Flood risk assessment of the Narmada Basin, India, under climate change scenarios

Floods exert a substantial influence on human welfare at the global level. There exists compelling scientific evidence indicating that occurrences of extreme flooding events are anticipated to increase in frequency in the forthcoming years, coinciding with heightened precipitation during such events. India ranks as the second most flood-affected nation globally, following Bangladesh, contributing to approximately one-fifth of the worldwide mortality attributed to floods (Agarwal & Narain 1991). Research has illustrated that the intensity of flooding is significantly responsive to temperature variations across numerous regions globally (Panagoulia & Dimou 1997; Menzel et al. 2002; Prudhomme et al. 2013). On a global scale, floods in Asian rivers are of paramount importance owing to the number of fatalities and individuals impacted (Jonkman 2005). Approximately 40 million hectares of terrain within the country are susceptible to flooding. The National Flood Commission reports that an average of 18.6 million hectares of land experiences flooding annually. Numerous additional studies have posited that climate change contributes to the elevation of flood risk by augmenting precipitation levels relative to the mean annual rainfall (Hirabayashi et al. 2008; Fleming et al. 2012). The assessment of climate change impacts on extreme events has received comparatively less scrutiny and is characterized by greater uncertainty (Dobler et al. 2012). Recent years have witnessed a plethora of studies focused on the assessment of climate change impacts and strategies for flood management (Artlert & Chaleeraktrakoon 2013). Flood events significantly hinder and obstruct developmental processes, thereby exacerbating the vulnerability of rural communities and, consequently, perpetuating and amplifying the incidence of poverty (Omuto et al. 2013). The considerable number of individuals impacted by flood disasters in India implies that a critical factor contributing to these repercussions may be insufficient resilience to contend with the event in terms of infrastructure, property, adaptation, awareness, and preparedness (Mohanty et al. 2020). The proposed methodologies aim to quantify the repercussions of climate change on urban river flood losses through the application of hydrological and hydrodynamic models. In lower elevations, particularly in flat regions, rivers overflow their banks resulting in the inundation of low-lying areas, submerging standing crops and property, and disrupting transportation networks. During flooding events, low-lying regions adjacent to rivers may remain inundated for weeks, and occasionally months, culminating in the complete devastation of agricultural land, residences, commercial establishments, and certain transportation routes (Thapa & Prasai 2022). Recent innovations in flood modelling facilitate the estimation of property-level flood inundation, thereby aiding in the assessment of actual flood risk when analyzing the implications of flood zone designations on property valuation (Shu et al. 2022). Climate change has precipitated an escalation in extreme storms and the associated uncertainties of flood risk in urban environments, creating significant challenges for urban flood mitigation (Choi et al. 2023). Several studies employ the Soil and Water Assessment Tool (SWAT) and the Hydrologic Engineering Centre's River Analysis System (HEC-RAS) models to evaluate prospective streamflow and flooding scenarios within the designated area. There are studies that utilize the SWAT and HEC-RAS models to analyze the future streamflow and flooding in the concerned area, using data from meteorological and discharge stations (Satriagasa et al. 2023). The big economic effect of a flood is damage to public and private property. Huge amounts of funds and support are required for relief and rehabilitation of flood-displaced people. Not only do floods affect the local economy of a community but also their businesses with other communities (Martina et al. 2023). Thus, for the aforementioned reasons, it is imperative to investigate the impact of climate change on the risk assessment of floods and the likelihood of their occurrence.

The initial phase in managing flood risk sustainably involves conducting a thorough flood risk assessment. Evaluating flood risk plays a pivotal role in assisting decision makers in implementing suitable measures to control flood risk and minimize the negative consequences of floods. An incremental strategy should be embraced, conducting assessments necessary for decision making at the regional, developmental, and local levels, as well as at the site-specific level. It is crucial in determining the importance of recognized risk based on flood hazard and vulnerability factors through the adoption of innovative methodologies, tools, information, and parameters (Paul et al. 2020). A comprehensive flood risk assessment relies on considering various factors encompassing the physical, social, economic, and environmental aspects (Grigg 2023). It is essential for the flood risk assessment to encompass all potential sources of flooding, including the impact of runoff from a development on flood risk locally and beyond the development site. Meteorological, hydrological, hydraulic, economic, social science, and ecological approaches should all be integrated into flood risk analysis. The key outcomes of flood risk analysis include risk maps and real-time flood forecasting and warning systems used as operational protocols. Risk maps typically offer data on flood likelihood, water level, flow rate, sediment movement, and more. Interactive web-based methods are emerging as a popular choice for presenting flood risk analysis results nowadays (Demir et al. 2023). Real-time flood forecasting and warning systems furnish details on ongoing flood situations to improve the lead time for preparatory actions such as reservoir management and evacuations (Alsabhan & Dudin 2023). These systems primarily involve meteorological and hydrological modelling. Assessing flood risk is crucial for examining flood risk concerns in depth and delivering a quantitative assessment of potential flood risk for a proposed or existing development or land earmarked for zoning, assessing its potential impact on flood risk in other areas, and the efficacy of any suggested mitigation measures.

