The occurrence of natural catastrophes has significantly increased in recent times due to various factors such as environmental degradation, climatic change, rapid population growth, and intensified and improper land uses [1
]. Some of the impacts of natural disasters on human settlements include severe injuries and fatalities, damage to properties and infrastructure, economic losses as well as social disruptions [2
]. Among natural disasters, flood deserves special attention because it is one of the most frequent and costliest in terms of human and economic losses: it constitutes 31% of the total economic damages resulting from natural disasters globally [3
Asian countries, including Malaysia, are highly vulnerable to floods because of their geographic locations and other factors [4
]. About 90% of natural disaster related injuries in Malaysia are caused by flood with average yearly cost of around USD205 million [5
]. According to Malaysia’s Department of Irrigation and Drainage, around 29,000 km2
of land area and over 4.82 million people or 22% of the total population are affected by floods annually [6
], with residents of low-lying areas, especially close to riverbanks, being the most susceptible to flash floods [7
]. Floods are mainly triggered by the occurrence of intense monsoonal rainfall, which ranges between 2000 to 3000 mm annually. The resultant huge concentration of run-off exceeding the absorptive capability of the natural drainage system cause severe damages [6
]. Several houses are being flooded, traffic flow interrupted, and lives are lost in extreme cases [8
Due to Malaysia’s high susceptibility to floods, various stakeholders are continuously seeking permanent solutions to the country’s flood menace [9
]. Indeed, an effective and systematic way of managing periodic flood disasters requires mapping flood susceptible areas and understanding potential flood triggering factors. This study, therefore, integrates GIS capability with experts’ judgments to help in assessing and mapping current and future flood incidences in Malaysia, targeting Perlis area as a case study. The analytical network process (ANP), a type of multi-criteria tool for decision making, is used to assess and rank experts’ input on the significance of various flood causing factors and possible mitigation measures. There is a dearth of published research that combined GIS-based ANP model and RS to forecast flood susceptible zones in Perlis, Malaysia. Recent studies that adopted the ANP method have not incorporated remote sensing microwave images such as Radar Satellite (RADARSAT) into the analysis [10
]. Moreover, Perlis is in the northern part of Malaysia that is regarded as the “rice bowl” of the country [13
]. The frequent flood occurrences in the area might affect the food security of the entire country. This study, therefore, bridges this gap by first reviewing relevant literature on flood forecasting, followed by a description of the study area in Section 3
. Section 4
and Section 5
, respectively, explain the methodology employed in the study and presents the findings. Section 6
discusses the results and Section 7
concludes the paper with some key recommendations.
2. Literature Review
Flood is a natural and persistent overflow of huge quantity of water that occurs as a result of intense and prolonged rainfall and ends in exceeding the absorptive capability of the ground, and capacity of rivers, streams, and coastal areas [14
]. A review of literature on flood vulnerability analysis has shown that researchers are increasingly using multi-criteria decision-making tools (MCDT) to assess the risk of flooding [15
]. For example, GIS has been used to compute flood vulnerability index based on social and physical criteria in several urban areas [16
]. One of the most common approaches of flood forecasting is integrating GIS with the analytic hierarchy process (AHP) and its many variants [17
Also, different RS tools have also been used to evaluate the vulnerability of an area to flooding. For instance, SPOT (Systeme Pour l’Observation de la Terre [Satellite for observation of Earth]) imageries were utilized along with a digital elevation model (DEM) in the delineation of the monsoon flood in Bangladesh [18
]. DEM has been used to analyze the vulnerability of a coastal industrial city to flooding due to predicted sea level rise [19
] and multi-sensor RS data has been utilized to analyze the causes of recurrent flash flood in Jeddah city, Saudi Arabia [20
]. In another study, Landsat Thematic Mapper band 7 with the Near Infrared band 4 have been used to delineate flooded areas [21
