Nonlinear characteristics and configurational pathways of direct economic losses of storm surges: empirical evidence from China’s coastal regions

Abstract

Introduction:

Driven by the dual forces of global climate change and sea-level rise, storm surge disasters have emerged as a primary external shock constraining the sustainable development of coastal socio-economic systems.

Methods:

Different from previous studies that mainly focus on average effects, this study uses panel data from 11 coastal provinces and municipalities in China from 2011 to 2023 and employs Tobit models, two-part models, panel threshold models, and fuzzy-set qualitative comparative analysis to examine the nonlinear characteristics and configurational pathways of direct economic losses from storm surges.

Results:

The results show that storm hazard intensity and coastal economic exposure are positively associated with direct economic losses from storm surges, whereas regional adaptive capacity shows a loss-mitigation effect. Direct economic losses exhibit evident nonlinear characteristics, and the loss-amplifying effect of storm hazard intensity increases once it exceeds the estimated thresholds. Low direct economic losses cannot be explained by any single necessary condition, but are associated with different combinations of conditions, mainly including Defense Strategy-Driven Pathway and Comprehensive Resilience-Driven Pathway.

Discussion:

These findings are of great significance for improving coastal disaster risk governance systems and enhancing differentiated disaster prevention and mitigation capacities.

1 Introduction

Driven by the dual forces of global climate change and sea-level rise, the frequency and intensity of extreme marine hazards have shown a significant increasing trend, constituting a major external shock that constrains the sustainable development of coastal economies (Ghorai and Sen, 2015; Calafat et al., 2022). The storm surges—characterized by their high destructiveness, extensive impact range, and rapid onset—have consistently emerged as the primary threat causing direct economic losses in coastal regions (Needham et al., 2015). This challenge is particularly acute in China (Fu et al., 2023). With a continental coastline spanning 18,000 km, China’s coastal regions serve as the engine of national economic growth, accounting for over 60% of the national GDP and 40% of the population (Wang et al., 2026). However, rapid urbanization has precipitated a significant trend of “seaward accumulation” of population and assets, thereby drastically exacerbating the risk exposure of these high-value areas (Li et al., 2016; Deb et al., 2024). According to the Bulletin of China Marine Disaster, the average annual direct economic losses caused by storm surges amount to billions of RMB, exhibiting a fluctuating upward trajectory (Lin et al., 2024). Therefore, understanding the characteristics of direct economic losses from storm surges is important for enhancing coastal disaster adaptation capacity and safeguarding marine economic security.

Existing studies have examined storm surge disasters and their economic consequences from multiple perspectives. On the one hand, studies have revealed the natural processes and spatial heterogeneity of storm surges through numerical simulation, hazard assessment, inundation risk analysis, and compound disaster risk assessment (Zhang et al., 2022; Fu et al., 2023; Li et al., 2023). On the other hand, as disaster research has gradually extended from physical hazard assessment to socioeconomic consequences, increasing attention has been paid to the estimation, prediction, and socioeconomic explanation of direct economic losses from storm surges. Relevant studies show that storm surge losses are associated not only with the physical intensity of hazard events, but also with economic exposure, regional development level, disaster prevention and mitigation capacity, and socioeconomic structure (Sui et al., 2023; Lin et al., 2023, 2024; Yin et al., 2024). These studies provide an important basis for understanding storm surge disaster losses. However, existing analyses still focus mainly on loss assessment, prediction accuracy, or the average effects of individual factors. Further discussion is needed on whether the relationship between direct economic losses and relevant conditions changes across different hazard-intensity stages, and whether lower-loss outcomes may arise from multiple combinations of conditions.

The disaster risk framework provides an important theoretical basis for understanding these issues. According to the disaster risk framework of the Intergovernmental Panel on Climate Change (IPCC), disaster risk is generally understood as the likelihood of adverse consequences arising from the interaction among hazards, exposure, and vulnerability, while adaptive capacity affects the preparedness, response, and recovery of exposed systems under disaster shocks¹. From this perspective, direct economic losses from storm surges can be understood as realized outcomes generated by the interaction between actual disaster processes and disaster risk conditions, including hazards, exposure, vulnerability, and adaptive capacity. Accordingly, direct economic losses may not follow a single stable linear relationship, but may exhibit stage-specific differences as hazard intensity changes. At the same time, lower losses may result from multiple combinations of conditions rather than from any single determinant. These considerations lead to two questions: whether direct economic losses from storm surges exhibit nonlinear characteristics across different hazard-intensity stages, and whether lower-loss outcomes can be explained by multiple combinations of conditions. Based on this understanding, this study focuses on direct economic losses from storm surges in China’s coastal regions. It first conducts a baseline analysis of direct economic losses based on the disaster risk framework. It then examines changes in loss relationships across different hazard-intensity stages to reveal the nonlinear characteristics of direct economic losses from storm surges. Finally, it introduces a configurational perspective to explore multiple pathways associated with lower direct economic losses.

The possible contributions of this study are threefold. First, from a theoretical perspective, this study links the disaster risk framework with direct economic losses from storm surges, providing an analytical logic based on hazards, exposure, vulnerability, and adaptive capacity for explaining loss outcomes. Second, from an empirical perspective, this study further examines nonlinear changes in loss relationships across different hazard-intensity stages on the basis of baseline analysis, thereby complementing existing studies that mainly focus on average effects or linear relationships. Third, from a methodological perspective, this study introduces configurational analysis to discuss multiple combinations of conditions associated with lower direct economic losses, providing supplementary evidence for understanding differences in storm surge losses across China’s coastal regions.

