Flood Inundation Modeling and Hazard Mapping under Uncertainty in the Sungai Johor Basin, Malaysia

Flooding can have devastating impacts on people’s livelihood, economy and the environment. An important instrument in flood management is floodplain maps, which assist land planners and local authorities in identifying flood-prone areas, and provide useful information for rescue and relief agencies for their operations. Developing floodplain maps often involves flood inundation modeling. This typically requires precipitation and stream flow data, topographic information, building a hydraulic model and calibration of its parameters. Often however, floodplain maps are built on a single model outcome without an explicit consideration of all the sources of uncertainty in the modeling process. The research presented in this thesis addresses the uncertainty in flood inundation modeling, which may arise from input data and hydraulic modeling approach. The study area is the Sungai Johor basin in Johor, Malaysia, an agriculture-dominated area. The present study analyses the modelling uncertainties arising from estimations of design flow, terrain data sets, geometric description in hydraulic models and different modeling approaches, and develops recommendations for practitioners. Explicit account for uncertainties and studying their impact in flood inundation mapping allow for more informed and effective decision making.

This thesis is dedicated to my late parents. Who born me, raised me, supported me, taught me and loved me.
Allahyarhamah Hajjah Halijah Bt Md. Yassin (06 August 1946 ~ 06 April 2015) Allahyarham Haji Md. Ali Bin Hj. Ismail (31 December 1935~ 11 June 2016 Al-Fatihah vii Summary Flood is a natural disaster that occurs almost regularly in Malaysia particularly during the monsoon seasons. Hence, it is of no surprise that flood is considered one of the most significant natural hazards in the country in terms of number of affected population, fatalities and economic damage. One of the efforts to minimize flood losses is providing useful information through floodplain inundation maps, i.e. spatial distribution of flood hazard. Traditionally, many modellers have used deterministic approaches in flood inundation modelling. Deterministic approaches are based on a single simulation with the "best fit model" and do not explicitly consider uncertainties in model parameters, terrain data, and model structure. When model results are then used to generate a flood hazard map, neglecting uncertainties may lead to precise, but inaccurate maps and lead to wrong or misleading information to decision makers. Thus, the scientific literature has recently proposed a number of probabilistic methods to recognize, assess and account for uncertainties affecting flood inundation modelling. In this context, this research work aims to contribute to this research work by further exploring the impact of various sources of uncertainty on the results of hydraulic models. The case study of this research is the Sungai Johor river basin in Malaysia. Both 1-D and 2-D hydraulic models were utilized.
In using 1-D hydraulic models, the geometric description of rivers and floodplains is performed by using a number of cross-sections, which play an important role in the accuracy of model results. In this work, criteria for cross-section spacing were tested and verified via numerical experiments.
Similarly, digital elevation models (DEMs) used as geometrical input significantly affect the results of flood inundation modelling exercises. DEM is essential input that Summary viii provides topographical data in flood inundation modelling. However, it can be derived from several sources either through remote sensing techniques (space-borne or air-borne imagery) or from more traditional ground survey methods. These DEMs are characterized by different precision and accuracy. This study quantified the effect of using different DEM data source and resolution in a 1-D hydraulic modelling of floods.
This study also explored the differences arising from the use of deterministic and uncertainty approaches in deriving design flood profiles and flood inundation maps.
To this end, the generalized likelihood uncertainty estimation (GLUE) technique was used and the uncertainty in model predictions was derived through Monte Carlo analysis. In particular, this work focused on impact of uncertain inflow data and roughness coefficients in the accuracy of flood inundation models.
As part of this research, 2-D hydraulic modelling software (LISFLOOD-FP) was also used to assess the effect of spatial data re-sampling (e.g. from high to low resolution) on model outcomes. This study evaluated two re-sampling techniques with combination of three different aggregation functions, i.e. minimum, maximum and mean values.
This research work has not only provided useful results, but has also suggested further research and improvement of flood risk and mapping practices. The knowledge generated by, as well as the findings of this thesis, will be transferred to other study areas in Malaysia.

Background
Flooding is the most significant natural hazard in Malaysia in terms of number of affected population, fatalities and economic damage. Since 1920, the country has experienced major flood events in 1926, 1963, 1965, 1967, 1969, 1971, 1973, 1979, 1983, 1988, 1993, 1998, and 2005   According to DID (2003), the total flood prone area in Malaysia is around 30,000 km 2 , while the country total area is 328,799 km 2 (see Figure 1.1). It is also estimated that as 2000, 22% of the total population of Malaysia, which counts 22.2 million people, lives in this flood prone area. In term of economic damage, as at 2000, the total Annual The flood hazard map developed for the Sungai Johor basin helps in assist in the assessment and management of flood risk, however not all uncertainties associated with this problem were considered, not all available data sources were used (like LiDAR), and the models used could be better fine-tuned. There are a number of ways to improve flood hazard mapping for Sungai Johor basin, and it is our intention to do it in this study.

Flood mapping
Flood mapping is an issue addressed in many countries. It is worth noting that the European Union (EU) has adopted a new directive known as EU Floods Directive (EU, 2007) that proposed a transition from traditional flood defence approaches to holistic flood risk management strategies . The main objective of the European directive is to reduce and manage flood risk by Several important parameters are required for performing hydraulic flood modelling such as topographic data, discharge data to provide model inflow and outflow as boundary conditions, estimation of the roughness coefficient and validation data (Bates 2004).
A substantial of research have been made to investigate the flood hazards, not only to understand the behaviour of flood flow (i.e. in river channel and floodplain), but also the characteristics of flood such as occurrences, magnitude and extent. Most of this effort was reasonably carried out by conducting the hydraulic modelling of floods (Horrit and Bates, 2002;Patro et al., 2009;Poretti and De Amicis, 2011;Crispino et al., 2014). Furthermore, the output from hydraulic modelling of floods for instance in estimation of inundations area and flood profile is useful information's for assisting the decision makers in flood relief planning and operations.
Although maps of flood hazard provide useful indication on the potentially inundated area and negative impact posed by flood, there is significant uncertainty associated with these maps . Unfortunately, although modellers are well aware that significant approximation affects flood hazard assessment and various methods to deal with uncertainty have been recently Introduction 5 developed, the awareness among environment and river basins agencies, authorities and engineering consultancies is still lacking as the advances in uncertainty analysis are hardly applied. To facilitate a wider application of these methods, the development of clear methods is therefore needed .

Uncertainty in flood hazard mapping
The most common representation of of flood inundation modelling results remains a deterministic approach based on a single simulation using a best fit model.
Unfortunately, this approach does not explicitly account for the uncertainties in the modelling process  and may lead to a precise but inaccurate hazard assessment , despite increasing knowledge in flood propagation and inundation processes.
Although ample literature has been discussed to identified the source of uncertainties in flood inundation mapping (Bales and Wagner, 2009;Domeneghetti et al., 2013;Dottori et al., 2013;Jung et al., 2013;, but to eliminate the uncertainties completely are impossible due to various limitations such as computational times, cost, technology and knowledge of the flood science itself.
Uncertainty in flood hazard mapping may arise from accuracy of topographic data flood hazard mapping. As an example, it's common to integrate between surveyed river cross-section data with existing topographic floodplain data.

Research questions
The proposed study aims to address the following research questions: i. How do many sources of uncertainty (e.g. hydrologic data, topographic data, and model selection) affect flood hazard mapping?
ii. What are the potentials and limitations of different data sources (including remote sensing) in supporting flood inundation modelling?
iii. How can we model uncertainty to better define safety levels in the design of flood protection structures?

