Flood hazards under changing climate: applying an analytical hierarchy process in the Mazam-Watrak River Basin, India

Flood is a hazardous phenomenon that is frequently occurring, and it results in immeasurable human suffering (Sharannya et al. 2018) throughout the globe. Nowadays, floods are one of the most common natural disasters (https://www.who.int/health-topics/floods#tab=tab_1). Floods affect a country's economy, as they damage agricultural, urban, and industrial areas (Merz et al. 2010). The change in the climate influences an increase in flood frequency and intensity (Vogel et al. 2013) and their destructive results require the identification and mapping of flood hazards (Rao et al. 2019) in climate change scenarios also.

As floods are natural disasters, it is not possible to prevent them completely but losses due to floods can be prevented through their evaluation and management strategies (Kourgialas & Karatzas 2011). Flash floods are caused by rapid and excessive rainfall that raises water heights quickly, rivers, streams, and channels (Archer et al. 2019; https://www.who.int/health-topics/floods#tab=tab_1). For planning and implementing the flood management strategies, the flood-prone areas must be identified as a priority (Tehrany et al. 2013) for early warning, facilitating quick response, and thus reducing the harmful impact of the floods.

Numerous studies have been carried out to assess the flood zone maps of different regions of the world and various techniques and theories have been applied to identify flood risks and to delineate them into maps, e.g., hydrological models, 1D and 2D hydrodynamic models (Patel et al. 2017), and coupled hydrologic and hydrodynamic models (Li et al. 2019); multiple criteria analysis (Sharma et al. 2018); machine learning models (Li et al. 2021).

The application of hydrologic and hydrodynamic models is limited by the availability of various data including long span hydrological and meteorological data (Hurtado-Pidal et al. 2020). In many countries, there is a lack of such data. In this context, the GIS-based AHP model proves to be an efficient method for delineating flood hazard zone in a large area (Dung et al. 2022), i.e., river basin, where the number of data variables, spatial variability of data and criteria are required to introduce in the model. This method does not restrict the input criteria. There is a facility in the AHP model in terms of flexibility for introducing the importance of each of the parameters by adjustment of the weight assigned to each parameter. In addition, in AHP, the decision makers' judgement for criteria importance can be checked through the consistency ratio, and thus there is a scope for improving the accuracy of the assessment (Dung et al. 2020). The combination of GIS-AHP can be widely used nowadays, covering large areas like river basins in a data-scarce environment (Luu et al. 2018) for estimating flood hazards (Rahmati et al. 2016) including a number of criteria for hydrological data, and physical geography data. There are two methods available for weighing the factors influencing the flood hazard, i.e., direct grading method (Zangemeister 1971) and pairwise comparison (Saaty 1980). Many researchers used pairwise comparison within AHP to assess flood hazard (Meyer et al. 2009), as pairwise comparison leads to the most reliable results (Grozavu et al. 2011).

Urban area (Mayaja et al. 2016), residential area, and agricultural land around the cities reducing the vegetation, and construction of the impervious surface, affect hydrology and accelerate the runoff from stormwater resulting in floods (Wang et al. 2018). Thus, land use and its changes have been two of the factors influencing flood and relevant hazards. Global warming increases the intensity of extreme precipitation events and the risk of flooding (Tabari 2020).

Flood is generally caused by normal rain over a long period of time or intense rainfall within a short time period in addition to the causes such as channel modification, urbanization, deforestation, and land cover (Fekadu 2018). In this study, a flood hazard area is identified in the Mazam-Watrak River basin in the state of Gujarat, India based on flood causal factors such as rainfall amount, elevation, slope, land use cover, and drainage density (DD). Each flood causal factor is classified into five classes as per susceptibility of the flood, namely very low, low, moderate, high, and very high susceptibility of the flood. Identification of the most influential factors using multi-criteria analysis is carried out. The map for each flood influencing factor has been developed into five classes using a pairwise comparison of the AHP (Lappas & Kallioras 2019) and the weight of each parameter. The flood hazard index ((FHI) has been identified for flood susceptibility in the GIS environment (Sepehri et al. 2019). This work is carried out in a GIS environment with limited field data.