A number of advanced tools and methods are available for flood modelling such as numerical flood models, remote sensing, and Geographical information system (GIS)-based flood models and machine learning-based techniques for simulation of floods. In this study, however, we have used the HEC-RAS model for its advantages over other tools and methods for simulating floods as data requirements of the HEC-RAS 2D model are less compared to other modelling techniques. The methodology proposed in this study requires calibration of only a few parameters, specifically roughness and rainfall loss coefficients, which simplifies the modelling process compared to traditional hydrological models. Moreover, the potential for the creation of distinctive flood models grounded in essential mathematical formulations that are easily parameterized through the utilization of global, open-source, and publicly available data has captured the interest of researchers. Statistical models engage both deterministic frameworks and probabilistic methodologies, thereby establishing mathematical correlations between a multitude of flood-inducing variables and the geographical distribution of documented flood incidents within the designated study region. The primary aim is to ascertain the likelihood of flood occurrences, necessitating an exhaustive database of flood inventories, which poses a significant limitation to their implementation. Furthermore, these methodologies are labour-intensive and necessitate meticulous scaling to accurately delineate flood-prone regions. Machine learning methodologies are frequently employed in the evaluation of flood risks (Mojaddadi et al. 2017). Nonetheless, they are also time-intensive and require advanced computing capabilities, as well as specialized software and stringent input parameters, thereby constraining their applicability, particularly in scenarios where the spatial repository of historical flood occurrences is not accessible.

In the current investigation, an evaluation of flood risk in the Narmada Basin is conducted under different climate change scenarios. The assessment focuses on seven gauge stations where peak flows are evaluated. Two Global Climate models downscaled to finer resolution regional climate models (RCMs) are chosen from the Coordinated Regional Climate Downscaling Experiment (CORDEX) South Asia domain based on their minimal variation in statistical parameters. These models are downscaled and the historical and future scenario data for a 30-year period (2026–2055) is extracted and bias corrected for all seven gauge locations. Subsequently, the future scenario data of the Radioactive concentration pathways (RCP) 4.5 and 8.5 are analyzed, and rainfall frequency analysis is conducted on the 30-year future data for both models. The intensity–duration–frequency (IDF) curves are generated for different return periods and future scenarios. From the IDF curves, peak rainfall values are obtained for each model and each gauge station. Furthermore, peak flows at each location are simulated using hydrological modelling with the help of the conceptual rainfall–runoff model called the Hydrologic Engineering Centre–Hydrologic Modelling System (HEC-HMS) after the model is calibrated and validated at multiple outlet points of the basin. These peak flows serve as input for hydraulic modelling for simulation of floods at various locations or dam sites. A dam-to-dam segment is selected along the main river, and flood modelling is carried out to identify the maximum inundation extent and depth of water along with other parameters such as velocity and water levels. Additionally, flood maps are prepared for various return periods and climate change scenarios. To mitigate flood risk and reduce the extent of flooding shown in the maps, fuzzy optimization is employed to optimize flood release from each reservoir.

The present study aims to evaluate the flood vulnerability of the Narmada Basin in light of climate change projections. Narmada, the largest westward-flowing river in India, traverses a significant portion of Madhya Pradesh, along with parts of Maharashtra and Gujarat. Geographically, the Narmada Basin spans from East Longitudes 72° 32′ to 81° 45′ and North Latitudes 21°20′ to 23° 45′. It courses through the Deccan trap amidst the Vindhya and Satpura Mountain Ranges before merging into the Gulf of Cambay in the Arabian Sea. Encompassing an area of 98,796 km², the basin predominantly covers 87% of Madhya Pradesh, 1.5% of Maharashtra, and 11.5% of Gujarat. Originating from the Amarkantak Plateau in Anuppur district, Madhya Pradesh, the Narmada delineates the historical demarcation between North and South India as it flows westward for approximately 1,312 km. The basin's landscape comprises 60% arable land, 35% forested areas, and 5% other land categories such as grasslands or wastelands. The precipitation patterns in the basin vary greatly, with the upper catchment receiving annual rainfall between 1,000 and 1,850 mm, contrasting with lower regions that receive around 650 mm or even less. The primary land uses within the basin are agriculture (56.9%), forest cover (32.8%), wetlands (6%), and built-up areas (1.13%). While the hilly terrains boast dense forestation, the plains, ranging from upper to lower regions, offer expansive and fertile grounds ideal for agricultural activities. Predominantly characterized by black soils, the Narmada basin also features coastal plains in Gujarat comprising alluvial clays with a surface layer of black soils.

