Enhancing real-time flood forecasting and warning system by integrating ensemble techniques and hydrologic model simulations

Floods are probably the most devastating form of natural disaster. Human life and property suffer horrifying damage, together with almost total environmental destruction. Flood hazards may become more frequent and extensive (Hirabayashi et al. 2013; Grundmann et al. 2023). Human activities and climate change-induced increases in rainfall intensity (Tabari 2020) result in longer dry seasons with a higher risk of flooding. Two important variables are land-use changes and climate change. Human-induced changes in land cover, such as forest clearance, urbanization, or agriculture alterations, can affect rainfall patterns.

These changes in surface features, such as evapotranspiration and surface runoff, affect the local climate, which then leads to rainfall patterns becoming increasingly more severe (Zhang et al. 2022; Kantharia et al. 2024). Civilizations are becoming more flood-prone due to rising economic activities and human populations, particularly in riverine areas (Sofia et al. 2017; Habibi et al. 2023). The United Nations International Strategy for Disaster Reduction (UNISDR) estimates that between 1995 and 2015, there were the maximum flood occurrences, taking the lives of about 157,000 people. Further, projections indicate that by 2050, the increasing severity and frequency of floods could lead to damage costing around USD 1 trillion or more each year (Huang et al. 2019; Hegdahl et al. 2023). One of the most valuable tools for understanding and controlling floods is the spatial modeling of floods at the watershed scale (Kedam et al. 2024). This enables us to reduce flood damage and implement suitable preventive measures.

Due to climate change, floods and droughts are becoming more frequent. This seriously impedes real-time water resources management in arid and semiarid areas (Hess 2020; Chuphal & Mishra 2023). In response to these concerns, many nations have constructed and managed multifunctional dams to stabilize water supplies and control floods. On the other hand, increasing variability in dam intake due to changes in climate and land use directly affects water level computation and dam operation (Arduino et al. 2005). In flood seasons, uncontrolled discharge can cause devastating downstream damage (Chen et al. 2023; Di et al. 2024). Even if only the minimum water supply required for everyday human use and aquatic habitats in the lower reaches is guaranteed, it would still be impossible to avoid this problem. So, the forecast of dam inflow is essential for the effective management of water resources and operation of dams.

In India, where monsoon rains and melting snowpacks frequently cause floods, the need for efficient flood forecasting and warning systems is critical (Aronica et al. 2018; Dube & Ashrit 2023). The western parts of India, particularly in the states surrounding the Sabarmati, Tapi, and Narmada Rivers, have experienced significant flooding. The Dharoi Dam on the Sabarmati River plays a vital role in water resource management and irrigation for north Gujarat. On the other hand, overly heavy rainfall can lead to dam releases, which pose a significant threat to the downstream city of Surat, as well as to numerous local communities and agriculture farms (Patel 2020).

A forecasting and early warning system for floods can reduce economic losses and, more importantly, save many lives (Gelete et al. 2023; Islam et al. 2023). Many studies (Palmer 2001; Arduino et al. 2005; Ushiyama et al. 2019) conducted abroad have proven the effectiveness of flood forecasting, and their methodologies and results could be helpful for research. This research paper intends to design a real-time flood forecasting and warning system for the Dharoi Dam. A hybrid approach, combining ensemble techniques and hydrological models, was used in this research. The first objective is to build a hydrological model for the Dharoi Dam catchment area. Hydrological modeling using HEC–HMS is an important tool for simulating flood events in any river basin using various input parameters such as precipitation data, basin geometry, soil type, roughness coefficient, slope, and more (Jamal et al. 2023; Lee et al. 2023). This hydrological model will be employed to simulate the rainfall–runoff processes by considering the local topography and land surface characteristics of a given area (Lopez et al. 2022). The next stage involves incorporating geographic information system (GIS) data and remote sensing information, which will enhance the model's precision.

To improve the predictive accuracy of the flood forecasting system, it is important to integrate ensemble techniques into the hydrological model. Previous studies (Aronica et al. 2018; Teja & Nanduri 2020) have also shown that ensemble methods, including multiple hydrological models and data-driven approaches, can be used together. Ensemble methods that are suitable for specific regions will undoubtedly become increasingly important in hydrology-related studies, especially for flood forecasting. Building a real-time flood forecasting system requires the assimilation of current meteorological and hydrological data (Ossandón et al. 2022; Mendes & Maia 2023). For real-time flood forecasting, it is important to use the most advanced data assimilation techniques available, including remote sensing and weather radar observations, to keep improving and updating flood forecasts.

