Development of a reservoir operation model determining the pre-release strategy for the flood events Open Access

Rising global temperatures intensify the hydrological cycle, driving more frequent and severe rainfall events, accelerating snowmelt, and increasing sea levels, all of which amplify the risk and impact of flooding (Lee et al. 2023a, 2023b). Reservoirs play a crucial role in mitigating flood risks and are a cornerstone of effective flood management (Yi & Yi 2024). However, with growing environmental awareness, there is a shift from constructing new reservoirs to optimizing the use of existing ones for flood control. One of the most effective approaches to enhancing flood management is investigating the potential of pre-release strategies (Chen et al. 2013). Reservoirs can strategically pre-store or prerelease flows for flood control or water conservation (Cheng & Chau 2001; Karaboga et al. 2008). Pre-storage is retaining more water in the reservoir to reduce the risk of water shortages in case the anticipated flood does not occur (Li et al. 2010). Pre-release mitigates flood severity by providing reservoir space to hold anticipated inflows during the severe storm events (Jain et al. 1992; Mediero et al. 2007).

The pre-release approach aims to make flood control storage available before storm events, thereby enhancing flood management and reducing flood damage (Ghobadi & Kaboli 2020; Rezay Nazarzadeh et al. 2020). This strategy helps avoid or reduce the use of emergency spillway gates, which can lead to unplanned, sudden increases in releases and potentially cause severe downstream damage (Krzysztofowicz & Duckstein 1979). Pre-releasing water is a common practice in various countries to manage flood risks effectively (Hidayah Ishak & Mustafa Hashim 2018; Endo et al. 2020; Fujita & Kanae 2023). Determining the pre-release strategy requires a careful balance between reducing flood risks and ensuring water conservation. Operational rules, for example, often aim to maintain sufficient flood control storage; however, this approach may necessitate the unnecessary opening of reservoir spillway gates, potentially leading to water shortages post-flood season (Liu et al. 2006). While releasing water in advance can reduce flood impacts, it also involves uncertainties from forecasted meteorological and hydrological data (Lu et al. 2022). Enhancing the accuracy of forecasted meteorological and hydrological data to improve reservoir flood control operations is a popular topic (Beven 1993; Wei et al. 2018).

The pre-release strategy has consistently demonstrated its effectiveness in managing reservoir storage, reducing flood risks, and preventing reservoir failure during storm events (Alfieri et al. 2017). Studies comparing scenarios with and without pre-release highlight its ability to manage storage efficiently during extreme inflows (Hossain et al. 2020). Decision support tools that utilize real-time flood forecasts further enhance pre-release planning, allowing operators to act proactively before events like typhoons. This approach significantly improves flood control outcomes without compromising water supply security (Chou & Wu 2013). Pre-release strategies also contribute to evenly distributed water releases, which reduce peak flows at downstream protection points, strengthen downstream flood resilience, and maintain reservoir safety (Wei et al. 2022). Recent advancements include the development of pre-release indices and refined scheduling models, which effectively prevent floods while maintaining water supply reliability. These models provide practical tools for reservoir operators to enhance flood management and safety (Huang et al. 2024). Furthermore, hierarchical pre-release flood operation rules, incorporating flood forecasts and uncertainty information, have been proposed to optimize real-time flood control. Multi-objective optimization models that account for multiple goals – such as flood control, power generation, and navigation – are being developed and optimized using advanced algorithms, offering a comprehensive approach to cascade reservoir operations (Liu et al. 2024).

Various methodologies have been developed to assess and enhance pre-release strategies in reservoir management. Many studies emphasize the role of inflow, precipitation, and surface runoff in revising reservoir intake estimations, as seen in the formulation of release equations based on probable maximum precipitation (Hossain et al. 2020). Pre-release strategies often aim to minimize downstream flood risks while achieving targeted water levels post-flood events (Nguyen et al. 2020). Optimization techniques have also been extensively explored, employing methods such as multi-objective optimization models (Dobson et al. 2019), fuzzy optimization models (Meng & Wu 2013), optimization-simulation models (Kim et al. 2021), stochastic reservoir operation optimization models (Celeste & Billib 2009), and artificial intelligence-based models (Chaves & Chang 2008). While these models perform well under specific conditions and constraints, their practical application poses challenges. For instance, optimization models can sometimes produce infeasible or impractical solutions, complicating their use in real-world operations (Chen et al. 2024). Effective decision-making must also consider reservoir flood control capacity to avoid exceeding downstream channel limits (Jiang et al. 2022). Moreover, prompt decision-making during flood events remains a critical operational challenge (Rezay Nazarzadeh et al. 2020).

