Unsteady flow analysis using hydrological and hydraulic models for real-time flood forecasting in the Vamsadhara river basin

Abbreviation

Description

1D

One-dimensional

2D

Two-dimensional

CWC

Central Water Commission

DEM

Digital elevation model

DGPS

Differential Geographical Positioning System

DTM

Digital Terrain Model

GCP

Ground Control Points

GD

Gauge discharge

GIS

Geographic Information System

HEC-HMS

Hydrologic Engineering Center's Hydrologic Modeling System

HEC-RAS

Hydrologic Engineering Center's River Analysis System

HFL

High flood level

IMD

Indian Meteorological Department

LISS

Linear Imaging Self-scanning Sensor

LULC

Land Use Land Cover

MIKE-SHE

Generalized River Modeling Package- Systeme Hydroloque Europeen

NRSC

National Remote Sensing Centre

NSE

Nash–Sutcliffe model efficiency coefficient

R²

Coefficient of determination

SANDRP

South Asia Network on Dams, Rivers, and People

SCS-CN

The Soil Conservation Service Curve Number

SCS-UH

The Soil Conservation Service Unit Hydrograph

SRTM

Shuttle Radar Topography Mission

TIN

Triangular Irregular Networks

USACE

United States Army Corps of Engineers

USDA

United States Department Of Agriculture

UTM

Universal Transverse Mercator

VRB

Vamsadhara river basin

WGS 1984

World Geodetic System84

WSEL/WSE

Water surface elevation

In a tropical country like India, floods are the most frequent and extensive natural catastrophes (Sahoo & Sreeja 2015). The coastal area surrounding India's east coast is a densely populated stretch that is vulnerable to natural calamities such as tropical cyclones (Sahoo & Bhaskaran 2018). According to a report by the Indian Meteorological Department (IMD), the Andhra Pradesh (AP) coast ranks as the second most cyclone-prone area after Odisha and has the second-largest vulnerable region to floods after Kerala. Tropical cyclones bring heavy rain and strong winds, causing devastation on most of the east-flowing rivers like Vamsadhara and Nagavalli, which record high daily rainfall. While such hydro-meteorological calamities as floods cannot be totally controlled, they may be regulated through the application of advanced geospatial tools and decision support systems to reduce risk and prevent it (Khan et al. 2014). The hydrological modeling system (HMS) is vital in analyzing the hydrological components and flood hydrographs by utilizing geospatial data to extract the input parameters essential for modeling from resources such as digital elevation models (DEM) and land use land cover (LULC). Furthermore, these pre-acquired flood hydrographs are integrated into hydraulic models to help in understanding the complexities of flood prediction and analysis. As a result, flood damage may be reduced by establishing an early warning system in advance and delivering a flood forecasting service to the public (Borsch & Simonov 2016). In a conventional flood warning system, the stream water levels are monitored and warnings are issued with maps of the area that correlate to the river water levels (Fang et al. 2008; Ostheimer 2012; Krajewski et al. 2016).

To construct a fully functioning hydraulic model, it is necessary to choose an acceptable hydrodynamic model as well as the input variables and boundary conditions that will be employed in the model. Researchers have differing perspectives on the choosing and selecting of hydrodynamic models, which include one-dimensional (1D) and two-dimensional (2D) models, which have been a focus of study in recent years. Horritt et al. (2007), Anees et al. (2016), and Montané et al. (2017) demonstrated the benefit of hydraulic modeling in simulating flood events, estimating vulnerable areas, and identifying flood hazard zones, all of which are required components of developing any flood and risk management strategy involving 1D or 2D flow. Earlier studies have shown that a continuously simulated hydrological and hydrodynamic 1D steady-flow model can predict flood hazard zones (Sindhu & Rao 2016; Mahapatra et al. 2021). As floods are so unpredictable, an unsteady flow analysis seems to be the best technique to visualize flood hazard zones. The unsteady flow analysis was used for inundation mapping because water can flow in both directions once it reaches the floodplain (Patel et al. 2017; Mihu-Pintilie et al. 2019; Shustikova et al. 2019). The hydrodynamic models solve shallow water equations and provide more accurate estimates of inundation time and duration (Teng et al. 2017). They can also anticipate flood velocity, water surface elevation (WSE), and flooded areas, laying the basis for mitigation strategies. Such an unsteady flow simulation can quantify the spatial risk by converting the flood depth and duration to meaningful and perceptible data. The flood depth is a significant factor in determining the extent to which the flood plain gets inundated (Neelz & Pender 2013). Villazon et al. (2009) employed Hydrologic Engineering Center's River Analysis System (HEC-RAS) to evaluate the unsteady flow process in the Pirai river during flood events and to quantify the peak discharge.