In particular, this study will contribute to the global knowledge base on flood prediction and early warning systems, while also addressing the more immediate needs of areas affected by flooding around the Dharoi Dam in India. By incorporating findings from recent research and technologies, this project aims to make a significant contribution toward deflecting disasters caused by floods. Through cooperation with other regions facing similar problems, it has the potential to serve as a model for similar projects in other areas.

Traditional flood forecasting depends mainly on hydrological models using historical data, rainfall forecasts, and river discharge measurements. While these techniques have proven valuable, their effectiveness can be limited, especially in complex areas such as the Dharoi Dam catchment area. Bridging the gap between existing flood forecasting capabilities and the developing requirements of flood-affected regions is important. This paper combines ensemble techniques with hydrological modeling to develop a real-time flood forecasting and warning system for the Dharoi Dam region. Flood forecasting is an area of research that has been advancing rapidly in recent years. Traditional flood forecasting methods have proven valuable, but their accuracy is limited; this can be especially problematic in an area as complex as the Dharoi Dam catchment area. Weather anomalies such as rainfall patterns, soil conditions, and land-use changes make flood predictions challenging to determine within a reasonable timeframe, as these factors are themselves unpredictable (Mosavi et al. 2018; Nguyen et al. 2021; Nti et al. 2021). Short-term forecasts that are delayed or inaccurate can result in the loss of life and property, damage to critical infrastructure, and significant inconvenience, disruption, or even distress to agriculture and livelihoods (Yin et al. 2023; Li et al. 2024). Therefore, there is an urgent necessity to bridge the gap between existing flood forecast capabilities and the needs of developing flood-prone areas, such as the Dharoi Dam region.

This research article proposes a novel way to forecasting floods and issuing warnings for the Dharoi Dam area by combining ensemble techniques with hydrological modeling. It fills the gap in flood forecasting capabilities, targeting to meet the urgent need for more accurate and timely flood forecasts. Drawing inspiration from successful international and national initiatives, this study seeks to devise a solution tailored to the region's unique environment. A sophisticated flood forecasting and warning system can reduce the severity of flooding in the Dharoi Dam area. It can prevent such human and economic disasters by issuing warnings in advance, giving communities enough time to prepare or evacuate if necessary. The system can also protect other critical infrastructure, such as the Dharoi Dam and downstream infrastructures and communities. Moreover, it can aid agriculture by helping farmers protect their crops and livelihoods. By preventing the floods from causing such devastation, the developed system can enhance overall resilience and well-being in the region.

Moreover, this system can help prevent the loss of human life during flood events. Accurate and timely information enables authorities to organize life-saving activities quickly and keep people safe in flood-affected areas. In addition, the system's ability to monitor water levels and predict flood patterns can help design effective flood prevention strategies for long-term resilience.

The flood forecasting system relies on remote sensing techniques, data collection, and analysis and modeling methods to collect and process weather, water, and dam inflow information. These data play a vital role in issuing real-time flood forecasts and warnings, which can then be disseminated to the general public by various media. Furthermore, historical data and forecasting methods make it more exact and reliable. Through continual monitoring and regular updating, the system remains up to date and produces reliable information in the event of flooding. The principal methodology adopted involves using ensemble data from the TIGGE platform, focusing on five major global data sets: ECMWF (European Centre for Medium-Range Weather Forecast) with 50 members, NCEP (National Centres for Environmental Prediction) with 30 members, UKMO (United Kingdom Meteorological Office) with 17 members, IMD (Indian Meteorological Department) with 20 members, and NCMRWF (National Centre for Medium-Range Weather Forecast) with 11 members. The ensemble data are used with a 5-day lead time. With the system drawing upon the strengths of various sources, it can fill gaps or uncertainties in the data and make better projections.