2 Theoretical analysis framework

Direct economic losses from storm surges are the outcome of interactions between marine hazard processes and coastal socioeconomic systems. Existing studies have discussed storm surges from multiple perspectives, including storm-surge formation mechanisms, disaster risk assessment, economic loss estimation, and regional adaptive capacity. Studies on tropical-cyclone-generated storm surges show that storm-surge intensity is closely related to wind fields, atmospheric pressure, storm tracks, storm structure, tides, waves, nearshore bathymetry, and coastal geomorphology (Needham et al., 2015; Pineau-Guillou et al., 2020; Lockwood et al., 2022). In China’s coastal regions, related studies have further developed storm-surge hazard indicators, geospatial risk-assessment methods, and disaster-chain analytical frameworks, revealing the complex relationships among natural forcing, coastal exposure conditions, and protective systems in storm-surge risk (Shi et al., 2013; Zhang et al., 2021; Fu et al., 2023; Wang et al., 2025).

Disaster risk research provides an important theoretical basis for understanding storm-surge losses. The IPCC disaster risk framework generally conceptualizes risk as the possibility of adverse consequences arising from the interaction among hazard, exposure, and vulnerability, while adaptive capacity may influence the preparation, response, and recovery processes of exposed systems under disaster shocks². Under this framework, risk emphasizes the possibility of adverse consequences before or during a disaster event, whereas loss refers to the realized outcome formed through the interaction between risk conditions and an actual hazard process. Therefore, direct economic losses from storm surges can be understood as economic consequences shaped by hazard pressure, coastal exposure, social sensitivity, and adaptive capacity.

Existing studies on storm-surge and coastal-disaster losses also support this “risk conditions-loss outcome” analytical logic. One stream of research focuses on estimating, forecasting, and spatially characterizing direct economic losses, using statistical models, machine learning, case-based reasoning, and hybrid forecasting methods to analyze the level and variation of storm-surge losses (Sun et al., 2021; Guo et al., 2022; Wang et al., 2022; Shi et al., 2024). Another stream examines socioeconomic differences in storm-surge losses and suggests that economic development, regional inequality, disaster prevention and mitigation capacity, and post-disaster recovery capacity may be associated with the magnitude and spatial differentiation of losses (Yi et al., 2021; Lin et al., 2023, 2024; Yin et al., 2024). These studies indicate that direct economic losses from storm surges may be related not only to the physical intensity of disaster events, but also to coastal economic agglomeration, population structure, infrastructure conditions, and emergency response capacity.

2.1 Direct economic losses from storm surges under the disaster risk framework

Based on the disaster risk framework, direct economic losses from storm surges can be understood as a process through which risk conditions are translated into realized loss outcomes. Existing studies regard storm-surge disasters as complex disaster processes shaped by the interaction between natural systems and socioeconomic systems (Shi et al., 2013), and indicate that disaster losses are generally related to hazard intensity, exposure, and disaster prevention and mitigation capacity (Zhang et al., 2021). Specifically, hazard reflects the natural pressure generated by storm-surge processes on coastal regions. Storm-surge intensity is usually associated with wind speed, atmospheric pressure, storm tracks, tides, and nearshore geomorphological conditions (Needham et al., 2015; Pineau-Guillou et al., 2020; Lockwood et al., 2022). Exposure reflects the population, assets, infrastructure, and economic activities that may be affected by disaster events. With coastal urbanization and the agglomeration of marine economic activities, populations, industries, and high-value assets have increasingly concentrated in coastal areas, which may increase the scale of potential losses (Li et al., 2016; Deb et al., 2024; Lin et al., 2023). Vulnerability reflects the degree to which exposed systems are susceptible to damage under disaster shocks, while adaptive capacity reflects the resources and capabilities through which regions may reduce damage, organize emergency response, and support post-disaster recovery. Existing studies show that economic development, disaster prevention and mitigation capacity, and regional recovery capacity may influence the spatial differences in storm-surge losses (Yi et al., 2021; Lin et al., 2024; Yin et al., 2024).

From the socioeconomic perspective, population structure may constitute an important aspect of social sensitivity. Disaster vulnerability research suggests that children and elderly people may face higher vulnerability and recovery pressure in disaster contexts (Yuan et al., 2019). A higher dependency burden usually indicates a larger share of elderly and young populations relative to the working-age population, which may increase the demand for care, evacuation assistance, emergency support, and post-disaster recovery resources. Thus, population-structure conditions may influence the process through which storm-surge shocks are translated into economic losses. In addition, information infrastructure may alter the translation from disaster shocks to loss outcomes by affecting warning dissemination, emergency coordination, public service provision, and recovery capacity (Yang et al., 2023; Xiong and Li, 2025).

2.2 Nonlinear characteristics of direct economic losses from storm surges

The transformation from risk conditions to direct economic losses may exhibit nonlinear characteristics. Existing studies on loss assessment and forecasting suggest that characterizing direct economic losses from storm surges usually requires the joint consideration of hazard processes, regional exposure, and socioeconomic conditions; a single linear relationship may be insufficient to capture the complexity of loss variation (Sun et al., 2021; Guo et al., 2022; Wang et al., 2022). From the natural-process perspective, storm-surge responses are closely related to storm intensity, storm structure, tide-wave interactions, and nearshore conditions, and storm-surge impacts may vary across different combinations of natural conditions (Li et al., 2022; Lockwood et al., 2022; Van Tien et al., 2025). From the socioeconomic-process perspective, when hazard intensity is relatively low, existing protective facilities, public services, and emergency responses may absorb part of the shock; when hazard intensity further increases, protective systems and emergency capacity may face greater pressure, and the loss risk of exposed assets and vulnerable groups may be amplified (Sun et al., 2015; Wang et al., 2025).