]. Similarly, advanced very-high-resolution radiometer (AVHRR) data has been employed for regional flood studies [22
], as well as amplitude change detection technique using multi-pass synthetic aperture radar (SAR) data for identifying flooded areas [23
]. Other studies have analyzed the depth of flood using AVHRR imageries from National Oceanic and Atmospheric Administration (NOAA) based on the tonal variation (the relative brightness or color) of the flooded water in the images [24
]. Other techniques such as artificial neural network (ANN), decision tree (DT), and adaptive neuro-fuzzy interface systems (ANFIS) model among others can deal with the issues of uncertainties in flood assessment but are sometimes complex and difficult to handle [26
In recent times, due to the need to handle the complex data required for flood vulnerability analysis in a more efficient way and account for likely uncertainties, various advanced machine learning algorithms and statistical models such as multivariate regression models have been explored [27
]. Other studies have used support vector machine models [28
], ANN model [29
], ANFIS [30
], a combination of fuzzy logic and support vector machine [31
], integrated machine learning and statistical models [32
], DT algorithm [33
], logistic regression [34
], and an ensemble of regression trees and support vector machine [35
]. Despite the recent focus on quantitative methods of flood vulnerability analysis, qualitative approaches such as the AHP and ANP are still being adopted [36
] due to their relative simplicity in computing flood influencing parameters. The weights of each of the parameter considered in an AHP model are determined by expert opinions unlike the quantitative methods that require rigorous training and computations [28
]. Though the AHP model has the drawbacks of expert bias and unsuitability for global application [28
], it could still be effectively applied on a regional or local scale. In a recent study of Riyadh, the AHP model was adopted in identifying flood vulnerability zones [38
Analytic Hierarch Process (AHP) and Analytic Network Process (ANP)
AHP is a mathematical-based model developed by Thomas Saaty [39
] in the early 1970s to simplify and improve the process of decision-making. It comprises six major steps: problem definition; decomposing the problem into a hierarchy; carrying out pair-wise comparison; calculating the maximum eigenvalues and eigenvectors; calculating the consistency ratio to check the consistency of the matrix; and aggregating the relative important weights of a decision to arrive at the final weighting for decision alternatives [40
]. Indeed, AHP has been recognized among researchers, practitioners, and decision-makers for its application in assessing various geo-hazards problems including flood and landslide susceptibility [41
], slope failure [42
], groundwater vulnerability [43
], and urban seismic vulnerability [44
]. However, some inconsistencies were pinpointed especially during the early and middle period of its discovery. For instance, as a top-down decision or hierarchy model, criteria and alternatives are assumed to be independent from each other, which rarely holds true in real-life situations [45
]. Moreover, bias could occur when the criteria and sub-criteria correlate with each other.
Therefore, to improve decision-making processes and overcome some of the shortcomings of AHP, particularly the assumed independence of decision factors, the ANP model was developed as an improved generalization of the AHP method to deal with the problem of inner-dependences between elements in a network. Similar to the AHP, ANP uses experts’ judgments as inputs for evaluating decision factors in multi-criteria decision-making problems such as flood susceptibility mapping. The ANP organizes elements into networks of clusters and nodes. Clustering is a means of classifying similar elements or factors within the same group. Elements contained in clusters are referred to as nodes and the ANP model supports comparison between nodes in the same cluster (inner-dependence) and with nodes in other clusters (outer-dependence). Nodes in typical Multi-Criteria Decision-Making (MCDM) applications have unequal degrees of influence and estimating the degree of influence or significance of each factor is essential for producing valid susceptibility maps. Therefore, it is necessary to perform intra and inter comparison of nodes and clusters in order to prioritize the various conditioning factors as a means of establishing their respective levels of significance.