With this methodology, the system is equipped to respond to many scenarios and provide crucial data for flood preparedness and response. Through this raw ensemble, data for July, August, and September 2023, when most rainfall occurs, were collected. This raw ensemble data was then postprocessed using several postprocessing methods, including bias correction, the BMA approach, cNLR, HXLR, HLR, LR, and OLR. This postprocessing generates postprocessed ensemble precipitation data, which gives more accurate and reliable information for predicting future rainfall patterns. This data allows meteorologists and researchers to analyze and understand the behavior of rainfall in the particular period from July to September 2023. Postprocessing of ensemble precipitation includes model fitting, model prediction, and model verification, which includes metrics such as the Brier score, RPS score, reliability diagram, and ROC-AUC curve. The postprocessing techniques improve the accuracy of the ensemble precipitation data by correcting model biases and uncertainties. These approaches enable meteorologists and researchers to readily determine the reliability of the data and make more accurate forecasts about future rainfall patterns within the chosen time period. These postprocessed ensemble precipitation data were used in a hydrological model. The hydrological model simulates flood events and shows the peak flow of particular flood events. The hydrological model was simulated multiple times for all ensemble members, including ECMWF, NCEP, UKMO, IMD, and NCMRWF (Figure 1). The postprocessed ensemble discharge values were used to develop confidence intervals using the bootstrapping sampling method. This method help establish upper and lower limits of the flood confidence interval. Based on bootstrapping sampling, threshold discharge levels were identified, along with the exceedance probability of floods. This information is helpful for the gate operation to safely release excess flood water downstream of the dam.

The research paper focuses primarily on the Sabarmati River and the Dharoi Dam, located in the state of Gujarat, India (Figure 2). The Sabarmati River, approximately 371 km in length, traverses the central region of Gujarat and functions as a crucial hydrological source for the area. The geographical coordinates of the location span from approximately 23.18° latitude north to 72.61° longitude east. The hydrological network of the Sabarmati River inclucdes numerous tributaries that substantially contribute to its discharge and overall catchment area. The Sabarmati River is augmented by several significant tributaries, namely, the Wakal, Harnav, and Sei Rivers (Patel & Yadav 2023b). The Sabarmati River watershed covers an estimated area of 21,674 km², with average annual rainfall of 787 mm (Patel & Yadav 2023a). There are main two river gauging stations located upstream of the dam, Kheroj and Jotasan, both managed by the Central Water Commission. Dharoi Dam, located in the Mehsana District of Gujarat State, is a versatile reservoir that has been built across the Sabarmati River.

The Dharoi Dam has a storage capacity of approximately 1,580 million cubic meters of water, facilitating various functions such as irrigation, water provision, and hydroelectric power production (Patel & Yadav 2022). Over the years, the region has experienced floods caused by monsoons, resulting in extensive inundation and significant harm to both human life and property.

Comprehending the patterns and underlying causes of these flood events is imperative to develop and implement efficient flood management and mitigation strategies. From a hydrological perspective, the study area has undergone variations in water levels and flow patterns, wherein the Dharoi Dam has assumed a significant role in regulating the flow of the Sabarmati River.

Developing real-time flood forecasting information is an important step in effectively managing flood risks. Decision-makers have what they need to know to prepare to implement recovery plans to reduce the suffering brought by flooding disasters. A new technique for increasing the accuracy and applicability of flood forecasts is using ensemble precipitation data sets. This research paper aims to examine this one particular technique of merging different kinds of precipitation forecasts from different sources, models, or methods. This integration aims to create a better and accurate picture of rainfall patterns. In recent years, there has been increasing recognition that ensemble precipitation data sets can reduce the uncertainty in rainfall forecasts (Yadav & Yadav 2023). Improtantly, these uncertainties affect the chance of flooding events. Ensemble precipitation data sets generally include a wide range of precipitation input data. In this study, inputs may include data from numerical weather prediction models, radar system observations, satellite precipitation estimates, and information gathered by ground-based weather stations. Therefore, by integrating different sources into a united ensemble, we can look at projected precipitation. Each ensemble member represents a predictive forecast, revealing the essential erratic and fluctuating nature of precipitation predictions.

The ensemble system can simulate all possible precipitation scenarios with statistical analysis and data assimilation techniques. If we do, we may learn much about the uncertainty surrounding the next rainstorm (Palmer 2001; Pandey et al. 2023). Another point about using ensemble precipitation data sets in real-time flood forecasting is that they can cover a broad range of outcomes. According to the ensemble approach, rainfall forecasts have errors and uncertainties. Thus, instead of issuing a single, deterministic forecast, it offers an array of possible flood forecasts. It is then open to decision-makers to objectively consider all possible courses of action and make plans accordingly, considering both the most likely and the least desirable possibilities. This uncertainty quantification has tremendous value for informed decision-making and risk management (Pln & Kolukula 2023; Sharma et al. 2023). In addition, the creation of ensemble precipitation data sets also has the potential to increase the accuracy of flood forecasts and reduce uncertainties. Combined with hydrological models, these ensembles give forecasters a better overall sense of what different rainfall scenarios will do to river flow and flood levels. Through the countless model simulations generated by different ensemble members, the system can cover the entire window of flood possibilities, which can greatly enhance the precision and reliability of flood prediction. Deploying ensemble precipitation data sets for online flood forecasts is a rather complex process (Ushiyama et al. 2019; Teja & Nanduri 2020). There are several steps in the process, including data collection, data quality control, data assimilation, and hydrological modeling.