Existing disaster-loss studies also provide empirical evidence for nonlinear analysis. Zhai and Jiang (2014) find that U.S. hurricane economic losses approximately follow a power-law relationship with maximum wind speed and storm size, suggesting that tropical-cyclone-related economic losses may not follow a simple linear relationship. Studies on storm-surge losses in China further show that economic development, disaster prevention and mitigation capacity, and regional differences may alter the way disaster shocks affect economic losses (Yi et al., 2021; Lin et al., 2023; Yin et al., 2024; Han et al., 2025). Therefore, direct economic losses from storm surges may not follow a single stable linear relationship, but may instead exhibit stage-specific differences under different hazard-intensity conditions.

2.3 Configurational pathways of storm surge disaster losses

Storm surge disaster economic losses may also exhibit configurational characteristics. The level of direct economic losses in coastal regions is usually not determined by a single condition, but may depend on the combination of hazard pressure, exposure level, social sensitivity, adaptive capacity, and information-support conditions. Configurational analysis emphasizes causal complexity, conjunctural causation, and equifinality, suggesting that the same outcome may be jointly produced by different combinations of conditions (Ragin, 2008; Fiss, 2011). This perspective is useful for explaining possible pathways through which different coastal regions may form similar low-loss outcomes under different risk conditions.

In recent years, configurational methods have been increasingly applied in marine and environmental management to analyze how multiple conditions jointly shape regional governance and resource-use outcomes. For example, existing studies have used configurational perspectives to examine port environmental efficiency, recreational fisheries performance, and regional environmental management, emphasizing the explanatory power of different condition combinations for outcome differences (Yang et al., 2025; Zhai et al., 2025). For direct economic losses from storm surges, configurational analysis can complement regression models that focus on average effects and further identify multiple combinations of conditions associated with lower loss outcomes. Therefore, lower direct economic losses from storm surges may arise from multiple configurational pathways. For example, a combination of weaker hazard pressure and lower exposure may correspond to lower losses; regions with higher exposure may also form relatively low losses under stronger adaptive capacity, lower social sensitivity, or better information-support conditions. Different pathways do not imply that any single factor has a universal determining role, but instead reflect the combinational differences of risk-related conditions across regions and disaster contexts.

2.4 Theoretical analysis framework

Based on the above literature, direct economic losses from storm surges, as the realized economic outcome of disaster risk shaped by hazard, exposure, vulnerability, and adaptive capacity, can be understood within an analytical framework from a multidimensional perspective of “disaster risk–nonlinear characteristics–configurational pathways” (Figure 1). First, the IPCC disaster risk framework provides a theoretical basis for explaining the relationships between hazard, exposure, vulnerability, adaptive capacity, and direct economic losses. Second, the transformation from risk conditions to loss outcomes may be affected by changes in hazard-intensity stages, thereby exhibiting nonlinear characteristics. Finally, lower loss outcomes may not originate from a single condition, but may be formed by combinations of multiple risk-related conditions.

Figure 1

3 Materials and methods

3.1 Variable selection

Variable selection in this study is based on the disaster risk framework and is designed to support the baseline analysis, nonlinear analysis, and configurational analysis of direct economic losses from storm surges. According to the disaster risk framework, disaster losses can be understood as realized outcomes generated by the interaction between hazard processes and exposed systems, and are generally associated with hazards, exposure, vulnerability, and adaptive capacity. Considering the availability of continuous provincial panel data for China’s coastal regions, this study uses direct economic losses from storm surges as the outcome variable, and selects storm hazard intensity, coastal economic exposure, socio-demographic vulnerability, and regional adaptive capacity to represent the core dimensions of hazard, exposure, vulnerability, and adaptive capacity, respectively. In addition, because information infrastructure may affect warning dissemination, emergency communication, and resource coordination, information support capacity is introduced as a moderating variable to examine whether it moderates the relationship between socio-demographic vulnerability and direct economic losses.

• Direct economic losses from storm surges (DEL) are used as the dependent variable and represent the realized economic consequences of storm surge events. According to the disaster risk framework, disaster losses reflect the combined effects of hazard intensity, exposure, vulnerability, and adaptive capacity. In China, direct economic losses mainly include damages to farmland, mariculture, housing, infrastructure, and vessels (Lin et al., 2023). Following the Natural Disaster Statistical Survey System, these losses are first estimated through preliminary disaster assessments and subsequently verified through comprehensive post-disaster investigations. Provincial annual data on direct economic losses are obtained from the China Marine Disaster Bulletin.

• Storm hazard intensity (SHI). In the disaster risk framework, storm hazard intensity refers to the physical hazard pressure generated by storm surge processes. Storm surge impacts are related to typhoon intensity, wind fields, atmospheric pressure, astronomical tides, waves, and nearshore hydrodynamic conditions (Needham et al., 2015; Pineau-Guillou et al., 2020; Lockwood et al., 2022). According to the China Marine Disaster Bulletin, typhoon-induced storm surges are the main source of direct economic losses from storm surges in China. Considering that the wind stress was the largest contributor to storm surges (91%) (Li et al., 2022), this study uses maximum wind speed as the proxy for storm hazard intensity. Specifically, typhoon track data from the China Meteorological Administration Typhoon Net are used to identify storm-surge-related typhoon events affecting China’s coastal provinces from 2011 to 2023. The track data include the latitude, longitude, time, and wind-speed information of typhoon centers at six-hour intervals. Following Guo and van de Lindt (2019), the Holland empirical wind-field model is used to estimate wind-speed series for the affected administrative regions. The wind-speed data are mainly obtained from the China Meteorological Administration Typhoon Net.

