Assessing Flood Risks of Small Reservoirs in Huangshan, Anhui Province, China

Abstract

Based on the regional disaster system theory, this study constructed a comprehensive flood risk indicator system for small reservoirs, covering the entire flood disaster process from three dimensions: hazard, vulnerability, and exposure. The analytic hierarchy process (AHP) and entropy weight method (EW) were used to determine indicator weights, and a risk assessment was conducted for small reservoirs in Huangshan City, Anhui Province, China. The results indicate that most reservoirs exhibit medium–low overall risk, yet distinct localized high-risk zones exist. High-economic-density areas such as Tunxi District, the central–eastern parts of Huangshan District, and the central and eastern parts of Qimen County have become high-risk clusters due to prominent exposure indicators (numbers of villages and medical facilities). Reservoirs in western and northern regions exhibit higher hazard levels, primarily driven by rainfall and catchment areas. Dam height and reservoir capacity are the main factors affecting vulnerability, with no significant spatial clustering for high-vulnerability reservoirs. Through the decoupling of three-dimensional indicators, this study reveals the differentiated driving mechanisms of “hazard–vulnerability–exposure,” providing a scientific basis for optimizing flood control engineering (e.g., reservoir capacity expansion, dam reinforcement) and risk zoning management (e.g., emergency evacuation planning in high-exposure areas) for small reservoirs.

1. Introduction

Reservoirs are an essential infrastructure for national economic and social development, playing key roles in flood control [1], water supply [2], irrigation [3], power generation [4], and ecological regulation [5]. By the end of 2023, China had built 94,877 reservoirs, of which 89,811 were small (total storage capacity between 0.1 and 10 million m³), accounting for 94.7% of all reservoirs [6] and covering 90% of county-level administrative regions. Multiple flood events result in loss of life and property [7], especially dam breach floods [8,9]. Small reservoirs are the most numerous, the most widely distributed, and the most prone to dam failures. In the over 60 years since the founding of the People’s Republic of China, 3462 reservoirs have been breached nationwide, of which 3336 are small reservoirs, accounting for 96.4% of all breaches [10]. From 2000 to 2021, a total of 88 dam breach incidents occurred nationwide, with breaches of small reservoirs accounting for as much as 90.1% [11]. Compared with medium- and large-scale reservoirs, small reservoirs also suffer from generally lower construction standards, poorer operation and maintenance practices, and a higher proportion of impaired or hazardous facilities. Most small reservoirs were built between the 1950s and 1970s. At the time of construction, they were affected by funding shortages, weak technical capabilities, and a lack of hydrological data. Consequently, reservoirs were built without rigorous design and construction processes, often following a “three-simultaneous” approach of surveying, designing, and constructing concurrently. Moreover, over 95% of the small reservoirs built during that period were earth–rock fill dams designed primarily for irrigation and water supply without detailed flood control or multipurpose benefit analyses, and thus essentially lack flood control functionality [12]. After completion, funding, staffing, technical support, operation, and maintenance, standards for these reservoirs remained far below those of medium- and large-scale reservoirs. After decades of operation, most small reservoirs have reached or even exceeded their service life, with serious safety and integrity issues. In recent years, the government has invested substantial funds in the hazard mitigation and reinforcement of reservoirs, achieving significant results. However, due to the large number of reservoirs, dam breach incidents still occur intermittently. The 2021 breach of the Zenglongchang Reservoir in Inner Mongolia and the 2022 mountain flood disaster in Sheyuegou, Qinghai, caused severe casualties and economic losses. Therefore, there is an urgent need to conduct flood risk research on small reservoirs.

Risk analysis and assessment support flood disaster management were carried out through three mechanisms: (1) identifying system-specific threats and vulnerabilities, (2) evaluating the probability and magnitude of risk events, and (3) informing non-structural interventions such as early warning systems and land use planning [13]. The formation of flood risk in small reservoirs involves the interaction of three levels: upstream catchment characteristics, reservoir inherent defects, and downstream safety impacts [14]. Because small reservoirs are located within relatively small catchment areas, their runoff generation processes typically feature rapid concentration, short inflow lag times, and swift reservoir water level rises [15]. However, small catchments exhibit high spatial heterogeneity, and the rainfall–runoff generation mechanisms are complex and highly nonlinear. This makes reservoir inflow flood forecasting difficult to perform accurately [16]. Furthermore, data collection is challenging: small reservoir catchments often have only a few rainfall gauges and lack necessary hydrological stations, resulting in incomplete inflow and outflow data that compromise the accuracy of risk assessments. At the reservoir-intrinsic risk level, most small reservoirs have low construction standards and incomplete design documentation, with key parameters—such as the reservoir level–storage and level–discharge relationships—remaining undefined [17]. In addition, most small reservoirs lack effective gate control and spillway facilities. Some dams exceed 15 m in height, presenting clear structural safety hazards that increase the risk of engineering failure. Small reservoirs are mostly located in mountain flood control zones; in the event of large-volume discharges or dam breaches, they can inflict devastating impacts on downstream villages, towns, and critical infrastructure [18]. Especially under the backdrop of intensifying global climate change [19], extreme weather events are occurring more frequently, rainfall intensity has increased significantly. Therefore, short-duration heavy rainfall in localized areas can easily exceed the designed storage and discharge capacities of small reservoirs. It not only exacerbates the frequency and extent of flood disasters, but also poses new challenges to the flood safety of small reservoirs.

