A GIS-Based Comparative Analysis of Frequency Ratio and Statistical Index Models for Flood Susceptibility Mapping in the Upper Krishna Basin, India

Abstract

The Upper Krishna Basin in Maharashtra (India) is highly vulnerable to floods. This study aimed to generate a flood susceptibility map for the basin using Frequency Ratio and Statistical Index models of flood analysis. The flood hazard inventory map was created by 370 flood locations in the Upper Krishna Basin and plotted using ArcGIS 10.1 software. The 259 flood locations (70%) were selected randomly as training samples for analysis of the flood models, and for validation purposes, the remaining 111 flood locations (30%) were used. Flood susceptibility analyses were performed based on 12 flood conditioning factors. These were elevation, slope, aspect, curvature, Topographic Wetness Index, Stream Power Index, rainfall, distance from the river, stream density, soil types, land use, and distance from the road. The Statistical Index model revealed that 38% of the area of the Upper Krishna Basin is in the high- to very-high-flood-susceptibility class. The precision of the flood susceptibility map was confirmed using the receiver operating characteristic and the area under the curve value method. The area under the curve showed a 66.89% success rate and a 68% prediction rate for the Frequency Ratio model. However, the Statistical Index model provided an 82.85% success rate and 83.23% prediction rate. The comparative analysis of the Frequency Ratio and Statistical Index models revealed that the Statistical Index model was the most suitable for flood susceptibility analysis and mapping flood-prone areas in the Upper Krishna Basin. The results obtained from this research can be helpful in flood disaster mitigation and hazard preparedness in the Upper Krishna Basin.

1. Introduction

Floods are well-known catastrophic hydro-meteorological disasters that cause huge damage to human life, settlements and industries, agriculture and animals, and the economy [1,2,3,4,5,6,7]. Globally, about 170 million people have been affected due to floods every year [8,9]. Under climate change, the frequency and magnitude of floods have increased significantly due to environmental degradation, excessive growth of the population, urbanization, and changing land use pattern [10,11]. According to Alfieri et al. [12], the frequency of floods has increased by over 40% in the last two decades throughout the world, which has caused damages of about 75 billion USD per year. Although measuring the magnitude and reporting the destruction because of inundation are enormously challenging [2], assessments and approximations of floods have significantly increased [13,14]. To minimize potential losses to the natural and artificial environment because of floods, flood prevention is essential [15]. Therefore, recognizing and plotting flood susceptibility zones significantly help with flood early warning systems, flood management and mitigation, emergency services, and disaster risk reduction [16,17,18,19]. Many scholars have conducted studies on Flood Susceptibility Mapping (FSM) using Remote Sensing (RS) and Geographical Information Systems (GISs) for flood hazard analysis [20,21,22,23,24,25,26,27].

The FSM includes environmental, geological, geomorphological, meteorological, topographical factors, and soil characteristics [28,29,30,31]. The Logistic Regression (LR), Frequency Ratio (FR), Weight of Evidence (WoE), Statistical Index (SI), Analytical Hierarchy Process (AHP), and Artificial Neural Network (ANN) models have been used by several researchers and government agencies to estimate flood susceptibility precisely [32,33,34,35,36,37,38,39]. Although numerous models have been used for flood susceptibility analysis (FSA), the results obtained by different models varied slightly from region to region. The meteorological and physical characteristics of the river basin are the main factors contributing to floods [40]. Therefore, testing and evaluation of these models can provide optimal and more reliable results [41]. Over the years, statistical methods (LR and SI) have been applied for FSA and natural hazards mapping, and the results of these models are accurate, reliable, and effective for flood disaster management and planning. FR and SI models are simple to run in a GIS environment and can develop valid FSM [27,34,36,37].

India receives more than 75% of its annual rainfall during the monsoon season from June to September [42]. During the monsoon season, flooding in the major rivers of India is the usual phenomenon, for example, the Ganga, Brahmaputra, Narmada, Tapi, Mahi, Godavari, Krishna, and Kaveri Rivers [11,43,44]. According to the NITI Aayog report [45], about 7.17 million hectares of areas of India are affected by floods annually. In addition, about 1.2 million houses are damaged by floods, 1654 human lives are lost, and 618,248 cattle lives are lost annually. The average annual economic losses are Rs 5649 crores. Previous scholars have specified that flood risk zone identification and defensive structural and non-structural risk reduction measures could efficiently decrease flood losses to some extent [46,47]. According to Ullah and Zhang [48], FSM is extensively used to determine flood-prone zones in flood management studies. Therefore, the foremost objective of the present research is a comparative analysis of the FR and SI models for FSM in the Upper Krishna Basin (UKB).