• Coastal economic exposure (CEE). Coastal economic exposure is used to represent the scale of economic activities and assets that may be affected by storm surge events. In the disaster risk framework, exposure generally refers to the presence of population, assets, infrastructure, and economic activities in areas potentially affected by hazards (Zhang et al., 2021). Gross ocean product reflects the total scale of ocean-related economic activities in a region, including marine fisheries, marine transportation, coastal tourism, and other marine industries. A higher gross ocean product usually indicates a greater concentration of marine economic activities and coastal assets, which may correspond to greater potential direct economic losses when storm surge events occur. Existing studies also suggest that the concentration of economic activities and the scale of the marine economy are closely related to storm surge economic losses (Lin et al., 2023). So, this paper uses gross ocean product as the proxy for coastal economic exposure. Data on gross ocean product are mainly obtained from the China Marine Economic Statistical Yearbook and the China Marine Economy Statistical Bulletin.

• Socio-demographic vulnerability (SDV). In the disaster risk framework, socio-demographic vulnerability is used to represent the population-structure dimension of vulnerability in exposed systems. Children, older adults, and other dependent groups may face higher needs for evacuation, care support, medical assistance, and post-disaster recovery during disaster events (Yuan et al., 2019). Following the vulnerability information in coastal storm surge analysis (Zhang et al., 2021), this study uses the dependency ratio as the proxy for socio-demographic vulnerability. The dependency ratio reflects the proportion of the non-working-age population relative to the working-age population, and can partially capture the pressure on public services, medical assistance, and labor supply during disaster response and recovery. The data required to calculate the dependency ratio are obtained from the China Statistical Yearbook.

• Regional adaptive capacity (RAC). Regional adaptive capacity is used to represent the overall ability of a region to prepare for, respond to, and recover from storm surge impacts in the disaster risk framework. Higher socioeconomic development can generally enhance adaptive capacity through improved public services, infrastructure construction, technological progress, emergency response, and post-disaster reconstruction. Previous studies have shown that economic development and disaster prevention and mitigation capacity are closely related to storm surge disaster losses (Guo and Li, 2020; Yi et al., 2021). This study uses per capita GDP as the proxy for regional adaptive capacity in the main analysis. Although per capita GDP does not directly measure specific protective engineering projects, monitoring and warning facilities, or emergency resources, it reflects general development resources, public service capacity, and the basis for post-disaster recovery at the provincial scale. To examine the robustness of proxy selection, the robustness tests further use disaster-monitoring facility indicators as alternative proxies. Data on per capita GDP are obtained from the China Statistical Yearbook.

• Information support capacity (ISC). Information support capacity is used as a moderating variable in this study to examine whether it affects the relationship between socio-demographic vulnerability and direct economic losses from storm surges. Storm surge disasters are sudden and time-sensitive, and information transmission efficiency and communication network conditions may affect risk warning, population transfer, emergency organization, and recovery coordination. For regions with higher socio-demographic vulnerability, stronger information support capacity may help improve disaster information transmission, optimize emergency resource allocation, and reduce pressure on vulnerable groups during disaster response and recovery. Therefore, this study uses the length of optical fiber cable lines as the proxy for information support capacity and introduces the interaction term between socio-demographic vulnerability and information support capacity into the model. Data on optical fiber cable length are obtained from relevant statistical yearbooks.

3.2 Data analysis

Based on the variable selection described above, this study uses data for 11 coastal provinces and municipalities from 2011 to 2023. Table 1 presents the descriptive statistics for the five key variables.

As shown in Table 1, DEL ranges from 0 to 87.26, with a mean of 6.1703 and a standard deviation of 14.0154, indicating substantial variation across provinces and years. SHI ranges from 0.0001 to 60.2317, with a mean of 11.7497. CEE has a mean of 6688.866 and ranges from 613.8 to 21059. SDV ranges from 19.30 to 55.82, while RAC has a mean of 75263.16 and ranges from 22234 to 207193. It is evident that all variables exhibit considerable variation. Given that DEL is the dependent variable in this study and contains a substantial number of zero values, a more detailed analysis of its distribution is conducted next.

Figure 2 visually confirms the non-normal characteristics of the direct economic loss. The plot is dominated by an extremely sharp peak at zero, indicating that for the vast majority of observations, storm surges resulted in negligible or no recorded direct economic loss. Conversely, the distribution exhibits a long right tail that extends far into the high-value territory. This “long tail” represents the low-frequency, high-impact events: severe storm surge disasters that, while infrequent, cause economic damages that far exceed the normal level. In summary, the distributional evidence provides the empirical rationale for combining a Tobit and a two-part model that separately examines whether a positive loss occurs and, conditional on occurrence, the magnitude of the loss.

Figure 2

3.3 Methods construction

Direct economic loss is a non-negative continuous monetary outcome with a substantial number of zeros. The empirical analysis therefore proceeds in two stages. First, the Tobit model is used as the main baseline specification to estimate the overall relationship between the four risk-related factors and the latent loss outcome subject to the lower limit of zero. SHI, CEE, SDV, and RAC are introduced sequentially so that the incremental explanatory contribution and coefficient stability of each dimension can be observed. Second, the two-part model extends the baseline analysis by decomposing the loss process into: (i) a Logit model for the probability that a positive loss is observed, Pr(DEL > 0); and (ii) an OLS model for ln(DEL + 1) among observations with DEL > 0. The two-part model thus distinguishes the factors associated with whether losses occur from those associated with the magnitude of recorded positive losses. The baseline regression model is specified as Equation 1:

where, DEL_it denotes the direct economic loss caused by storm surge disasters; SHI_it denotes storm hazard intensity;CEE_it denotes coastal economic exposure; SDV_it denotes socio-demographic vulnerability;RAC_it denotes regional adaptive capacity; and the final term is a random disturbance. The empirical estimates moves from the Tobit estimates of the overall loss relationship to the two-part decomposition of loss occurrence and positive-loss magnitude.