Aim and research objective
The general aim of this study is to develop a model-based methodological framework allowing for flood mapping and thus assisting public administrator in making appropriate decisions under uncertainty, with application to Sungai Johor basin.
The specific objectives of this research are as follows: i. To identify the most relevant sources of uncertainty associated with the generation/development of flood hazard maps.
ii. To develop and integrate the necessary models and data sources (including remote sensing data) allowing for accurate description and prediction of the natural processes leading to flooding, and thus supporting flood mapping. iii.
To identify the source/effect of uncertainty related to safety levels of flood protection structures.

Dissertation Structure
This thesis is organised in eight chapters. The first three chapters are general.
Chapter 1 provides an overview of the research with concise explanations of its Chapter 6 addresses the research question number two by using different sources of DEMs (with different resolutions) and remote sensing data in 1-D hydraulic modelling.
Chapter 7 addresses the third research question by comparing deterministic and probabilistic approaches for floodplain mapping using 1-D hydraulic modelling.
Lastly, Chapter 8 summaries the findings and presents the conclusion and recommendations.

What is floods
Flood is a natural hazard that resulted from combination of hydrological and meteorological factors. It occurs when a normally dry land areas are temporary inundated due to overflowing of water at the natural or artificial confines of a river, including groundwater caused by prolonged or heavy rainfall. (Wisner et al., 2004;Martini and Loat, 2007;Klijn 2009). Hydrologists define flood as a sudden increase in water discharge that caused a sudden peak in the water level. Once flood is over, the water level will drop back to near-constant base flow or no flow. As summarized by Martini and Loat (2007), flooding is when water and/or sediments exist at unwanted areas other than the water body. Whereas, Ward (1978), defined flood as a body of water which is not normally submerged.

Types of flood
Flood can be categorized into different types based on location of occurrence and what cause them. The major ones are as described below.

River flood
River flood occurs when a river basin is filled with too much water that is more than the capacity of the river channel. River flood is considered as an expected event as it usually occurs seasonally, normally during rainy seasons. The surplus water overflows the river banks and runs into adjoining low-lying lands.

Coastal flood
Flood that occurs in coastal area due to the drive of the ocean waters inland is known as coastal flood. Natural phenomenon such as tropical storm, hurricane or intense offshore low pressure can cause unusually high amount of the ocean water to be driven towards the land resulting in the coastal flood. Similarly, tidal sea waves that happen due to earthquake or volcanic activities in the sea can also caused coastal flood.

Urban flood
Heavy rainfall and changes in the runoff behaviours are the most common reasons for urban floods. The changes in the runoff behaviours is mostly due to the development of the land to buildings and paved roads which have less absorbing ability compared to an undeveloped area or natural fields. The rainfall runoff in the urban areas can be as high as six times than that in a natural fields. As a result, roads become rapid rivers and basements as death traps when they are filled with water.

Flash flood
Flash floods occur when a large amount of water flood within short period of time.
Normally it occurs locally and suddenly without or with little warning. Flash floods could happen due to immoderate rainfall or a sudden release of water from a dam.
This research will focus and discuss on river flood and the extent of the flood to the adjacent area along the river.

Flood prone areas
The areas adjacent to a river prone to flooding can be defined as floodplain and floodway. A flood area that is deep with high flow velocities with presence of debris flow that can cause possible erosion is identified as floodway. There should be no development allowed to take place within the floodway area except for critically necessary infrastructure such as bridges (UNISDR 2002).
A floodplain on the other hand represents the areas surrounding the river channel

Hazard and flood hazard
It is important to understand and be accustomed with the terms and terminology used in disaster management. However, there are different definitions and terminologies used implicated in term of hazard and flood hazard. Below are the defining term of hazard and flood hazard. property, livelihoods, infrastructure and services, social and economic disruption or environmental degradation. Samuels et al. (2009) defined hazard as a physical event, phenomenon or human activity with the potential but not necessarily lead to harm.

Flood hazard
Flood is one of the most commonly occurred environmental hazards that may not necessarily caused by natural events but can also be due to or aggravated by human activities such as deforestation, pollution or uncontrolled urbanization that changes or disrupts the natural landscape.
According to ISDR, only a few hazards, such as earthquakes and hurricanes, are true natural hazards. Flood is categorized as a socio-natural or unnatural hazards where a naturally original disaster aggravated by human factors (ISDR 2009).
In general, flood hazard is the result from a combination of physical exposure represented by the type of flood and their statistical pattern at a particular site, and human vulnerability to geophysical processes. Human vulnerability is associated with keys socio economy such as the numbers of people at risk on the floodplain and the ability of the population to anticipate and cope with the hazard. Merz et al. (2007) defined flood hazard as the exceedance probability of a potentially damaging flood event in a particular area within a specific period of time. However, this statement does not represent the consequences of such floods to community, environment or development.
A flood hazard statements should taken into account the depth of the process that goes beyond a flood frequency curve such as the inundation depth, flow velocity,

Flood modelling
Flood modelling is a simplification of the real situation event. A flood model of a particular river basin for example simulates the real flood events that have occurred using the actual hydrological input data, the basin's hydraulic characteristics and boundary conditions. These modelling are able to show effects on the results based on different boundary conditions or input data. Hence by simulation, the behaviour of the flood risk or hydraulic characteristics at a certain period of time can be determined and investigated.
In the development of flood mapping, with recent advances in technology whereby computation time has been tremendously reduced, it is becoming necessary to simulate flood inundations in the flood plains caused by different magnitudes of flood events. Nowadays, different types of inundation models exist and approaches have been made by various researchers by using various hydrodynamic modelling models . One of the most important developed tools for hydraulic modelling is geographical information system or GIS that allows one or two dimensional representation of computed hydraulic parameters. Variety of software has been used widely for dynamic 1-D flow simulation in rivers such as MIKE 11, HEC RAS, SOBEK-1-D etc. Even though the 1-D models are simple to use and provide information on bulk flow characteristics, it is however fail to provide information particularly on the flow field. A 2-D model whereas require substantial computer time to provide the information.
As there is a limitation of using 1-D or 2-D numerical models, attempt have been made to couple 1-D river flow models and 2-D floodplain flow models. The coupled between two numerical models offer a great advantage for real time simulation of flood events. Among that coupled models known is SOBEK The flow in the channel can be presented as: where Qc is flow in the channel and Q is total flow. Here, determines how flow is partitioned between the floodplain and channel, based on the conveyance of Kc and

Kf.
Where is calculated as while Kc is represents as conveyance in the channel and Kf is conveyance in the floodplain. Conveyance is defined as where P is wetted perimeter, A is cross-section area and n represents Manning's n roughness coefficient. From the above equation, the 1-D equation can be written as follows: where Ac and Af is the cross sectional area of the flow of the channel and floodplain, xc and xf are the distances along the channel and floodplain, R is hydraulic radius (A/P) and S is the friction slope. The finite difference method was utilized for discretion of the equations 2.4 and 2.5 and solved using a four-point implicit method.

LisFlood-FP
The LISFLOOD-Floodplain or also known as LISFLOOD-FP is a hydraulic model originally developed by Bates and De Roo (2000). This model has been broadly tested and compared with other models in determine the standard of the model (Neal et al., 2012). Furthermore, the stability of the original numerical solver by Bates et al. 2010 for low friction condition has been improvised by de Almeida et al. (2012). This LISFLOOD-FP works on a 2-D regular grid structure and simulates water flow by solving the shallow water equations in 1-D, without the convective acceleration terms from the momentum equations .
To calculate the flow, Q between cells, equation 2.7 is used: Where q is the flux between cells from previous iteration, g is gravity, hflow is the maximum depth of flow between cells, ∆t is the model time-step, h is the water depth in each cell, z is elevation, ∆x is the cell width and n is a friction coefficient.
Having established the discharge across all four boundaries of a cell, the cell water depth (h) is updated using equation 2.8: Where cell are indexed in two-dimensions using i and j. To enhance the model robustness, the time step, t which is controlled by shallow water Courant-Friedrich-Levy (CFL) condition was introduced in the LISFLOOD-FP formulation: where α is a coefficient typically defined between 0.3 and 0.7 .