The trends in flooding have increased globally because climate change impacted the trend in rainfall (Chandole et al. 2019), and changes in land use (Kourgialas & Karatzas 2011). The UNESCO Division of Water Sciences initiated a working group on identifying the relative role of climatic variability and land cover changes on floods and low flow as a function of a spatial scale (Blöschl et al. 2007). To estimate the impact of climate change on flood hazards or the flood susceptible area, two factors, i.e., rainfall and land use, have been considered in this study for two scenarios: base scenario and advanced scenario, and thus, flood hazard impacts are assessed under changing climate also. The flood hazards are considered in the base scenario, i.e., in the data span from 1961 to 1989, and also in the climate change scenario, i.e., 1990–2016.

The aim of the present study in the Mazam-Watrak River basin is to (1) develop the flood hazard map in base and advanced scenarios, considering five factors, i.e., rainfall amount, slope, elevation, stream DD, and land use land cover (LULC) and (2) understand the impact of changing climate due to changes in rainfall and land use changes on the flood hazard zones.

Daily rainfall data at the rain gauge stations have been collected for this study from the State Water Data Center (SWDC) Gandhinagar, Gujarat, and Indian Meteorology Department (IMD), Pune. Average annual rainfall distribution over the river basin for the base scenario and advanced scenario has been carried out using Landsat data. The Landsat-8 OLI/TIRS C1 Level 2 images of Feb 2016 for the advanced scenario, and Landsat-5 TM C1 Level 1 of Feb 1989 for the base scenario have been used to generate the LULC through the web link https://earthexplorer.usgs.gov/. Population density of the districts in this basin varies from ca. 200 to a maximum of ca. 500 per km⁻² (https://indiawris.gov.in/downloads/Sabarmati%20Basin.pdf). Maximum and minimum elevation in the basin is 378 and 7 m., respectively. The map of the river gauging station is shown in Figure 1(b). There are five river gauging stations installed in the Mazam-Watrak River basin for which data are maintained by the State Water Data Centre (SWDC). The daily peak discharge data at all four stations, other than Rellawada are available from 1981 to 2016. Whereas at station Rellawada, the peak discharge data are available from 1997. Thus, the peak discharge data are not available for adequate data length and at an adequate number of stations in this basin to assess the flood hazards in the base and climate change scenarios. Thus, the AHP method will prove the most efficient one to assess flood hazards under this set of data. The Mazam-Watrak River basin falls in the hot semi-arid ecoregion with alluvium derived soils. Most of the land is covered with agricultural and pasture lands. The LULC has been changing with development in the basin through human activities and growing urbanization, and deforestation (Showqi et al. 2014). Also, the impact of climate change on rainfall amount in this basin has been revealed (Joshi & Makhasana 2020). Thus, the impacts of climate change and changing LULC are assessed on flood hazard in this basin in this study.

The most influencing flood-producing factors have been considered in this study. There is no such recommendation or standard available for which factors should be included in the flood hazard assessment (Tehrany et al. 2014a). For the study in the Yasooj region, Iran (Rahmati et al. 2016), four factors, i.e., slope in percent, distance from the river, LULC and altitude were considered. In some other research by Kazakis et al. (2015), the factors of flow accumulation, distance from drainage network, elevation and slope, land use, rainfall intensity and geology were considered. While, in the study by Seejata et al. (2018), six factors, namely rainfall depth, slope, DD, land use, elevation, and soil were considered. In the present study, the most used and effective factors, namely slope in percentage, rainfall depth, DD of the streams, LULC, and elevation have been selected in the data-scarce environment. All these factors have been converted to a raster grid with 30 × 30 m cells for application of the AHP method. LULC has been influenced by urbanization, commercial activities, industrialization, deforestation, and man-made activities and ultimately one of the driving factors of the regional climate change affecting flood and its hazard. Also, the rainfall is influenced by changing climate. Climate impacts may occur at larger scales, so one would expect them to be apparent in both small and large catchments and be consistent in a region (Blöschl et al. 2007). Thus, the changes in LULC and rainfall amount have been identified in this study to assess the impact of climate change on flood hazards. The other factors influencing floods, namely elevation, slope, and DD will not be much influenced by changing climate.

