Typhoon Storm Surges in the Guangdong Hong Kong Macao Greater Bay Area Based on the ADCIRC Model ^†

Abstract

The Guangdong Hong Kong Macao Greater Bay Area is a core economic region in China with a high incidence of typhoon storm surges. Its low-lying terrain and dense river networks make it vulnerable to severe disasters when typhoons overlap with astronomical tides. This study integrates typhoon, terrain, and tide level data from 2000 to 2024 to construct an ADCIRC (Advanced Circulation Model) v54.01 numerical model, identify risk factors and high-risk areas, and design and verify the effectiveness of coordinated prevention and control countermeasures. Results show that the model has reliable simulation accuracy with MAE < 0.2 m and RMSE < 0.3 m; typhoon intensity and terrain elevation are the dominant factors, with high-risk areas concentrated on the west bank of the Pearl River Estuary and Dongguan Water Town; the comprehensive “engineering + non-engineering” measures can reduce the inundation area by 60% and the inundation rate of high-risk areas from 85% to 22%, providing technical support for regional disaster prevention and control. The novelty of this study lies in the integrated approach of combining grey relational analysis and multiple linear regression to quantify the contribution of key influencing factors, coupled with scenario-based evaluation of coordinated engineering and non-engineering measures tailored to the complex terrain and river network characteristics of the GBA.

1. Introduction

The GBA is economically developed but frequently affected by typhoon storm surges, with seawater intrusion and waterlogging seriously threatening infrastructure and residents’ safety. Existing studies have shortcomings such as insufficient model adaptability and lack of countermeasure coordination. Focusing on the core of disaster prevention and control, this study conducts three aspects of work: constructing an ADCIRC model v54.01 suitable for terrain, identifying high-risk areas, and designing quantitatively effective prevention and control countermeasures. The technical route is “data preprocessing→model construction→risk identification→countermeasure design→effect evaluation”.

2. Model Construction and Data Processing

2.1. Data Preprocessing

Core data includes hourly parameters of 3 typical typhoons (“Haiou”, “Mangkhut”, “Yagi”), 30 m resolution DEM terrain data, astronomical tide data from 3 tide gauge stations (separating M2, S2, K1, O1 tidal components), and Manning coefficient assignment for 4 types of land use.

The 30 m DEM data was obtained from the National Geomatics Center of China (NGCC), Ministry of Natural Resources, with a release year of 2021 and vertical accuracy of ±2 m, covering the entire GBA region. The 30 m land use data was derived from the Resource and Environment Science and Data Center (RESDC), Chinese Academy of Sciences, with a data year of 2020 and overall classification accuracy of over 90%. Both datasets were preprocessed for consistency with the model grid to meet the requirements of storm surge inundation simulation.

To ensure the accuracy of the ADCIRC model, Manning coefficients were assigned based on land use types, with reference to typical regional characteristics. As shown in

Table 1. Manning coefficient values for different land use types.

Land Use Type	Manning Coefficient n	Typical Area
Construction land	0.020–0.025	Nansha, Guangzhou
Cultivated land	0.025–0.030	Dongguan Water Town
Water area	0.012–0.015	Pearl River Estuary
Forest land	0.030–0.035	Dinghu, Zhaoqing

, the Manning coefficient for construction land in Nansha, Guangzhou, was set to 0.020–0.025, while that for cultivated land of Dongguan Water Town was 0.025–0.030; these values were calibrated to match the surface roughness of different land cover types.

The assignment of Manning coefficients to each model grid adopts a coupled approach of “lookup table-land use data”. Specifically, spatial overlay analysis was performed between the model grids and the 2020 30 m land use map, enabling each grid to be matched to one of the four land use types: construction land, cultivated land, water area, and forest land. Subsequently, the corresponding Manning coefficient range for each type was retrieved from Table 1, and the mean value of the range was adopted as the final coefficient for the grid, balancing computational stability and accuracy. For core areas such as Nansha and Dongguan Water Town, the coefficients were further calibrated to 0.022 for construction land and 0.028 for cultivated land based on the validation results of Typhoon “Mangkhut”, ensuring consistency with local surface roughness characteristics.