GIS environment
A hydraulic model is intended to represents the flood physical processes over time of a river channel or flood plain as realistic as possible that able to provide acceptably accurate output for different scenarios to its user (Pullar and Springer, 2000). With GIS, a hydraulic model is presented in a spatial or geographical manner that would allow the model to analyze, predict and solve engineering problems in a more powerful and comprehensive way.
Many modelling application uses GIS as the database manager and visualization tools through the use of Windows Graphical User Interfaces (GUIs) making the output easier to understand by its users. The benefits of GIS integrated modelling are tremendous.
With the integration of GIS in these modelling, some of the techniques or procedures from the manual flood hazard mapping processes may need to be modified or changed, among others include search method, governing algorithms, data requirements and flood inundation extent and depth. (Noman et al., 2001). In GIS, data can be extracted, combined with others or reformatted if needed for various modelling processes and even used to generate other inputs as required by the models (Robbins and Phipps, 1996). It is important as suggested by Noman et al., (2001) that the integration of GIS in a hydraulic model should be made in such a way to allow automatic data transfer without jeopardizing the ability to replace the hydraulic model with the alternative ones. The data exchange system between hydraulic model and GIS software was first developed by Evans (1998) using HEC-RAS as the study package. The system enable HEC-RAS to import cross-section coordinates from a terrain model to develop channel and reach geometry and exports the data back to a GIS upon completion of the hydraulic calculations for comparison with the terrain model. In 1998, ESRI further translated and improved Evans' code and with some added utilities enhance its use. The result was an ArcviewGIS extension called AVRas. In general Arcview GIS allows user to work with maps and geographic information.
Study by Tate et al. (1999) to improve the HEC-RAS model's accuracy led to the development of Avenue scripts for Arcview GIS that incorporates data such field survey, stream geometry and control structures into a GIS-based terrain model. A very accurate digital orthography was used to develop the terrain model in this study.
Using this Avenue scripts , applied GIS techniques to create from linear cross-sections a continuous river bathymetry in the form of a 3D mesh, and integrated this bathymetry with the floodplain topography using a simple smoothing algorithms.

Input data for flood modelling
The performance of any model can only be as good as the data it uses to parameterize it and to calibrate and validate. While models should be selected based on the characteristics of the problem in hand, it is also clear that models of different complexity have different data requirements, and in practice this may constrain user's choice in model selection. Asselman et al. (2009), highlighted the data required by any hydraulic model are in principal the boundary condition, initial condition, topography data, friction data and hydraulic data for use in model validation. Whereas, Methods et al. (2007) noted the data required for flood modelling are hydraulic boundary determinations, geometric data, discharge data, roughness data and calibration and validation.
In general, modelling of floodplain flooding requires high quality input data, which should include rainfall, a digital terrain model, land use and calibration. Rainfall data should ideally be provided by a dense network of rain gauges and/or the weather radar. Both sources are important, the former is generally considered as more accurate, whilst the latter typically has higher spatial resolution, which enables advanced applications such as nowcasting (quick precipitation forecasting). The followings are the available techniques in obtaining a DEM for a flood modelling : i. aerial stereo-photogrammetry (Baltsavias, 1999;Westaway et al., 2003), ii. airborne laser altimetry or LiDAR and iii. airborne Synthetic Aperture Radar interferometry (Hodgson et al., 2003) and iv. Radar interferometry from sensors mounted on spaceborne platforms, in particular the Shuttle Radar Topography Mission (SRTM) data (Rabus et al., 2003). Currently, LiDAR is the most used techniques in the hydraulic modelling literature (Marks and Bates, 2000;French, 2003;Charlton et al., 2003;Cook and Merwade, 2009).
Land-use data is used to automatically parameterise variables such as roofed and other impervious areas, surface roughness, etc. Land-use images can be provided by remote sensing. Where this technology is not available, key features such as streets, car parks, housing, green areas, etc. can be distinguished from the existing maps and the corresponding imperviousness and roughness can be assigned to each surface type.
Finally, calibration is required in floodplain models to match the flood extent and the flood depth. It is usually difficult to properly calibrate, verify and validate a flooding model due to a lack of comprehensive sets of observed data. Observed flood extents such as flood mark, photos or video record can be invaluable for calibration.

Definition of uncertainty
In recent years, flood disasters have contributed to the realization that the future is in inherently uncertain. In flood risk management, one of the crucial issues is how to deal with the uncertainty. Generally, uncertainties are associated with human behavior, organizations and social system which make it more difficult to predict future vulnerability of area to flooding. Uncertainty reduces the strength of confidence in the estimated cause and effect chain.
Different authors give different interpretation and definition of uncertainty and other related terms such as error, risk and ignorance. Walker et al. (2003) defined uncertainty as the deviation from the ideal complete determinism of knowledge of a Literature Review 21 relevant system that is not achievable. Pappenberger et al. (2005), meanwhile described uncertainty in a more general concept that reflects the lack of sureness about something that can be as little as just a short of complete sureness to an almost completely lack of conviction about the results. Whereas Refsgaard et al. (2007) on the other hand describes uncertainty as the result of the given information being incomplete or blurred, inaccurate, unreliable or inconclusive, or potentially falsely judge that led a person to be uncertain or lacks confidence about the specific outcomes of an event.

Types of uncertainty
There are many ways to differentiate uncertainties. Apel et al. (2004) and Thieken (2005, 2009), classified uncertainty into two types known as aleatory uncertainty and epistemic uncertainty (see Table 2.1).
From the point of view of model-based decision support, Walker et al. (2003) distinguish uncertainty into three dimensions as follows.
o The location within the model where the uncertainty shows itself ; o The level where the uncertainty manifests itself along the spectrum of different levels of knowledge between determinism and total ignorance in terms of statistical uncertainty, scenario uncertainty and recognised uncertainty; o Nature of uncertainty whether due to the lack or imperfection of knowledge or the inherent variability of the case being described in the study.  • Subjective uncertainty, • Lack of knowledge/limited knowledge uncertainty, • Ignorance, • Specification error, and • Type-B uncertainty 2.6.3 Sources of uncertainty Prinos et al. 2008 have identified the likely sources of uncertainties for each element with the variable, and divided it into three types, known as model uncertainty, parameter uncertainty and data uncertainty. Nevertheless, the uncertainties in flood risk management such as in simulation modelling are principally due to natural variability and knowledge uncertainty. In flood risk mapping, source of uncertainties arising from several factors such as model approach (Cobby et al., 2003;Horritt and Bates, 2002;Horritt et al., 2006;Tayefi et al., 2007), topography (Casas et al., 2006), friction coefficient (Aronica et al., 2002); grid cell size (Werner, 2001), or flow characteristic (Purvis et al., 2008).

Types and content of flood map
In the past, government agencies implemented engineering solutions such as dams, levees, seawalls and others in the attempt to reduce flood damage to the communities. However, these solutions often did not reduce flood damage costs and property loss, nor discourage continued development within the flood-prone area. Now, there is an action to transform the management of flood from a conventional flood defence solutions to a flood risk management approach. In Europe, the European Parliament has adopted a new Flood Directive with the main objective is to establish a framework to assess and manage flood risk (EU, 2007). One of the directive tasks is to produce flood hazards maps and risk maps in every state that will form the basis of a flood risk management plans in the future. Thus, to achieve this directive, flood mapping has become a priority and an important aspect for the EU members.
In the field of flood risk management, the confusion is not only arising in use of risk related definition, but also in the naming of different flood maps (de Moel et al., 2009). For instance, Merz et al. (2007), proposed four type of flood map namely as  Table 2.3). In general, flood map can be defined as a map presents the area prone to flooding at one or more floods with given return periods.  (Merz et al., 2007)

Content
• Flood extent according to probability classes, according to past events.
• Degree of danger.
Purpose and use • Land-use planning and land management Watershed management.
• Water management planning.
• Hazard assessment on local level.
• Emergency planning and management.
• Planning of technical measures.
• Overall awareness building.
Target group/use • National, regional or local land-use planning.