Thus, the ANP allows the interdependences between decision factors to be considered, allowing network model structuring, which is good for reflecting interdependences between real-life factors such as those that influence flood occurrence. A key innovation of the ANP is its network structure, which enables interactions between elements of different clusters and dependences between the elements in the same cluster to be considered. ANP’s strength lies in its ability to handle the complexities of everyday decision-making problems, which many other decision systems might not handle effectively [40
]. ANP is a relatively simple, intuitive approach that can measure all tangible, intangible, individual, and shared values or criteria in the flood susceptibility model. Its ease of use and facilitation of better communication among multiple diverse stakeholders makes it appealing to a broad range of users, particularly for addressing complex real-life problems such as flooding [46
]. ANP’s structured analysis facilitates continuous conversations leading to the gathering of relevant information from every participant in the group thereby enabling the achievement of consensus. Such a robust model that enables input from multiple diverse stakeholders and aggregation of different flood conditioning factors, including indirect dependencies between the factors, can aid decision results that are likely to be more consistent with real-world decision situations [47
]. Hence, these reasons influence the choice of ANP for this study. Figure 1
illustrates the structural difference between the AHP and ANP.
This study illustrates the robustness of coupling ANP as a decision-making tool with geospatial technologies in flood disaster susceptibility mapping, based on experts’ judgments. By using the ANP model, two shortcomings of the AHP technique were avoided namely its assumption of decision elements to be independent, which rarely occur in a real-life situation and the issue of rank reversal. Thus, ANP enabled the modeling of the interdependence of the factors responsible for flood disaster occurrence.
The findings of the study indicate that the flood vulnerable areas in Perlis are situated in the southern part of the state. According to our findings, alluvium deposit, paddy fields, flat slope, and low humidity clay soil are the most important factors influencing flood generation in the southern part of the study area with susceptibility levels of 34.1% (27,773.7 ha), 49% (39,863 ha), 83.0% (68,934.6 ha), and 40.7% (33,220.4 ha) respectively. This corresponds with the results of Adiat et al. [50
] that conducted a similar study in Kedah State bordering the southern part of Perlis. To produce their flood susceptibility map, the authors utilized similar factors adopted in the present study such as soil type, geology, rainfall, and slope. The presence of these factors in various proportions within a location trigger flood. For instance, similar geology types were confirmed to be present in Perlis in the study of Arafin and Lee [67
], which prevent the percolation of storm water into the ground resulting in water overflow. Despite using different MCDM approaches for different study areas with similar flood conditioning factors, these studies present similar outcomes. The prioritization of the conditioning factors by the respective experts also highlighted the significance of land use, geology, rainfall, and slope in influencing a region’s flood susceptibility. According to their findings, the highly susceptible flood areas in Kedahare located at the border with the southern part of Perlis, corresponding to this present study’s outcome. Similarly, Azizat and Omar [68
] considered factors such as geology, soil type, rainfall, and slope (DEM) in mapping Perlis’ flood hazard areas. Frequency ratio, statistical index, and Poisson method were used in that study and the flood susceptibility results correspond with the result of this present study. The similarity in results is likely due to the consistency in experts’ judgements regarding flood triggering factors at all levels for the separate studies. Outcomes of integrated MCDM-GIS models are largely influenced by the prioritization of the flood triggering factors by experts. This underscores the significance of carefully selecting experts with sound knowledge of the terrain for reliable hazard susceptibility mapping. In this instance, the outcome of the experts-model based map corresponds with the susceptibility levels of the study areas from field observation and Azizat and Omar’s [68
] integrated methods, thus confirming the reliability of the integrated ANP model adopted in this study.