2. Study Area

The Krishna Basin is the second-largest river basin in Peninsular India [49] (refer to Figure 1a). The river has its source near Jor village (Mahabaleshwar) at an elevation of 1337 m in Maharashtra. The river flows for a length of 1400 km and joins the Bay of Bengal in Andhra Pradesh. The Krishna Basin extends over the parts of Maharashtra, Karnataka, Telangana, and Andhra Pradesh. The total area of the Krishna Basin is 2.59 Lakh sq. km. The present study is mainly confined to the UKB up to the Almatti Dam in Karnataka [50] (refer to Figure 1b). The UKB covers parts of Satara, Sangli, and Kolhapur districts in Maharashtra and Belagavi, Bagalkote, and Vijaypura districts in Karnataka (refer to Figure 1c). The catchment area above Almatti dam is approximately 35,925 km² [50]. The major tributaries of Krishna in the upper reaches are Koyna, Warna, Panchganga, Dudhganga, and Ghataprabha. The UKB is roughly triangular. It is entirely covered by Deccan trap lava flows. The alluvium mainly occurs along the banks of Krishna River and its major tributaries and thickness vary from 2 to 20 m. The average annual rainfall (AAR) of the UKB is 1347 mm. Nevertheless, annual rainfall varies from 600 mm to 6208 mm. According to the 2011 census, the total population in the UKB occupying three districts in Maharashtra is about 8,170,973 [51]. In the UKB, the Krishna River and its tributaries are susceptible to recurrent flooding during the monsoon season). Some of the recent devastating floods that occurred in the UKB were in 2005, 2006, 2009, 2011, and 2019. Recently, a disastrous flood hit the UKB in August 2019. The list of the locations affected by the 2019 flood was obtained from various published reports [52,53]. The flood data were classified into training and testing datasets.

3. Materials and Methods

Flood conditioning factors (FCFs) data were obtained from various sources and

Table 1. Parameters used for flood susceptibility mapping, their sources, and descriptions.

Flood Conditioning Factors	Data Type	Descriptions	Source
Elevation	Raster grid	Derived from ASTER DEM (30 m × 30 m) using ArcGIS	USGS https://earthexplorer.usgs.gov (accessed on 23 August 2022)
Slope
Aspect
Curvature
TWI
SPI
Rainfall	Attribute data	Derived from raingauge rainfall data and converted into raster data with 30 m × 30 m cell size	Department of Agriculture, Maharashtra, India Meteorological Department, and Karnataka Sate Natural Disaster Monitoring Center, India
Distance from the river	Vector data (Line)	Derived from stream networks of the UKB (30 m × 30 m) using ArcGIS	USGS https://earthexplorer.usgs.gov (accessed on 23 August 2022)
Stream density	Raster grid	Derived from ASTER DEM (30 m × 30 m) using fill, flow accumulation, drainage density command in ArcGIS	USGS https://earthexplorer.usgs.gov (accessed on 23 August 2022)
Soil types	Vector data (Polygon)	Digital soil map of the world-ESRI shape file	FAO http://www.fao.org (accessed on 22 August)
Land use	Raster grid	Landsat 8 OLI/TIRS, 30 m × 30 m	USGS https://earthexplorer.usgs.gov (accessed on 25 August 2022)
Distance from the road	Vector data (Line)	Derived from road networks of the district and converted into raster data with 30 m × 30 m cell size	DIVA-GIS https://www.diva-gis.org › gdata (accessed on 23 August 2022)
Flood inventory database	Vector data (Point)	Google Earth and Reports	Electronic Media (News), Print Media (Newspaper), Social Media, and Published Reports

depicts the details. The FSM output was validated by the receiver operating characteristics (ROC) curve. Figure 2 shows a detailed methodology used for FSA.

3.1. Flood Location Data

Flood location data are critical to investigate the association between flood-influencing aspects and flood incidences. The flood inventory map was prepared using the 370 flood locations affected by the August 2019 flood in the UKB (refer to Figure 3). The flood location data were obtained through various reports. The flood sites were classified randomly as training (70%) and (30%) testing sites to build and evaluate the flood models. Accordingly, 259 (111) flood locations were used as training (testing) data.