Two prerequisites for risk evaluation studies are flood mapping and hazard assessment. Previous studies (Bhandari et al. 2017; Kayyun & Dagher 2018; Pathan & Agnihotri 2020) assessed the computational approaches to use for flood hazard and stream flow modeling. Mani et al. (2014) and Kalimisetty et al. (2021) developed a method for estimating floods by combining 1D and 2D flow simulations using MIKE-SHE. Sarchani et al. (2020) found that combining 1D/2D hydraulic models gives a more realistic depiction of floodplain area, flow rates, and water depths at high peak discharge. Ongdas et al. (2020) discovered submerged regions during flood occurrences with varying return times on the river Yesil in Kazakhstan. Following the flow of inundation mapping, hazard quantification through the generation of flood maps that regulate flood risk management techniques plays a critical role. These flood inundation maps developed in the Geographic Information System (GIS) are regarded to be reliant on the output provided by hydraulic model simulation results used to assess flood hazards. The development of the digital terrain model (DTM) and GIS had a profound influence on the ability to characterize and comprehend the consequences of extremely varied boundary conditions on the hydrologic response (Ramírez 2000). Several previous studies (Dewan et al. 2007; Giardino et al. 2012) evaluated, verified, and assessed flood extent and damage using remote sensing techniques in combination with auxiliary field measurements in GIS platforms. With improved processing capabilities and meteorological and terrain data, the usage of hydrological and hydrodynamic models for inundation zone identification and hazard assessment has expanded significantly (Andrysiak & Maidment 2000). Talukdar et al. (2021) investigated the Baitarani river basin in Odisha using a dynamic wave momentum equation to combine hydrologic and FLOW-2D models. The study found that the Bhadrak district is most susceptible to Baitarani river flooding during major floods, showing that the 2D model mapping was adequately accurate. Surwase et al. (2019) utilized HEC-RAS to model flooding along the Mahanadi River in Odisha in 2008. The model's flood extent and flood hazard map were verified using a satellite image, which produced good results. Khattak et al.’s (2016) study demonstrates the use of hydraulic models and the GIS to construct flood extent maps for a stretch of Pakistan's Kabul river. GIS approaches, HEC-RAS, and multi-criteria decision analysis are promising for developing realistic flood extent predictions that are more effective in flood risk zonation (Ouma & Tateishi 2014; Rahmati et al. 2015; Mishra & Satapathy 2021). Based on literature reviews, it may be stated that combining hydrological and hydrodynamic models enhances flood hazard response and management. This strategy is based on flood predictions to minimize loss of life and property. Flood risk analyses on large scales are rare and suffer from constraints due to a lack of data and methods to accurately reflect hydrodynamic processes (Vorogushyn et al. 2018). In this context, acquiring high-resolution elevation data and satellite data during storm occurrences for calibration and validation is a challenging and crucial component of the study.