2.2. Core Construction of the ADCIRC Model

Model calibration and validation use the mean absolute error (MAE) and root mean square error (RMSE) to quantify the accuracy, with the formulas as follows:

$$\mathrm{MAE}={\displaystyle \frac{1}{N}}{\displaystyle \sum _{k=1}^N|}{S}_k-O_k|$$

(1)

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{N}}{\displaystyle \sum _{k=1}^N{(S_k-O_k)}^2}}$$

(2)

$N$ is the number of samples; $S_k$ is the simulated storm surge value at the k-th time step; and $O_k$ is the measured storm surge value at the k-th time step. Approximately 85,000 unstructured triangular meshes are adopted. Astronomical tide component parameters are input at the open offshore boundary, and the Jelesnianski wind field and Hybrid pressure field models are used for the typhoon forcing boundary [1]. Consistent with the calibrated Manning coefficients for core areas in Section 2.1, the ADCIRC model achieves reliable accuracy after validation with Typhoon “Mangkhut”, with MAE < 0.2 m and RMSE < 0.3 m.

3. Risk Identification and Delineation of High-Risk Areas

3.1. Model Simulation Results

Typhoon “Mangkhut” exhibits the most significant impact: the maximum storm surge at Nansha Station, Guangzhou, reaches 2.8 m, with a maximum inundation area of 820 km² [2]; the maximum inundation areas of “Haiou” and “Yagi” are 580 km² and 420 km², respectively [3]. High-risk areas are concentrated on the west bank of the Pearl River Estuary, Dongguan Water Town, and Dapeng Peninsula, Shenzhen, as illustrated in Figure 1.

3.2. Quantitative Analysis of Influencing Factors

Sequence standardization: Due to the different dimensions of various influencing factors, mean normalization is first performed on the reference sequence and the comparison sequences [4]:

$$x_i^{\prime }(k)={\displaystyle \frac{x_i(k)}{{\overline{x}}_i}},\hspace{1em}i=0,1,2,3,4$$

(3)

$${\overline{x}}_i={\displaystyle \frac{1}{m}}{\displaystyle \sum _{k=1}^m{x}_i}(k)$$

(4)

$\mathrm{m}$ is the number of typhoon cases.

Calculation of relational coefficients:

$${\zeta }_i(k)={\displaystyle \frac{{\min }_i {\min }_k|x_0^{\prime }(k)-x_i^{\prime }(k)|+\rho {\max }_i {\max }_k|x_0^{\prime }(k)-x_i^{\prime }(k)|}{|x_0^{\prime }(k)-x_i^{\prime }(k)|+\rho {\max }_i {\max }_k|x_0^{\prime }(k)-x_i^{\prime }(k)|}}$$

(5)

$\rho$ is the distinguishing coefficient.

Grey relational analysis and multiple linear regression were adopted to quantify the influence of various factors [5]. The grey relational degree ranking is typhoon intensity (0.85) > path distance (0.72) > astronomical tide level (0.68) > terrain elevation (0.65), while the contribution rate ranking is typhoon intensity (61%) > path distance (19%) > astronomical tide level (15%) > terrain elevation (5%). Notably, terrain elevation dominates the spatial distribution of inundation, and the specific contribution rate differences among all influencing factors are intuitively illustrated in Figure 2.

3.3. Delineation of High-Risk Areas

Three major high-risk areas are delineated: the west bank of the Pearl River Estuary, Dongguan Water Town, and Dapeng Peninsula, Shenzhen-Daya Bay, Huizhou, with a total area of approximately 120 km².

4. Design and Effect Evaluation of Prevention and Control Countermeasures

4.1. Coordinated Countermeasure System

4.1.1. Engineering Measures

Seawall construction: Construct 50 km of 3.3 m-high C30 reinforced concrete seawalls on the west bank of the Pearl River Estuary, and reinforce the existing seawalls in Humen, Dongguan, to 3.0 m [6].