Many scholars at home and abroad have employed methods based on historical data [20], hydrological simulation [21], indicator systems [22,23], machine learning [24], and others to evaluate flood risk in reservoirs. Wu proposed risk analysis model integrates with the advanced first-order and second-moment method in the Keelung River [25], and Dutta used a physically based distributed hydrological model for flood inundation and a geographical information system (GIS)-based raster model for flood loss estimation in the Ichinomiya River [26]. Huang applied the Latin hypercube sampling method for flood forecasting uncertainty in the Three Gorges Reservoir [27]. However, most of these assessments target medium- and large-scale reservoirs or river basins, with relatively few studies focusing on small reservoirs, and they generally require high data precision. Small reservoirs generally suffer from data deficiencies and insufficient data availability. Accurately representing reservoirs in basin- and watershed-scale models remains challenging due to limited data on small reservoir operations [28]. Some studies have assessed the flood risk of reservoirs by examining the disaster-generating environment, dam structure, reservoir operation, flood discharge measures, and downstream resilience [29,30]. However, these approaches fail to consider the risk factors and events spanning the entire flood disaster chain. This fragmentation of the risk chain hinders comprehensive flood risk management by decision-makers. Hence, a method that requires minimal data, offers satisfactory accuracy, and enables rapid assessment is warranted.

Huangshan City is located in the central heavy-rainfall zone of southern Anhui Province, China, with mountainous terrain covering 80% of its area and a dense river network. Since 1950, numerous hydraulic projects—including a large number of small reservoirs—have been constructed, making the region a typical flood-prone area. The spatial distribution of small reservoirs overlaps with high-risk zones for flash floods, particularly in Xiuning and Shexian Counties. Huangshan City has repeatedly suffered severe flood events in recent years. In June 2024, 179,000 residents were affected, and direct economic losses exceeded CNY 500 million. In July 2020, a flash flood inundated Shexian County overnight and affected 206,000 people. The event severely damaged power and road networks, municipal facilities, healthcare, and schools. The National College Entrance Examination was postponed, and economic losses surpassed CNY 3.5 billion. Although no small reservoir breach has yet occurred in Huangshan, the city has frequently faced high-water-level operations, spillway risk, and secondary flood hazards. For example, in June 2024, Yaoyuan Reservoir in Tunxi District exceeded its flood control limit, directly affecting over 40 downstream households and forcing the temporary closure of the historic Tunxi Old Street cultural district. Based on this background, this study takes Huangshan City as a case study, and proposes a flood control risk identification and assessment approach for small reservoirs. Our approach for evaluating the overall risk comprehensively considers upstream catchment characteristics, small reservoirs, and downstream safety. By accounting for the incomplete data typical of small reservoirs, it constructs a comprehensive risk evaluation system based on hazard, vulnerability, and exposure. This study aims to establish a flood risk assessment method suited to small reservoirs and to provide technical support and decision making assistance for risk identification and routine safe operation management.

2. Materials and Methods

This study developed a flood risk assessment framework for small reservoirs by integrating meteorological, socio-economic, and geospatial data. The framework systematically decomposes dam breach flood risk factors through three analytical dimensions with an evaluation index system established based on regional disaster system theory [31]. Based on the analytic hierarchy process (AHP) and entropy weight method (EW), the comprehensive weights of each index were calculated. This study established a dam breach flood risk assessment model, including hazard, vulnerability, and exposure. The research methodology and framework are illustrated in Figure 1.

2.1. Index System Construction

2.1.1. Flood Risk Identification

Small reservoirs intercept and store runoff generated within their upstream catchment areas, serving downstream needs, such as water supply, agricultural irrigation, and ecological development. This study identified three risk factors. (1) Upstream catchment characteristics: precipitation, catchment area, and topographic conditions are the main factors affecting runoff. Precipitation causes a sharp increase in surface runoff and water volume, which has an impact on flood generation within the catchment [32]. Topography regulates hydrological processes, while the catchment area determines the total volume of water converging into the reservoir. Advanced rainfall monitoring infrastructure can make precipitation tracking and the early detection of rainfall anomalies more precise, and support precise disaster risk management. Land use and land cover (LUCC) mainly influence rainfall–runoff processes. According to the data obtained from the previous investigation at the hydrological bureau, the vegetation conditions of the small reservoir watersheds in the study area are relatively good and stable for a long time. There is a relatively stable relationship between rainfall and runoff. The differences in the impact of the underlying surface conditions on the runoff generation and convergence of different reservoir basins are relatively small, and mainly affected by rainfall conditions and topographic conditions. Therefore, land use change indicators were not selected. However, in the study area encompassing small reservoirs, the underlying surface exhibits well-vegetated conditions that have remained stable over extended periods. This stability has established a consistent rainfall–runoff correlation, with minimal variation in runoff generation and concentration processes across different reservoir watersheds attributable to underlying surface heterogeneity. Consequently, the dominant controlling factors are rainfall characteristics and topographic parameters rather than LUCC dynamics, justifying the exclusion of land use change indicators in this analysis. (2) Reservoir defects: the risks of the reservoir itself are from structural and functional vulnerabilities, specifically including dam stability, storage capacity, dam height, completeness of flood-releasing facilities, and the level of operation and maintenance management. Dam breach discharge, reflecting the hydraulic characteristics following a dam failure, manifests the reservoir’s engineering deficiencies under extreme conditions. These factors collectively determine the reservoir’s resilience against external hydrological stressors. (3) Downstream impacts of dam breach: downstream areas act as the receptors of dam-breach flood risk, which is measured by the spatial distribution of affected villages and critical infrastructure. The more villages and towns affected by disasters, the more inter-jurisdictional coordination bodies are required, which increases the complexity of collaboration. And some critical areas such as schools, hospitals, and nursing homes are places where vulnerable groups are concentrated. Due to the difficulty of evacuation in emergencies and the potential public safety hazards, these areas require special attention.