Floods are frequent in the Vamsadhara river basin (VRB) and have caused havoc in 1999, 2004, 2006, 2009, 2010, 2013, 2014, and 2018 (https://www.cwcvsk.gov.in/strmlist.html). The severity of rains that occurred in 2006, 2010 (Jal), and 2013 (Phallin) was exceptionally high, affecting both the Odisha and AP coasts. According to reports, floods in 2006 affected 18,912 villages, 67.39 lakh people, and 4.90 lakh acres of agricultural land. These floods hit all major rivers, including Vamsadhara, which were flooded, posing a threat to life and property in 27 districts (https://health.odisha.gov.in/PDF/Disaster_Management.pdf). The 2010 Jal cyclone brought heavy rainfall to the region, affecting 2,73,483 people and causing major agricultural damage in AP (https://apsdma.ap.gov.in/common_mns/DM_plans/DDMP_pdf/District%20DM%20plan-Srikakulam%20District%202015.pdf). The extremely violent cyclone storm ‘Phailin’ that battered Odisha's coast in 2013 severely destroyed houses, agriculture, and communication infrastructure. (https://documents1.worldbank.org/curated/en/168471468257979992/pdf/838860WP0P14880Box0382116B00PUBLIC0.pdf). However, no substantial hydrological studies have been conducted in the majority of the VRB so far. This necessitates the construction of a watershed model and the categorization of flood-inundated areas in the watershed in order to evaluate and forecast flood risk. The significance of this research is to generate the simplest and most reliable model for predicting the affected region of the large flood plain in future emergency scenarios.

To fill this void, the research aims to investigate flood forecasting and analysis approaches by integrating hydrological and hydrodynamic models for stream flow in the VRB and mapping flood-prone areas. The hydrological modeling system (HEC-HMS) was utilized to generate peak flood flow hydrographs at defined sites. These hydrographs were then used to run a hydrodynamic model, notably in the HEC-RAS, to anticipate floods. These two models were computed and verified individually for these three major storm events from 2006, 2010, and 2013, utilizing input datasets, observation data, simulation techniques, and statistical model performances. The integrated applications are being used to simulate floods and determine the degree of flooding. Thus, this method is used in the VRB, which has historical significance when it comes to flooding events.

To calculate the excess rainfall for each basin, it was necessary to use a mathematical equation that represented rainfall loss or to transform the excess rainfall into direct runoff. HEC-HMS has a total of 12 distinct loss models, some of which are intended particularly for event simulation and others for continuous simulation (USACE 2008). Among the methods for simulation investigations, gridded loss methods and soil moisture accounting loss methods are not favored since they need a large number of parameters. The Soil Conservation Service Curve Number (SCS-CN) technique was chosen from the other loss methods for event-based simulation studies. As Sardoii et al. (2012) suggest, SCS-CN yields better outcomes than the other loss methods. The watershed coefficient is referred to as the curve number (CN). It is based on the characteristics of a watershed, such as land use and land cover (wasteland, agriculture, forest, and, more recently, urban (USDA-SCS 1986)). It has a range of 0–100, and it is employed to calculate the amount of rain that falls on the catchment. This method is simple to apply because it requires only a few watershed parameters, such as the CN number and initial abstraction (Wagener & Wheater 2006). Less optimized parameters have a key influence in increasing the accuracy of any model. The model performs better because there are fewer parameters that increase accuracy. The risk of uncertainty and the number of predictor variables go up when there are more parameters, which might result in over-parameterization. In Table 1, Equations (1) and (2) provide information on how much rain penetrates the surface and when it flows off.

Table 1

Equations used in the present study

Equation .	Formulae .	Description .
1		where Q denotes runoff depth, P represents monthly rainfall, S denotes maximum possible retention, and CN indicates curve number
2
3		where Qpeak denotes peak discharge, C denotes conversion constant, Tpeak denotes time to peak, D denotes duration, Tlag denotes time lag, S denotes slope, CN denotes curve number, A denotes area, and L denotes hydraulic length
4
5
6		where S denotes storage in the river reach, K denotes travel time, and x denotes weighting factor varied between 0 and 0.5.
7		where Z1 and Z2 refer to the height of the main channel inverts, Y1 and Y2 refer to the water level at cross-sections, V1 and V2 refer to average velocities, and refer to velocity weighing coefficients, g refers to gravity acceleration, and he energy head loss
8		where t denotes time, H denotes the water surface elevation, h denotes the water depth, q denotes a source or sinks term, and u and v denote the X and Y velocity components, respectively
9		where V denotes velocity equation given by Manning's roughness coefficient (n), R denotes the hydraulic radius, and H denotes the water surface elevation