The relationship between SHI and DEL may be nonlinear: the associations between the explanatory variables and loss may change after hazard intensity crosses a critical level. To examine this possibility, we employ a panel threshold regression model in which SHI serves as the threshold variable. We first estimate a comprehensive model to test whether a SHI threshold remains after controlling for CEE, SDV, and RAC. We then estimate separate threshold specifications in which SHI, CEE, SDV, and RAC are entered in turn as regime-dependent explanatory variables, allowing their coefficients to differ below and above the estimated SHI threshold. The panel threshold model is specified as Equations 1–6:

where, the meanings of the existing indicators are as described above. SHI_it is the threshold variable, indicates the threshold value to be estimated, is the index function. This model divides the samples into two groups based on the magnitude relationship between the threshold value and the threshold variable SHI, and obtains different regression parameters and through regression.

At present, the GridSearch method is usually utilized to continuously attempt the possible values of the threshold, thereby obtaining the threshold value that minimizes the sum of squares of the regression residuals. Regarding the test of the significance of the threshold effect, Hansen proposed a likelihood ratio statistic for the null hypothesis and used the self-sampling method (Bootstrap) to simulate its asymptotic distribution, thereby calculating the asymptotically effective probability value P to test its significance. Regarding the verification of the authenticity of the threshold value, Hansen proposed the LR statistic for the null hypothesis and provided the formula for calculating its rejection domain. When , reject the null hypothesis, where is the significance level.

The panel threshold model identifies how the relationships between factors and DEL under different SHI, but it does not identify complete combinations of conditions associated with low-loss outcomes. Such pathways may arise through conjunctural causation and equifinality rather than through a single dominant factor. We therefore employ fsQCA as a complementary configurational method to identify alternative combinations of SHI, CEE, SDV, and RAC that are sufficient for low DEL.

Specifically, DEL is used as the outcome and SHI, CEE, SDV, and RAC are used as antecedent conditions. Because fsQCA operates on set memberships rather than raw variables, all measures are calibrated to fuzzy-set scores ranging from 0 to 1. Following Ragin’s (2008) direct calibration method, the 95th percentile, median, and 5th percentile are used as the anchors for full membership, the cross-over point, and full non-membership, respectively. The resulting calibration anchors for the 2011–2023 sample are reported in Table 2.

After constructing the truth table, the explanatory power of each configuration is evaluated by calculating its consistency and coverage scores. Consistency measures the degree to which a configuration is a reliable sufficient condition for the outcome. Coverage, in turn, assesses the empirical relevance of the configuration, or the extent to which it accounts for the outcome. In the process of configuration analysis of DEL caused by storm surges, suppose Y represents the outcome variable and X represents the antecedent variable. From the perspective of sets, if the antecedent variable X is included in the outcome variable Y, it indicates that the antecedent variable X is a sufficient condition for the outcome variable Y. Conversely, if the antecedent variable X contains the outcome variable, it indicates that the antecedent variable is a necessary condition for the result variable. Suppose there is n sample, then the calculation formulas for the consistency and coverage of the combination of the h variables are as Equations 7–8:

where, the outcome variable is the DEL caused by the storm surge disaster, and the antecedent variables are the SHI, CEE, SDV, and RAC. X_i indicates the membership degree of the h antecedent variable combination in the i sample; Y_i represents the membership degree of the outcome variable in the i sample; min(X_i,Y_i) represents the minimum value of the combination of antecedent variables and the membership degree of the outcome variable.

4 Results and discussion

4.1 Basic results

4.1.1 Baseline regression results

Before estimating the econometric models, pairwise correlations and multicollinearity diagnostics were conducted to examine the relationships among the explanatory variables and to ensure the reliability of subsequent regression analyses. Table 3 reports the correlation matrix. The largest absolute correlation coefficient among the explanatory variables is 0.4915, observed between CEE and RAC, while the correlation between the DEL and SHI reaches 0.4202. All pairwise correlation coefficients are well below the commonly accepted threshold of 0.8, suggesting that no severe linear dependence exists among the explanatory variables.

To further evaluate multicollinearity, variance inflation factors (VIFs) were calculated and are reported in Table 4. The largest VIF is 1.34 for CEE, and the mean VIF is only 1.18, both substantially below the conventional threshold of 10. These results indicate that multicollinearity is unlikely to distort coefficient estimation or inflate standard errors, providing a reliable basis for subsequent regression analyses.

Table 5 presents the baseline Tobit regression results for DEL caused by storm surges using panel data from 11 coastal provinces and municipalities during 2011–2023. Overall, the coefficient signs and significance levels of the variables are relatively stable, and the estimates based on conventional standard errors are broadly consistent with those using province-clustered robust standard errors, indicating that the baseline findings are unlikely to be affected by within-province error correlation. In addition, as CEE, SDV, and RAC are introduced sequentially, the pseudo R² increases from 0.0477 to 0.0814, while the log likelihood improves from −439.292 to −423.762. These results indicate that direct economic losses from storm surges are not determined solely by hazard intensity, but arise from the combined effects of hazard, economic exposure, population structure, and adaptive capacity.

As shown in Table 5, SHI is positive and significant at the 1% level in all models, with coefficients ranging from 0.7228 to 0.8015. Its coefficient does not change substantially as additional variables are introduced. This indicates that storm hazard intensity is the most stable physical factor affecting direct economic losses and that its effect is relatively independent of regional socioeconomic conditions. Higher wind speeds generally imply stronger wind stress, greater storm surges, and higher risks of coastal inundation and infrastructure damage, thereby significantly increasing DEL. The coefficient of CEE remains positive throughout and becomes significant at the 1% level after regional adaptive capacity is included, indicating that the concentration of coastal economic activities and assets amplifies storm surge losses. The marked increase in its significance after controlling for RAC suggests that economic development has the dual effects of increasing exposure and enhancing adaptive capacity. On the one hand, the concentration of industries, infrastructure, and assets in coastal areas increases the value potentially exposed to damage. On the other hand, economic development can also strengthen regional disaster prevention and mitigation capacity. RAC is significantly negative in the full model, indicating that a stronger socioeconomic foundation helps reduce the actual losses caused by storm surges. Higher adaptive capacity generally implies greater investment in protective measures, more complete infrastructure and public services, and stronger emergency response and post-disaster recovery capabilities, all of which can jointly reduce losses before, during, and after disasters. SDV is significantly negative in the full Tobit model, indicating a negative overall association with direct economic losses after the other risk dimensions are controlled for. However, this result requires further examination using the subsequent models.