2.1.2. Flood Risk Assessment Indicator System

The selection of flood risk evaluation indicators for small reservoirs follows the principles of systematicity, scientific rigor, data availability, and representativeness (Figure 2). Building on the scientific foundation of existing studies and following disaster system theory, we divided indicators into three criterion levels: hazard, vulnerability, and exposure.

In this study, the hazard is quantified by both the intensity of the disaster and the environmental conditions under which it forms [33]. The specific indicators comprise five metrics: precipitation station density, catchment area, topographic wetness index (TWI), annual average precipitation, and annual maximum 24-hour precipitation.

• Precipitation station density: This represents the region’s capacity for precipitation data collection. A higher density corresponds to greater accuracy in rainfall monitoring and flood forecasting. It is expressed as the number of precipitation stations per unit area and calculated as the ratio of the number of rainfall stations to the respective county or district.
• Catchment area: The size of the catchment area determines the total runoff volume following precipitation. The catchment area of each reservoir was calculated in ArcGIS by coding in the Python 2.7 extension module based on runoff-concentration principles.
• TWI: TWI is a key metric for quantifying the impact of terrain variability on hydrological processes. Higher TWI values indicate that the area has a larger slope catchment area or lower hydraulic gradient and is more susceptible to flooding hazards [32]. The calculation formula is shown by Equation (1):

  $$\mathrm{T}\mathrm{W}\mathrm{I}=\ln \left({\displaystyle \frac{\alpha }{\tan \beta }}\right),$$

  (1)

  where $\alpha$ represents the specific catchment area and $\beta$ is the catchment slope gradient (expressed in radians).
• Annual average precipitation and annual maximum 24-hour precipitation: Annual average precipitation reflects regional hydrological baselines, while annual maximum 24-hour precipitation is an indicator of the intensity of extreme rainfall events. Both parameters were vectorized, respectively, based on Huangshan City’s multi-year average precipitation contour map and multi-year average maximum 24-hour precipitation contour map, and then obtained by spatial interpolation using the Topo to Raster command. By updating indicators such as the average annual precipitation or the annual maximum 24-hour precipitation, the dynamic assessment of the flood risk of reservoirs can be achieved.

Vulnerability indicators include dam height, storage capacity, dam types, flood-releasing facilities, operation and maintenance management level, and dam-breach discharge.

• Dam height: Dam height determines the potential energy release level of the reservoir. The higher the dam height, the greater the flood energy after the dam failure.
• Storage capacity. Reservoir capacity denotes water storage capability; larger capacity results in a greater flood volume release during a breach.
• Dam types: Dam types are classified by structure into concrete dams, masonry dams, and earth rock fill dams, with stability-based grading assigned values of 3, 2, and 1, respectively.
• Flood-releasing facilities: Flood-releasing facilities are first categorized by the presence of gates, with gated reservoirs assigned the highest safety value of 3; for ungated reservoirs, those with diversion or flood tunnels are assigned 2, and those with only standard spillways are assigned 1.
• Operation and maintenance management refers to the routine management, maintenance, and upkeep of reservoirs to ensure their safe, stable, and efficient operation. It encompasses reservoir operation scheduling, safety management, ecological protection, and the maintenance and renewal of auxiliary facilities. Due to the lack of detailed operation and maintenance data, agricultural, forestry, and water expenditure, including projects directly related to reservoir operation and maintenance, was used as a proxy. To minimize interference from annual expenditure fluctuations, the average agricultural, forestry, and water expenditure over the past five years was used as the evaluation data.
• Dam breach discharge refers to the maximum downstream flow intensity during a reservoir failure. It inversely reflects the comprehensive vulnerability arising from structural defects, management shortcomings, and environmental adaptability [34,35,36], serving as the “pressure threshold” in the disaster chain.

Exposure is defined as the people, property, systems, or other entities that may be harmed in a hazard-prone area [37]. Villages, as population aggregation units and carriers of social functions, are counted to directly reflect the spatial density and distribution extent of receptors within the dam breach flood threat zone. Nursing homes, schools, and hospitals reflect the distribution of critical public facilities. Hospitals, as essential lifeline infrastructure, could face system paralysis if inundated, exacerbating secondary disasters. Therefore, exposure is represented by the number of villages, nursing homes, schools, and hospitals located within a 3 km radius of the reservoir, with counts conducted in ArcGIS.

2.2. Indicator Weight Processing

AHP and EW are the most common methods for risk assessment [38,39,40,41]. AHP hierarchical analysis is a multi-criteria, multi-objective decision making method proposed by Thomas L. Saaty, an American operations researcher, in the 1970s [42]. The decision problem is decomposed into objective, criterion, and program layers, with element weights determined through pairwise comparisons, enabling the ranking of decision programs based on their advantages and disadvantages. Since respondents assign relative importance to each indicator, human subjectivity significantly influences the results [43]. EW is an evaluation method by objectively assigning weights to the degree of variability of the values of each indicator [38]. It can effectively avoid the interference of human subjective factors, but due to overreliance on objective data, it may lead to the inconsistency of the results with the real experience.