Flood hazard map in Malaysia
In  For this study, to characterize the site into hazardous classes, the inundation water depths used with modifications after the standards of Japan International Corporation Agency (JICA) were adopted to generate flood hazard maps (see Table   2.7). These water depth classes were based on human characteristics in conformance with the Japanese Flood Fighting  Administratively, the State of Johor is divided into eight districts in which most of Johor River is located entirely within the Kota Tinggi district (see Figure 3.1).
The administrative town of this district is also known as Kota Tinggi and has a total administration area of 3,500 km 2 . 65% of its border is surrounded by the sea. As  River, Telor River, Panti River, Tembioh River, Permandi River, Seluyut River, Berangan River, Tiram River and Lebam River (see Figure 3.3).

Climate
Johor River Basin receives equatorial climate with constant high temperatures and high relative humidity. The temperatures vary little throughout the year with an annual average temperature of 27ºC and a mean relative humidity of 82%. As shown in Table 3.1, the rainfall pattern in Johor River Basin is influenced by two monsoon regimes, Northeast Monsoon and Southwest Monsoon. Southwest May -September Winds below 15 knots affecting west coast area.
The rainfall is usually greatest in the months of November to January but rain falls in all months of the year. Rainstorms are short, intense and generally have a limited spatial extent. The rainfall distribution over the catchment of Johor River is 3,000 mm of the north and north-west, and about 2,000 mm in the coastal region with an average of about 2,500 mm.

Land use
The land use conditions within the basin as a whole is covered by the natural forest, Most of the agriculture area is planted with oil palm and rubber tree. Other land uses are for built up area, swamp, water body and others (see Figure 3.5).

Flood issues
The Johor River Basin has been subjected to repetitive flooding during its history.
Most of the severe floods occurred during the north-east monsoon period, which brought large volume of runoff to the relatively large basin of Johor River. During December 2006 and January 2007 floods, it has been reported that Typhoon Utor has been associated to the series of floods that hit Malaysia, Singapore and Indonesia.  Besides the recent two major floods, Kota Tinggi town also experienced occurrences of major historical floods in December 1948, December 1969, January 1970, November 1979, December 1982, November 1989, December 1983, December 1991, December 2003, December 2004and March 2004(DID 2009).
Study area and data availability 39

Hydrological data
There are about 23 numbers of rainfall stations, 5 numbers of water level station and 1 numbers of stream flow station within the Johor River Basin. All the hydrological stations were maintained by Department of Irrigation and Drainage, Malaysia (DID) (see Figure 3.9).

Topography data
For this study area, five type of topography data which are available either commercial or non-commercial (freely available). The following will describe the type of topography data used in this study.

e. River cross section survey
The survey was carried out through a ground survey method across the river at average width range from 70 to 300 m with the spacing between cross section at approximately 1000 m. The purpose of this survey is to enable sufficient coverage data for the area under water which may not be captured by others survey method such as radar wave and laser altimetry.

43
There are also three numbers of bridges are identified located along this study reach.
Information available for these bridges includes (  Despite the development of a plethora of two-dimensional (2-D) hydraulic models (e.g. Franques and Yannitell, 1974;Cunge 1975;Horrit and Bates, 2002;Horrit et al., 2006;Tayefi et al., 2007;Cook and Merwade, 2009;Castellarin et al., 2011), onedimensional (1-D) hydraulic models remain widely used for flood risk studies (Baptist et al., 2004;Hooijer et al., 2004;. A number of studies have showed that the performance of 1-D model are often very close to the one of a 2-D model provided that the topography of the river and floodplain is properly represented (e.g. Horrit and Bates, 2002;Castellarin et al., 2009;Cook and Merwade, 2009).
In terms of topographical data, 1-D modelling requires a certain number of crosssections to represent the river channel and its surrounding topography (Cook and Merwade, 2009). Nevertheless, only a few guidelines are available to assist the modeller in determining the location or spacing between the cross-sections (Samuels 1990;1995;Castellarin et al., 2009;Fewtrell et al., 2011). Common sense suggests that locating a river cross-section at every river bend and at upstream/downstream of every structure across the river may produce better result.
However, having cross-section data at finer spacing or at every river meander (i.e. river bend) and at upstream/downstream of structures (e.g. bridges) will increase the cost of the topographical surveys. Samuels (1990) recommended certain guidelines for the selection of cross-section locations for 1-D hydraulic models. These guidelines were based on a combination of common sense, practical experience and mathematical equations. Samuels' guidelines aimed to provide a correct reproduction of the backwater effects and an accurate representation of the physical waves. In general, Samuels (1990) emphasised to include the following locations of river cross-sections: (ii) At the upstream/downstream of the river structures (i.e. bridge or weir); (iii) At all discharge and water level stations along the reach; and (iv) At sites which are important to the modeller.
In addition to these ʹʹessentialʹʹ cross-sections, Samuels (1990) recommended the following equations in determining the spacing of the cross-sections: where ∆x is the spacing between two cross-section spacing; B is bankfull surface width of the main channel and k is constant (with the recommended range between 10 to 20).
For an appropriate estimation of backwater effects in subcritical flow, Samuels (1990) suggested: where D = bankfull depth of flow, and s = main channel slope. Over this length, the backwater upstream of a control (as well other disturbance) decays to less than 0.1 of the original value. Finally, the minimum spacing between the cross-sections can be determined due the influence of the rounding error as follows: where q represents the number of decimal digits precision, d is the digits lost due to cancellation of the leading digits of the stage values; s is average surface slope and Ɛ s is relative error on surface slope. However, it was noted that, these guidelines were based on the condition of the rivers in United Kingdom.
To test the optimal cross-sectional spacing in 1-D hydrodynamic model, Castellarin et al. (2009) presented two case studies in United Kingdom and Italy. The main objective of their study was to evaluate the efficiency of the 1-D model by adopting some of the guidelines and equations recommended by Samuels (1990). As a result, for both selected river reach, the inaccuracy of the model in terms of mean absolute error (MAE) was found to increase as the spacing between the cross-sections increases. However, differences between the models were found to be relatively small (less than 0.10 m) and within the accuracy of computational hydraulic models (Samuels 1995;Di Baldassarre and Claps, 2011). Castellarin et al. (2009) concluded that while the minimum spacing of cross-section adopted through the Equations 4.2 and Equations 4.3 were appropriate for describing the hydraulic behaviour of the 1-D hydraulic modelling: the role of cross-sections spacing 49 river reaches being studied, it was found inadequate when the hydraulic models is to be used for designing flood mitigation structures. The main limitation of the study by Castellarin et al. (2009) was that numerical results were not compared against observations. Instead, the results from the hydraulic model with the highest number of cross-sections were used as a benchmark/reference. (2009)  This chapter aims to explore the impact of cross-section spacing by using real world observations. In particular, the performance of different 1-D hydraulic models based on different cross-section spacing is evaluated (for the first time) using independent calibration and validation data of flood water level observations.

