Flood Modelling Using Integration of Multi-data Analysis and HEC-RAS Model in Mata Allo River, Sulawesi

. The amount of rainfall in a watershed with steep slopes, small cross-sectional areas, and less water catchment areas. This will cause an increase in water discharge in the river which can cause flooding. These characteristics can be found in Mata Allo River, Enrekang Regency. To identify the most flood-hit areas, the simulating model can be done utilizing the HEC-RAS program. Use of Satellite Imagery Data such as Sentinel-2 for extracting land use data information, and Sentinel-1 for data extraction of actual water bodies/rivers. The analysis is carried out by integrating the interpretation results from multi-sensor images with the results of modeling the flood inundation area using HEC-RAS. Based on the analysis results, the land use classification accuracy is 82.9% for Sentinel-2 data using the random forest algorithm. While for the actual extraction of water bodies using Sentinel-1 imagery was 89.6%. Approaching the threshold value between water and non-water bodies is taken using -13.39. The inundation area in the study area reached 87.66ha at the largest discharge model. The most affected land use after integrating each data is built-up land, most of which are settlements covering an area of 47.26ha.


Introduction
The territory of Indonesia is located in a wet tropical climate zone with quite high rainfall [1]. If detailed, about 80 percent of disasters in Indonesia are classified as hydrometeorological disasters such as floods, landslides, and hurricanes. Floods require serious attention from various parties because they contribute 37 percent of 143 of all disasters that occur nationally [2]. Floods occur when the volume of water flowing in drainage channels or rivers exceeds the flow capacity and absorption capacity of the surrounding dry land [3], [4]. Every year, the intensity and area of flood areas continue to increase due to environmental damage caused by humans, so the surface runoff rate increases, and the area of water catchment area decreases, which almost occurs in all watersheds in Indonesia. [5], [6]. One of the efforts to reduce the risk of flood disasters is flood spatial data management. With spatial flood information, decision-making in spatial planning for affected areas will be easier.
The Mata Allo River which passes through Enrekang Regency is one of the contributors to floodwater runoff yearly when there is an increase in rain intensity [7]. One of the causes of this flood is the change in land use in the upstream area of the river [8]. The area that is always affected is the City of Enrekang in the Juppandang district. Almost every year the area experiences floods with different flood intensities. One of the major flood events occurred on April 29, 2019, which caused the City of Enrekang to be flooded [2]. * Corresponding author: ucasideng@unm.ac.id Spatial modeling in mapping flood hazard and vulnerability simulating the coverage of inundation areas and their potential impacts [9]. Integrating spatial data, such as remote sensing data [10] is widely used, because the additional information obtained from remote sensing data will strengthen the results and sharpen the spatial information of the analysis results carried out [11]- [13]. The lack of spatial data that provides information about the condition of the area that has the potential to be affected by flooding will cause gradual losses when a flood disaster occurs again. So that a need for spatial information on flood disaster areas that can be a reference or reference in disaster mitigation activities. Seeing the dynamics of the development of hydrological studies, especially on the phenomenon of urban flooding, now the development of the study is leading to a spatial (spatial) based study. Spatial-based studies cannot be separated from the role of Geographic Information Systems (GIS) as a supporting tool.
Moreover, the function of GIS can present a form of modeling of a hydrological phenomenon and the phenomenon of flooding in urban areas [14], [15]. Spatial modeling can be done by utilizing Geographic Information Systems (GIS) and Remote Sensing (RS) as information providers in the spatial modeling input. One of the programs in flood modeling is the HEC-RAS program. Using this program, the flood inundation area can be simulated based on the river discharge at the time of the flood [16], [17].
In this study, we approach the integration of multisensor images (optical and radar) on flood modeling results by utilizing DEM data and river discharge results from field measurements during flood events using the HEC-RAS program.

Study Area
The location of the flood modeling is in Juppandang Village, Enrekang District, Enrekang Regency. Astronomically at 3°33'17.22" S -3°34'23.81" S and 119°45'59.91" E -119°47'14.10" E. In this area, there is the Mata Allo River, where on the west side there is an urban center that is densely populated with settlements. The residential area is flood-prone every year when rainfall increases [2]. More details can be seen on the following map.