The combination of AHP and EW addresses two critical limitations in conventional risk assessment: subjective biases from expert judgments and overreliance on datasets. By balancing qualitative expertise with quantitative data-driven adjustments, the results can be more accurate and consistent with the actual situation [44]. It has a wide range of applications in the field of risk assessment [45]. Therefore, in this paper, the combination of AHP and EW weighting was used to evaluate the risk of dam failure in small reservoirs.

2.2.1. Standardization

Each evaluation index cannot be directly calculated and needs to be standardized due to the differences in units and dimensions. Firstly, the evaluation indicators were classified into two categories—positive and negative—based on their respective impacts on flood risk. The classification results are shown in

Table 1. Evaluation indicators for the positive or negative impact on the risk of dam breach floods.

Objective Level	Standardized Level	Index Level	Index Attribute
Flood risk of small reservoir dam breach	Hazard (H)	Precipitation station density	N
		Catchment area	P
		TWI	P
		Annual average precipitation	P
		Annual maximum 24-hour precipitation	P
	Vulnerability (V)	Dam height	P
		Storage capacity	P
		Dam types	N
		Flood-releasing facilities	N
		Operation and maintenance	N
		Dam breach discharge	P
	Exposure (E)	Number of villages within 3 km	P
		Number of hospitals within 3 km	P
		Number of schools within 3 km	P
		Number of nursing homes within 3 km	P

. Positive indicators (e.g., dam height) were positively correlated with increased risk, while negative indicators (e.g., maintenance level) were inversely related. This research used the extreme value method to standardize the index data.

For positive indicators (P), Equation (2) was used:

$$Y_{ij}={\displaystyle \frac{{\mathrm{X}}_{ij}-X_{min}}{X_{max}-X_{min}}},$$

(2)

For negative indicators (N), Equation (3) was used:

$$Y_{ij}={\displaystyle \frac{{\mathrm{X}}_{max}-X_{ij}}{X_{max}-X_{min}}},$$

(3)

where $Y_{ij}$ is the standardized value in the range [0, 1]. ${\mathrm{X}}_{ij}$ is the original value of the $j$-th evaluation index of the $i$-th reservoir, $i$ = 1, 2, …, m; $j$ = 1, 2, …, n, n and m are the number of indexes and the number of samples, respectively. $X_{max}$ and $X_{min}$ are the maximum and minimum values of the $j$-th evaluation index, respectively.

2.2.2. Analytical Hierarchy Process

First, we constructed a three-level hierarchical model comprising the goal (flood risk), the three criteria (hazard factors, vulnerability factors, and exposure factors), and their corresponding subcriteria (Figure 1). A judgment matrix for each criterion was then assembled by performing pairwise comparisons among its factors. Ten senior experts were invited to score these pairwise comparisons using the standard Saaty 1–9 scale. In this scale, 1 denotes equal importance; 3, 5, 7, and 9 represent, respectively, “slightly more important”, “clearly more important”, “strongly more important”, and “extremely more important” for the first factor over the second; and the even numbers 2, 4, 6, and 8 serve as intermediate values.

Second, the maximum eigenvalue of the judgment matrix and its corresponding eigenvectors are calculated. The third step calculates the consistency of the result. The consistency of the judgment matrix can be verified by the consistency ratio (CR). The CR calculation formula is shown in Equation (4):

$$\left\{\begin{array}{c}CR={\displaystyle \frac{CI}{RI}}\\ CI={\displaystyle \frac{{\lambda }_{max}-\mathrm{n}}{\mathrm{n}-1}}\end{array}\right.,$$

(4)

where $CR$ is the consistency ratio, $CI$ is the consistency index, and $RI$ is the stochastic consistency index. ${\lambda }_{max}$ is the maximum eigenvalue of the matrix, and n is the number of indicators/subindicators. If $CR$ is less than 0.1, the consistency test is artificially passed, otherwise, the comparison matrix needs to be rebuilt [46].

2.2.3. Entropy Weight Method

Information entropy can quantify an indicator’s uncertainty or disorder: the greater the indicator’s variation, the higher its information entropy, and the smaller its weight in the system. The information entropy of the j-th evaluation indicator is calculated by Equation (5):

$$\left\{\begin{array}{c}{e}_j=-{\displaystyle \frac{1}{\ln \left(m\right)}}{\sum }_{i=1}^m{p}_{ij}\ln p_{ij}\\ p_{ij}={\displaystyle \frac{Y_{ij}}{{\sum }_{i=1}^m{Y}_{ij}}}\end{array}\right.,$$

(5)

where $e_j$ represents the information entropy of the j-th indicator, and $p_{ij}$ denotes the proportion of each sample’s value among all values for indicator $j$. To avoid $p_{ij}$ being zero and affecting subsequent calculations, 0.001 is uniformly added to the standardized value,$Y_{ij}$; thus, the information entropy formula can be expressed as Equation (6):

$$\left\{\begin{array}{c}{e}_j=-{\displaystyle \frac{1}{\ln \left(m\right)}}{\sum }_{i=1}^m{p}_{ij}\ln p_{ij}\\ p_{ij}={\displaystyle \frac{Y_{ij}+0.001}{{\sum }_{i=1}^m{(Y}_{ij}+0.001)}}\end{array}\right.,$$

(6)

Entropy weights are computed using Equation (7):

$${\omega }_j={\displaystyle \frac{1-e_j}{k-{\sum }_{j=1}^n{e}_j}},$$

(7)

where ${\omega }_j$ is the weight of the j-th, and $k$ is the number of indicators. The larger the ${\omega }_j$ is, the more important the indicator is in decision making.