In this research, six postprocessing methods were used for bias correction of ensemble precipitation data. Bayesian model averaging (BMA) is a statistical technique for postprocessing ensemble weather forecasts that involves combining various models or submodels to produce a more accurate and reliable forecast. The fundamental principle of BMA is to assign weights to each model in the ensemble according to its past reliability and performance and then use these weights to produce an ensemble forecast that represents a weighted average of the individual model forecasts. Censored non-homogeneous logistic regression (cNLR) is a statistical method used to postprocess ensemble precipitation data. It considers censored observations, which are observations or values of precipitation that are below a certain detection threshold or limit of detection. Data on precipitation frequently contains censored observations, especially in remote or sparsely populated areas where reliable or frequent precipitation measurements are scarce. Logistic regression is a statistical method used for modeling the relationship between a binary dependent variable (i.e., a variable that can take on only two possible values) and one or more independent variables.

Implementing a real-time flood forecasting and warning system for the Dharoi Dam signifies a significant advancement in addressing flood-related hazards and safeguarding downstream communities and infrastructure. To evaluate the effectiveness of this system, a hybrid methodology was utilized, which involved integrating ensemble techniques into hydrological models. A comprehensive analysis of the system's performance for 2023 was conducted, with particular emphasis on July, August, and September. These lead times ranged from 1 to 5 days. Scatter plots, histograms, spread-skill plots, and verification rank histograms were analyzed during the model fitting process, and the results from these procedures played an important role in evaluating the system's predictive ability.

Among all lead times tested, the 3-day lead time was particularly promising, with an exceptionally low Brier score of 0.0006. A Brier score of zero suggests a high degree of forecast authenticity. If a Brier score is close to zero, then the forecasts, made previously with a 3-day lead time before the event, are indeed accurate, indicating that the system has effectively predicted the occurrence of a flood. The average Brier scores for the 1-day and 5-day lead times were slightly higher at 0.12, indicating moderate accuracy. The 2-day and 4-day lead times, however, showed Brier scores ranging from 0.08 to 0.10. The variations in Brier scores observed at different lead times indicate the accuracy of the system in medium-term forecasting, which are of special interest for the 3-day forecast lead time.

The reliability diagram is an important assessment of whether a model is calibrated. In this case, the reliability diagram was used for measuring the performance of the 3-day advance forecast, which turned out to be better. While the other lead times did not perform optimally, a look at the chart of the 3-day lead time shows how the predicted probabilities lined up well with the observed outcomes. The reliability chart showed that the system operates in a well-defined probability background. Further support for the system's success was lent by assessment of the ROC curve and the area under the curve (AUC).

The AUC values were more than 0.90 for the 3-day lead time and between 0.85 and 0.93 for other lead times. These AUC values for 3-day lead time show that the system can accurately discern between flood and nonflood events and is robust when it comes to prediction.

Through performing a comparison between different postprocessing methods, it was discovered that logistic regression (logreg), BMA, and hybrid logistic regression (hlogreg) were respectively superior to all others, including conditional nonlinear regression (cNLR), ordinary linear regression (OLR), and hybrid extreme learning regression (HXLR). This observation highlights the importance of the ensemble approach in increasing forecasting accuracy. All three models logreg, BMA and hlogreg, scored higher in terms of the Brier score and Reliability diagram. The choice of these methods as the most effective postprocessing methods is a useful reference for establishing the flood forecasting system.

The model was evaluated using the RPS, which achieved a commendable value of 0.09 for the 3-day lead time. This indicates the high level of accuracy in the forecasts and their consistent alignment with the actual observed outcomes. Among the postprocessing methods, logreg, hlogreg, and cNLR demonstrated strong performance with RPS scores close to the ideal score of 0. On the other hand, while still performing reasonably well, BMA, OLR, and HXLR achieved slightly higher RPS scores, falling in the range of 0.14. The validation using the RPS score serves as a supplementary metric for validating the superior performance of the ensemble techniques, specifically logreg and hlogreg, in improving the accuracy of flood forecasts.

