Flash Flood Susceptibility Mapping at Andungbiru Watershed, East Java Using AHP-Information Weighted Method

Flash floods are among the most frequent natural disasters caused by heavy rain associated with a severe thunderstorm, which leads to social and economic losses in infrastructure and agriculture. Therefore, this research aims to map flash flood potential susceptibility (FFPS) in the Pekalen watershed, using Geographic Information System (GIS) technology and statistical analysis to reduce the risk of flooding. The opinion and experience of an expert on the weight assessment method were carried out using the Analytical Hierarchy Process (AHP). Furthermore, the probability statistical methods and GIS were used in flash flood areas in the Pekalen watershed in Andungbiru, Probolinggo village. This study was carried out using geomorphological factors, namely elevation, slope, stream power index, and topographic wetness index, with a resolution of 30 m. Thematic map scale of the land use, river density, distance to the river, rainfall, and geology is in the ratio of is in a ratio of 1:25,000. Imagery processing was carried out using Landsat 8 30 m x 30 m resolution imagery, such as the Normalized Difference Vegetation Index. The result showed that the model map of FFPS obtained low 8%, low 23%, moderate 27%, moderate to high 26%, high 13%, and very high 2% index values. The next stage of modeling analysis led to validation using statistic receiver operating Characteristic Curve (ROC) of area Under Curve (AUC) with a value of 90.15. In conclusion, the factors that significantly trigger flash floods are distance to the river, land use, and slope.


Introduction
Flash floods are one of the natural disasters affecting human life, leading to death and natural and artificial environmental damage (Guzzetti et al., 2005;Penning-Rowsell et al., 2005;Salvati et al., 2010). According to Du et al. (2013), the process of measuring runoff areas and damage caused by flash floods is challenging. This is because flash flood disasters generally have varying intervals and durations (Tehrany et al., 2015).
Several research methods use AHP as a disaster modeling map extension. AHP is also used to calculate weight assessment based on the opinion and experience of experts (Bui et al., 2019;Elkhrachy, 2015;He et al., 2019). The research on potential mapping disasters using opinions and combination statistics produced good validation values, namely AHP-Weighted Information content AUC 0.893, AHP method 0.821, and information Content 0.842 (He et al., 2019). This study used the AHP-weighted information content method, which combines the expert experience method and frequency distribution statistics. The model results are validated using statistics, while the area below the curve (AUC) is used to calculate the accuracy of the flash flood model. Mapping potential flash flood-prone areas using the AHP-Weighted Information Content method is essential for sustainable planning (Costache et al., 2020), protecting life, and maintaining a river condition environment (Zhang et al., 2015).
AHP is a method used for various kinds of disasters, such as landslides, flash floods, etc. This is an interesting research due to its complex nature, making it possible to carry out a spatial approach during the flash flood modeling process (He et al., 2019). There has never been a mapping of the potential for flash floods in the Pekalen watershed, therefore, it is imperative to use the AHP method combined with information content weights to determine the factors that influence the accuracy of the modeling results.
Vulnerability is needed to determine the appropriate methods and analysis needed to map a potential flash flood (Cloke & Pappenberger, 2009). Flash flood modeling utilizes GIS mapping and remote sensing methods (Pradhan & Shafie, 2009). The purpose of this study is to apply the AHP-weighted Information Content method at in the Pekalen watershed using a geographic information system (GIS). This method was used to determine the distribution of potential flash flood events based on factors that have been calculated from their various weights using expert knowledge and mathematical analysis.

Study Area
This research was carried out in Pekalen Watershed, Probolinggo, East Java, with coordinates 1130.23 E-1130.35 E and 70.55'0 -80.2'3. Figure 1 shows the flash flood event area is in the village, with the area of the Upper Pekalen watershed covering 8,530 hectares.
The flash flood event occurred at an elevation of 608 meters above sea level.