Although numerous pre-release operation plans have been developed using various criteria and successfully applied to reservoir flood control, there is limited research on reservoir operation models that determine release volumes based on inflow while also considering downstream (control point) flood conditions in real time. Few studies have explored the timing and quantity of target pre-release in detail. To address these gaps, this study proposes an innovative approach that incorporates the target reservoir elevation by determining how much the water level should be pre-adjusted (controlled) to minimize downstream damage, taking inflow into account. This approach generates a reference graph that assists in determining the target reservoir elevation after the pre-release. A key advantage of this approach is its adaptability; as the forecasted flood event changes in real-time, the reference graph is recalculated, enabling operators to adjust release volumes dynamically to achieve the target reservoir elevation.

The main objectives are (A) developing a reservoir operation model that guides the operators to reduce peak release in flood season, (B) generating reference tables and graphs to determine the target reservoir elevation after pre-release, and (C) proposing practical applications for prompt decision-making in real-time reservoir operation. We selected the Seomjingang reservoir for this study because its downstream areas are prone to flooding. The Seomjingang reservoir has a historical record of not fully utilizing its flood control storage, as its previous model did not account for target pre-release during the flood season; in contrast, this study introduces a model that incorporates pre-release strategies. By integrating forecasted inflow and downstream control point conditions, this study provides a more efficient approach to reservoir operation, including determining the maximum release at a specific water level. We conduct three case studies on severe flood events in the Seomjingang reservoir and validate the model using data from the Hapcheon reservoir.

Most reservoirs in South Korea lack established pre-release strategies to prepare for upcoming floods. The novelty of this study is developing strategies that determine the timing and quantity of target pre-release, accounting for both reservoir inflow and downstream conditions. It also provides user-friendly reference graphs and tables to support operators in making prompt and reliable reservoir operation decisions during flood events. These tools dynamically update with new forecasted inflow data at each time step, enabling real-time decision-making. The proposed pre-release strategy can potentially reduce the risk of downstream flooding and reservoir failure. The potential impact of this study lies in its contribution to improved flood management practices and reservoir operation strategies. With the newly developed pre-release strategy, the flood control storage can be fully utilized for more effective flood management.

The five major river basins in South Korea are the Han River, Nakdong River, Geum River, Yeongsan River, and Seomjin River. In South Korea, 70% of the annual precipitation falls during the flood season (June–September), making flood control particularly challenging. With two-thirds of the country covered by mountains, heavy storms quickly flow downstream in a short period. To address these challenges, South Korea built 20 multipurpose reservoirs and three flood control reservoirs. These reservoirs store water during storms and release it after the storm passes, ensuring sufficient storage for the next floods.

The Seomjin River basin has two multipurpose reservoirs (Juam and Seomjingang reservoirs) and one hydropower reservoir (Boseonggang reservoir). Juam reservoir is downstream of the Boseonggang reservoir supplies industrial and municipal water to downstream areas and serves as a flood control reservoir (60 MCM) during the flood season. Boseonggang reservoir is at the uppermost reach of the Boseong River (the largest tributary of the Seomjin River). As a hydropower reservoir, it has limited flood control capacity (1 MCM). During the heavy rainfall, it releases water through the spillway into the Boseong River.

In South Korea, the design flood for the reservoir is determined during the reservoir planning stage. The design flood ensures structural safety and prevents reservoir failure due to overtopping. The design flood is crucial for setting the flood water level and flood control storage capacities. The 200-year flood frequency is commonly used as the standard for the design flood. In some cases, we occasionally use smaller design floods than the 200-year flood frequency. The Seomjingang reservoir has a design flood of 2,953.0 CMS (200-year flood frequency), with a design release (of 1,810.0 CMS). The design release is the maximum release estimated through reservoir flood routing. The design flood for the stream is the maximum discharge that flows in the channel considering the upstream reservoir. All downstream sections of the Seomjin River use a 100-year frequency for design flood. The use of a 200-year flood frequency for the reservoir, compared to a 100-year flood frequency for downstream rivers, can cause conflicts in flood management. The flood stages for streams are classified into concern, caution, alert, and emergency stages (Table 2). The concern stage is when water reaches a level that overtops the intake facilities. The caution stage is the level at which a flood watch is issued. The alert stage is the level at which a flood warning is issued. The emergency stage is the river level that exceeds the designed flood level.