Hydraulic Modelling
To perform hydraulic modelling, the Manningʹs n roughness coefficient for the river channel ranged from 0.020 to 0.080 m -1/3 s, while n for flood plain from 0.030 to 0.100 m -1/3 s (Chow 1959). In order to test the sensitivity of the results to Manningʹs n roughness coefficient, the models were run using various n values within those ranges. In the absence of any knowledge of the prior distribution of the model parameters, a random distribution was assumed to select 5000 sets of Manning's n roughness coefficients value within these ranges for each hydraulic model. Apart

Cross-section spacing
In

Results and Discussion
This section is divided into three parts. The first part describes the performance of Then, the five best fit models (i.e. the five models using the calibrated Manningʹs n roughness coefficient) were used to simulate the January 2007 flood (validation event). Table 4.3 shows the average MAE of each model for the simulated water levels at the cross-section X and cross-section Y. It was found that the MAE value for different models does not change significantly as spacing between the cross-sections increased from 1,000 m to 8,000 m. The results of this study as summarised in Table 4.3, confirm that the equations recommended by Samuels (1990) provide a proper indication about cross-section spacing. In fact, although the models derived from different spacing of crosssections, the differences of MAE is between 0.02 m and 0.13 m.

Comparing flood water profiles and inundation maps
The results of the hydraulic models can be used to produce flood water profiles (Brandimarte and Di Baldassarre, 2012) as well as flood inundation maps  to support flood risk studies. The former are useful for the design of flood protection measures (e.g. dikes and levees), while the latter can allow the identification of flood prone areas. water profiles pointed out that, although small differences were found in terms of performance in simulating flood water levels at the cross-section X (see previous section 4.3.1), that significant differences arise in terms of the spatial distribution of the flood water profiles. This is caused by the absence of topographical information, and therefore geometric input and output, along the river. This outcome suggests that a detailed topographical survey is required when hydraulic modelling is meant to support the design of flood mitigation structures, such as levees (see also, Castellarin et al., 2009).   Table 4.4), which is less than the accuracy of LiDAR data (around 0.30 m).
In contrast, the resulting flood inundation patterns are rather different (see Figure   4.5) as the flood extent mapping is strongly affected by the cross-sections used as geometric input of the hydraulic model.    The performance of the five models (characterised by different cross-section spacing ranging from around 1000 m to around 8000 m; see Table 4.1) did not show significant differences. For instance, the differences in model performance were between 0.02 m and 0.13 m for the validation event of January 2007. This confirms the findings by Castellarin et al. (2009) as the models developed according to Samuels (1990) performed as good as the models with a higher number of crosssections.
Although model performance at the water level station X and Y is not significantly Chapter 5 2-D hydraulic modelling: the role of digital elevation models

Introduction
The prediction of inundation area and knowledge flood dynamics is a useful tool in floodplain and flood risk management (Horritt et al., 2007). In the past, to use a one dimensional (1-D) hydraulic model is a common practice among hydraulic modeller compared to the two dimensional (2-D) hydraulic models. This is due to the fact that 2-D models have been confined by the inadequacy of high quality topographic data (Horrit and Bates, 2001;Mark and Bates, 2000;Yu and Lane, 2006;Fewtrell et al., 2008  study are consistent across model scales and contrary to previous findings conducted by Yu and Lane (2006).
In particular, the result presented in this chapter aims to fulfil two objectives. The first objective is to evaluate the effect of different re-scaling techniques on model performance and inundation area. Secondly, to investigate the influence of different resolution on the model performance and inundation area which will be evaluated relative to benchmarking high resolution DEMs.

Differentiation of DEMs re-sampling technique
In this chapter, the aggregate functions are applied using Spatial Analysis Tools in In order to test the significance of scale, the original LiDAR DEM with resolution of 1 m was re-sampled into three coarser resolution raster DEMs at 30, 90 and 120 m. As shown in Table 5.1, for each re-sampled DEMs namely as Raster Type A and Raster Type B, two different re-sampling techniques were applied. In total, six models were developed according to the six sets of DEMs with different resolution and different re-sampling technique (see Table 5.2). In order to assess the sensitivity of the different models to the model parameters, the Manning's n roughness coefficients for all the models were randomly distributed from 0.02 to 0.08 m -1/3 s for the river channel, and between 0.03 and 0.

Model Calibration and validation
The results of the calibration exercise in simulating 2006 flood event are shown in   In addition, this analysis showed that the accuracy of the models based on DEM Raster Type A is ascending as the resolution of DEMs decreased from 30 m to 120 m.
Results from the 120 m model (see Jhr A120, Table 5.3) are surprisingly good, indicating that a low resolution model is capable of making reasonable estimation of water levels.   The reduced inundation area seen in Raster Type A models is due to the fact that higher resolution data enable sudden changes in elevation to be recorded more accurately compared to lower resolution data. A higher resolution DEMs would therefore detect and describe small sinks and peaks more precisely whereas lower resolution DEMs may even be unable to capture these small sinks or peaks.

Flood simulation
Nevertheless, differences in inundation areas due to the changes in resolution are relatively small, unlike the differences from using raster from different re-sampling technique. In summary, this part of analysis suggests that although resolutions play some roles in the determining flood prediction results, it is not an influential factor, as inundation areas remained almost the same with similar sizes and shapes for DEMs at different resolution levels but from the same source (see model based Raster Type B, Figure 5.3d, 5.3e and 5.3f). By contrast, DEMs from different re-sampling technique produced vital discrepancies in simulation results. Therefore, how DEMs are re-sampled is critical in assessing flood impacts. Hydrology and Earth System Sciences. 19, 1-13.

Introduction
In hydraulic modelling of floods, one of the most fundamental input data is the geometric description of the floodplains and river channels often provided in the form of digital elevation models (DEM). During the past decades, there has been a significant change in data collection for topographic mapping technique, from conventional ground survey to remote sensing techniques (i.e. radar wave and laser altimetry; e.g. Mark and Bates, 2000;Castellarin et al., 2009). This shift has a number of advantages in terms of processing efficiency, cost effectiveness and accuracy (Bates 2012; Di Baldassarre and Uhlenbrook, 2012).
DEMs can be acquired from many sources of topographic information ranging from the high resolution and accurate, but costly, light detection and ranging (LiDAR) survey obtained from lower altitude to low-cost, and coarse resolution, space-borne data, such as advanced spaceborne thermal emission and reflection radiometer (ASTER) and shuttle radar topography mission (SRTM). DEMs can also be developed   Tarekegn et al. (2010) carried out on a study area in Ethiopia used a DEM which was generated from ASTER image. Integration between remote sensing and GIS technique were needed to construct the floodplain terrain and channel bathymetry.
From the results obtained, they concluded that the ASTER DEM is able to simulate the observed flooding pattern and inundated area extends with reasonable accuracy.
Nevertheless, they also highlighted the need of advanced GIS processing knowledge when developing a digital representation of the floodplain and channel terrain.

Digital Elevation Model
The required input data for the HEC RAS include the geometry of the floodplain and the river, which is provided by a number of cross sections. We identified several sources of DEM data for our study area (details are given below) with different spatial resolution and accuracy: i. DEMs derived from an original 1 m LiDAR dataset (obtained from DID).  were used to explore the impact of different topographic information on the hydraulic modelling of floods (see Table 6.1 and Figure 6.1).

Model calibration and validation
Then, data from two recent major flood events that occurred along the Johor River in To assess the sensitivity of the different models to the model parameters, the Manning's n roughness coefficients for all the models were sampled uniformly from 0.02 to 0.08 m -1/3 s for the river channel, and between 0.03 and 0.10 m -1/3 s for the where WSERef i denotes the WSE simulated by the reference model, WSEDEM i the WSE estimated by the models based on each test DEMs (see Table 6.1), and x corresponds to the total number of cross-sections where model results were compared.
For comparison of the inundation area, the F statistic (e.g. Horrit et al., 2007;  Where A refers to area that is both observed and predicted as inundated, B refers to area of inundation present in model but absent in observation, and C refers to area of inundation present in observation but absent in model. A value of 1 means a perfect match between observed and predicted areas of inundation, and a lower F indicates discrepancy between observed and predicted.