Rainfall is one of the most influential factors affecting flooding over the basin. Climate change influences global warming which affects the rate of temperature rises. Warmer air can hold more water vapor and causes extreme rainfall and the occurrence of frequent flood events. The abrupt changes in hydro-meteorological time series could be due to the climate regime such as local changes in the environment (extreme rainfall, urbanization, and deforestation). In this study, regional environmental changes have been considered in the parameters of rainfall and land use change patterns. The distribution of mean annual rainfall was considered in a few studies by Gazi et al. (2019). Time series data (1961–2016) at the rain gauge station for the rainfall depth have been collected from the SWDC and Indian Meteorological Department (IMD). For this study area, and for the same data span, the annual rainfall demonstrates an increasing trend, and the magnitude of the annual rainfall for various recurrence intervals in the advanced scenario increases (Joshi et al. 2019). The time span 1961–1989 has been considered as the base scenario and time span 1990–2016 has been considered as an advanced scenario for assessing the flood hazard under changing climate. The average annual rainfall is categorized into five classes: (1) very low, low, medium, high, and very high as shown in Table 1. The annual minimum, average and maximum rainfall over the data span averaged over the study area for the base scenario and advanced scenario is 274.6; 737.5; 1,469.8; and 299.2; 818.9; and 1,825.7 mm, respectively.

The slope is a surface indicator for the identification of flood susceptibility. Elevation and slope parameters are considered in most of the studies as they are the most influential parameters for flood hazard (Rahmati et al. 2016; Seejata et al. 2018). The higher the slope, the greater the velocity of flow, and the lower the possibility of infiltration. Thus, slope in percent plays an important role in the surface runoff volume for the estimation of flood susceptibility areas. The digital elevation model (DEM) is converted into the slope map and elevation map using the Arc-GIS spatial conversion tool. The higher slope value indicates the steeper terrain, and the lower value of the slope is the flatter terrain. The elevation of the river basin indicates the flow direction following the terrain. The slope and elevation have been classified into five classes as shown in Table 2.

The influence of the land use parameter on flood was considered in many studies, e.g., Stefanidis & Stathis (2013), Lappas & Kallioras (2019), and Gaňová et al. (2014). LULC is an important factor to identify those zones that have shown high susceptibility to flooding.

Agricultural land, residential areas, and roads, which are mostly made of impervious surfaces and bare lands, increase the storm runoff (Tehrany et al. 2014b). In the present study, the LULC maps have been categorized into five land use classes, i.e., forest, agriculture, pasture, water bodies, and urban for the base scenario (1989) and advanced scenario (2016). The distribution of LULC in the Mazam-Watrak River basin for the years 1986 and 2016 has been created through classification in Arc-GIS. The Landsat-8 OLI/TIRS C1 Level 2 images of the year 2016 (advanced scenario), and Landsat-4-5 TM C1 Level 1 for the year 1989 (base scenario) have been developed through the web link https://earthexplorer.usgs.gov/. LULC distribution of the Mazam-Watrak River basin is carried out in Arc-GIS 10.3. The LULC classifications in base scenario and advanced scenario are shown in Table 3.

The DD is the total length of the river reach or a stream over the entire river basin divided by the total area of the drainage basin (Gebre & Getahun 2015). The DD parameter is considered in most of the studies, as it is the most influential parameter for flood hazard (Rahmati et al. 2016; Seejata et al. 2018). In this study, DEM is used to compute the drainage basin using the spatial analyst tool. DD has been classified into five classes, i.e., very low, low, moderate, high, and very high as shown in Table 5.

Elevation, average annual rainfall, slope, DD, and land use LULC; these five factors have been considered for delineating the flood hazard zone using the multi-criteria evaluation techniques in a GIS. The flood-prone area of the Mazam-Watrak River basin has been classified into five categories, namely very high, high, moderate, low, and very low. The flood hazard maps are generated for base and advanced scenarios. From Geospatial, the raster file has been generated for all the input parameters. The AHP of the multi-criteria evaluation technique has been used to identify the weight of each parameter in each category. A consistency check has been applied to identify whether the given weight is consistent or not. The weight of the flood generating factors is carried out in five classes as the area under the very high, high, moderate, low, and very low susceptibility of the flooding.