3.2. Flood Controling Factors

The FCFs were selected based on the area’s physical characteristics, data availability, and literature reviews [54,55,56,57,58]. The analyses of the relationship between flood-controlling elements and flood occurrences required a precise flood inventory map [59]. The flood conditioning parameters comprised elevation, slope, aspect, curvature, Topographic Wetness Index (TWI), Stream Power Index (SPI), rainfall, distance from rivers, stream density, soil types, land use, and distance from the road (refer to Table 1). All factors were converted into raster format with 30 × 30 m spatial resolution. The scaling factors (elevation, slope, TWI, SPI, rainfall, stream density, and distance from the rivers and road) were classified into six classes using the natural breaks classification (refer to Figure 4). The elevation is the most recurrently used key factor for FSM. The lower-elevation areas have a high possibility of flooding and are more susceptible to flood disasters [37,60]. The UKB elevation map was prepared using the Digital Elevation Model (DEM) (refer to Figure 4a). The elevation in the UKB stretched between 425 m and 1435 m. According to Souissi et al. [61], the velocity of surface runoff and the infiltration rate of water are affected by the slope. Therefore, ground slope is essential for FSA and FSM (refer to Table 1). The slope of the basin varied from 0 degrees to 70.61 degrees (refer to Figure 4b). The aspect is one of the crucial FCFs [62]. It influences the received rainfall and shows the direction of the maximum slope of the area surface. The aspect map of the UKB was classified into the 9 classes, and the aspect map of the UKB was created in ArcGIS (refer to Figure 4c). Curvature was categorized into three classes: concave, flat, and convex. According to Young and Mutchler [63], a convex slope can generate considerably extra runoff as compared to a concave slope. Hudson and Kesel [64] (2000) revealed that an area with a curvature value from 1.0 to 2.0 is more susceptible to flooding. The curvature map of the UKB was prepared using ArcGIS 10.1 (refer to Figure 4d). TWI represents the cumulative flow at any point within the watershed. It also denotes the spatial distribution of wetness in the basin. TWI has significance in the FSA and flood mapping [24,36,37,54]. TWI was computed using Equation (1) [23,65] and a TWI map was prepared from the DEM in ArcGIS 10.1 (refer to Figure 4e).

$$\mathrm{TWI}=\ln \left[\frac{A_s}{tan B}\right]$$

(1)

where $A_s$ is the cumulative area to a specific pixel and $B$ is the slope angle in degrees. According to Jebur et al. [66], SPI refers to the rate of the erosive power of surface runoff in a specified site in a river basin and can be given in Equation (2). SPI was reclassified into 6 classes and the SPI map for the UKB was generated in ArcGIS 10.1 software (refer to Figure 4f).

$$\mathrm{SPI}=A_{\mathrm{s}}\times tanB$$

(2)

Rainfall is one of the FCFs that is widely used in flood prediction studies [35,36]. The intensity of rainfall in a short period over an area can cause devastating floods [67]. The long-term rainfall data for the UKB were collected from various sources (refer to Table 1). Several scholars have used the Inverse Distance Weighting (IDW) method for spatial and temporal analysis of rainfall [68,69]. Accordingly, the rainfall distribution map of the UKB based on annual average rainfall (AAR) was prepared (refer to Figure 4g). Distance from the rivers controls the magnitude and spread of floods. There is an inverse relation between the distance from the streams and flood spread [61]. In order to measure distance from the rivers, the Euclidean Distance tool in ArcGIS software was used, and the stream density map of the UKB was generated using the DEM (refer to Figure 4h). According to Shekhar and Pandey [70], the risk of flooding increases when drainage density is high. Accordingly, a stream density map of the UKB was prepared using ArcGIS (refer to Figure 4i). Soil types, texture, and organic matter content affect the rates of surface runoff infiltration in the ground [71,72]. Therefore, the soils in the basin play a significant role in flood magnitude and extent. In addition, the water-holding capacity of soils (soil moisture) depends on the characteristics of soils. Soil type and characteristics were studied from the digital global soil data in shapefile format prepared by the Food and Agriculture Organization (FAO) of the United Nations (refer to Figure 4j). Hölting and Coldewey [73] pointed out that floods are also determined by land cover types. The LC affects surface runoff, infiltration rates, time of concentration of rainwater in the basin, and potential flood extent [74,75]. Nevertheless, areas with dense vegetation cover the decrease in the rate of runoff and increase in the percolation rate, whereas a built-up (urban) area obstructs water intrusion into the ground and accelerates the runoff [76,77]. Therefore, LU data of the UKB were obtained from the Landsat-8 images of the 2021 using random forest classification in the Google Earth Engine platform. The LU data were classified into six classes of land use: agriculture, built-up area, forest land, open land, shrub land, and water bodies (refer to Figure 4k). Lastly, distance from the road was derived with the Euclidean Distance tool in ArcGIS and classified into six categories to determine the distance of specific pixels to the nearest road line (refer to Figure 4l).