2.2.4. Integrated Weight Method

After determining weights via AHP and EW, the integrated weight for each indicator can be calculated using Equation (8):

$$W_j={\displaystyle \frac{{\omega }_j{*\omega }_j^{\prime }}{\sum _{j=1}^n{\omega }_j{*\omega }_j^{\prime }}},$$

(8)

where $W_j$ represents the integrated weight of the j-th indicator, ${\omega }_j$ is the AHP-derived weight, and ${\omega }_j^{\prime }$ is its EW-derived weight.

2.2.5. Flood Risk Assessment

The purpose of risk assessment is to identify current or future high-risk areas [34,47], support mitigation measures and decision-making [48], and evaluate the effectiveness of mitigation measures under potential hazard scenarios [49,50]. Many organizations, institutions, groups, and researchers have established various risk assessment frameworks, commonly including “Hazard–Vulnerability (H-V)” [51,52], “Hazard–Exposure–Vulnerability (H-E-V)” [53,54,55], and “Susceptibility–Vulnerability–Recovery (S-V-R)” [56]. According to the Intergovernmental Panel on Climate Change definition of risk [57], this study adopts the “Hazard–Exposure–Vulnerability (H-E-V)” model for reservoir breach flood risk assessment [41], calculated by Equations (9)–(12):

$$R=FHI+FVI+FEI$$

(9)

$$FHI={\sum }_{j=1}^5{W}_{Hj}\times Y_{Hj},$$

(10)

$$FVI={\sum }_{j=1}^6{W}_{Vj}\times Y_{Vj},$$

(11)

$$FEI={\sum }_{j=1}^4{W}_{Ej}\times Y_{Ej},$$

(12)

where $R$ is the comprehensive flood risk of reservoir breaches; and $FHI$, $FVI$, and $FEI$ represent the risks of hazard, exposure, and vulnerability. $W_{Hj}$, $W_{Vj}$ and $W_{Ej}$ are the values of hazard, exposure, and vulnerability indices, and $Y_{Hj}$, $Y_{Vj}$ and $Y_{Ej}$ represent the standardized values for each subindicator.

2.3. Dam Breach Analysis

The Schoklitsch formula is an empirical equation used to calculate the instantaneous dam-breach flood discharge by estimating peak flow from the dam’s geometric dimensions and water level conditions. Due to its minimal parameter requirements and ease of data acquisition, it has seen widespread application in hydraulic engineering [34]. Small reservoirs in the Huangshan region are predominantly earth–rock fill dams; assuming an instantaneous breach, the breach flood discharge is calculated as Equation (13):

$$Q_{max}={\displaystyle \frac{8}{27}}\sqrt{g}{\left({\displaystyle \frac{B}{b}}\right)}^{{\displaystyle \frac{1}{4}}}b{H}^{{\displaystyle \frac{3}{2}}},$$

(13)

where $Q_{max}$ is the maximum breach discharge (m³/s), $g$ is the gravitational acceleration (9.8 m³/s), $B$ is the dam crest length (m), $b$ is the average breach width (m) and set b at 60% of the dam length, and $H$ is the upstream water depth at the dam crest.

3. Case Study

3.1. Study Area

Huangshan City is located in southern Anhui Province, between 117°02′–118°55′ E and 29°24′–30°24′ N (Figure 3). It currently administers three urban districts (Tunxi, Huangshan, and Huizhou) and four counties (She, Xiuning, Qimen, and Yi). The terrain is predominantly mountainous and hilly, rising gradually from north to south, with limited arable land. The Huangshan region is traversed by numerous rivers, including the Qingyi and Qiupu systems that flow into the Yangtze River, the Chang and Lean systems that discharge southwest into Poyang Lake, and the Xin’an River system that drains eastward into the Qiantang River Basin. The Xin’an River Basin covers 5615 km² (57.3% of the city’s area), while the Yangtze River Basin accounts for 4192 km² (42.7%). Huangshan City lies within the subtropical monsoon climatic zone, characterized by mild temperatures and distinct seasons. The annual average temperature is 16 °C, with minimal interannual variation. Influenced by prevailing northeast and southwest monsoon winds, the city’s annual average precipitation ranges from approximately 1700 to 2000 mm. Rainfall is predominantly concentrated from April to July, accounting for 56% of the annual total [58]. There are currently 184 small reservoirs in the city, most of which were constructed between the 1950s and 1970s. These dams are primarily earth–rock fill, with spillways typically ungated; over one-third have dam heights exceeding 15 m.

After decades of operation, some reservoirs have entered a state of structural distress, and a breach could have severe consequences for downstream areas. Therefore, understanding the flood risk associated with small reservoir breaches is critically important for regional safety and development.