Equation .	Formulae .	Description .
1		where Q denotes runoff depth, P represents monthly rainfall, S denotes maximum possible retention, and CN indicates curve number
2
3		where Qpeak denotes peak discharge, C denotes conversion constant, Tpeak denotes time to peak, D denotes duration, Tlag denotes time lag, S denotes slope, CN denotes curve number, A denotes area, and L denotes hydraulic length
4
5
6		where S denotes storage in the river reach, K denotes travel time, and x denotes weighting factor varied between 0 and 0.5.
7		where Z1 and Z2 refer to the height of the main channel inverts, Y1 and Y2 refer to the water level at cross-sections, V1 and V2 refer to average velocities, and refer to velocity weighing coefficients, g refers to gravity acceleration, and he energy head loss
8		where t denotes time, H denotes the water surface elevation, h denotes the water depth, q denotes a source or sinks term, and u and v denote the X and Y velocity components, respectively
9		where V denotes velocity equation given by Manning's roughness coefficient (n), R denotes the hydraulic radius, and H denotes the water surface elevation

View Large

The transform model replicates the process of excess precipitation into direct runoff from a watershed. This excess precipitation has a uniform geographical distribution and a steady intensity throughout time (USACE 2008). HEC-HMS provides a total of eight distinct transformation models. Among them, the SCS Unit Hydrograph (UH) model was selected since it is extensively used and requires minimal parameters to calculate runoff rates. When compared to other models that are more difficult and need more inputs for most catchments, it simply takes one input, i.e., lag time. Ramly & Tahir (2015) conclude that this SCS-UH approach requires fewer parameters and can be used to accurately simulate stream flows in different watersheds around the world. Equations (3)–(5) and Table 1 are used to calculate the peak discharge for each basin using the SCS-UH. Khodaparast et al. (2010) suggest that SCS-UH is more consistent and yields superior results over other transform models. The Muskingum routing technique is a lumped flow routing methodology derived for eight basin reaches using Equation (6) and (Table 1). Since its debut in 1930, this approach has been mostly used to analyze river engineering in order to determine hydrologic routing variables (Tewolde & Smithers 2006). One of the drawbacks of using the aforementioned approaches to construct a hydrological model is the unavailability of data.

The standard step approach for uniform flow analysis in HEC-RAS is used to construct water surface profiles from one cross-section to the next by resolving the energy equation using an iterative process termed the standard step method, as presented in Equation (7) and Table 1. Most of the flood flows are somewhat diffusive, so they may be modeled using the diffusive-wave method (Ponce 1989). To compute 2D unsteady flow in HEC-RAS, there are two methods, the Diffusive-wave method and the Full momentum method. The 2D diffusive-wave method was chosen over the other method because it enables computations to run faster and steadily. This approach allows for realistic modeling of the majority of 2D modeling scenarios, such as flood modeling, where inertial forces predominate over frictional and other forces. Typically, the diffusive-wave equation may be applied to larger river systems. The use of the diffusive-wave model provides a sufficiently precise answer if the underlying assumptions are met; it mostly computes approximate global estimations (i.e., flood extents). HEC-RAS will employ both continuity and diffusive-wave versions of the momentum equation to compute WSE at a point in time. Equations (8) and (9) and (Table 1) indicate the unsteady differential mass conservation (continuity) form and Manning's formula in the diffusion-wave equation.

Efficiency measures of the model are determined by assessing its performance both before and after calibration, as well as its accuracy during validation, using the criteria mentioned below (Moriasi et al. 2013). Equation (10) shows the Nash–Sutcliffe model efficiency (NSE), and the coefficient of determination (R²) in Equation (11) is used in evaluating the model's overall performance.

The HEC-HMS modeling system was used to simulate 90 days of daily rainfall over 3 years (2006 for calibration, 2010 and 2013 for validation). The hydrologic parameters in the model were subjected to sensitivity analysis, and three parameters were shown to be the most sensitive. CN, travel time (K), and runoff coefficient (x) were also shown to be the most sensitive parameters in other studies based on the parameter sensitivity analysis (Sardoii et al. 2012; Tahmasbinejad et al. 2012). During the calibration process, the pre-set values for these parameters were fine-tuned to the measured value by comparing simulated results with real-time data that had acceptable range indicators that were confirmed by the model's statistical performance (Mahapatra et al. 2021), while fitting the computed discharge values with historical data at gauge stations in the basin. In Table 3, the calibrated parameters CN, K, and x were computed for the run in 2006. The NSE and R² indices were utilized to do statistical analysis. It was determined that the calibration procedure was effective in terms of both visual and statistical comparisons. The model was run using predefined values of the calibrated model in the validation phase, and the corresponding values obtained were evaluated statistically in the same way as the calibration process was done. Thus, the model can be utilized as a real-time flood forecast model because it is calibrated and further evaluated with 2 years of real-time data that corresponded well with the observed data.