In the current investigation, the primary objective is to ascertain the influence of climate change on the precipitation and streamflow within a river basin. To achieve this, future climate projection datasets are essential for the study area, in conjunction with recorded precipitation and streamflow data from the seven Central Water Commission (CWC) gauging stations. The recorded gridded precipitation data are sourced from the Indian Meteorological Department (IMD) at a spatial resolution of 0.25 degree × 0.25 degree, encompassing 64 years of rainfall records (1951–2014). The recorded precipitation data are imperative for the hydrological modelling of the watershed as input parameters. Simulated historical and anticipated daily precipitation data are derived from the driving General Circulation Models (GCMs), the Ireland Centre for High-end Computing–European Consortium (ICHEC EC) EARTH and the Canadian Centre for Climate Modelling and Analysis–Canadian Earth System Model version 2 (CCCma-CanESM2), as part of the CORDEX–South Asia experiments pertaining to the South Asian domain, addressing both RCP 4.5 and RCP 8.5 scenarios over a 30-year period (2026–2055). These data are crucial for assessing the impacts of climate change. The downscaled regional climate models (RCMs) utilized for the GCM ICHEC EC EARTH and CCCma-CanESM2 are the Rossby Centre Regional Atmospheric Climate Model (RCA4) and the Regional Climate Model version 4 (RegCM4–4), respectively, featuring a resolution of 0.44° × 0.44°, equating to approximately 50 km. The ICHEC EC EARTH model is affiliated with the Swedish Meteorological and Hydrological Institute (SMHI), while the CCCma-CanESM2 model was developed by the Indian Institute of Tropical Meteorology (IITM). The downscaled RCM data for the driving GCMs is acquired from CORDEX experiments in net-CDF format, which constitutes a four-dimensional array of data. The retrieved data are subsequently extracted and subjected to bias correction utilizing software known as CM-Hyd, which employs an appropriate bias correction technique prior to its application as input for hydrological analysis. The recorded streamflow data are essential for the calibration and validation of the hydrological model implemented at various outlets of the basin and are procured from the India Water Resources Information System (INDIA-WRIS).

A digital elevation model (DEM) with a resolution of 90 m is necessary for hydrological modelling to delineate sub-basins within the catchment area and also serves as input for the HEC-RAS 2D model for terrain modelling. This DEM is obtained from the Shuttle Radar Topography Mission (SRTM) data available through the USGS Earth Explorer. Given that the basin encompasses an area exceeding 98,000 km², a 90-m DEM is deemed appropriate for hydrological modelling of such an expansive catchment. Decadal land use and land cover (LULC) data from 2006 to 2015 with a resolution of 10 m are required as input for the HEC-RAS model, wherein the LULC map must be uploaded along with all corresponding Manning roughness values. This data are sourced from the National Remote Sensing Centre (NRSC). Data pertaining to dams and reservoirs are essential for conducting flood modelling and mitigation studies as a dam-to-dam approach for analysis is employed in this research. The requisite data are acquired from the office of the Narmada Control Authority in Indore.

The objective of the present study is to evaluate first the impact of climate change on streamflow and precipitation of the Narmada Basin and subsequently to make an assessment of flood hazard due to climate change scenarios in the basin and mitigate the floods using optimization of peak flow releases from the reservoirs.