3.3. Frequency Ratio (FR) Model

The FR is one of the widely used models in various natural hazard (flood, landslide, and forest fire) analyses and mapping [78,79,80,81]. It was used to identify the probabilistic relationship between flood occurrence and FCF. The FR of each conditioning factor class can be expressed as the ratio of the class percentage and the total flood percentage. A greater bivariate probability indicates a stronger correlation, whereas a low FR value means a weak correlation. The FR value is expressed in Equation (3) as follows [34].

$$FR= \frac{N_{ij}}{P_{ij}}/\frac{N}{P}$$

(3)

where FR is the frequency ratio of the sub-classes, $N_{ij}$ is the number of flood locations, $P_{ij}$ is the total number of flood locations in the study area, N is the total number of pixels for each class of the factor, and P is the total number of pixels in the study area. Then, to calculate the flood susceptibility index (FSI), the FR values of each class were combined following Equation (4).

$$FS{I}_{FR}={\displaystyle \sum }\mathit{FR}$$

(4)

where $FS{I}_{FR}$ is the flood susceptibility index of the FR method and $\sum \mathit{FR}$ is the total FR values of all classes from each conditioning factor.

3.4. Statistical Index (SI) Model

The statistical index (SI) model was developed by van Westen [82] for landslide susceptibility mapping. SI weight values are calculated using the natural logarithm of the flood existence in each class of the conditioning factor divided by the total flood density in the study area. The higher positive SI value means a strong relationship between flood occurrence and the class of conditioning factors. The negative value shows the low possibility of flooding the class [83]. In this study, the SI weight values of each class of each conditioning factor were obtained using the following Equation (5) [82].

$$W_{ij}=\mathit{ln}\left(\frac{D_{ij}}{D}\right) =ln\left[ \left(\frac{N_{ij}}{S_{ij}}/\frac{N}{S}\right)\right]$$

(5)

where $W_{ij}$ is the statistical index value given to the i class of j factors; $D_{ij}$ is the density of floods over the i class of j factor; D is the total density of floods over the entire area; $N_{ij}$is the number of floods in i class of j factor and is the total number of floods in the study area; $S_{ij}$ is the number of pixels in i class of j factor. N is the total number of flood pixels of the study area and S is the total number of pixels of the study area.

The $W_{ij}$ weights were given to conditioning factors using a reclassification method. Finally, the reclassified factors were summed up to generate the flood susceptibility index (FSI) using a raster calculator in the ArcGIS 10.1 software. The FSI calculation is shown by the following Equation (6).

$$FS{I}_{SI}={\displaystyle \sum }{W}_{ij}$$

(6)

where $FS{I}_{SI}$ is the flood susceptibility index of the SI method and $\sum W_{ij}$ are the total $W_{ij}$ values of all classes from each conditioning factor.

3.5. Models Validation

Validation is an essential component in natural hazard susceptibility mapping to check the effectiveness and applicability of the model. This study applied the ROC to signify the graphical association of FSI and the cumulative percentage of flood occurrence. Frequently, ROC and area under ROC curves (AUC) are applied to evaluate the binary response model such as a logistics model [84,85]. They were used for both training data (70%) and testing data (30%). The AUC was also used to calculate the success and prediction rates for training and testing data. The computed results were depicted in the percentage of the study area categorized as FSI (x-axis) versus the cumulative percent of flood occurrence (y-axis). The AUC was calculated using the following equation [86].

$$AUC={\displaystyle \sum }_{i=0}^n\left(x_i-x_{i-1}\right)y_i-[\frac{\left(x_i-x_{i-1}\right)\left(y_i-y_{i-1}\right)}{2}]$$

(7)

where $x_i$is the percentage of the area and $y_i$ is the area of the flood. AUC values vary between a maximum value of 1 or 100% and 0.5 or 50%. According to Silalahi et al. [87], based on the AUC value, results can be classified as a very good model (0.9), a good model (0.8–0.9), a medium or reasonable model (0.7–0.8), and a poor model (<0.6).