FOM IV makes releases based on predetermined amounts by the reservoir level alone, disregarding inflow volume. This method uses a predefined spillway rule curve (SRC), which is based on the design flood. The SRC considers the size of the reservoir, flood control capacity, and spillway gate release capabilities to determine release for varying reservoir elevations. However, this approach may not accurately reflect actual flood conditions if the inflow significantly differs from the design flood.

We selected three significant flood events from 2010 (case study 1), 2011 (case study 2), and 2020 (case study 3) due to their severity and the availability of data at 10-min intervals. These events were analyzed to evaluate the applicability of the developed model and reference graph, ensuring they can effectively guide pre-release strategies for severe floods. By applying the model to these cases, we demonstrated its potential for practical use in flood management.

Input data includes reservoir characteristics, a table of storage by reservoir elevation, and a table of maximum release by reservoir elevation. Upon selecting the name of a multipurpose reservoir, the system internally retrieves the unique code of the reservoir and loads the data for the selected reservoir stored within the model. The Seomjingang reservoir has daily data from 1975, hourly data from 1992, and 10-min data from 2010. The developed model uses 10-min inflow data but can also accommodate hourly inflow data.

The discharge data for the flood events was provided by the Flood Control Office (https://www.hrfco.go.kr/river/resources01.do). The public can access 10-min release data through the Han River Flood Control Office (http://hrfco.go.kr/main.do). The unique codes of the reservoirs and historical hourly reservoir inflow data are available in the National Water Resources Management Information System (http://www.wamis.go.kr/) (Table 4). Input data for this article is publicly accessible (Yi & Yi 2024).

Based on user input, the model loads reservoir characteristics including low, restricted, normal high, flood, and maximum elevations. The table of storage by elevation lists the relation between storage volumes (MCM) and reservoir elevations (EL.m). Updated every 10 years following sedimentation surveys, this table is accessible on the Korea Water Resources Corporation (https://www.water.or.kr/). The table of maximum release by elevation contains maximum release based on the reservoir elevation, influenced by the location of spillways and their types.

The FOM model generates both common and additional outputs. The common outputs include reservoir operation results for the specified inflow period, such as releases, reservoir elevations, storage volumes, peak release, reference graph (peak release vs. target reservoir elevation) and a reference table. When users use functions 1 and 2 in FOM II, it produces the ratio as an additional output.

A reference graph is used to determine the target reservoir elevation after pre-release. This is a plot of the peak release on the x-axis and the target reservoir elevation on the y-axis. Once the reservoir elevation is reduced to the target elevation through pre-release, operators can make the corresponding peak release to make full use of the flood control storage. The following are the example steps for generating the reference graph and table for the Seomjingang reservoir. First, we run FOM V via an iterative process, using 100.0 CMS peak-release intervals as an example, to generate a series of target reservoir elevations after pre-releases. Second, we turn these results into the reference graph and table which are updated when new inflow forecasts become available. Third, peak release from the reservoir is determined by considering the downstream flood conditions. Fourth, from the reference graph, we read the target reservoir elevation (y-axis) using the peak release (x-axis). Fifth, pre-releases are made until the actual reservoir elevation reaches the designated target reservoir elevation. Finally, operators use the reference graph to adjust pre-releases based on the updating inflow forecasts and downstream flood conditions.

The final product is the reference graph and a corresponding reference table. The reference graph plots the peak release against the target reservoir elevation. We define the target reservoir elevation as the elevation after it has been lowered through pre-release, prior to starting flood operations. Flood control in real situations can be challenging for the reservoir operators as it requires urgent decision-making during the abrupt flood event.