In HEC-RAS Mapper, a spatially varying land classification data set is created and then connected to the 2D flow area mesh using Manning's n values. Despite the widespread availability of traditional Manning's n values, predicting the value that should be utilized to simulate the flood on a large scale is difficult. This is mostly due to the diversity of roughness levels, which may vary by region (Schumann et al. 2007). As a consequence, this model must be simulated against a genuine flood occurrence before producing the necessary results, or an actual survey of the area must be conducted to determine the friction coefficient (Hunter et al. 2007), while the latter is unsuitable for large regions. Several studies, however, approximated Manning's roughness coefficient (n) with a respectable degree of precision using a look-up database of land use land cover applied to a hydrodynamic numerical model (Hossain et al. 2009; Lumbroso & Gaume 2012). Manning's roughness coefficient (n) is optimized in 10% increments from its base value throughout each simulation to produce calibrated roughness values ranging from 0.030 to 0.050, as mentioned in Table 4.

After providing the model with inputs, including unsteady flow data, it is then made to run for simulation. The theta (θ) parameters are examined, i.e., a weighting factor of 1.0 is selected for the research area, which varies from 0.6 to 1.0. As stated by Pappenberger et al. (2005), the probability of obtaining a successful model result is significantly increased when the weighting factor is set to 1.0. Water surface tolerance is set to 0.003 m, and the algorithm is run for 20 iterations with an initial condition ramp-up fraction of 0.1. After simulation, the velocity, depth, and WSE profiles are imported into Arc GIS to generate flood inundation mapping for the three flood events.

The model was validated at two GD stations during the 2010 and 2013 floods, as shown in Table 6. Figure 9(c) and 9(d) compare the computed and actual discharge at both the GD stations during the 2010 flood event, which match well at the peak. Similarly, the hydrograph generated between the calculated and observed discharge during the 2013 flood event is shown in Figure 9(e) and 9(f), and it is very similar to historical data at both the Gunupur and Kashinagar gauges. In Table 6, the findings exhibit a low degree of variance within ±20% acceptable level (Sabzevari et al. 2009) from the observed data, which is acceptable. The validated model performances in 2010 with E were 0.81 and 0.80. However, the 2013 model's performance is E equal to 0.84 and 0.82, which are satisfactory and acceptable. For the 2010 flood, R² values of 0.72 and 0.75 were obtained. However, they were 0.70 and 0.73 for the 2013 flood, which is deemed to be acceptable. Furthermore, the HEC-RAS model was fed with peak flow data derived from the standardized model at designated locations within the watershed.

The HEC-RAS model created for the VRB was calibrated and verified at two gauging stations on reaches R140 and R150 at cross-sections RS 3000 and RS 24000 for flow by comparing simulated values with observed data. Chow et al. (1988) recommended a roughness range of 0.035–0.065 for the river stretch and 0.08–0.15 for the overflow area. In the modeling process, the roughness coefficient (n) value of 0.035 is employed until the final fitted values nearly match the values at gauge stations. Table 7 shows the results of the calibration for the floods on 30 September 2006. The statistical indicators NSE and R² show that correlations for both calibration and validation are within an acceptable range of the values listed in Tables 7 and 8. The validation phase was carried out for two peak occurrences: the floods of 19 September 2010 and 28 October 2013. As shown in Table 8, the discrepancy between actual and simulated water surface levels was slight, so the findings are regarded as good. Thus, it can be said that the RAS model was able to accurately predict stream flow in the Vamsadhara river watershed.