4. Results and Discussion

4.1. Flood Susceptibility Analysis by the Frequency Ratio Model

The result of the FR was obtained for each sub-class of all flood-controlling factors and is summarized in

Table 2. Result of the FR and SI model for different classes of the factors.

Factors	Class	No. Pixels	Area (%)	Floods Pixels	Flood (%)	Frequency Ratio (FR)	Stastical Index (SI)
Elevation (m)	425–587	13685981	37.58	225	86.87	2.31	84
	587–663	9331587	25.62	28	10.81	0.42	−86
	663–747	6918565	19.00	5	1.93	0.10	−229
	747–858	4221014	11.59	1	0.39	0.03	−340
	858–1031	1703324	4.68	0	0.00	0.00	0
	1031–1435	559906	1.54	0	0.00	0.00	0
Slope (degree)	0–3.05	12859597	35.31	69	26.64	0.75	−28
	3.05–6.65	13741170	37.73	135	52.12	1.38	32
	6.65–11.35	5703779	15.66	50	19.31	1.23	21
	11.35–18.01	2382527	6.54	5	1.93	0.30	−122
	18.01–27.14	1226979	3.37	0	0.00	0.00	0
	27.14–70.61	506325	1.39	0	0.00	0.00	0
Aspect	Flat	3989717	10.95	19	7.34	0.67	−40
	North	7311702	20.08	26	10.04	0.50	−69
	Northeast	3589296	9.86	25	9.65	0.98	−2
	East	3794366	10.42	27	10.42	1.00	0
	Southeast	4130598	11.34	36	13.90	1.23	20
	South	3568439	9.80	25	9.65	0.99	−1
	Southwest	3628161	9.96	43	16.60	1.67	51
	West	3274717	8.99	32	12.36	1.37	32
	Northwest	3133381	8.60	26	10.04	1.17	15
Curvature	Convex	3903486	10.72	16	6.18	0.58	−55
	Flat	20857583	57.27	146	56.37	0.98	−2
	Concave	11659308	32.01	97	37.45	1.17	16
Topographic Wetness Index (TWI)	2.36–6.23	11479372	31.52	117	45.17	1.43	36
	6.23–7.83	13937655	38.27	106	40.93	1.07	7
	7.83–9.81	6107429	16.77	25	9.65	0.58	−55
	9.81–12.40	3013928	8.28	7	2.70	0.33	−112
	12.40–15.98	1604271	4.40	4	1.54	0.35	−105
	15.98–27.71	277722	0.76	0	0.00	0.00	0
Stream Power Index (SPI)	−13.82–−6.32	15348035	42.14	159	61.39	1.46	38
	−6.32–−1.92	5347891	14.68	34	13.13	0.89	−11
	−1.92–−0.14	8268133	22.70	48	18.53	0.82	−20
	−0.14–2.00	4989706	13.70	13	5.02	0.37	−100
	2.00–5.21	2027553	5.57	5	1.93	0.35	−106
	5.21–16.51	439059	1.21	0	0.00	0.00	0
Rainfall (mm)	470–826	16998500	46.67	155	59.85	1.28	25
	826–1267	9985786	27.42	85	32.82	1.20	18
	1267–1875	6621008	18.18	19	7.34	0.40	−91
	1875–2736	1759703	4.83	0	0.00	0.00	0
	2736–4037	777959	2.14	0	0.00	0.00	0
	4037–5821	277421	0.76	0	0.00	0.00	0
Distance from rivers (m)	0–2282	10494750	28.82	226	87.26	3.03	111
	2282–4979	9609658	26.39	27	10.42	0.40	−93
	4979–7884	7927304	21.77	3	1.16	0.05	−293
	7884–11,411	4973052	13.65	2	0.77	0.06	−287
	11,411–15,768	2584543	7.10	1	0.39	0.05	0
	15,768–26,349	831070	2.28	0	0.00	0.00	0
Stream density (km/sq.km)	0.05–0.39	5594652	15.36	2	0.77	0.05	−299
	0.39–0.57	9414604	25.85	3	1.16	0.04	−311
	0.57–0.73	7693109	21.12	13	5.02	0.24	−144
	0.73–0.92	6566155	18.03	55	21.24	1.18	16
	0.92–1.13	4719547	12.96	95	36.68	2.83	104
	1.13–1.58	2432310	6.68	91	35.14	5.26	166
Soil Types	Ap	5527500	15.18	9	3.47	0.23	−147
	Bv	1943925	5.34	1	0.39	0.07	−263
	Hh	3172624	8.71	35	13.51	1.55	44
	l	28852	0.08	0	0.00	0.00	0
	Lc	2468282	6.78	1	0.39	0.06	−287
	Nd	3963555	10.88	6	2.32	0.21	−155
	Ne	120372	0.33	0	0.00	0.00	0
	Vc	15251970	41.88	129	49.81	1.19	17
	Vp	3943297	10.83	78	30.12	2.78	102
Land use	Agriculture	14170930	38.91	51	19.69	0.51	−68
	Built-up/Urban	1431623	3.93	136	52.51	13.36	259
	Forest	2561431	7.03	0	0.00	0.00	0
	Open Land	10295513	28.27	62	23.94	0.85	−17
	Shrub Land	7096236	19.49	10	3.86	0.20	−162
	Water Bodies	863184	2.37	0	0.00	0.00	0
Distance from road (m)	0–1286	12581331	34.54	101	39.00	1.13	12
	1286–2916	11188412	30.72	92	35.52	1.16	15
	2916–4804	7174183	19.70	43	16.60	0.84	−17
	4804–7291	3823147	10.50	20	7.72	0.74	−31
	7291–11,238	1329214	3.65	3	1.16	0.32	−115
	11,238–21,875	324090	0.89	0	0.00	0.00	0