Thereafter, sensitivity analysis, which is a collection of methods used for evaluating the degree of sensitivity of the spatial multi-criteria model output to slight changes in the input values was performed [69
]. It is recommended that a sensitivity analysis should be carried out after every multi-criteria analysis to monitor/verify the stability of results because of possible uncertainties in the experts’ judgments [70
]. There are two common methods of sensitivity analysis namely: (i) modification of weights obtained from the experts, and (ii) the assumption of equal weights for all factors. The main purpose of the sensitivity analysis is to examine how sensitive the choices are in relation to the changes in the criteria weights. This is quite imperative in the situation where uncertainties exist in the definition of the importance of different factor(s). The answers to the questions of ‘what if analysis’ can be addressed through a sensitivity analysis. Hence, the sensitivity analysis is recommended for all kinds of multi-criteria analysis. However, because of the sophisticated network structure of the ANP model, this kind of analysis is hardly conducted [71
]. In this study therefore, to test the robustness of spatial flood susceptibility model, one-at-a-time (OAT) sensitivity analysis was used by modifying the weights of each criterion. The OAT is the most common and simplest technique used in changing the weight of one criterion at a time in order to understand the effect of the change on the outputs. Thus, the weights were modified by running 22 scenarios of the flood susceptible zones. Different sets of values of the sub-criteria were systematically altered to explore their impacts on the overall weights, which were subsequently input into the GIS software to generate different flood susceptibility maps.
Based on the sensitivity analysis results, only six out of the total scenarios correspond with the developed spatial flood susceptibility map: alluvium, paddy fields, major rainfall, moderate rainfall, flat slope, and low humidity clay soil as demonstrated in the Scenario 1, Scenario 8, Scenario 12, Scenario 13, Scenario 15, and Scenario 21 respectively (Table 6
). These scenarios were chosen based on their conformity with the original spatial flood forecasting model developed herein as shown in Appendix A Figure A1
, Figure A2
, Figure A3
, Figure A4
, Figure A5
and Figure A6
. This is because they showed almost the same trend with the developed model. In summary, the modified scenarios verified the strength of the study’s spatial flood susceptibility model. The remaining 16 scenarios that show negative results are not presented here.
Similarly, reliability and validity analyses of the spatial flood forecasting model were carried out using two forms of validations namely: (i) superimposing the flood extent model over the spatial flood susceptibility model using GIS to determine matches between the flood extent information and the flood susceptibility model, (ii) comparing the flood susceptibility model with the spatial information of the study area such as the slope, land use, outcrops, and soil type data layers. In validating the spatial flood susceptibility model using real-life flood incident in the study area, radar satellite images of the November 2010 flooding were used. Figure 7
illustrates the side by side views of the spatial flood susceptibility model and the flood extent model developed based on the real-life flood incident of the study area. From the comparison of these images in Figure 7
A,B, it is observed that the flooded areas were found to be scattered over the areas forecasted to be very highly and highly susceptible to flooding. Similarly, Figure 7
A was superimposed over Figure 7
B to observe the relationship between the two as shown in Figure 8
. Table 7
presents the susceptibility levels of the spatial flood susceptibility model as well as the magnitude of the flood extent model.
An analysis of the superimposed image indicates that 38.4% (30,924.6 ha) of the flooded areas are dispersed over the areas identified as very highly susceptible to flooding and 19.0% (15,341 ha) in highly susceptible to flooding (Figure 8
). Therefore, this shows a significant match between the spatial ANP flood susceptibility model with the real-life flood extent model of the study area.
Moreover, the spatial information obtained were also used as the second form of validation in this study. Alluvium deposit, paddy fields, flat slope, and low humidity clay soil data layers were as well used in validating the spatial flood susceptibility model in this study. This was done by superimposing the flood extent model with the spatial information of the study area. Figure 9
shows the major flooded areas were found to be in the major paddy fields areas, represented in dark-grey color, which formed 49% (39,863 ha) of the total study area (Table 8
Therefore, this study’s findings indicate that the area is characterized by paddy fields overburdened by clay, which is impervious; hence, the percentage of permeation of rainfall in the study area will be very low thus increasing its susceptibility to flooding. Similarly, regarding the geology, flat slope, and low humidity clay soil of the area, which cannot be changed, it implies that the area might have to live with flood incidences. This further explains the reason why the flood susceptibility of each accounts for 34.1% (alluvium), 49% (paddy fields), 83.0% (flat slope), and 40.7% (low humidity clay soil) respectively.