Uncertainty Estimation -GLUE analysis
In hydraulic modelling, multiple sources of uncertainty can emerge from several factors, such as model structure, topography, and friction coefficients (Aronica et al., 2002;Trigg et al., 2009;Brandimarte and Baldassarre, 2012;Dottori et al., 2013). A methodological approach to estimate the uncertainty is the generalised likelihood uncertainty estimation (GLUE) methodology (Beven and Binley, 1992), a variant of Monte Carlo simulation. Although some aspects of this methodology are criticized in several papers (e.g. Hunter et al., 2005;Mantovan and Todini, 2006;Montanari 2005;Stedinger et al., 2008), it is still widely used in hydrological modelling because of its easiness in implementation and a common-sense approach to use only a set of the "best" models for uncertainty analysis (e.g. Hunter et al., 2005;Shrestha et al., 2009;Vázquez et al., 2009;Krueger et al., 2010;Jung and Merwade, 2012;Brandimarte and Woldeyes, 2013).
According to the GLUE framework (Beven and Binley, 1992), each simulation, i, is associated to the (generalized) likelihood weight, W i , ranging from 0 to 1. The weight, W i is expressed as a function of the measure fit, Ԑ i , of the behavioural models.   The RMSE value of the other DEMs is 4.66 m for contour maps (Jhr T20), 7.01 m for ASTER (Jhr A30) and 6.47 m for SRTM (Jhr S90). It's also noticeably that the RMSE of the SRTM DEM for this particular study area is within the average height accuracy 1-D hydraulic modelling: the role of digital elevation models 89 found in other SRTM literature either global or at particular continent (see Table 6.3).

Quality of DEMs compared with the reference points
It is proven that this type of DEM gives an acceptable result when used in large scale flood modelling (e.g. Patro et al., 2009;Paiva et al., 2011;Yan et al., 2013). Despite having the lowest vertical accuracies, the ASTER and contour DEMs are still widely used in the field of hydraulic flood research as they are globally available and free (e.g. Tarekegn et al., 2010;Wang et al., 2011;Gichamo et al., 2012). The differences in the vertical accuracies may partly due to the lack of information in topographical flats areas such as floodplains.
In terms of ME values for all LiDAR DEMs models show the positive value indicating the overestimation of the ground elevation, while the other source of DEMs (contour maps, ASTER and SRTM) indicating the opposite value.
Although LiDAR DEM gives the lowest error, it is useful to note that this type of DEM has a number of limitations as highlighted in the several papers (see Sun et al., 2003;Casas et al., 2006;Schumann et al., 2008): ii. its availability is very much limited by economic constraint, iii. its inability to capture the river bed elevations due to the fact the laser does not penetrate the water surface, and iv. its incapability to penetrate the ground surface in densely vegetated areas especially for the tropical region. However, the further use of each DEM in this study is subject to its performance in the hydraulic flood modelling during the calibration and validation stages, which are described in the following sub-section.   Figure 6.3g are due to model instabilities) and was therefore eliminated from further analysis. The best-fit models, using the optimum Manningʹs n roughness coefficients (Table   6.4), were then used to simulate the January 2007 flood event for model validation.
This was carried out for all models except ASTER based model due to its poor performance (see Figure 6.3g). The results of this first analysis suggest that the reduction in the resolution of LiDAR DEMs (from 1 m to 90 m) does not significantly affect the model performance.
However, the use of topographic contour maps (Jhr T20) and SRTM (Jhr S90) DEMs as geometric input to the hydraulic model produces a slight increase of model errors.
For instance, Jhr L90 and Jhr S90 have the same resolution (90 m), but the different accuracy results into increased (tough not remarkably) errors in model validation (from 0.39 m to 0.60 m). This limited degradation of model performance (Table 6.4), in spite of the much lower accuracy of topographic input (Table 6.2) can be attributed to the fact that models are compared to water levels observed in two cross-sections.
A spatially distributed analysis (comparing the simulated flood extent and flood water profile along the river) might show more significant differences (see Section 6.4.3).

Quantifying the effect of the topographic data source on the water surface elevation and inundation area
Inundation area (sensitivity analysis) are substantially different (see Figure 6.4 and Table 6.5). Furthermore, the aforementioned measure of fit F was found to decrease for both decreasing resolution and lowering accuracy. This sensitivity analysis also shows that the results of flood inundation models are more affected by the accuracy of the DEM used as topographic input than its resolution. Table 6.5 shows the comparison between the different models in terms of simulating flood extent.
Water surface elevation  (Figure 6.4). Whereas, flood water profiles simulated by the models based on topographic contour maps and SRTM DEMs [see Figure 6.5(e) and Figure   6.5(f)] are rather different.  Table 6.6 shows the summary of MAE in terms of water surface elevation simulated by the models. To better interpret the differences that have emerged in comparing the results of models based on different topographic data, a set of numerical experiments were carried out to explore the uncertainty in model parameters. As mentioned, we varied the Manning's n roughness coefficient between 0.02 and 0.08 m -1/3 s, for the river channel, and from 0.03 to 0.10 m -1/3 s, for the floodplain, with steps 0.0025 m -1/3 s. Then, a number of simulations are eliminated as described in Section 6.3.4.

Conclusions
This chapter assessed how different DEMs (derived by various sources of topographic information or diverse resolutions) affect the output of hydraulic modelling. A reach of the Johor River, Malaysia, was used as the test site. The sources of DEMs were LiDAR at 1 m resolution, topographic contour maps at 20 m resolution, ASTER data at 30 m resolution, and SRTM data at 90 m resolution. The LiDAR DEM was also re-sampled from its original resolution dataset to 2, 20, 30 and 90 m cell size. Different models were built by using them as geometric input data.
The performance of the five LiDAR-based models (characterised by different resolutions ranging from 1 to 90 m; see Table 6.4) did not show significant differences. Neither in the exercise of independent calibration and validation based on water level observations in an internal cross section, nor in the sensitivity analysis of simulated flood profiles and inundation areas. Another striking result of our study is that the model based on ASTER data completely failed because of major inaccuracies of the DEM. In contrast, the models based on SRTM data and topographic contour maps did relatively well in the validation exercise as they provided a mean absolute error of 0.6 m, which is only slightly higher the ones obtained with LiDAR-based models (all around 0.4 m). However, this outcome could be attributed to the fact that validation could only be performed by using the water level observed in a two internal crosssections. As a matter of fact, higher discrepancies emerged when LiDAR-based models are compared to the models based on SRTM data or topographic contour maps in terms of inundation areas or flood water profiles. These differences were found to be relevant even when parameter uncertainty is accounted for.
The chapter also showed that, to support flood inundation models, the quality and accuracy of the DEM is more relevant than the resolution and precision of the DEM.
For instance, the model based on the 90 m DEM obtained by re-sampling the LiDAR data performed better than model based on the 90 m DEM obtained from SRTM data. These outcomes are unavoidably associated to the specific test site, but the methodology proposed here can allow a comprehensive assessment of the impact of diverse topographic data on hydraulic modelling of floods for different rivers around the world.
Chapter 7 Uncertainty in simulating design flood profiles and inundation maps on the Johor River, Malaysia