3.2. Data

This study focuses on small reservoirs in Huangshan City that are still in normal operation, collecting environmental, socio-economic, and basic geographic data of the study area, as well as the reservoirs’ own characteristics and engineering protection measures. Meteorological data include the long-term average precipitation and the long-term maximum 24-hour precipitation in 2023 for Huangshan City, sourced from the Anhui Province Water Resources Bulletin. Precipitation stations in 2023 were obtained from the Hydrological Yearbook. Digital Elevation Model (DEM) data in 2019 were acquired from the National Earth System Science Data Center (https://www.geodata.cn, accessed on 28 December 2024) at a 30 m resolution. Reservoir characteristic data—including dam height, storage capacity, dam types, and spillway measures—were obtained from field measurements and statistics. Data on funding for reservoir operation and maintenance in 2023 were sourced from the Huangshan City Statistical Yearbook. Data on nursing homes in 2024 were obtained from eldercare institutions registered by the Ministry of Civil Affairs of the People’s Republic of China (https://yanglao.mca.gov.cn/, accessed on 18 February 2025), including information on location, institution name, unified social credit code, address, and number of beds. Using the names of eldercare institutions nationwide, spatial point data were derived through a reverse address-lookup coordinate tool. Point of Interest (POI) data were obtained from OMS (openstreetmap.org/, accessed on 16 November 2022), from which 46,474 POI records for Huangshan City in 2022 were collected, and locations of kindergartens, primary schools, secondary schools, vocational schools, and hospitals were extracted. The data sources are shown in

Table 2. Data sources.

List of Data Source	Time	Source	Note
Precipitation station	2023	The Hydrological Yearbook	Historical data
Digital Elevation Model (DEM)	2019	The National Earth System Science Data Center (https://www.geodata.cn, accessed on 28 December 2024)	Historical data 30 m
Annual average precipitation	2023	Anhui Province Water Resources Bulletin	Historical data
Annual maximum 24-hour precipitation	2023
Storage capacity	2024	Field measurements and statistics	Field-based data
Dam types	2024
Dam height	2024
Flood-releasing facilities	2024
Operation and maintenance	2023	Huangshan Statistical Yearbook 2024	Historical data
Villages	2021	Based on publicly available administrative division data	Historical data
Nursing homes	2024	the Ministry of Civil Affairs of the People’s Republic of China (https://yanglao.mca.gov.cn/, accessed on 18 February 2025)	Historical data
Schools and hospitals	2022	OMS (openstreetmap.org/, accessed on 16 November 2022)	Historical data

.

4. Results

4.1. Indicator Weight and Analysis

Based on the established hierarchical structure, indicator weights were calculated following the AHP and EW algorithms, and all AHP-derived weights passed the consistency test. The calculations yield the respective weights for each indicator, as shown in

Table 3. Comprehensive weight determination.

Objective Level	Standardized Level	Standardized Level Weight by AHP	Standardized Level Weight by EW	Index Level	Index Attribute	Index Level Weight	${\mathit{\omega }}_{\mathit{j}}$	${\mathit{\omega }}_{\mathit{j}}^{\prime }$	${\mathit{W}}_{\mathit{j}}$
Flood risk of small reservoir dam breach	Hazard (H)	0.3430	0.3154	Precipitation station density	N	0.1336	0.0458	0.0105	0.0090
				Catchment area	P	0.1104	0.0379	0.2260	0.1600
				TWI	P	0.0600	0.0206	0.0430	0.0165
				Annual average precipitation	P	0.2862	0.0982	0.0150	0.0275
				Annual maximum 24-hour precipitation	P	0.4099	0.1406	0.0209	0.0548
	Vulnerability (V)	0.4453	0.2659	Storage capacity	P	0.1043	0.0464	0.1572	0.1365
				Dam types	N	0.2422	0.1079	0.0067	0.0136
				Dam height	P	0.1559	0.0694	0.0165	0.0215
				Flood-releasing facilities	N	0.1260	0.0561	0.0057	0.0059
				Operation and maintenance	N	0.0743	0.0331	0.0237	0.0146
				Dam breach discharge	P	0.2973	0.1324	0.0561	0.1388
	Exposure (E)	0.2117	0.4187	Number of villages within 3 km	P	0.3115	0.0659	0.0185	0.0228
				Number of nursing homes within 3 km	P	0.2406	0.0509	0.1503	0.1431
				Number of schools within 3 km	P	0.1556	0.0329	0.0995	0.0613
				Number of hospitals within 3 km	P	0.2923	0.0619	0.1504	0.1740

. At the criterion level, AHP-derived weights are hazard (0.343), vulnerability (0.4453), and exposure (0.2117), indicating that the reservoir’s structural vulnerability is the primary factor influencing flood risk. At the indicator level, the annual maximum 24-hour precipitation (0.1406) and annual average precipitation (0.0982) emerged as dominant predictors of flood hazard intensity. Dam breach discharge (0.1324) and dam types (0.0694) are the key indicators of vulnerability. The number of villages within 3 km (0.0659) and the number of hospitals within 3 km (0.0619) are the primary indicators for exposure.

However, entropy weight method results show that exposure (0.4187) is the most significant factor to consider in reservoir flood risk. Catchment area (0.2260) and storage capacity (0.1572) are key indicators influencing hazard and vulnerability, with weights far exceeding those of other peer indicators. Among exposure indicators, the number of nursing homes within 3 km (0.1503) and the number of hospitals within 3 km (0.1504) are the primary indicators to consider.

The integrated weight calculation results indicate that the importance of hazard (0.2679), vulnerability (0.3309), and exposure (0.4012) increases in that order.

4.2. Hazard of Flood Risk

The hazard zoning of reservoir flood risk is shown in Figure 4. The reservoir risk levels, indicated from red to dark blue, represent an increase in the overall risk. Within the study area, only one reservoir is classified as high risk, and eight reservoirs are classified as medium–high risk. The number of reservoirs at medium risk, medium–low risk, and low risk is 65, 53, and 47, respectively. Reservoirs of the same risk level tend to cluster spatially, with higher hazards in the northwest and lower hazards in the southeast; some reservoir hazard distributions overlap with administrative boundaries. In Qimen and Yi counties, all reservoirs are rated medium or medium–high risk, whereas in the neighboring Tunxi District, all reservoirs are rated low risk. Reservoirs in the northern part of Huangshan District are classified as low risk; hazard levels increase progressively from north to south, with a sudden medium–high risk transition in the eastern part of the district.