. Table 2 shows that the maximum (2.31) FR value with a high flood area (86.87%) for lower elevation class (425–587 m) proves the high probability of flooding in the lower-elevation area of the UKB. Conversely, zero values of FR were observed for the higher elevation classes (858–1031 m and 1031–1435 m) in the basin that signified that these areas have less probability of flooding. Normally, as the elevation increases, the FR value decreases, revealing a high positive relation between FR and flooding events [35]. The slope of the UKB basin has a relationship with flooding events. The steeper slopes produce runoff more quickly toward the downslope. FR values above 1.00 were observed for slopes between 3.05° and 6.65° (1.38) and slopes from 6.65° to 11.35° (1.23), which has the highest susceptibility to flooding. The maximum flood area (52.12%) was observed for a slope varying between 3.05° and 6.65° in the UKB (refer to Table 2). An FR value greater than one is significant for support of flooding [5]. Zero FR values were noted for the higher slopes above 18.01°, which have the lowest susceptibility to flooding in the UKB (refer to Table 2). According to Jaiswal et al. [88], a high slope ultimately increases the flood probability of the flood by increasing the swiftness of discharge, and a high-slope area has less time to infiltrate water in the ground. The slope curvature parameter of floods has a significant effect on flood susceptibility. The maximum FR values of the aspect parameter of flood control were noted for aspects facing southwest (1.67), west (1.37), and northwest (1.17). Therefore, these slope directions had the greatest control on flood occurrence in the basin (refer to Table 2). The concave slopes had the highest FR value (1.17) followed by flat areas with a value of 0.98. About 89% of flood areas were concave and flat in the UKB (refer to Table 2). Khosravi et al. [35] observed maximum flood areas for concave and flat slope areas in the Haraz watershed. In the case of TWI, the first two classes (2.36–6.23 and 6.23–7.83) had an FR of more than 1, which revealed a higher probability of flood occurrences in these zones in the UKB (refer to Table 2). Sharif et al. [89] stated that TWI signifies the influence of landscape on the location and extent of saturated source zones in producing surface runoff in a basin. The maximum FR value (1.46) was observed for the first class of SPI, which varied between −13.82 and −6.32. However, for the remaining classes, the FR was less than one. According to Khosravi et al. [35] the smaller the SPI, the greater the probability of flooding in the basin.