Introduction
In hydraulic modelling of floods, estimating the potential flood extent, i.e.
inundation map, and maximum water levels, i.e. flood profile, corresponding to a river discharge with a given return period, i.e. design flood, is an important step for assisting decision makers in flood risk management. A traditional representation of simulated results in flood modelling is based on a single simulation used as the best estimate. This approach, which can be called as "deterministic", does not explicitly account for the uncertainties in both the estimation of the design flood  and the hydraulic modelling process  and may lead to an inaccurate hazard assessment as highlighted in the recent literature (e.g. Beven, 2009;Domeneghetti et al., 2013;Dottori et al., 2013). For design flood profile, this uncertainty is sometimes accounted for by adding arbitrary freeboard heights (of e.g. 1 foot or 1 meter; Brandimarte and Di Baldassarre, 2012) to the simulated maximum water levels.
Several sources of uncertainty have been shown to significantly affect the estimation of design flood profiles and inundation maps, including high flow data, hydraulic model parameters (e.g. Manning's n roughness coefficients) and topography data.
High flow data are considered one of the most uncertain variables in hydraulic modelling of floods (Pappenberger et al., 2006). As an example, Di Baldassarre and Another important source of uncertainty, which gives a significant impact in hydraulic modelling, are the friction parameters, i.e. Manning's n roughness coefficients. Different Manning's n are often used to represent the channel and floodplain friction conditions and they are treated as calibration parameters (Horritt, 2005). Although the roughness coefficient is theoretically different from one point to another, most studies use lumped values of roughness for the channel or floodplain or even the entire cross-sections (e.g. Pappenberger et al., 2005). There are several techniques to assess uncertainty in hydraulic modelling of floods, such as Bayesian forecasting (Krzystofowich, 1999(Krzystofowich, , 2002, a methodology using a fuzzy extension principle (Maskey et al., 2004), parameter estimation by sequential testing (PEST) (Liu et al., 2005) and generalized likelihood uncertainty estimation (GLUE) (Blazkova and Beven, 2009;Yatheenradas et al., 2008). Among the several uncertainty techniques listed above, the GLUE methodology proposed by Beven and Binley (1992) is one of the most common methods to represent uncertainty in hydraulic and hydrological predictions. This technique has been widely used in various studies related to uncertainty, which include flood inundation mapping and design flood profile (Romanowich and Beven, 1998;Pappenberger et al., 2005;Brandimarte and Wolyes, 2013).
The objective of this paper is to explore and quantify the difference arising between deterministic approaches and uncertainty analysis in simulating design flood profile and flood inundation maps. This study focuses on the uncertainty in hydraulic model parameters (Manning's n roughness coefficients) and in the estimation of the design flood (1-in-100 year flood hydrograph), as both uncertainties have shown to be significant in the application of 1-D hydraulic modelling (Pappenberger et al., 2008;Brandimarte and Woldeyes, 2013). The topographic uncertainty was neglected in this research, which is supported with high quality topographic data of Digital Elevation Model (DEM), LiDAR (Light Detection and Ranging). Yet, it is important to note that the impact of topographic uncertainty in hydraulic modelling of floods has been recently discussed in Md Ali et al., (2015).

Estimation of design flood profile
The current approach of estimating safety level of the flood protection structure (e.g. dykes, levees) is by considering a freeboard height with the design flood profile.
Whereas, the normal approach to determine the design flood profile is by simulating the calibrated hydraulic model with the design flood hydrograph. In Malaysia, DID highlighted in the manual handbook that the minimum freeboard height to consider when designing the flood protection structure is 1 meter above the 1-in-100 year design flood profile. Thus, for the specific application example, the 'best fit' was run in unsteady conditions using the estimated Q 100 at upstream boundary condition (i.e. Rantau Panjang; see Figure 4.1).
However, several scientific literatures concern that the use of the 'best fit' model with single design flood estimation in prediction of design flood profiles might misrepresent the existence of any sources of uncertainties in hydraulic modelling simulation (i.e. Beven and Freer, 2001;Beven, 2006;Yan et al., 2013). Therefore, to avoid overlooking the existence of any sources of uncertainties which might affect the simulated design flood profile, the use of the probabilistic approach is recommended.

108
A number of quantitative analyses of uncertainty in hydraulic modelling have been previously reported, but these studies typically focused on only one source of uncertainty, such as the friction coefficient (Aronica et al., 2002;Bates et al., 2004), grid cell size (Werner, 2001), the quality of the DEM used (Casas et al., 2006) or flow characteristic (Purvis et al., 2008). This is mainly attributed to the fact that conducting a detail analysis that considers all source of uncertainty in the estimation of the design flood profile is not a simple task. The work would be computationally infeasible and it would require strong assumptions on the nature of errors. In addition, the data required to apply rigorous methods of uncertainty analysis are seldomly available (Beven, 2006).

Estimation of design flood profile with uncertainty
As mentioned above, Manning's n roughness coefficient was selected as the first variable of uncertainty in hydraulic modelling to be considered in the research. The model was run in a Monte Carlo framework to assess the parameter uncertainty using GLUE. For this scenario, given the accuracy of the used calibration data, all the couples of Manning's n roughness coefficient (river channel and floodplain) giving a MAE larger than 1-m are rejected. This choice of the rejection criteria for the MAE is against the requirement made by DID which determined that the minimum freeboard of any flood defence structure (i.e. dykes, levees) for main river should not be less than 1-m above 1-in-100 years flood event. In addition, the Manning's n roughness coefficients on the floodplain which are smaller than the channel are also eliminated. By adopting the GLUE methodology framework, all the models which passed these rejection criteria are considered as 'behavioural', and are used to simulate 1-in-100 years flood event.
Uncertainty in simulating design flood profiles and inundation maps on the Johor River, Malaysia

109
According to the GLUE framework (Beven and Binley, 1992), each simulation, i, is associated to the (generalized) likelihood weight, W i , ranging from 0 to 1. The weight, W i is expressed as a function of the measure fit, Ɛ i , of the behavioural models.
where, Ɛ max and Ɛ min are the maximum and minimum value of MAE of behavioural models. Then, the likelihood weights are the cumulative sum of 1 and the weighted 5 th , 50 th and 95 th percentiles. The likelihood weights were calculated as follow: Although some aspects of this methodology are criticized in several papers (e.g. Hunter et al., 2005;Mantovan and Todini, 2006;Montanari 2005;Stedinger et al., 2008), it is still widely used in hydrological modelling because of its easiness in implementation and a common-sense approach to use only a set of the "best fit" models for uncertainty analysis (e.g. Shrestha et al., 2009;Vázquez et al., 2009;Krueger et al., 2010).
The second variable of uncertainty in this research is the uncertainty in design flood induced by data error. Given that our research aims to evaluate two different approaches of design policies, the analysis was limited to hydraulic modelling only. .3 and 6.4 proposed by Kuczera (1996) and.
According to Kuczera (1996) the systematic data error can be described as Where Q indicates the observed value, Q' refers to the true value of river discharge, Q a represents the river discharge value overspill in times of high rainfall, and α is a positive value coefficient. Here, the Q a value is 300 m 3 s -1 and α = ± 50%.

Simulation of flood inundation maps
To generate the deterministic 1-in-100 year flood inundation map, a 1-in-100 year design flood hydrograph was used as an input (i.e. boundary condition) with the 'best fit' model. The 'best fit' model was selected from the calibration process (see Section 7.2.1). By integration in a GIS environment, the result of the 'best fit' model was then presented in term of 1-in100 year flood inundation map.