Data Used
This study uses two types of images, radar images, and optical images. For radar images, use Sentinel-1 Image and Sentinel-2 for optical images. The image can be obtained at https://scihub.copernicus.eu/dhus/#/home. In addition, DEM (Digital Elevation Model) data is also prepared for flood analysis. These data can be obtained from DEMNAS Indonesia with an 8.2m resolution at https://tanahair.indonesia.go.id/demnas (specifically for the Indonesian region). For more details regarding the details of data usage, see the following table, Based on information on flood events by the National Disaster Management Agency in Enrekang Regency. April 29, 2019, was recorded as one of the major flood events in Enrekang Regency. Therefore, this study tries to simulate and model the flooding that occurred on that date. By utilizing sentinel-2, sentinel-1, and DEM images to determine the affected area.

Data Analysis
Data analysis for Radar Images and optical images, both use the SNAP program as the default analysis program. Can be downloaded at https://step.esa.int/. As for the DEM data in modeling flood discharge, the whole is carried out on the HEC-RAS program (can be obtained at https://www.hec.usace.army.mil/). Before analyzing each of the data used. First, for the image data, both sentinel-2 and sentinel-1, corrections are made to improve the existing data. For Sentinel-2 data itself, corrections are made such as reprojection to match the Sentinel-1 data, and Rayleigh correction to reduce bias in the atmosphere. Meanwhile, the atmospheric correction was not carried out, because the data obtained had gone through the TOA (Top of Atmosphere) to BOA (Bottom of Atmosphere) process. For Sentinel-1 data itself, special corrections are needed, such as applying the Apply Orbit File to update the metadata on the image because the metadata of Sentinel-1 products is generally less accurate [18]. Then apply Thermal Noise Removal to reduce noise. Calibration to produce sigma band output on either VV or VH polarization data (depending on the data used). Furthermore, Terrain Correction involves DEM and orbit files to correct errors from the SAR system such as layover, shadow, and foreshortening. So that the geometric representation of the image corresponds to the field coordinates. And converting data from sigma naught (σ°) to Digital Number (DN) which is in units of decibel (dB) as a backscattering coefficient. And lastly, subset or cut the image to fit the scope of the study area. At the same time reducing the computational load [10].
For a more complete analysis procedure, see the diagram of the data analysis process below.

Fig. 2. Analysis flow chart
Land use classification uses the Random Forest method which is included in the Supervised Classification system. Based on the ability of the Random Forest algorithm to take into account all aspects of input and computation that are repeated as needed [19]. to produce more accurate classification results, both are used for the classification of diverse land cover [10], [20], [21].
Extraction of water bodies using radar imagery [22] by setting limits or thresholds on backscattering values [23]- [25]. By determining the value boundary between water bodies and non-water bodies, it will be easier when the extraction process is actual water bodies or rivers. The date of recording the image used is also the dry phase for paddy farming. so that it will not interfere with the actual extraction of water bodies.
Modeling using the HEC-RAS program goes through several stages. Such as making Terrain data from DEM data which is used as the basis for forming a river cross-section model and the elevation of the study area. make cross-section data along the study river, to find out the 2D cross-sectional model. So that at the time of flood modeling with river discharge input, it is known that the increase in river water level occurred [17].

Data Visualization
Utilization of Remote Sensing data with different sensors such as optics and radar is very helpful in analyzing land cover to disasters [26]- [28]. In this study, Sentinel-2 and Sentinel-1 data are used, as optical sensors and radar sensors. In land cover data visualization for interpretation purposes, it is enough to use TrueColor Composite [29], this is because, at the study location, it is quite clear that the differences between each object are visible. For Sentinel-1 data using VV polarization due to the extraction of water bodies [30], not paying too much attention to Horizontal scattering as for land use. So the use of VV Polarization will be more optimal for the extraction of water bodies [13]. Meanwhile, the DEM data which is converted into Terrain Model is intended to facilitate the determination of River Reach [31] or the main flow of the Mata Allo River, and the determination of the Bank as the left and right boundary of the river. The appearance of the data used can be seen in Figure 3 below,

Waterbodies Extraction
Analysis using Sentinel-1 Image data by utilizing VV polarization. Extraction of water bodies is carried out by determining the boundary value between the backscatter value of water objects and non-water objects [23], [24]. Based on the backscattering curve value, it was found that the water body value was below -13.39dB ( Figure  4). The results of determining the limit or threshold, it is then applied to separate the backscatter value of water objects. So that the data is obtained as shown in Figure  5. This method is a fairly efficient way of extracting water bodies [22] by exploiting the backscattering response of the highly responsive Sentinel-1 image polarization to water bodies [32]. So that it is easy to distinguish between water bodies and non-water bodies. Many other studies have used radar imagery, especially for Sentinel-1 to map water bodies [18], [22], [33], [34]