The GCM models chosen for this study are ICHEC EC EARTH and CCCma-CanESM2 based on their statistical parameters and minimal variability for said parameters within a specific region. As mentioned previously, these GCM models are further downscaled to the RCM RCA-4 and RegCM4-4 of resolution 50 km. The historical and future scenario data produced by these two models are acquired and adjusted for biases utilizing the CM-Hyd software. The data simulated by the GCM models are bias corrected through the application of a linear scaling method. CM-Hyd serves as a tool for the retrieval and bias correction of data derived from global and Regional climate models, with the aim of providing representative simulated climate data tailored to the study location. A 30-year span of precipitation data (2026–2055) is extracted, bias corrected, and subjected to rainfall frequency analysis using Gumbel's extreme value distribution to develop IDF Curves for the aforementioned models. IDF curves are generated for future scenarios under RCP 4.5 and RCP 8.5, considering various return periods for both GCM models. The IDF curves yield the highest rainfall intensities and maximum daily rainfall values for different return periods.

Hydrological modelling is carried out in the basin utilizing the HEC-HMS model, with calibration and validation performed at various gauge stations within the basin. Subsequently, the calibrated model is utilized to assess peak discharge at different gauge locations for various return periods and scenarios in both models. The determination of peak discharge at each gauge location is essential to conduct hydraulic modelling, specifically 2D flood modelling in HEC-RAS 2D, in order to generate flood inundation depth and risk maps for different return periods. The calibration and validation of the HEC-RAS 2D Model is also undertaken, enabling the acquisition of parameters such as velocity and water surface elevation (WSE) in the analysis. Flood risk evaluation is conducted from one dam to another along the Narmada River. Following the generation of risk maps or inundation maps, the subsequent stage involves optimizing dam releases during peak flows for the purposes of mitigation and flood control, a process facilitated through the application of fuzzy optimization.

Flow data in form of a hydrograph are entered for upstream and for downstream of the river stretch and normal depth data are entered. All other flow data for the tributaries that were obtained from sub-basin flows of the HEC-HMS model are also entered. HEC-RAS solves continuity and momentum equations (2D Saint-Venant equation) and HEC-RAS 2D modelling can be done by using both full momentum and Diffusive wave equation. In diffusive wave approximation, gravity and friction are assumed to be dominating forces acting on control volume. The diffusion wave approximation approach is used for computing the flow field in the 2D mesh, as it leads to shorter computation time and may reduce model instability. Upon completion of the unsteady flow simulation, the outcomes pertaining to flood inundation and mapping become accessible. The analysis yields information on inundation depth, area, velocity, as well as WSE at any given point within the river and its flooded vicinity. Consequently, outcomes are generated for various segments of the river spanning from one dam to another. The HEC-RAS 2D model used in the study needs to be calibrated and validated at all dam-to-dam stretches of the river using the water level data available at all dam sites. Before using the original peak discharges estimated, the HEC-RAS model is calibrated for Manning's roughness n values using real-time dam releases data obtained from the Narmada Control Authority.

Subsequently, the application of fuzzy logic for the optimization of dam releases in the context of mitigation and flood control is pursued. MATLAB R2023b version is employed for this particular objective. Utilizing a fuzzy inference system (FIS), the optimal release for each reservoir is determined by considering storage and inflow as input variables and release as the output variable. To facilitate this process, five triangular membership functions are assigned to all inputs and outputs, categorized as very low, low, medium, high, and very high. The maximum and minimum ranges for both inputs and outputs are established based on historical data and the storage capacity of the reservoir.

Table 1 shows the operating rule of the FIS for the purpose of optimizing the reservoir release or outflow.

Hence, all inputs related to different rules will yield a specific outcome based on the membership functions and input range. The resultant output determines the optimum release for each reservoir. By utilizing the optimal release flow values, flood modelling is conducted again using the HEC-RAS 2D model. This aims to determine the scope of inundation, depth, and other relevant factors for the optimal release. Subsequently, a comparison of various parameters is undertaken to evaluate the influence of optimizing reservoir releases on the extent and depth of inundation for different scenarios and return periods.

The HEC-RAS 2D model is established, followed by the execution of the unsteady flow simulation to examine inundation and flood mapping outcomes for every section of the river between dams. The flood mapping analysis is conducted specifically for maximum discharges originating from individual reservoirs subject to the impact of climatic variations. The findings from the flood modelling exercise are presented in a tabular format, illustrating distinct river segments.

Before performing flood modelling with the peak flows from the reservoir, the HEC-RAS 2D model is utilized to perform flood modelling with the original real-time flows from each reservoir in order to calibrate and validate the model for Manning's roughness (n values) updated with the land cover map. Calibration and validation of the model are done for each dam-to-dam segment by taking water level values at the gauge stations whose observed water levels are known. Once calibrated and validated, the hydraulic model is then used for flood modelling using peak flows from the reservoirs. The following tables provide information of various parameters such as inundation area, maximum water depth, maximum velocity, and maximum WSE obtained from flood modelling for various dam-to-dam segments using peak flows.