To reduce the uncertainty in spatial distribution, it is necessary to give initial weights to every subclass notified under various criteria (Sepehri et al. 2019). In this study, each factor/parameter is classified into five classes and weights for every subclass have been given. The thematic layer of each flood producing factor was classified. The weight gives the ranges of flood susceptibility within each factor. In this study, rank/class (R) was assigned to each class according to the order of the influence of the class on flood hazard. Ranks (class) (R) of 1–5 have been adopted, where rank/class 1, 2, 3, 4, 5, respectively, represented very low, low, moderate, high, and very high flood hazard potential.

It is perceived from Figures 3 and 4 that, in base scenarios, the susceptibility of flooding is low and very low in most of the basin, while in the advanced scenario, the susceptibility of flooding is very high downstream of the river reach because the area with high rainfall range proliferates in the advanced scenario and reaches the downstream and some portion in the middle-eastern part of the study area. The value of mean annual rainfall is in the higher range in the advanced scenario as compared to the baseline scenario.

Figures 4 and 5 display the study area in each class of slope and elevation, respectively.

The basin is identified in an elevation map recognizing the highest elevation of 378 m in upstream reaches, and elevation of 7 m downstream of the river reaches. The upstream river reaches have been detected through the highest slope. Table 2 shows the Natural Break's classification (Dash & Sar 2020) of the elevation and slope over the Mazam-Watrak River basin.

Table 3 displays area (%) in each land classification and weight of each class of LULC. It is perceived from Figures 7 and 8 and Table 3 that forest and agriculture cover within the very low flood susceptibility condenses in advanced scenarios. While urban, water bodies and pasture land are covered through high susceptibility, and flood proliferates in the basin in advanced scenarios.

DEM is used to compute the drainage basin using the spatial analyst tool by considering the circle for density module calculation and classifying it into the five classes as per Natural Break Classification. Table 4 displays the classification of the DD and weight of each class.

The analytical hierarchy process (AHP) provides a systematic approach for assessing and integrating the impacts of various factors (Stefanidis & Stathis 2013). AHP is a powerful mathematical technique that was introduced by Saaty in 1980 (Siddayao et al. 2014). In this pairwise comparison of the priority, a vector is carried out, and thus this method assesses two criteria at one time (Bandi et al. 2019; Gazi et al. 2019). AHP simplifies the complex multi decision-making process which brings about a more systematic approach for base decisions made without disregarding opposing views (Yusof & Salleh 2013).

As per Goepel (2019) in AHP, a balanced scale has to be generalized. Also, it improves weight uncertainty and weight dispersion. Gaňová et al. (2014) analyzed the flood hazard in river basins of eastern Slovakia using the MCA, specifically for the AHP and GIS for the purpose of flood risk management. Many other researchers have also carried out studies using the AHP techniques that generally utilize the GIS environment for the multi-criteria decision-making process (Rahmati et al. 2016; Rincón et al. 2018; and Phrakonkham et al. 2019).

As described in section 4, In this study, each parameter is classified into five classes and weights to every subclass have been given. The thematic layer of each flood producing factor was classified. The weight gives the ranges of flood susceptibility within each factor. In this study, ranks (class) (R) of 1–5 have been adopted, where ranks 1, 2, 3, 4, and 5 respectively, represented very low, low, moderate, high, and very high flood hazard potential.

Further, the rating of each flood influencing factor is calculated by pairwise comparison in AHP to calculate the factor's final weight to generate the flood hazard map. The relative significance is set from 1 to 9 indicating the relative importance as shown in Table 5.

In order to check the reliability of the obtained ratings/weights, the consistency ratio (CR) must be computed. In AHP, the consistency ratio must be <0.1, so as to be accepted. Otherwise, it is necessary to review the subjective judgments and recalculate the weights. In case the consistency ratio is not accepted (i.e., CR > 0.1), it identifies the most inconsistent judgment derived from pairwise comparisons and proposed the changes, in order to achieve the acceptable limits (Stefanidis & Stathis 2013).

The subscripts W and NR indicate weights and normalized ratings for each factor elevation (E), slope (S), LULC, average annual rainfall (R) and DD, respectively. All thematic layers were aggregated in the GIS environment using the WLC method based on the above equation. The produced map was classified into five classes as very low, low, moderate, high, very high flood hazards as per natural break classification (Dash & Sar 2020).