Table 6 further decomposes DEL into two processes: whether a positive loss occurs and the magnitude of the loss conditional on its occurrence. The results of Part I show that SHI is significantly positive in all models, indicating that greater storm hazard intensity increases the probability of observing positive direct economic losses. CEE becomes significantly positive after RAC is controlled for, suggesting that higher economic exposure not only increases potential loss magnitude but also raises the likelihood that observable economic losses will occur. RAC is significantly negative, indicating that regional adaptive capacity can reduce the probability of loss occurrence through risk prevention, protective engineering, and emergency response.

The results of Part II are broadly consistent with those of Part I. SHI significantly increases the magnitude of losses among observations with positive losses, indicating that hazard intensity not only determines whether losses occur but also further amplifies disaster consequences. CEE is significantly positive in the full model, suggesting that once losses occur, a high concentration of coastal economic activities and assets is translated into greater monetary losses. RAC significantly reduces conditional loss magnitude, indicating that adaptive capacity plays not only a preventive role before disasters but also limits the expansion of losses through emergency rescue, infrastructure resilience, and post-disaster recovery. The pseudo R² and R² of the full models increase to 0.3704 and 0.4030, respectively, indicating that SHI, CEE, SDV, and RAC have relatively strong explanatory power for loss magnitude.

Unlike the Tobit results, SDV is not statistically significant in either part of the two-part model, indicating that its negative relationship is not consistently reflected in either the probability of loss occurrence or conditional loss magnitude. Therefore, the negative coefficient in the Tobit model should not be interpreted as evidence that socio-demographic vulnerability itself reduces disaster losses. A more plausible explanation is that a higher dependency ratio may induce an endogenous adaptive response. Regions with larger dependent populations, including older adults and children, may face stronger social demands and policy pressures, prompting them to strengthen disaster preparedness education, population evacuation, emergency services, and protective infrastructure. However, whether these compensatory measures can operate effectively depends on the timely dissemination of warning information, the efficient coordination of emergency resources, and the rapid coverage of vulnerable groups. Therefore, the effect of SDV on DEL may be conditionally moderated by regional information support capacity rather than manifesting as a stable independent effect. On this basis, the subsequent analysis introduces an interaction term between SDV and ISC to examine whether ISC moderates the effect of socio-demographic vulnerability on direct economic losses from storm surges.

Table 7 reports the estimated interaction effects between SDV and ISC. After the interaction term is introduced, the signs and significance levels of the core variables remain generally stable: SHI and CEE are significantly positive across all models, whereas RAC remains significantly negative. This indicates that the previous conclusions regarding hazard intensity, economic exposure, and regional adaptive capacity are not substantially altered by the inclusion of the interaction term.

Table 7 shows that the coefficient of the interaction term between SDV and ISC is −0.0172 and is significant at the 10% level in both Tobit models. The result remains valid when province-clustered robust standard errors are used. This indicates that, as information support capacity improves, the marginal effect of socio-demographic vulnerability on direct economic losses becomes more negative, providing preliminary support for the view that information infrastructure can enhance compensatory adaptation in vulnerable regions. More developed information and communication conditions can improve the efficiency of warning dissemination, facilitate the timely identification and coverage of key groups such as older adults and children, and promote the coordinated allocation of evacuation services, emergency assistance, and rescue resources, thereby reducing the likelihood that vulnerable population structures are translated into actual economic losses. However, although the interaction coefficients are also negative in both parts of the two-part model, neither reaches statistical significance. This suggests that the moderating effect of information support capacity is mainly reflected in the overall loss relationship captured by the Tobit model, while there is insufficient evidence to confirm that it significantly affects either the probability of loss occurrence or the magnitude of losses conditional on their occurrence. Overall, information support capacity may constitute an important condition shaping the effect of SDV.

4.1.2 Robustness test for the baseline results

To examine the robustness of the estimated determinants of direct economic losses from storm surges, this study conducts a series of tests by adjusting the sample period, replacing key variables, treating extreme values, and using lagged explanatory variables.

First, empirical results may be sensitive to the choice of sample period, as different time windows may yield different estimation results (Wang et al., 2023). Therefore, Column (1) in Table 8 shortens the sample period from 2011–2023 to 2011–2021 while retaining the baseline variable definitions. The results show that SHI and CEE remain positive and significant at the 1% level, SDV remains negative and significant at the 10% level, and RAC remains negative and significant at the 1% level. The signs and significance levels of the main variables are broadly consistent with the baseline regression results. This indicates that the main findings are not driven by the additional observations for 2022 and 2023 and remain valid over the earlier common sample period.

