Flood Susceptibility Assessment (FSA) Using GIS-Based Frequency Ratio (FR) Model in Kota Belud, Sabah, Malaysia

Floods are the most common and destructive natural disasters that harm human health and natural environments [1, 2]. Flood hazard was becoming one of the dangerous natural hazards in Kota Belud, Sabah, Malaysia, resulting in extensive damage to properties infrastructures and impacting the local social economy. In 2017, this disaster affected approximately 4,441 people in 62 localities. The increase in the flood frequency and magnitude is related to heavy rains caused by monsoons or typhoons from neighboring countries, but it is also associated with the effects of the Ranau 2015 earthquake, with a magnitude of 6.0 on the Richter's scale. This situation causes the riverbed to become shallower due to it being filled with sediment deposits resulting from debris and woodpiles [3-11].

FSA mapping is an essential component of early warning systems or strategies for preventing and mitigating future flood situations, which aids in reducing the adverse effects of flood hazards. FSA mapping can be considered one of the risk assessment methods [12-14]. Risk is the product of susceptibility, hazard, and elements at risk, and it relates to the potential loss or damage produced by an event within that area. The susceptibility indicates the probability of a specific type occurring in each location, whereas the hazard refers to the likelihood of an event of a specific type and magnitude occurring in each location within a reference period [12]. This means that susceptibility can be used to predict the spatial occurrence of an event, whereas hazard can be used to predict the Spatio-temporal occurrence of an event. Even though the terms susceptibility and hazard have different meanings, they are always used interchangeably in geohazard research [15-19].

This study will use a statistical-bivariate Frequency Ratio model to produce a FSA map. This model was selected because the statistical-bivariate method has advantages over the statistical-multivariate method in terms of the techniques used for correlation analysis between the two variables [20-22]. This method combines the views of experts or the results of previous researchers in the selection of factors causing floods with the correlation of statistical techniques [16-19]. This method can study and evaluate flood hazards separately for each parameter used to map flood hazards. The integration of the statistical approach and the geographic information system aids in determining the value of the produced weights [1, 23, 24]. Hence, the primary goal of this research is to determine the Flood Susceptibility Level (FSL) using a GIS-based Frequency Ratio (FR) model.

It is critical to examine the occurrence of previous flood events in order to forecast future floods in the study area. The study utilized a geospatial-based methodology for FSA mapping. The GIS-based FR method has been used to analyse the flood-prone area. The methods used in this study aim to conduct the FSA consisting of three stages; (stage 1) flood inventory – which is the process to identify and characterize the flood event from a field survey along with handheld GPS, (stage 2) flood causative conditioning factors – their area eight parameters were used as the causative factors to flooding (elevation, slope curvature, slope angle, topography wetness index, drainage density, drainage proximity, land use, and soil type) and (stage 3) Frequency Ratio (FR) model - the FSA was defined using a Geographical Information System (GIS) based-bivariate statistical analysis, and then validate the model using the area Under the Curve (AUC).

3.1 Flood Hazard Inventory (FHI)

The delineating flooded area is the most critical component whenever dealing with flood susceptibility mapping. Initial identification of flooded area locations from the field survey and published reports from government agencies may be used to obtain significantly. The field observations were carried out during and after the 2017 flood events to collect accurate data for FSA mapping (Figure 1).

A total of 100 flood locations were identified and located on the map. The training dataset for the model was the total number of flood locations from the 2017 field survey, and previous flood data from 2016 (54 flood locations) was used to validate the FSA model (Figure 2).

1.jpg

Figure 1. Field investigation during and after the flood events of 2017

2.png

Figure 2. Flood inventory map in the study area

3.2 Flood causative factors

The study utilized a geospatial-based methodology for flood susceptibility mapping. Selection of the valuable parameter for the flood hazard map in any area is essential, and it is challenging to choose factors commonly used in flood susceptibility mapping [26]. ArcGIS software prepared the eight physical factors contributing to flood occurrences, such as elevation, slope curvature, slope angle, topography wetness index, drainage density, drainage proximity, land use, and soil type. All of the causative factors in the FR model were adopted after doing a literature review and conducting a reconnaissance survey to understand the flood conditions that frequently cause floods in the study area. (Figure 3). The more reliable the conditioning parameters, the more accurate the flood susceptibility zonation will be.

Digital Elevation Model (DEM) is the primary source to build the model because it contains information about topographic attributes that can help flood analysis [27]. Interferometric Synthetic-Aperture Radar (IFSAR) data with the 5m spatial resolution are used and processed for the DEM. Timbalai RSO Borneo (meters) was used to rectify the dataset. This study used DEM to extract elevation, slope angle, slope curvature, and topographic wetness index.

3.png

Figure 3. Flood causative factors used in FR model

In the analysis of assessing and mapping flood susceptibility, one of the crucial parameters in flood control is topographic elevation [28]. The elevation was classified to 5 classes; <5m, 6-10m, 11-20m, 21-30m and >30m. The Topographic Wetness Index (TWI) shows the total accumulation of water flow at any point in the drainage basin and the ability of water to flow down by gravity [26]. TWI has a significant effect on flood mapping [29]. The TWI has no unit and was classified into five classes: <5, 5-6, 6-8, 8-10, >10 using the natural break jenk method.