The proposed methodology for the delineation of the flood hazard area can be a useful tool for the mitigation of the devasting impacts of floods. However, this model utilizes the relative rating of each factor for influencing the flood and weighting of each factor in various classes, the validation techniques of the model developed for this study area should be supported by the historical flood events data and its analysis.

To validate the results of this model developed in AHP, the latest information/data for villages under flooding have been collected for Sabarkantha and Mehmdabad districts, from which these rivers are passing (https://wrd.guj.nic.in>floodwarningbook_pdf_2021), and it is compared with the flood hazard map obtained for the advanced scenario.

The proposed methodology suggests a pairwise comparison of the importance of each factor for inducing flood using a 5 × 5 matrix, where diagonal elements are equal to 1 as shown in Table 6. The values of each row characterize the importance of inducing flood between two factors. The first row of Table 6 illustrates the importance of elevation in regard to the other factors which are placed in the columns. Iswandi et al. (2016) concluded that the weight of the elevation is highest from the expert's assessment in determining the hazard over Padang West Sumatera, Indonesia. Also, the flooded areas are often located in low elevations (Kazakis et al. 2015). As per the sensitivity analysis of the regions in Greece, it was concluded that elevation and slope have the biggest influence on flood hazard mapping in the studied region (Kazakis et al. 2015). In this study also, elevation is considered the prominent factor for causing flood hazards. The rainfall and slope of the terrain are indirectly associated with the elevation. Therefore, rainfall and slope have been assigned the rate value 1.5 in the row of elevation. Rainfall is more important to DD and LULC and thus, DD and LULC are assigned the rate values 3 (Seejata et al. 2018) and 2 in the row of rainfall. Elevation and slope are more important to LULC and have been assigned rate values of 3 and 2, respectively, as per the study by Rahmati et al. (2016).

The row for average annual rainfall describes the importance of rainfall for producing the flood to the other factors. Therefore, the row has the inverse values of the pairwise comparison (e.g., 1/1.5 for rainfall to elevation). The last row of Table 6 shows the summation of the rating values of each factor. Normalized rates of factors according to the principle of eigenvectors were considered and are shown in Table 7.

Table 7 includes the normalized rating values (i.e., the ratio of rating value to the summation of the rating values of the factor shown in Table 6), their mean and eventually the corresponding average normalized rate of each factor.

For consistency check, the judgement matrix (in Table 6) is multiplied by the criteria weight of Table 7 and the weighted sum is obtained. The ratio of the weighted sum to criteria weights is calculated and shown in Table 8.

Thus, the consistency ratio of the judgment matrix is obtained as 4.4%, which is less than 10%, so the judgment matrix of the pairwise comparison matrix shown in Table 6 is reasonably consistent.

On comparing the area encroached on by each flood hazard zone for the various flood susceptibilities, in base and advanced scenarios, it is seen in Table 10 that the area in very low, low, and moderate categories of flood susceptibility reduces in an advanced scenario. The area encroached on in high and very high categories of flood susceptibility increases in advanced scenarios. Thus, it can be concluded that larger areas are influenced by high to very high categories of flood susceptibility, in advanced scenarios.

It is seen from Figures 10 and 11 that the flooding hazard in the upstream area of the Mazam-Watrak River basin is very low, while downstream of the river basin is highly affected by flooding in both scenarios. Also, the area encroached on by high flood hazards belongs to the middle-eastern basin, i.e., on the Watrak River side in comparison to the Middle-Western side, i.e., the area covered by the Mazam River, which is a right bank tributary of the Watrak River. Furthermore, from Table 10, it is perceived that the areas in low flood hazard categories in the base scenario reduce in advanced scenarios, while areas proliferate in high and very high flood hazard categories in advanced scenarios.

Furthermore, from Figures 10 and 11, it is seen that the areas under very high flood hazards are those which belong to low elevation, and are flat in slope as seen in Figures 5 and 6. It is further observed from Figure 4 that areas in downstream reaches and the middle-eastern part of the basin belong to the high to very high rainfall categories in advanced scenarios. Thus, flood hazard maps confirmed that elevation and slope and high rainfall played a significant role in identifying the flood hazard areas.