4.3. Vulnerability of Flood Risk

The vulnerability distribution of reservoir flood risk is shown in Figure 5. Within the study area, the numbers of reservoirs classified as high, medium–high, medium, medium–low, and low vulnerability are 6, 5, 15, 58, and 90, respectively. Reservoirs located in the southeastern Tunxi District, Xiuning County, and She County are generally of low or medium–low vulnerability. Although highly vulnerable reservoirs (n = 6) show no basin-wide clustering, 83% (5/6) are spatially concentrated within headwater tributaries of the main river system, suggesting localized geomorphic controls on vulnerability.

4.4. Exposure of Flood Risk

The exposure distribution downstream of reservoirs is shown in Figure 6. In the study area, the numbers of reservoirs with high, medium–high, medium, medium–low, and low exposure are 8, 9, 23, 39, and 95, respectively. Reservoirs with high exposure are predominantly clustered in three administrative units: Tunxi District, Yi County, and Huangshan District. These regions spatially align with the Xin’an River Basin, the Yi County Valley Basin, and the Huangshan Piedmont Basin. Schools, hospitals, and nursing homes extracted from Huangshan City POI data are also predominantly distributed in these areas.

4.5. Comprehensive Flood Risk

Figure 7 shows the integrated flood risk distribution for small reservoirs in Huangshan City. The numbers of small reservoirs in the low, medium–low, medium, medium–high, and high–risk categories are 24, 70, 33, 27, and 20, respectively. As shown, the integrated flood risk distribution of small reservoirs in Huangshan City exhibits clear spatial heterogeneity. Low- and medium–low-risk reservoirs are primarily concentrated in Xiuning County, the southern of Huangshan City. Medium-risk reservoirs are concentrated in Yi County. Reservoirs at the junction of Tunxi and Huizhou districts and in northern Huangshan District face higher risk. This may be related to the low elevation, flat terrain, and high rainfall in these areas, which promote water accumulation. Additionally, as economic and administrative centers of Huangshan City with concentrated key infrastructure (schools, hospitals, nursing homes), they are highly susceptible to breach risks.

5. Discussion

This study developed an integrated framework to evaluate the flood risk of over 180 small reservoirs in Huangshan City, considering hazard, vulnerability, and exposure. Overall, most reservoirs in Huangshan City fall into low to lower–medium risk categories, yet distinct spatial variations are evident. The comprehensive flood risk of reservoirs reflects the combined influence of natural factors (elevation, terrain, watershed hydrology) and socio-economic factors (distribution of critical infrastructure).

Higher-hazard reservoirs are primarily located in the relatively low-elevation, flatter areas of northwest Huangshan City, such as Huangshan District, Xiuning County, and Yi County. In contrast, Tunxi District is mainly composed of low-risk reservoirs. The occurrence of this situation can be attributed to the patterns of local topography and rainfall conditions. Intense rainfall and steep terrain rapidly increase surface runoff, thereby elevating flood risk [59]. The Huangshan peak range spans Huangshan District, Qimen County, and Yi County, reaching elevations up to 1800 m, while the small reservoirs are mostly situated between 100 m and 400 m elevation. Such pronounced elevation differentials accelerate runoff formation, and when combined with high rainfall totals and intensities in the region, increase flood process variability and thus elevate risk levels. Small reservoirs in She County, at 100–300 m elevation, and on flat terrain, receive lower rainfall totals and intensities than those in the northwest, resulting in uniformly lower risk levels. Although Xikou Reservoir in Xiuning County and the neighboring Hejia Reservoir share similar rainfall and terrain characteristics, Xikou exhibits much higher hazard, underscoring the dominant influence of rainfall distribution and watershed hydrology. Flat terrain may help reduce peak flows and flood risk [60].

Highly vulnerable reservoirs are generally located at the heads of tributaries and are characterized by large storage capacity, high dam height, and spillway systems lacking gates. The operation of spillway gates is crucial for cascade flood control [61]. Of the six highly vulnerable reservoirs, five have storage capacities exceeding 5 million m³, and one is only 100,000 m³ short of the medium-size reservoir threshold. Regarding dam height, all six exceed 30 m and two exceed 52 m, surpassing the international small reservoir standard of 15 m height. However, China classifies small reservoirs by storage capacity, and these taller dams were still designed to small reservoir standards, resulting in a mismatch between design specifications and actual risk exposures, thereby weakening resilience [62].

High-exposure reservoirs are concentrated in the administrative and economic centers—Tunxi District, Huizhou District, and Huangshan District—coinciding with dense distributions of POI facilities (schools and hospitals), confirming that “at-risk populations and critical infrastructure are core exposure indicators” [63]. In the event of a break, these areas would face elevated risks of casualties and property loss.