After examining flood inundation extents in Ras Mapper, flood risk maps are generated for flood occurrences by exporting the raster inundation layers and processing them in Arc GIS. Flood risk maps show the geographical distribution of risk, which is considered as the likelihood of occurrence of an event multiplied by the consequences. These maps of flooding may be created using deterministic or probabilistic approaches (Bates et al. 2004; Merz et al. 2007; Baldassarre et al. 2009a). The VRB, with its braided rivers and frequent flash floods, requires the adoption of a probabilistic depiction of flood extent (Alfonso et al. 2016).

The suitability of different models will vary depending on the required degree of precision and the complexities of the region to be depicted. If the goal of the work is to find the inundated area that will be flooded at each cross-section, a 1D model may be enough, like the Vamsadhara river. However, the 1D method's usefulness depends on the modeled region. A 1D model is unlikely to accurately estimate the maximum inundation extent in more complex regions or for large river basins since the inundation extent is determined by how flow propagates along floodplain boundaries. To predict the maximum flood extent, 2D or 1D–2D models can be used. These models will need more time to set up and compute than the 1D model, but they may be essential in more complex regions. When it comes to flooding dynamics and comprehensive information about velocities on the floodplain, a 2D or 1D–2D model is necessary. This is owing to the intrinsic limits of 1D modeling, such as the stability concerns with the 1D Vamsadhara river model, which were connected to cross-sectional spacing, substantial energy losses within the river channel, and flow being limited between channel banks.

The features of the floodplain, such as the floodplain behind the levee, its shape, and rural and urban floods, will have a significant influence on whatever model is chosen and how this model is developed.

Case I: The 1D model is difficult to use to represent flow in a river if the main channel is constrained by a levee, but both 1D–2D and 2D models may be used to describe the channel–floodplain interaction where the floodplain and river are separated by a levee or other barrier. The results utilizing a purely 2D model will probably be quite sensitive to the breaklines used to capture the barriers. The banks are represented clearly in the 1D–2D case with the lateral structures, and there is no risk of leakage due to 2D cell displacement, as is the case with pure 2D models. However, for this type of case, the model's performance and stability would be significantly impacted by the parameters defining the lateral structures.

Case II: If the landscape around the river is V-shaped and there is no barrier separating the floodplain from the river, it is probable that a less complex model is required. One of the limitations of 1D modeling is that for each cross-section, a single water surface is calculated. This simplification will be more realistic if the floodplain is V-shaped than if it is confined by a levee. Thus, V-shaped floodplains are best represented by a 1D modeling technique. A 2D model may also be used as an alternative. It is not necessary to use breaklines or smaller cell sizes to record the bank elevations if there are no obstructions between the river and the floodplain. In some situations, the sub-grid approach makes it possible to use very large computational mesh sizes. This means that calculation times can be kept under control even at the catchment scale. Using a 1D–2D technique for a V-shaped catchment might be challenging since there is no apparent demarcation between the main channel (which should be represented in 1D) and the floodplain. As a result, locating acceptable places for the lateral structures will be problematic. The flow will likely be predominant in the direction of the stream in a V-shaped valley, so a combined 1D–2D method is not required to represent this sort of terrain.

Case III: In the case of a simple or rural floodplain with few obstructions to the flow of water, it may be modeled using a 2D or 1D–2D model with a large cell size (>20 m). Important topographical features may be defined using breaklines. A 1D model may be enough if there are few obstacles to the flow and the friction coefficient can be used to find the obstacles.

Case IV: Multiple barriers slow the flow and influence the flood propagation in urban floodplains. This research does not investigate urban flooding caused by rivers. Literature suggests that 2D models of urban floodplains are sensitive to mesh resolution and that the use of larger cell sizes may reduce model precision because obstacles cannot be accurately modeled (Yu & Lane 2006a). Studies show that using sub-grid terrain representation could make it possible for cells to be bigger (Yu & Lane 2006b, 2011).

This Vamsadhara model may be utilized instantaneously for similar types of basins where rapid forecasting is required in a short time span to prevent losses on the ground. A similar investigation might be performed on different river basins in the future to evaluate river flow modeling using different 2D models. In addition, future studies should focus on 2D models of meandering rivers.