Simulation of flood inundation maps with uncertainty
In the studies of extent validation, GLUE methodology dictates that the actual flood event's water boundaries and its performance measurement must be presence in order to estimate the Monte Carlo ensemble of likeliness of the predicted outputs against the observed data (Horritt 2006). A particular measurement of likelihood used in the calculation of the extents of forecasted flood event is presented in Horrit et al. (2007) and Di  as: Wherein A refers to the size of area of both the actual and simulated data that is overlapped; B is the area that is supposedly dry but inundated in the calibration; and, C is the area in the model that supposed to be flooded, but not. A, B and C denotes as the numbers of pixels correctly predicted as wet and dry in reference and predicted inundation maps. A dry and wet pixels are represented by code, either as value 0 or 1. The closer the derivation value of F is to 1, the more likely the outcome of the model output is comparable to the actual flood extents. Whereas if the value closer to At this stage, all the behavioural models from the above were used to simulate the 1in-100 year flood event. Then, the likelihood weights, L i were rescaled to a cumulative sum of 1 and the weighted 5 th , 50 th and 95 th percentiles, according to the values of F. The likelihood weight, L i is expressed as a function of: Where (  It is found that the MAE value for the validation is 0.44 m. However, the use of freeboard definition in this approach may lead to wrong assumption should the design flood structure (i.e. levee) which is adopting 1-in-100

Calibration and validation
year design flood profile with 1-m freeboard be assumed as able to protect the flood prone area from extreme flood events of more than 1-in-100 year flood. This assumption have to be corrected by understanding that the freeboard does not provide an additional safety level to the flood prone area but rather to account for the overall uncertainty which may not have been considered during the hydraulic modelling.  Table 7.1 and Figure 7.3b, 7.3c and 7.3d. The statistics presented in the  Figure 7.3b, 7.3c and 7.3d are also computed using the data at each crosssection from all simulations for specific variable, or the combination of the two variables. Combined uncertainties produced the largest deviation in both minimum However, for this specific research, it was found that the design flood profile with 1m freeboard is higher than the 95th percentile uncertainty bound for each uncertainty variables either individual or combined. Hence, the standard freeboard overestimates the overall uncertainty.

Simulation of flood inundation map with deterministic approach
As shown in Figure 7.   As an additional, the comparison between deterministic and probabilistic flood inundation maps which may affect flood management are also being considered in this research.  Furthermore, there are no guidelines in defining or quantifying the uncertainties that are used for data collection, model simulation and creating flood inundation maps.
For example, is the result less than 5th uncertainty bound is acceptable? Thus, to promote the usage of probabilistic maps among the decision-makers, a modeller should provide the supporting information on the interpretation of distinction between the overall probabilities and those associated with respective uncertainty scenarios.

Conclusions
The The following conclusions are drawn from this specific case research: • The uncertainty from the Manning's n roughness coefficient has greater influence to flood inundation map than the uncertainty from inflow data.
This finding confirms the importance of this variable in the overall flood modelling process, but contrasts with previous studies (e.g. Brandimarte and Di Baldassarre, 2012;Domeneghetti et al., 2013), which found a larger impact of the uncertainty in inflow data.

Conclusion
The impacts of flooding are devastating in terms of human displacements, loss of life and damage in infrastructure and property. One of the efforts to minimize losses is providing early information to the community about risks through flood inundation maps (i.e. flood hazard map). These maps do not only identify future flood prone areas, but also provide useful information for rescue and relief agency, land planners and local authority.
Developing a flood inundation map involves a series of procedure which require hydrological studies (i.e. to estimate the return period of flood event) and hydraulic modelling (i.e. for estimation of water surface elevations and inundation area).
In normal practise, the result of flood inundation modelling is based on a deterministic approach (i.e. single simulation) without considering some uncertainty, regardless of the way the maps was created. However, without considering the uncertainties (e.g. model parameters, the terrain data and model structure), the result of the flood inundation modelling may not be accurate and this may lead to wrong information given on hazard assessments. hydraulic models for flood inundation mapping over one-dimensional (1-D) in simulating flow dynamics in floodplains (Horrit and Bates, 2002;Cobby et al., 2003;Horrit et al., 2006;Tayefi et al., 2007).
Nevertheless, there is still a doubt regarding the selection of the right modelling approach (either 1-D or 2-D) in prediction of accurate flood inundation extent when other factors such as the availability of high accuracy topography are limited.
Although with availability of detailed and accurate topographic data from LiDAR in the floodplain, the geometric description of the topography (such as cross-section spacing and resolution) on a hydraulic model can still gives a significant impact to the model output.
This thesis intends to contribute further in the knowledge of how uncertainty affects the flood inundation maps. The case study in this thesis is the Johor river reach in Johor, Malaysia, which is an agricultural-dominated area. It was chosen because of its comprehensive data available. Another unique aspect of this thesis is that the performance of the hydraulic models is evaluated using independent calibration and validation data of an observed flood water levels.
However, the intend of the thesis is not to certify any process, or to nullify the current state of process in creating flood hazard map, but rather to emphasize the impact of uncertainties in these maps. The next paragraphs will summarize the approach taken within this thesis and highlight the main conclusion of the research. Chapter 5: 2-D hydraulic modelling: The role of digital elevation models.

Research question
In hydraulic flood modelling, important input data is represented in the topographical dataset of floodplain and channel. With high resolution DEMs such as LiDAR becoming more readily available, more precise results can be produced such as in predicting flood inundation extent. However, the computational cost increases as the DEMs resolution becomes finer. Furthermore, computing hardware capabilities is still a limiting factor, more so for large research area. Therefore, there is a need for the DEMs to be aggregated to a coarser resolution. With the availability of GIS software, aggregation process become easier, but without understanding of the aggregation process of DEMs, there is a possibility of loss of determining data that could compromise the accuracy and reliability of the results.
To address this, the LISFLOOD-FP model integrated with ARC View to examine how sensitive the LISFLOOD-FP model was to the different aggregation technique and resolution while predicting the performance of the model and inundation area. Based on the algorithm available in GIS, two types of DEMs were developed from three different aggregation algorithm function (mean, maximum and minimum). The first DEM, the function of min was specified for channel, and floodplain as function of max, whereas for the second DEM, both channel and floodplain DEMs was specified as function of mean. Then, the selected aggregation algorithms are used to aggregate the LiDAR DEMs from 1 m to 30, 90 and 120 m in order to simulate the model performance and inundation area of the research area.
The model predicted water level was calibrated and validated against water level gauge data. Results indicate that the re-sampled DEM (30, 90, and 120 m) from different aggregation algorithm function give significant differences in calibration and validation analyses. In term or inundation area, the results from LISFLOOD-FP show that the different of aggregation algorithm function give significant differences between the models. The result from the analysis clearly showed that by selecting the proper aggregation function will provide much better accuracy of model results. This delineation may provide misleading information to the public or emergency relief agency that they are within the hazard area or not. By considering the uncertainty variables in flood hydraulic modelling, more information may provide and could be used to guide mitigation toward to higher risk area instead of all exposed area.

Research question
Results of design flood profile show that the uncertainty in the roughness coefficient, inflow and combine of both created different uncertainty bound. For roughness coefficient, the uncertainty bound is 1.06 m compare to inflow is 0.95 m. However, when combines the two parameter, the uncertainty bound is 1.42 m. In this research,

Recommendations
Information on flood inundation area is crucial in flood risk management. Not only important to rescue and relief agency during floods, but also to the planner when evaluate the propose land development in floodplain zone. Creating flood inundation maps involves hydrologic and hydraulic modelling; and topographical terrain analyses. However, this process is affected amongst other by input data, type of model used and river geometry in the model. o To further evaluate the effects of geometric data in 1-D hydraulic modelling, more variability of inclusion and exclusion of cross-section (e.g. at the river bend, width of cross-sections) should consider. Moreover, the characteristic of the river (e.g. length of the river, slope, and geomorphology) and varying upstream/downstream boundary conditions (e.g. releasing water from a dam, tide level, barrage) might give different results. In addition, research should be done in other river reach/catchment with different land use characteristics where the upstream of this research area is dominated by agriculture.
o Another area to explore is the uncertainty analyses techniques. In this thesis the GLUE methodology were adopted to perform the uncertainty analysis.
Although these approaches have been debated in certain scientific research, in contrast it still widely used due to it easiness in implementation and