Slope curvature is classified into three types: convex, concave, and horizontal. Slope curvature also influences infiltration and the rate of surface water runoff. Positive curvature represents that the slope gradient is convex in an upward direction, whereas negative curvature indicates that the slope gradient is concave upwardly, and the slope gradient will be flat if the value is zero [30].

In addition to topographic altitude, one of the essential topographic factors in hydrological studies is slope angle, which is essential in controlling the flow of water at the water's surface. The slope angle controls the surface runoff [29]. Low gradient slopes are more prone to flooding than high gradient slopes [31]. The slope angle was classified into five classes based on the slope classification published by the Department of Mineral and Geoscience Malaysia (JMG, 2006); flat (<5°), gentle (5°-15°), moderate (15°-25°), steep (25°-35°) and very steep (>35°).

Land use directly or indirectly affects the rate of infiltration, evapotranspiration, and generation of surface water runoff [29]. The land use map was prepared from the topography map and was classified into six classes: barren land, commercial, cultivation area, forest, public infrastructure, and residential area. Infiltration is affected by the spatial variability of soil type and regulates overland flows and flooding [29]. Therefore, soil type was prepared based on soil data derived from the Agriculture Department of Sabah (JPNS, 1976).

In general, flood events usually occur around drainage systems. Therefore, taking into account the buffer zone (distance) to drainage is an important parameter to ensure the accuracy of flood hazard maps (Sumit Das, 2018). Furthermore, the lowest point in an area is often closely related to the presence of a drainage system [29]. The higher the drainage density, the larger the catchment area susceptible to erosion, resulting in sedimentation at the lower ground [31]. The drainage density was calculated by dividing the total length of the drainage channel in the watershed by the watershed's total area. The topography map of the study area could be used to derive data, drainage density, and a proximity map.

3.3 Frequency Ratio (FR)

The FR model can be used to quickly assess geospatially the probabilistic relationship of databases with various levels of classification between dependent or independent variables [32]. For the calculation of the FR, the ratio of the range of flood occurrence zone and the flood-unaffected area is determined for each class or type of factor, and the ratio of this area is calculated in each factor category to the overall range. The description of this definition is explained simply using the following equation as Eq. (1):

$F R=\frac{\text { Percentage flood area }(\%)}{\text { Percentage pixels number in domain }(\%)}$                   (1)

The FR model is used to predict the association between previous flood events and the variables (or parameters) that influence those events [33]. If the result of the FR value obtained is below the value of 1 (one), then this value indicates that the correlation between flood events and the variable is weak. If the opposite happens, the FR value exceeds the value of 1 (one), then this value indicates the strong correlation between the variable and the occurrence of flooding. These calculations were made using the training data set as well as the validation data set to obtain the accuracy of the model generated through this method. Both of these data sets will go through the same steps in generating this model. To construct a flood susceptibility map using this method, the values obtained through the calculation of the FR for each factor (variable) will be summed in a "raster calculator" using ArcGIS software to obtain the final result, as shown in the equation Eq. (2). The produced map was classified into 5 zones ranging from very low to very high susceptibility zones using natural break method.

F S M=$\Sigma$ frequency ratio in each factor type              (2)

By comparing it to the flood inventory map, the FSM produced by the FR model for the study area was validated. In this study, the success rate curve was calculated to assess the efficacy of the FR model and factors used to predict floods. To generate the success rate curve, the relative ranks for each prediction pattern were calculated and sorted in descending order by calculating the index values of all cells in the study area. To calculate the percentage of floods in each susceptible class, the 100 classes were overlaid and intersected with the training dataset used to generate the model. The success rate curve was then constructed by plotting the susceptible classes from highest to lowest values on the X-axis and the cumulative percentage of flood occurrences on the Y-axis.

By calculating the prediction rate, the susceptibility map was also evaluated in terms of predictive power and validity. To create the prediction rate curve, the flood validation dataset was used instead of the flood training dataset, which was created using the same data integration and representation processes described above. If the slope of the curve graph is steeper, it indicates that there are more floods in the highly susceptibility category. The AUC range is varied, from 0.6 to 1.0. The highest accuracy has a value of 1.0, thus indicating that the model's performance is very satisfactory in predicting the occurrence of disasters without any tendency. Therefore, AUC values closer to 1.0 indicate that the model has high accuracy and reliability.

This study shows that combining statistical bivariate FR and GIS technology provides a valuable decision-making tool for FSA mapping by allowing for consistent and effective use of spatial information. The training dataset consisted of 100 flood locations inundated in 2017, while the validation dataset consisted of 54 flood locations from the 2016 flood report. Eight conditioning parameters were used as independent variables (elevation, slope curvature, slope angle, topography wetness index, drainage density, drainage proximity, land use, and soil type).

Lastly, a FSA map was generated and classified into five classes (low, low, moderate, high, and very high). All layers used in this study were resampled into 5x5 meter spatial resolution to standardize the final map. The FSA map shows that the very high flood susceptibility area was located in the main river channel and around the district center. The AUC curve showed that this model is trustable, with a success and prediction rate of 0.89 and 0.82, respectively. The small number of flood locations used for modeling training and validation was one of the study's limitations. However, depending on the data availability, the method used in this study can be easily applied in other fields where different factors can be considered. As a result, this FR model has been successfully tested and is appropriate for selecting land-use suitability, controlling, and managing flood hazard/risk in the study area.