The work proposes a conceptual method for merging the hydrologic model HEC-HMS with the hydrodynamic model and delineating flood inundation zones in a 2D unsteady flow for the VRB using HEC-RAS. The research also showed that the 2D diffusion-wave approach is extremely effective for determining WSE, depth, and velocity. Its animation capabilities aid in visual comprehension. The HMS and RAS models were calibrated and validated using three flood events from 2006, 2010, and 2013. As a result, based on statistical analyses, the generated models may be considered to have provided an acceptable quantified output. The study also examined the feasibility of using HEC-RAS for flood risk assessments. The expected flood regions for upstream and downstream of the basin allocated to danger zones were computed and found to be reasonable based on visual observation and historical data. Further vulnerability assessments could not be performed owing to a lack of data for the areas. Although it is possible to verify the accuracy of flood risk maps by delineating inundated areas over historical data (e.g., satellite images), this could not be done during flood events due to a lack of available satellite images. The Srikakulam district collectorate's (AP) data were utilized to verify the accuracy of the maps of the villages downstream of the basin, and the findings indicate that the maps are reasonably good. Thus, for flood forecasting in the VRB and for other hydro-meteorologically similar river basins, this calibrated model can be used as a reference model. Using this proposed model, stakeholders may examine various mitigation measures and approaches in response to anticipated flooding situations.

Despite continuous research, flood modeling at high spatial and temporal resolutions remains a substantial difficulty in hydrologic and hydraulic studies. Modeling applications provided for employing hydrologic models have significant drawbacks, including: (a) during flood events, model findings also revealed that simulated values were somewhat higher than observed levels; this discrepancy may be due to input data (DEM, rainfall, soil, and LULC) and a small subset of observation data (stage and discharge). Second, delineating the basin model into more than 17 sub-basins may aid in improving model accuracy (Hromadka et al. 1988). (b) Simulated data are exported and used as input in the hydrodynamic model: flow hydrographs derived from the hydrological model can be used as boundary conditions in hydrodynamic modeling. As a result, hydrodynamic modeling has its own set of inherent uncertainties, such as the quality of topographical data and the effect of roughness coefficient on flow levels, velocity, etc. Numerous investigations have proven that even small topographic inaccuracies may significantly affect model findings (Bates et al. 1997; Aronica et al. 1998; Nicholas & Walling 1998; Wilson 2004). Each model is simply a rough approximation of the actual geometry, which depends on the effective roughness coefficient. Minimizing the uncertainty in surface roughness can considerably improve the calculation of flood extent on landscapes (Aronica et al. 1998; Marks & Bates 2000). According to Hunter et al. (2007), an actual survey of river cross-sections in the area is required to quantify the friction coefficient and hence increases the model's accuracy relative to the projected value. Large-scale surveys are becoming more difficult and impractical to undertake. (c) Flood-inundated regions’ delineation in unsteady flow analysis: simulated hydrographs generated by calibrated models were used as input for hydrodynamic modeling. Uncertainties include the need for a high-resolution DEM to adequately depict the river flood plain, setting up the model, and model parameters such as friction, flow data, and geometry inputs for the model (e.g., flood plain and channel). It is also important to consider how long the computations will take when selecting mesh size and DEM resolution (Shustikova et al. 2019). Flood risk assessment is complex owing to the enormous number of variables that must be evaluated, as well as the fact that risk is dynamic, influenced by various factors such as climate change, model setup, and so on (Costabile et al. 2015).

Furthermore, as previously stated, the inaccuracies in the simulations from the integrated modeling system of flooding assessment derive from a variety of sources of uncertainty, which most likely spread in an uncertain and quasi-way. The next step of investigation must be focused on the evaluation and measurement of these flaws, as well as how they spread across the modeling system.

The authors are grateful to the officials and members of the India Meteorological Department, Survey of India (Bhubaneswar), Central Water Commission (Bhubaneswar), NRSC (Hyderabad), Srikakulam Collectorate Office (Disaster Management Plan), and Water Resources Department (Visakhapatnam) for providing data and other pertinent information.

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

The authors declare there is no conflict.