Maximum FR values (above 1) were observed for low-AAR classes (from 470 to 1267 mm) and the FR was zero for the areas with AAR varying between 1875 mm and 5821 mm in the basin (refer to Table 2). This shows that areas with low AAR are more susceptible to flooding as compared to high-AAR areas in the UKB. The distance from the river flood condition factor denoted that the first class of this parameter had the highest FR value (3.03), which suggests that areas from a stream to 2282 m distance are more susceptible to flooding in the UKB (refer to Table 2). In the case of stream density, the high FR values (greater than 1) were noted for areas with stream density varying between 0.73 km/sq.km and 1.58 km/sq.km. This specified that areas with high stream density are more susceptible to flooding than those with less stream density (refer to Table 2). In the UKB, the maximum FR value (2.78) was identified for Pellic Vertisols (Vp) followed by Chromic Vertisols (Vc), which was 1.19 (refer to Table 2). Vertisols soils have a high moisture-holding capacity due to the maximum clay content, which varies between 40 and 80% [90,91]. Therefore, the areas of the UKB with Vertisols soils (Vc and Vp) are more vulnerable to flooding than other types of soils. Land use is a very crucial aspect of the FSA of the UKB. A high FR value (13.36) was noted for the built-up area with about 53% of the flood area (refer to Table 2). The built-up area increases the surface runoff rate and decreases the infiltration rate. This suggests that growth in the urban or built-up area increases flood susceptibility in the UKB basin. The highest FR values for distance from the road were identified for distance classes of 0–1286 m (FR = 1.13) and 1286–2986 m (FR = 1.16), which denoted high flood susceptibility (refer to Table 2). The FSM of the UKB (refer to Figure 5a) was created by categorizing the FSI obtained from the FSA using the FR model. The FSI was classified into five susceptible zones, namely very low susceptibility, low susceptibility, moderate susceptibility, high susceptibility, and very high susceptibility. A proportion of 3% of the area of the UKB fell in the very-high-susceptibility class, while 12% of the area of the basin fell under the high-susceptibility class. Proportions of 25%, 35%, and 25% of the areas of the basin were identified in the moderate, low, and very low classes of susceptibility, respectively (refer to Figure 5a). Overall, areas with low elevation, low slopes, flat curvature, maximum stream density, maximum TWI, low SPI, near the river, highly urbanized, and covered by Vertisols are more susceptible to flooding in the UKB.

4.2. Flood Susceptibility Analysis by the Statistical Index Model

SI model results are summarized in Table 2. Higher positive SI values specify the maximum possibility of flood incidence, and negative values denote a low chance of flood existence in the region. The highest (84) SI value was noted for heights ranging between 425 and 587, whereas with the elevation increase, the SI values either became negative or zero (refer to Table 2). A negative SI value (−28) was observed for a varying slope between 0° and 3.05°. The maximum SI values of the slope class 3.05°–6.65° (SI = 32) and 6.65°–11.35° (SI = 21) revealed that areas between these slopes are more susceptible to flooding (refer to Table 2). A positive SI (16) was obtained for the concave curvature, while negative SI values were identified for the convex (−55) and flat (−2) curvatures. This specified that concave surfaces were more vulnerable to flooding in the UKB (refer to Table 2). The highest value of SI (36) was obtained for the TWI flood factor class between 2.36 and 6.23, and negative and zero SI values were found for TWIs above 7.83 (refer to Table 2). For the SPI, the highest SI (38) was obtained for the class of −13.82 to −6.32, and the negative or zero SI values were noted for the rest of the SPI classes (refer to Table 2). In this study, the highest SI (25) value was observed in the lowest rainfall class (470–826 mm), which was the lowland areas in the east of the UKB. The distance from the river of 0–2282 m displayed the highest SI value (111), while other classes were obtained as negative and zero SI values (refer to Table 2). It is obvious that with the increase in distance from the river, the flood possibility decreases [35]. The area with low stream density gained negative SI values (−299, −311, and −144), while the highest SI value (166) was obtained for the stream density class varying between 1.13 and 1.58 km/sq.km (refer to Table 2). According to the soil type factor, the Vertisols soil showed its influence on flood occurrence with high SI values for Vp (SI = 102) and Vc (SI = 17). With the high content of clay and moisture storage, the area with these soil types is more at risk for flooding. In contrast, negative SI values were observed for the Plinthic Acrisols (Ap), Vertic Cambisols (Bv), Chromic Luvisols (Lc), and Dystric Nitosols (Nd) soils (refer to Table 2). In the case of the land use factor, the highest SI (259) was derived for the built-up area (urban), which occupied 13.36% of the flood area, while negative SI values were observed for agricultural land and open land (refer to Table 2). The maximum SI (15) values observed for the second class (1286 m–2916 m) of the distance from the road showed a highly susceptible area for flooding in the basin. Inclusive, all the flood-controlling factors with high SI increased the probability of flood occurrence in the UKB. Likewise, for the FR model, the FSI was classified into five classes, namely very high, high, moderate, low, and very low (refer to Figure 5b). The FSM generated with the SI model revealed that 16% of the area of the basin was very highly susceptible to floods, whereas 22% of the area of the basin was under a highly susceptible flood zone. The moderate- and low-flood susceptibility classes occupied the same area (25%) of the basin and 13% of the area fell into very low-flood-susceptibility classes (refer to Figure 5b).