Upon completion of 2D flood modelling in HEC-RAS to analyse peak discharges originating from reservoirs, the optimal discharge from each reservoir is determined through fuzzy optimization. Subsequently, flood modelling is conducted for the optimal discharges from reservoirs along each segment of the river. Table 6 displays the peak flow rates and optimized flow rates from individual reservoirs under various climate change scenarios

The table illustrates that the reservoir releases have been significantly decreased for all reservoirs through the application of fuzzy optimization techniques in order to facilitate flood mitigation under both RCP 4.5 and RCP 8.5 scenarios. Consequently, subsequent flood modelling is conducted using these optimized release values to determine the scope of the inundation area and maximum depth. Both the inundation area and maximum depth resulting from optimal reservoir releases and peak releases are provided in the following table providing insights into the reduction in inundation area and maximum water depth achieved through optimized releases.

After flood modelling is done with the optimized flow, a comparison is made between the parameters for both peak flow and optimized flows (Table 7).

The comparison of the maximum inundation area and maximum water depth in the river for peak flows and optimized flows from the reservoirs across three distinct stretches under various return periods and future scenarios related to climate change are illustrated in the previous table. The modelling of flood risk under climate change scenarios provides valuable insights into the extent and magnitude of inundation for a given study area, enabling subsequent mitigation measures against floods. Analysis of the table and the accompanying graph reveals that the reduction in inundation area from the Bargi-to-ISP stretch is relatively minor, approximately 10%, in comparison to the other two stretches. Conversely, the ISP-to-OSP stretch experiences a reduction in inundation area exceeding 20%, whereas the OSP-to-SSP stretch shows a reduction of less than 20% for all considered return periods and future scenarios. Notably, the maximum water depth remains consistent for the same return period across different scenarios within a specific stretch. Consequently, the optimization of flow from reservoirs emerges as a viable approach for flood mitigation. The graph further demonstrates that the inundation area escalates with an increased return period for both RCP 4.5 and RCP 8.5 scenarios.

This research delves into evaluating the potential flood risk under different climate change scenarios in the Narmada Basin. It introduces a method to create flood risk maps and determine inundation extent, as well as suggests ways to lessen potential floods by adjusting reservoir releases using fuzzy optimization to minimize inundation considering dam-to-dam sections of the river. Hydraulic modelling proves to be a versatile tool for generating flood risk maps, determining inundation extent, and providing related information.

The findings show that as the return period increases from 20 years to 100 years, the flood inundation extent and maximum water depth rise for peak flow releases from the reservoirs, with higher values observed for the RCP 8.5 scenario compared to the RCP 4.5 scenario. Comparing the inundation extent for peak flows and optimal releases from each reservoir, a decrease in inundation extent is noted for optimal releases, ranging around 10% for the Bargi Dam-to-Indira Sagar Dam segment due to a 50–57% reduction in peak flow for different return periods. For the Indira Sagar Dam-to-Omkareshwar Dam section, the reduction in inundation area varies from 20 to 23% for different return periods and RCP scenarios due to a 70% reduction in peak flows through optimization. In the Omkareshwar Sagar Dam-to-Sardar Sarovar Dam segment, where peak flow reduction is around 86–90% with optimal release, the inundation area decreases by 17–18% for the RCP 4.5 scenario and 12–14% for the RCP 8.5 scenario. While optimizing flood release from reservoirs can help mitigate potential floods, there are other effective flood management strategies that can be implemented for flood mitigation.

The study also has certain limitations, which include use of two global climate models instead of Multi Model ensembles (MMEs) that reduce uncertainties in climate change projections. Hence, it may be recommended that the study be carried out with a greater number of global climate models (GCMs) for greater accuracy in predictions. Also, use of semi-distributed or distributed hydrological models such as the SWAT will yield better results for hydrological simulation, but those are data-intensive models and require huge amounts of data. Also, flood modelling in the present study is done for higher return periods of 20–100 years, which are critical. It can also be extended to smaller return periods of 2–10 years to understand the effect of climate change for flood risk assessments.

The following are the links for online repositories:

1. https://www.imdpune.gov.in- Indian Metereological Department, Pune-gridded rainfall data (0.250x0.250).

2. https://esgf-data.dkrz.de: Cordex South Asia Domain Data download for GCMs.

3. https://indiawris.gov.in: Streamflow data at gage stations.

The authors declare no conflict of interest.