Research Design
The predetermined parameters used to determine flash floods are Elevation, Slope, SPI, TWI, Land use, River Density, Distance to River, NDVI, short-term heavy rain, and geology.
These parameters are essential because each region has different regional morphological characteristics (Elkhrachy, 2015). Data were collected from the inventory of flash flood events obtained from the survey at the site. The results consist of 70% training data (63 ha) and 30% validation (27 ha) (Kia et al., 2012).
The selection of factors mapping the potential of flash floods is essential in creating an insecurity map to determine the Elevation, Slope, TWI (Costache et al., 2020), SPI, River Density, rainfall, NDVI (Bui et al., 2019), Land use (Costache et al., 2020), Geology (Cao et al., 2016), and distance from the river (Popa et al., 2019). Table 1 is a general description of data source and resolution information. Furthermore, this study comprises geomorphological and hydrological factors, with   SPI is a topographic prediction used to model the strength of a water flow. The topographic wetness index (TWI) is a spatial concept that models the accumulation of flow from each basin and the ability of water to move due to its slope and gravity. SPI and TWI modeling is derived from DEM and processed using tools in SAGA GIS, along with similarities from TWI (Eq.1) and SPI (Eq. 2) modeling (Moore et al., 1991). where the As and β denote specific water catchment areas (m 2 /m) and slope angles in degrees. TWI shows the amount of flow accumulated at any point in the catchment area and the tendency of water to descend to the slope using gravitational force (Moore et al., 1991).
d. An NDVI (ratio) is a measure that describes the vegetation characteristics of an area, factors that affect the surface flow, and the infiltration capabilities of a site (Figure 8).
According to Tehrany et al. (2013), areas with less dense vegetation are considered more prone to flooding.
e. Distance to the river (meter) is a factor that is lower from the thematic map layer of the river (Figure 7). This is a factor that shows Rainfall (mm) is also considered an essential variable in the assessment of flash flood vulnerability ( Figure 9). Therefore, precipitation is used to produce a map of rainfall distribution using the Inverse Distance Weighted (IDW) method (Bartier & Keller, 1996). = 1 1 + 2 2 + 3 3 + 4 4 + 5 5 + 6 6 + 7 7 + 8 8 + 9 9 + 10 10 = 1 1 + 2 2 + 3 3 + 4 4 + 5 5 + 6 6 + 7 7 + 8 8 + This study uses the statistical analysis method to quantitatively calculate the relationship between flash floods and causal factors based on a spatial modeling approach. It also focuses on the probability of frequency distribution with the approach of geodynamic environmental data. The method combines probability data with the opinions of experts that need to be verified.

Results and Discussion
Calculating AHP results using weighted Information content determined the weight value of each class factor. AHP is a method that uses expert opinion, such as individuals that are professionals in the field of disaster and GIS, to determine the input from the weight of each factor. The opinions of these experts include the National Disaster Relief Agency, National Disaster Management Authority, hydrologists, GIS experts, and the Development Planning Agency at Sub-National Level. Figure 13 is the result of an AHP weighted information content calculation.

Figure 13. AHP-Weighted information content calculation results
In Figure 13, the weight calculation shows that the factor with the most significant negative value is the distance to river factor, which is in the second class with a weight value of -5.42. This is followed by the land-use and marble factors with negative values -3.        The calculation obtained a map of the spread of flash flood insecurity with each index class, as shown in Figure 24. The study area in Table 3 shows that the highest percentage value is moderate 27%, followed by moderate to high 26%, Low 23%, High 13%, Very Low 8%, and very high with only 2%. The spatial distribution shows that the research area is fairly vulnerable because moderate to high values dominate 27% and 26%, respectively.
The results of the flash flood encoding using the AHP method can carry out the validation process using the graph below the AUC curve, as shown in Figure 25. The statistic method for the validation process used the analysis method under the curve (AUC) to obtain an event model value of 90.15%, as shown in Figure 25. This value indicates that the modeling made based on the probability of frequency distribution has proximity between flash flood events models.
In addition, the increase in performance (AUC value) is supported by a selection factor that is adjusted to the type of flash flood events from similar studies, which have an AUC value above 80%. The resulting flood hazard map is 30 m x 30 m resolution at a scale of 1:25,000. This map can be used as a reference in making flood risk models and regional spatial alignment, therefore, they are able to offer risk.

Conclusion
In conclusion, the flash flood potential vulnerability map using the AHP-Weighted Information Content method gives a significant result. Furthermore, adjusting the resulting map of each layer produced a resolution of 30 m x 30 m on a scale of 1:25,000. Of the ten factors analyzed using AHP-Weighted Information Content, those that significantly trigger flash floods are distance to the river, land use, and slope with a weight value of four. This shows the area with a high probability of occurring in the rice field features. Meanwhile, factors that have weak influence are TWI, SPI, and land-use factors. In this study, land use has a high value because it is a factor that affects runoff on the surface. The flash flood susceptibility potential map can be used as a reference in flood mitigation, with a resolution scale of 30 m x 30 m. Furthermore, research on modeling, it required information on the previous occurrence to determine the temporal phenomena that occur. The group number of each factor needs to be varied to test the dominance of the triggering factors for a flash flood.