4.3. Flood Susceptibility Model Validation and Comparison

Validation of flood susceptibility mapping is essential for flood hazard research. In this research, the FSIs obtained from FR and SI models were validated to compare their performance. ROC-AUC was applied by comparing FSI results with training data (70%) and testing data (30%) to generate success and prediction rates (refer to Figure 6). This method is simple and represents a reliable result, frequently used by many flood hazard studies [34,48]. The success and prediction rate curves are portrayed in Figure 6. The AUC calculation revealed the 66.89% and 82.85% success rates for FR and SI models, respectively. The result of ROC-AUC showed 68% and 83.23% prediction rates for the FR and SI models, respectively. The results validated by the AUC showed an accuracy of the success rate of 82.85% and an accuracy of the prediction rate curve of 83.23%, which specifies a good-quality susceptibility map obtained from the SI model [87]. Therefore, the flood susceptibility map of the UKB obtained by the SI model can be utilized for managing and preventing future flood damages to the people and infrastructure of the UKB.

4.4. Application of the Research

The UKB was severely affected by the disastrous flood in August 2019. Many parts of the Satara, Sangli, and Kolhapur districts in (Maharashtra), and the Belagavi, Bagalkote, and Vijaypura districts (Karnataka) were affected by the flood. The UKB received heavy rainfall from 27 July 2019 to 13 August 2019. Some of the stations located in the UKB recorded nearly 600% (Koyna, Mahabaleshwar) to 1500% (Sangli and Kolhapur) excess rainfall against average normal rain between 27 July and 13 August 2019 (18 days) [52]. Consequently, during a short period, an enormous discharge was produced by the major rivers (Krishna, Warna, and Panchganga) in the Satara, Sangli, and Kolhapur districts in Maharashtra (refer to Figure 7). The Krishna River, Warna River, and Panchganga River in the Sangli and Kolhapur districts became severely flooded on 5–6 August 2019. Extensive inundation in many parts of the Sangli and Kolhapur districts during the August 2019 flood was observed (refer to Figure 8). In the Sangli district, 58 villages covering about 15.17% of the area were affected. In Kolhapur district, 215 villages covering 21.20% of the area were affected [52]. In Sangli and Kolhapur districts, more than one lakh people from 20,000 families were moved to safety by the local people and the National Disaster Response Force (NDRF). About 29 boats in Kolhapur, 43 in Sangli, and 17 boats in Satara were busy evacuating people. In the Kolhapur district, 89 roads were underwater [91]. According to the Vadnere committee report [52], about 1.66 lakh hectares of seasonal and perennial crops were affected in 28 tehsils of Satara, Sangli, and Kolhapur districts. The estimated losses were Rs 3475 crores in these districts [52]. Figure 7 shows that most of the settlements are in the very-high-flood-susceptibility area of the UKB. Therefore, the present study is more beneficial to minimize the losses of infrastructures and settlements, animal and agriculture, and loss of human and animal lives.

5. Conclusions

The FSM is an essential aspect to understand the scenario of the flood hazard, vulnerability, and risk in hazard management. Therefore, it is indispensable to create the most accurate and trustworthy FSM of the UKB for flood prevention, management, and mitigation. There are several models for flood susceptibility analysis, but FR and SI models have been most frequently used in flood modeling [37]. Accordingly, FR and SI models were used for flood modeling in the UKB, and the FSM was produced by coalescing all of the weight conditioning factors. The model validation outcomes from ROC-AUC denoted that the FSM generated from the SI model is more reliable as compared to the FR model. The prediction rate of the SI model (83.23%) was greater than the FR model (68%). Therefore, the flood prediction accuracy and superior performance of the SI model are higher than the FR model. The FSM maps obtained for the UKB can be significantly supportive in understanding flood risks, flood possibilities, and flood preparedness to minimize the disastrous effects of floods in the UKB.