The postprocessed ensemble precipitation data, as shown in Figure 9, are used in the hydrological model to develop various flood scenarios. Figure 9 indicates that using raw ensembles the hydrological model provides huge uncertainty in the predicted results. To reduce this uncertainty and biases from the ensemble precipitation data, its postprocessing is the most important step in this research. For this research, cNLR methods show the best performance, and their verification indices, including the Brier score, RPS score, ROC-AUC curve, show good results. So, the cNLR method was adopted, and postprocessed precipitation data by using cNLR methods were used in the hydrological model.

The impressive efficiency of the 2-day and 3-day lead time can be attributed to their optimal combination of temporal proximity and the system's effective assimilation of real-time data. The selection of a 3-day lead time can be regarded as a pragmatic compromise, allowing for sufficient time to implement flood preparedness measures while ensuring a high level of prediction accuracy. The provided data is of great importance to emergency response agencies and the affected population, as it enables prompt evacuation and allocation of resources. It is imperative to recognize and evaluate the strengths and limitations of the proposed system. The ensemble method, in which different modeling techniques are combined, has already been proven to lead to marked improvements in the accuracy of forecasts. This research underlines the importance of this approach and points to logistic regression, BMA, and hybrid logistic regression as possible focal points for future research and development of flood forecasting systems. On the other hand, the results also show us that longer lead times, such as 5 days, can cause problems in terms of accurate forecasts. More research is needed to increase the accuracy of long-term forecasts. In addition, the system's performance depends upon the nature of the available data. One must emphasize the timely and accurate collection of data to ensure and enhance system performance. Therefore, investments in data acquisition and monitoring systems are necessary to maintain the flood forecasting system in the long run.

This visualization helps decision-makers understand the range and likelihood of different flood scenarios, enabling them to make informed decisions about flood preparedness and response. The insights gained from the boxplots are invaluable for dam and reservoir operations. By understanding the probability of exceeding critical discharge thresholds, operators can implement more effective water management strategies, such as controlled water releases, to mitigate downstream flood risks. Additionally, the ability to anticipate extreme discharge events allows for better planning and coordination with emergency services, ultimately enhancing the resilience of the community to flood events. In summary, this is a robust tool for flood prediction analysis, utilizing ensemble precipitation data to calculate and determine exceedance probabilities for different discharge thresholds. The resulting boxplots offer a clear depiction of the flood risk, aiding decision-makers in implementing effective flood management and dam operation strategies. By providing a detailed understanding of the likelihood of various discharge scenarios, this analysis offers proactive measures to mitigate flood impacts and protect vulnerable communities.

Figure 17 indicates that a few ensemble members with the lowest probability will hit a 1-day, 4-day, and 5-day discharge of 900 cumec. In contrast, the peak discharge exceedance probability is very low for 2-day and 3-day lead times. It is concluded that 900 cumec will be the maximum inflow to the dam with the least probability. These findings will assist decision-makers in operating the dam gate and maintaining the reservoir level, taking into account the probability of exceedance discharge.

To conclude, ensemble techniques combined with hydrological modeling have proven highly effective in real-time flood forecasting and warning for the Dharoi Dam and for 2-day and 3-day lead times. This research also places great emphasis on postprocessing; in particular, the cNLR method has shown outstanding performance, while logistic regression, BMA, and hybrid logistic regression have shown good performance. This performance is based on their verification indices such as the Brier score, reliability diagram, ROC-AUC plot, and RPS score value. The following results are important in increasing flood preparedness, early response, and protecting life and property in this area. However, there are still problems with longer lead times and data quality. The limitation of this study is that it is only applied to smaller river basins and semiarid regions. The results may change if the model is applied to larger river basins and different regions across the globe. This research will provide a foundation for further research and development for large-scale study. The research is an essential step toward achieving practical disaster risk reduction and community safety.

In 2023, a hybrid technique combining ensemble techniques and hydrological modeling simulations was developed for real-time flood forecasting and warning for the Dharoi Dam. This marked a big advancement in flood forecasting and control measures. The study results provide information on the effectiveness of ensemble methods in conjunction with hydrological models, particularly when 2-day and 3-day lead times are used. Further, the advanced application of postprocessing techniques, such as cNLR, logistic regression (logreg), hybrid logistic regression (hlogreg), and BMA, has shown excellent results. It is possible to estimate the impact of flood in advance by adopting this postprocessed ensemble precipitation data and using it in a hydrological model, which will provide the probable possibilities of flood. This will be helpful in flood preparedness, emergency rescue work, and life and property protection in river basin areas. Based on processing of ensemble data, it has been concluded that logreg, hlogreg, and cNLR methods show good results as per the rank probability scores and ROC-AUC plots. The logreg and hlogreg methods show the maximum values of 0.9309 and 0.9306 for 3-day lead time, respectively. Postprocessed ensemble precipitation data was imported into the HEC–HMS model, and for the 1–5-day lead time, it has been seen from postprocessing results that 3-day and 2-day lead times performed the best. Hydrological model results show that 2-day and 3-day lead times of ECMWF, NCEP, UKMO, IMD, and NCRMWF have a close agreement with observed inflow for the flood event of September 2023. The maximum inflow was 800 cumec for the September 19, 2023, which is simulated by the hydrological model using postprocessed ensemble data. The correlation value for the 3-day lead time is 0.86, which indicated an excellent performance of the inflow prediction model.