On this basis, following Yin et al. (2023), who conducted robustness checks by replacing key explanatory variables, Columns (2)–(4) in Table 8 re-estimate the model using alternative indicators. Column (2) in Table 8 replaces maximum wind speed with a dummy variable indicating whether a storm surge disaster occurred (Wu et al., 2022). The coefficient of the disaster-occurrence dummy is 9.1677 and is positive and significant at the 1% level, while CEE remains significantly positive and RAC remains significantly negative. Drawing on Nohrstedt et al. (2022), who used disaster frequency to analyze disaster impacts, Column (3) in Table 8 further replaces maximum wind speed with storm surge disaster frequency. The coefficient of disaster frequency is 5.8162 and remains positive and significant at the 1% level, while the directions and significance levels of the other variables are also broadly consistent with the baseline results. Therefore, whether storm hazard intensity is measured by maximum wind speed, disaster occurrence, or disaster frequency, its positive effect on direct economic losses remains stable, indicating that the baseline conclusion does not depend on a particular measure of SHI. Column (4) in Table 8 replaces per capita GDP with the number of marine, tidal, and meteorological observation stations per unit coastline length to provide a more direct measure of regional storm surge monitoring, protection, and adaptive capacity. The coefficient of coastal observation-station density is −21.4084 and is significant at the 1% level. Meanwhile, SHI and CEE remain significantly positive, whereas SDV remains significantly negative. This result indicates that denser disaster-monitoring facilities can strengthen hazard identification, monitoring and early warning, and emergency response, thereby reducing direct economic losses from storm surges. It also provides further evidence of the loss-mitigation effect of regional adaptive capacity.

Third, to reduce the potential influence of outliers or extreme observations on the empirical results, following Balan and Norden (2024), all explanatory variables are winsorized at the 1% level in both tails, and the Tobit model is re-estimated. The results in Column (5) in Table 8 show that, after winsorization, SHI and CEE remain significantly positive, while SDV and RAC remain significantly negative. Their coefficient signs and significance levels are consistent with the baseline estimates, indicating that the main findings are not driven by a small number of extreme observations.

Finally, the baseline model may be subject to endogeneity arising from omitted variables or reverse causality. To partially mitigate contemporaneous reverse effects, Column (6) in Table 8 re-estimates the model using one-period-lagged explanatory variables. The results show that lagged SHI and CEE remain positive and significant at the 1% level, lagged SDV remains negative and significant at the 10% level, and lagged RAC remains negative and significant at the 1% level. Although the use of lagged variables cannot completely eliminate endogeneity caused by omitted variables or reverse causality, the signs and significance levels of the main variables remain stable after the explanatory variables are temporally separated from current direct economic losses, further supporting the baseline estimates.

Across the six robustness checks, the hazard indicator remains positive, CEE remains positive, and both SDV and the adaptive-capacity measure retain negative coefficients with statistical significance. In particular, the shortened-sample estimate shows that the main pattern predates the 2022–2023 sample extension, while the observation-station proxy confirms the negative association between disaster-specific monitoring capacity and direct economic losses over the common 2011–2021 period. Although the magnitudes are not directly comparable with the 2011–2023 baseline estimates because the sample period and, in some columns, the variable definitions differ, the stability of coefficient signs and significance supports the substantive conclusions across alternative time windows, measurements, outlier treatments, and temporal specifications.

Overall, the results of the six robustness checks are broadly consistent with the baseline regression. SHI and CEE consistently exert significant positive effects on DEL, whereas SDV and RAC retain significant negative associations. Although coefficient magnitudes are not directly comparable across specifications because of differences in sample periods and variable definitions, the stability of their signs and statistical significance indicates that the main conclusions do not depend on a particular sample period, variable measurement, outlier-treatment method, or contemporaneous model specification.

4.2 Analysis of threshold effect

4.2.1 Test of threshold effect

Testing for threshold effects is a necessary prerequisite for determining the applicability of the panel threshold model. Before estimating the threshold regressions, it is necessary to examine whether the relationships between the explanatory variables and DEL change significantly when storm hazard intensity crosses a particular level and whether the estimated thresholds can be reliably identified. In this study, SHI is specified as the threshold variable. The threshold values are determined by minimizing the residual sum of squares, and their statistical significance is assessed using 300 bootstrap replications. The results are reported in Table 9.

Table 9 shows that the estimated SHI threshold is 33.1157 m/s in the comprehensive model incorporating SHI, CEE, SDV, and RAC simultaneously. In the separate models that allow the coefficient of each explanatory variable to vary across SHI regimes, the SHI model identifies the same threshold of 33.1157 m/s, while the CEE, SDV, and RAC models identify a threshold of 34.7958 m/s. The threshold effects in all models are significant at the 1% level, indicating that the relationships between the different explanatory variables and DEL from storm surges change significantly when SHI exceeds the corresponding threshold values. In addition, both threshold values fall within the Grade 12 wind-speed range specified in China’s national wind-scale standard and correspond to the wind-speed range of typhoon-intensity tropical cyclones³, which gives the estimated thresholds a meaningful meteorological interpretation.

To assess the precision of the threshold estimates, Figure 3 plots the LR profiles for the four separate regime-dependent variables. Each plot is labeled by the variable whose coefficient is allowed to differ across the SHI threshold.

Figure 3

(a), (b), (c), and (d) in Figure 3 respectively depict the threshold estimates and likelihood ratio functions for SHI within the context of impact on DEL from DEL, CEE, SDV and RAC. The horizontal axis represents the threshold estimates of SHI, while the vertical axis denotes the values of the likelihood ratio test statistic. The dashed red line indicates the critical value at the 95% confidence level. As established by Hansen (1999), the null hypothesis is rejected when the likelihood ratio . Figure 3 demonstrates that all likelihood ratio values corresponding to the thresholds of SHI all fall significantly below the critical value. This confirms the consistency between the estimated thresholds and their true levels, validating the robustness of the results.

4.2.2 Analysis of the threshold regression estimates

Based on the separate panel threshold models, Table 10 reports how the coefficients of SHI, CEE, SDV, and RAC differ below and above the estimated SHI thresholds. Overall, the effects of SHI, CEE, SDV, and RAC change when SHI crosses the corresponding threshold values, indicating that influences to DEL exhibit clear nonlinear characteristics.