Although high-risk reservoirs are few, they are located in key socio-economic areas; a breach could severely impact downstream populations, public services, and infrastructure, potentially triggering cascade failures. Flood management measures should be tailored to the risk characteristics of different regions. In high-hazard areas, enhanced rainfall and runoff monitoring should improve forecasting accuracy, reduce hydrological uncertainties, and mitigate potential hazard impacts [64]. Increased funding, strengthened engineering maintenance, and regular inspections—especially of dam structures, spillways, and gates—are recommended for highly vulnerable reservoirs. Reservoirs with large-capacity and tall dams should establish comprehensive operation and maintenance mechanisms [65]. In high-exposure regions, emergency plans should be prioritized, community outreach and training conducted, flood awareness raised, and emergency response capabilities enhanced [63]. Culvert and IoT-based warning systems are also effective measures to reduce flood risks [66]. Additionally, development in floodplain areas around small reservoirs should be restricted through careful planning and land use regulation to reduce flood impacts on infrastructure [67]. Sediment should be regularly removed from reservoirs to maintain storage capacity, prevent spillway blockage, and ensure unobstructed flood discharge [68].

The model based on the “Hazard–Vulnerability–Exposure” differential driving mechanism, while valuable, has several limitations affecting risk assessment. First, most of the indicators used are static, limiting the ability to capture the dynamic nature of flood risk. Dynamic factors such as population growth, land use changes, and climate change were not [69]. Currently, we have not conducted localized climate projections for Huangshan City or obtained related results. Dynamic flood risk assessment is achieved by updating indicators such as the average annual rainfall or the maximum 24-hour rainfall. In future research, we will consider projected rainfall scenarios under different emission pathways (e.g., RCP4.5, RCP8.5), as well as changes in downstream population and land use, and re-evaluate flood risk for small reservoirs. Second, although 2D hydrodynamic modeling can accurately simulate dam failure and inundation, it takes a long time and requires high-precision topographic data (such as DEM), that are difficult to obtain, making it unsuitable for rapid assessment [70]. In the future, a simplified 1D/2D coupled modeling approach can be used to simulate localized flooding at key high-risk reservoirs. By combining the 2D model outputs with the indicator-based assessment results, a tiered evaluation framework of “rapid screening + fine simulation” can be established. Third, both the choice of evaluation criteria and the weighting method can exert a noticeable influence on the results. The AHP and entropy methods yield notably different weights, especially for vulnerability and exposure. In AHP, experts tend to assign high weights to structural safety indicators (such as dam breach discharge, dam types, and dam height) because of their direct relevance to engineering integrity. By contrast, entropy weights for these same indicators are low, since their normalized values show minimal variability across Huangshan’s small reservoirs. Downstream exposure indicators (e.g., number of hospitals or nursing homes within 3 km) display greater dispersion in the raw data and thus receive higher entropy weights. By averaging AHP and entropy allocations (see Equation (8)), no single weighting scheme can dominate the composite risk index. This combined approach reduces over-sensitivity to either purely subjective judgments or purely data-driven differences, thereby enhancing the robustness and reasonableness of our results. Fourth, due to the data constraints, some engineering parameters, such as the reservoir capacity-to-outlet capacity ratio, were not included as vulnerability indicators. Detailed engineering data (such as stage-storage relationships and stage-discharge relationships) will be collected to improve the evaluation system in future research. Future research can integrate quantitative analysis with dam breach simulation regulations [71]. Finally, cascading effects are a systemic risk propagation phenomenon in reservoir flood risk management [72], whereby the failure of a single facility can trigger a chain reaction through hydrological–engineering coupling mechanisms [73]. And an upstream reservoir failure has a chain effect, such as the compounded flood effect downstream [74,75]. Given the complexity of cascading failures among small reservoir clusters, this study did not consider such effects at present. In future research, we will investigate them. Social factors, particularly public health, infrastructure, and community resilience, remain critically underdeveloped [76]. Future research will incorporate complex non-linear relationships among hazard, vulnerability, and exposure [77], as well as feedback mechanisms from implemented flood control measures [78] into the assessment system to improve accuracy. Additionally, future applications of the differential driving model should be adapted and localized to regional specifics [79].

6. Conclusions

Based on regional disaster system theory, this study organically integrated upstream runoff conditions, reservoir intrinsic characteristics, and downstream socio-economic exposure to develop a flood risk assessment indicator system for small reservoirs. By assigning indicator weights using AHP and EW, we quantified hazard, vulnerability, exposure, and integrated risk indices, and classified flood risk levels for over 180 small reservoirs in Huangshan City, Anhui Province. The results indicate that the overall reservoir flood risk in Huangshan City is low to low–medium. However, localized high-risk clusters are primarily in densely populated, economically developed zones: Tunxi District, central–eastern Huangshan District, and central–eastern Qimen County. High-hazard reservoirs are concentrated in the northwest part of Huangshan, with rainfall intensity and catchment area being the dominant influencing factors. Vulnerability is primarily affected by dam height and storage capacity, without obvious spatial clustering. High exposure is mainly found in economic centers such as Tunxi District and Huizhou District. A detailed analysis of the various indicators enabled the identification of key risk zones and weaknesses in current flood risk management strategies, for which targeted recommendations were proposed.

Compared with existing assessment frameworks, this model integrates three-dimensional spatial dynamics across the entire disaster continuum, from precipitation anomalies to downstream inundation. By coupling hydrological factors, reservoir structural characteristics, and socio-economic exposure factors, it enables rapid flood risk classification using easily obtainable indicators, thereby circumventing computationally intensive two-dimensional hydrodynamic simulations. In this study, indicator selection prioritized data availability and applicability by drawing from public datasets, remote sensing products, and expert evaluations, making the framework adaptable to data-scarce regions. This model can support local managers in inspections, maintenance, and emergency planning. Although future work should incorporate dynamic climate projections and cascade-failure simulations to further refine its accuracy, the framework’s excellent scalability and high practicality make it an invaluable tool for small reservoir risk governance.