It is observed from Figure 12 developed from the historical data of flood-affected villages within and very near the study area that flood-affected villages lie in the downstream and middle-eastern part of the basin.

Furthermore, Figures 3 and 4 show the rainfall amounts for the high and very high categories at Malpur town in the advanced scenario, and it is identified as the urban area, which is weighted higher score in the classification of the LULC map in the Figures 7 and 8. Figures 10 and 11 show that the area in and around the town of Malpur suffers high to very high flood hazards. Also, from Figure 12, it is visualized that flood-affected villages are just downstream reaches to Malpur town. Thus, it is confirmed that in addition to elevation and slope, average rainfall and LULC have also played a role in identifying the flood hazard areas in this study and results are validated from Figure 12. Thus, the results of the model developed in this study are verified with the physical factors selected, their weights in different classes and their relative importance in the pairwise matrix.

The physical factors selected in this study may be insufficient to truly simulate the flood hazards extent in the study area. The other studies were reported in the literature in which the physical factors like flow accumulation, soil type (Seejata et al. 2018), and distance from the river had been considered. However, the geology is the least affecting parameter identified by Kazakis et al. 2015, and the flood hazard map developed in this study has been validated to the historical data of the flood-affected villages in the study area. The limitation of this study is also that the weighing process in the pairwise matrix to determine the relative importance of the used factors that have been followed in this study is based on reviewing the previous literature. However, the importance of the factors are referred from the studies, which included the sensitivity analysis of the weight of the different factors. The relative importance of each factor has been incorporated in this study based on the previous literature but the results of this study have also confirmed the elevation, slope, rainfall amount, and LULC, i.e. factors which are ranked highest in this study, are also the factors affecting the floods as per the flood warning manual prepared from historical flood data.

The aim of this research is to delineate a flood hazard map in the Mazam-Watrak River basin in western India. The flood hazard map has been delineated in the base scenario and the advanced scenario to assess the impact of changing climate. Five parameters/flood-producing factors, namely slope in percent, elevation, rainfall, DD, and LULC have been considered in this study. However, the type of soil and flow accumulation could have been considered. The rainfall and LULC are subjected to changes over time, the LULC and rainfall map have been prepared in a base scenario and in an advanced scenario. There are other factors such as temperature and evaporation, which are influenced by changing climate, but they are not considered to directly cause impact on floods. In this study, the factor rainfall depth, which is directly influenced by changing climate, and LULC influencing the regional changes in the climate have been considered. The importance of each factor influencing the flood is referred from the past studies, which included the sensitivity analysis of the weight of different factors. The relative importance of each factor has been incorporated in this study based on the previous literature, but the results of this study have also confirmed that the elevation, slope, rainfall depth, LULC, i.e. factors which are ranked highest in this study, are also the factors affecting the floods as per the flood-warning manual prepared from historical flood data.

It has been identified that the flood hazard zones proliferate in a high to very high category in an advanced scenario; and flood hazard zone in a very low to moderate category condenses in an advanced scenario. Thus, the human, man-made, and developmental activities affecting LULC and increasing rainfall over time have been causing high to very high flood hazards due to changing climate. The flood hazard area exhibited in ‘high’ and ‘very high’ categories in the developed model in this study has been confirmed with the flood-prone villages in the flood-warning manual of the Government of Gujarat, India. Downstream reaches of the Mazam-Watrak River basin have been susceptible to the very high and high categories of flood hazards in both scenarios. It is concluded that the AHP model can be efficiently applied for the assessment of the flood hazards zone in the data-scarce environment. A similar study can be applied to other catchments and river basins for the assessment of the flood hazard zones.

Authors are thankful to the Department of Climate Change and Department of Higher Education of the state Government of Gujarat of India for providing the funds for the climate change project under the principal investigator. Authors are thankful to the Climate Change Department and Higher and Technical Education Department of the state government of Gujarat, India for funding the research project CCSGS (Grant No. PRJ 10-2017168640118, Principal investigator, Dr Geeta S. Joshi). Authors are thankful to the reviewers, who provided the comments for the revision of this paper.

For map generation and spatial distribution, ArcGIS Desktop 10.5.1 software is used.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.