We analyzed and presented the results for the Seomjingang reservoir using FOM V. The final outputs, including the reference graph and table, were generated at 100 CMS intervals through an iterative process. We analyzed three significant flood events from 2010 (case study 1), 2011 (case study 2), and 2020 (case study 3) for the Seomjingang reservoir and validated the model using the 2020 flood event in the Hapcheon reservoir.

We selected the Hapcheon reservoir as a new site for model validation. The Hapcheon reservoir is a multipurpose reservoir located in the Nakdong River Basin. It supplies water to downstream areas and manages flood control using the FOM II method. Like the Seomjingang reservoir, the Hapcheon reservoir released a significant volume of water during the 2020 flood, causing severe flood damage in downstream regions. To validate the developed model, the 2020 flood event at the Hapcheon reservoir (7–9 August 2020) was selected for analysis.

We developed a reservoir operation model to guide operators in reducing peak releases from reservoirs during the flood season. The model generates reference tables and graphs through an iterative process to determine the target reservoir elevation after pre-release and offers a practical application for prompt decision-making in real-time reservoir operations. Once operators identify the target reservoir elevation, they initiate pre-release operations until the desired elevation is reached. The newly developed model guides operators in determining the target reservoir elevation for pre-release by incorporating forecasted inflow and downstream flood conditions.

The developed reservoir simulation model can run five FOMs. We used the historical inflow data of 2010, 2011, and 2020 from the Seomjingang reservoir to demonstrate developing the reference graph and table using FOM V. These final products were used to determine the target reservoir elevation to ensure that the peak release does not exceed the design release while taking the downstream flood conditions into account. Once operators determine the target reservoir elevation, they start the pre-release until they reach that elevation before beginning flood operations. When updated forecasted inflow data become available, such as every 10-min, the model recalculates the target reservoir elevation. Based on this new information, we then adjust the release accordingly. Overforecasting the inflow could lead to excessive pre-release, lowering reservoir levels too much and risking the water supply during subsequent dry periods. This approach offers real-time decision-making that allows for proactive reservoir operation, helping to reduce the downstream flood damage and ensuring an adequate water supply for the upcoming dry season.

This study has several limitations and areas for future improvement. One limitation is the lack of consideration for flow routing from the reservoir to the downstream control point, as the downstream flow was monitored at a water level station. However, in cases where no monitoring stations are available downstream, incorporating flow routing becomes essential. Additionally, uncertainties in this study stem from the accuracy of inflow forecasting and the exclusion of the time required to open and close spillway gates, which could impact real-time applicability (Kim et al. 2018). Future research should focus on incorporating flow routing to better capture the dynamics of reservoir releases reaching the control point. Furthermore, developing a new FOM is essential, as different FOMs can lead to varying levels of downstream damage from the same flood event.

The effectiveness of the pre-release varies significantly across different factors such as peak inflows, spillway release capacities, locations, downstream design flood, reservoirs, and basin characteristics. One of the uncertainties in the pre-release strategy comes from unreliable meteorological and hydrological forecasting. When these forecasts inaccurately predict the storm events, the reservoir release may be underestimated or overestimated. Future research includes obtaining reliable forecasting data as early as possible prior to a flood event.

Flood control during actual flood events poses challenges for reservoir operators due to the need for rapid decision-making. We developed an approach that provides operators with a tool for making prompt operation decisions to better protect downstream regions from flooding and prevent dam failure during storm events. This study not only improves how reservoirs are managed during the flood event but also contributes to broader discussions on flood management, marking a shift towards more secure, efficient, and flexible reservoir operation strategies.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

E.L. conceptualized the study, conducted data curation, formal analysis, investigation, methodology development, and resource management, and also handled project administration, software development, validation, visualization, writing of the original draft, and review and editing of the article. S.Y. contributed to resources, visualization, writing of the original draft, and review and editing of the article. J.J. was responsible for methodology, resources, software, and also took part in the review and editing of the article. J.H. handled software and participated in the review and editing of the article. S.L. contributed to resources and the review and editing of the article. J.Y. managed resources and participated in the review and editing of the article. J.Y. was involved in the conceptualization, methodology, project administration, resources, software, supervision, writing of the original draft, and review and editing of the article.

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

The authors declare there is no conflict.