Flood detention basin: Set up a 15 km² flood detention basin in Machong, Dongguan, with a flood storage capacity of 6 million m³ to divert floods from the Pearl River Estuary.

4.1.2. Non-Engineering Measures

Early warning optimization: Shorten the early warning time from 6 h to 4 h and build a four-level push system [7];

Emergency response: Formulate three-level response standards (Level I: surge > 2.0 m, evacuation within 2 h; Level II: 1.5–2.0 m, evacuation within 4 h; Level III: ≤1.5 m, issue prompts);

Land management: Delineate three types of development zones: prohibited, restricted, and safe.

4.2. Effect Evaluation

Taking Typhoon “Mangkhut” as the benchmark, four scenarios are set for simulation, and the comprehensive measures achieve the optimal disaster reduction effect.

To quantitatively evaluate the disaster reduction effectiveness of different prevention and control strategies, we simulated four scenarios for Typhoon “Mangkhut” and summarized the results in

Table 2. Comparison of disaster reduction effects under different scenarios.

Scenario	Maximum Inundation Area (km2)	Average Inundation Depth (m)	Disaster Reduction Rate (%)
Current benchmark	820	1.6	—
Engineering measures only	532	1.0	40.0
Non-engineering measures only	758	1.5	29.3
Comprehensive measures	408	0.7	60.0

. As shown in Table 2, the comprehensive “engineering + non-engineering” measures achieve the optimal performance: the maximum inundation area is reduced from 820 km² to 408 km², and the average inundation depth decreases from 1.6 m to 0.7 m, which is significantly better than single engineering or non-engineering measures. These quantitative findings are further visualized in Figure 3, which clearly illustrates the trends in inundation area, depth, and disaster reduction rate across scenarios, highlighting the synergistic effect of comprehensive measures.

5. Model Verification and Reliability Analysis

5.1. Accuracy Verification

The MAE of core tide gauge stations is 0.13–0.16 m, and the RMSE is 0.16–0.19 m; the coincidence degree between the simulated inundation area and remote sensing results is 82%, reaching 87% in high-risk areas, with accuracy meeting specifications [8]. The temporal consistency between simulated and measured storm surge values at Nansha Station, along with the residual distribution, is further visualized in Figure 4, providing direct evidence of the model’s high temporal simulation accuracy.

5.2. Stability and Sensitivity

Sensitivity coefficient formula:

$$S={\displaystyle \frac{\mathsf{\Delta }Y/Y}{\mathsf{\Delta }X/X}}$$

(6)

$X$ is the input parameter, $\mathsf{\Delta }X$ is the variation of $X$ (set to ±20%), $Y$ is the output indicator, and $\mathsf{\Delta }Y$ is the corresponding variation of $Y$. The larger the value of $|\mathrm{S}|$, the higher the sensitivity of the output indicator to the input parameter [9].

When the Manning coefficient and mesh resolution fluctuate, the output change rate is <10%, indicating good model stability [10]. The sensitivity coefficients of core input parameters are as follows: typhoon intensity (1.0), astronomical tide level (0.39), and path distance (0.51). This ranking is consistent with the earlier influencing factor analysis, and the specific comparison of sensitivity coefficients is intuitively presented in Figure 5.

6. Conclusions

The constructed ADCIRC model can accurately simulate the storm surge process with MAE < 0.2 m and RMSE < 0.3 m, providing reliable support for risk identification.

Typhoon intensity and terrain elevation are the core influencing factors, with high-risk areas concentrated on the west bank of the Pearl River Estuary and other regions.

The comprehensive measures of “seawall + flood detention basin + early warning + management” can achieve a 60% disaster reduction rate, effectively improving the resilience of regional disaster prevention and control.

The research results can be extended to coastal high-risk areas such as the Yangtze River Delta and Bohai Bay, and the runoff and rainfall modules can be coupled to improve accuracy in the future.