Land Cover Classification
Land cover classification using the Random Forest algorithm [20], [35] which is run on the SNAP program for processing Sentinel-2 data. Based on the results of the interpretation using TrueColor composites, 8 types of land cover were determined as shown in Table 2. The results of the classification using Sentinel-2 data with Bands 2, 3, 4, and 8 as reference Bands for classification input, can be seen in Figure 5. It can be seen that the distribution of land cover in the form of agriculture is spread in the western area, while for settlements it is spread along the Mata Allo River as the study location. Meanwhile, shrubs and forests are still quite extensive surrounding settlements and hills. The dominant land use is forest, followed by shrubs, agricultural land, and built-up land

Accuracy of Classification and Waterbodies Extraction
Testing the accuracy of classification and identification of water bodies using the Confusion Matrix/Error Matrix method with Kappa Hat Coefficient. The results of the classification using sentinel-2 data with the Random Forest algorithm of 82.9% with a Kappa Hat Coefficient of 0.80. As for the results of the extraction of water bodies using Sentinel-1 data with the determination of the threshold value, the mapping obtained is as large as compared to High-Resolution Image from Google Earth 89.6% with Kappa Hat Coefficient 0.81.

Flood Modelling
Flood modeling with the HEC-RAS program, utilizing Terrain data formed from 8.2m resolution DEM data. As well as inputting river discharge data from the results of field measurements. Field measurements by a discharge measuring device installed by the Enrekang Regency Government in the Mata Allo River. The maximum discharge at the time of the flood incident was 73.23m 3 /s at 12:00 PM on 29 April 2019. And the normal discharge was 8.49m 3 /s at 12:00PM on 28 April 2019 around the upstream of the modeling as shown in Figure 7. As for 12:00 AM on 29 April 2019 based on the recording of the river flow rate of 64.74m 3 /s. This means that there is a drastic increase in discharge (6-8 times) from normal discharge. If we look at the water level rise in the simulation and modeling data based on the measurement discharge data ( Table 3). From normal discharge to maximum discharge, there was an increase in water level of 4.63m. If adjusted to the height of the river embankment at the study site. Such a large increase in water level can cause flooding in residential areas as shown in Figure 6. Based on reports from the National Disaster Management Agency and reports from the Public Health Crisis Center of the Ministry of Health of the Republic of Indonesia. The water level rise started around 04.00 AM on 29 April 2019. Very high intensity of rainfall [36], [37] and due to changes in land use in upstream areas [8], [38] become one of the factors causing flooding in the study area. To see the flow characteristics of each cross-section formed, see Appendix 1.  7 below shows the condition of the crosssection of the river in the upstream part of the study river and the downstream area of the study river which is located in the urban area of Enrekang Regency. You can see the difference in water level when viewed from Bank Level. The bank-level shows the river boundary if its height is crossed by the water level, it is included in the overflow class [16], [39]. At maximum discharge conditions, almost all areas on the right and left of the river experience overflow. Apart from rainfall intensity, land use changes, are based on existing river crosssectional models. This cross-section is included in the small category to accommodate the existing river flow when compared to the cross-sectional area of the Saddang River which is to the west of the Mata Allo River. In addition, the slopes in the upper reaches of the Mata Allo River have included slopes that cause water accumulation to accelerate [40].

Flood Affected Area
Based on the results of land cover analysis using Sentinel-2 data and extraction of water bodies from Sentinel-1 data, as well as flood area simulations using DEM discharge and Terrain data. Then the results of the analysis of areas affected by flooding in Juppandang District, Enrekang Regency are obtained as presented in Figure 9. It was found that the most affected areas related to human activities and people's livelihoods were built-up areas which reached 47.26ha of the affected area. Followed by agricultural land of 1,184Ha, while Shrub of 27.09Ha which can be connected to plantation land. The total area of the flooded area is 87.66 ha. More details can be seen in Table 4 below. The results from the affected areas are not much different from the results of flood event reports issued by the local government, such as from the National Disaster Management Agency. As for the accuracy of simulation and modeling results, it cannot be done accurately due to past flood events, and there are no reports on the related water level. However, it can be adjusted with the results of interviews from several sources of affected residents.

Conclusion
Based on the analysis results obtained, the use of an integrated approach between sentinel-2 and sentinel-1 images on the results of flood modeling or simulation with the HEC-RAS program is very possible and helps provide additional information. However, to produce better mapping accuracy, it is better to have the availability of hourly discharge data so that there are no data input errors. In addition, with the simulation results and flood modeling, it is hoped that the government can move to overcome this problem. And for future research to input more complete simulation and modeling data, such as sedimentation conditions, amount of rainfall, and other parameters. 73. 23