The results of the research show that for the 3-day lead time, the ECMWF and IMD ensemble discharge forecasts are accurate and consistent with observed data. In contrast, the overall performance of the NCEP and NCMRWF ensemble discharge forecasts is average. In view of this contrast, ECMWF and IMD stand out as reliable choices for real-time flood forecasting for the Dharoi Dam. This aspect of the study suggests that the advantages and disadvantages of different forecasting organizations should be factored in when setting up a real-time flood forecasting system.

This research is important in building a reliable flood forecasting and warning system, with mundane applications in disaster risk reduction and community security. Combining ensemble forecasting with hydrological modeling and applying the most advanced postprocessing techniques can potentially overhaul flood forecasting and management in the Dharoi Dam area. The successful results of this research in minimizing the lead time for flood forecasting, with the help of postprocessing techniques like logistic regression, hybrid logistic regression, and BMA, indicate that the data refinement and statistical methods can lower the inevitable uncertainties of flood forecasting. The 3-day lead time yields impressive results, but longer lead times remain a problematic aspect of flood forecasting. This problem calls for more research and development in increasing forecast accuracy for long lead times. In addition, the quality and reliability of input data remain critical factors in determining the quality of the forecasting system, so concerns about improving the quality of data and the amount of data needed to maintain the forecasting system will remain. The enhanced accuracy of flood forecasts made possible by ensemble techniques and hydrological modeling has the potential to save lives and safeguard vital infrastructure. Especially in regions prone to such natural disasters, flood forecasting and early warning systems are essential for mitigating the devastation caused by floods.

This study demonstrates significant advancements in real-time flood forecasting for the Dharoi Dam but with several limitations. The accuracy of the flood forecasting system heavily relies on the quality and resolution of the input data from various ensemble models. Any inaccuracies or delays in this data can impact the forecasting results. While the study focuses on a specific river basin and the methodology is designed to be adaptable, its applicability to other regions with different hydrological and climatic conditions may require further validation and adjustments. Additionally, the study's scope was limited to a maximum lead time of 5 days, and longer lead times, which are crucial for early preparedness and mitigation, were not extensively explored. Finally, the postprocessing techniques, although effective, introduce additional computational complexity and may require significant computational resources for real-time application.

Building on the findings of this research, several areas offer potential for future exploration and development. One key area is the extension of lead times beyond 5 days while maintaining high accuracy, which would greatly enhance early warning capabilities. Research can also focus on improving the integration and assimilation of real-time data from diverse sources, such as remote sensing, ground observations, and IoT-based monitoring systems, to enhance the robustness and accuracy of flood forecasts. Further development of postprocessing techniques to reduce computational load and improve real-time applicability is another promising avenue. Additionally, expanding the validation of the proposed methodology across various river basins with different hydrological and climatic conditions will help establish its generalizability. Finally, incorporating socio-economic factors into flood forecasting models can improve the decision-making process for disaster management and community resilience, ensuring that forecasts are scientifically accurate, practically relevant, and actionable for affected communities.

The authors thank the Civil Engineering Department, Sardar Vallabhbhai National Institute of Technology, Surat for providing an opportunity to do research work. The authors are also thankful to CWC-Gandhinagar, the State Water Data Centre-Gandhinagar, and Executive Engineer-Dharoi Dam for their valuable support in data provision and guidance in this project. The authors also extend their deepest gratitude to Mr Gaurav Ninama, Assistant Engineer at the Dharoi Dam, for their huge support in data provision and technical guidance in this research work and all those who have directly and indirectly funneled them into this research work. The authors are thankful to the Civil Engineering Department, Institute of Technology, and Nirma University for supporting this research work.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. On behalf of all authors, the corresponding author states that there is no conflict of interest.

The authors declare that they received no funding for this research, and there are no potential conflicts of interest in this paper.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.