of computation on the Cloud in which a
Condor (http://research.cs.wisc.edu/condor/) cluster of nodes were deployed as a set of virtual
machines instances on the Amazon Cloud. Each instance was a standard Ubuntu Linux image
with the addition of the Condor deployment configured to use the large scratch space
provided with these images. A set of parameter sweep jobs were deployed by modifying the
original source code such that each job could be instantiated by passing a single integer
number as part of the command line arguments to the program. This caused the correct
configuration files to be selected. A simple script was used to wrap each job and would first
decompress the files needed for each run before executing the main program and then
compressing the results back up before returning the results to a central Condor computer
on the Cloud. These files were then staged back to computers within Newcastle University.
2. Standard Datasets needed for “CityCat” simulations “CityCat” uses standard datasets for simulations. For the topography the digital terrain model (DTM) is used (see Fig. 1) and the numerical grid is generated automatically using the cell sizes of the DTM.
In addition to the terrain data, “CityCAT” uses the buildings layer from the OS- MasterMap (see Figure2) in order to exclude the buildings footprint from the computational grid. The cells which are removed from the computational grid are characterised as building and depending on different options/properties, the water captured on the roof of each building is either directly distributed to the neighbouring cells or slowly released and distributed to the neighbouring cells if a green roof is introduced. Exclusion of the buildings from the computational domain (Fig.3) improves the ability of the model to capture realistically the flow patterns in urban areas while the different options for the roof
drainage enable assessment of the different adaptation techniques for flood risk reduction. In most of the simulations within this project the buildings were cut-out of the computational domain and the rainfall falling on the roofs was directly distributed to the neighbouring cells. Hydrodynamic simulations are driven by boundary conditions which are “sources” of water in the model. Usually, these are time dependent functions which describe rainfall over the domain or water entering the domain from an external source. External sources of water can either introduce some volume of water in the domain (for example from a burst pipe) or they can force the water level at the boundary of the
Figure 2 -An example of Master Map coverage. Solid representation
Figure 1 - Terrain given by a Digital Terrain Model (DTM)
domain to follow a certain condition (e.g. river water level or tide). In this project most of the simulations were driven by rainfall as most of urban flooding is of that origin. However, two examples of forced water level boundary condition were also tested to explore the
model’s ability to simulate such complex flows.
3. Rainfall events A set of storm events, for different return periods and different storm durations, have been
created following the standard procedure from the Flood Estimation Handbook (FEH) and a
summer profile rainstorm located at Newcastle was used. Modelling different return periods
is necessary if we want to assess different level of risk as a rainfall event with a return
period of n years has probability of 1/n to happen in any given year. The storms with higher
return period are less likely to happen but they are larger in magnitude so should they
happen the flooding would be more likely. Different rainfall durations are used because, it is
not clear which storm duration would be critical for a given situation. The shorter duration
rainfall events usually have larger rainfall intensity but the overall rainfall volume is not very
large, whereas rainfall events with larger duration have larger volumes of rainwater.
Susceptibility to flooding of any particular area is determined by its topography and other
features and depends on the combination of these two factors (rain intensity and total
volume of rain) in a complex way. Therefore, being able to model an extensive range of