Specifically, when the maximum wind speed does not exceed 33.1157 m/s, the coefficient of SHI is 0.0260 and statistically insignificant. Once this threshold is exceeded, the coefficient increases to 0.6638 and becomes significant at the 1% level, indicating that the marginal loss-amplifying effect of SHI becomes substantially stronger in the high-intensity regime. For CEE, the coefficient is −0.0020 and significant at the 1% level when SHI does not exceed 34.7958 m/s; above the threshold, it becomes positive but statistically insignificant. This suggests that the relationship between CEE and DEL is not consistently monotonic. Under lower-intensity conditions, improvements in infrastructure and risk-management capacity associated with economic development may partially offset the exposure effect resulting from asset concentration, whereas this negative relationship no longer holds in the high-intensity regime. SDV is negative and only weakly significant when SHI does not exceed 34.7958 m/s. Above the threshold, its coefficient rises to 0.4791 and becomes significant at the 5% level. This indicates that, under lower-intensity conditions, compensatory measures such as early warning, evacuation, and public services may alleviate the adverse effects of SDV. Under high-intensity hazards, however, the disadvantages faced by dependent populations, including older adults and children, in accessing information, evacuating, and recovering from disasters become more pronounced, thereby significantly amplifying DEL. RAC is significantly negative when SHI does not exceed 34.7958 m/s, indicating that RAC effectively reduces DEL. Above the threshold, its coefficient becomes positive but remains statistically insignificant, suggesting that the stable loss-mitigation effects of conventional monitoring and warning systems, protective infrastructure, and emergency resources weaken under high-intensity shocks.

To sum up, once SHI exceeds the corresponding threshold values, the loss-amplifying effect of SHI increases significantly, SDV shifts from a potential risk factor to a significant loss amplifier, and the protective effect of RAC becomes less stable. Storm surge risk governance should therefore adopt intensity-specific strategies: strengthening routine disaster-prevention capacity in the lower-intensity regime, initiating the early evacuation of vulnerable populations and protection of critical assets as SHI approaches the threshold values, and prioritizing emergency resource coordination and rapid post-disaster recovery once the thresholds are exceeded.

4.3 Configurational pathways toward low DEL

4.3.1 Analysis of necessary conditions

Before conducting the configurational analysis, this study first performs a necessity test on the antecedent conditions of low DEL from storm surges to determine whether there is any single condition that must be present when the low-loss outcome occurs. Conventionally, when the consistency of an antecedent condition exceeds 0.90, it can be regarded as a necessary condition for the occurrence of the outcome. Table 11 reports the necessity test results for each antecedent condition and its absence with respect to low DEL from storm surges.

As shown in Table 11, among the antecedent conditions of low direct economic losses from storm surges, low storm hazard intensity (~SHI) has the highest consistency, at 0.7543, with a coverage of 0.8351. It is followed by low coastal economic exposure (~CEE), whose consistency and coverage are 0.6642 and 0.7737, respectively. In contrast, high storm hazard intensity (SHI) has the lowest consistency, at only 0.4308. This indicates that low storm hazard intensity is the most common antecedent condition among cases with low direct economic losses. However, its consistency remains well below the necessity threshold of 0.90, and the consistency values of the other conditions and their absence also fail to reach this threshold.

Therefore, no antecedent condition alone constitutes a necessary condition for low DEL from storm surges. The low-loss outcome is not independently determined by any single factor but is jointly shaped by multiple conditions. It is therefore necessary to conduct a further configurational sufficiency analysis to identify multiple combinations of conditions that can lead to low DEL.

4.3.2 Causal configuration analysis

To identify combinations of conditions that can produce low DEL from storm surges, this study further conducts a configurational sufficiency analysis. Following Yang et al. (2025), the case-frequency threshold is set to 1, the raw-consistency threshold to 0.80, and the PRI-consistency threshold to 0.70. FsQCA analysis can generate complex, intermediate, and parsimonious solutions. This study identifies core and peripheral conditions by comparing the nested relationship between the intermediate and parsimonious solutions: conditions appearing in both the intermediate and parsimonious solutions are identified as core conditions, whereas conditions appearing only in the intermediate solution are identified as peripheral conditions (Meng et al., 2024).

Table 12 reports two sufficient configurational pathways for achieving low DEL from storm surges. The overall solution consistency is 0.8899, indicating that, among the cases conforming to the two configurational pathways, 88.99% of the configurational membership is consistent with the low direct economic loss outcome. The overall solution coverage is 0.6751, indicating that the two pathways jointly explain 67.51% of the low-loss outcome. According to the criterion that the consistency of an individual configuration should not be lower than 0.75 (Li et al., 2023), the consistency values of L1 and L2 are 0.8955 and 0.9261, respectively, both substantially higher than the minimum threshold. This indicates that the two configurational pathways have a high degree of reliability.

Based on the results in Table 12, this study identifies two configurational pathways that can achieve low direct economic losses from storm surges, which are named the L1 Defense Strategy-Driven Pathway and the L2 Comprehensive Resilience-Driven Pathway.

1. Defense Strategy-Driven Pathway (L1).

Configuration L1 takes low storm hazard intensity and low coastal economic exposure as its core conditions, expressed as ~SHI and ~CEE. The consistency of this pathway is 0.8955, its raw coverage is 0.5458, and its unique coverage is 0.2430, giving it a broader range of case explanation among the two pathways.

In this pathway, typical cases include Hainan in 2012, 2017, and 2018, and Hebei from 2013 to 2015, where low SHI constitutes an important prerequisite for achieving low direct economic losses, indicating that the storm surge disaster experienced by the region has not yet reached an extreme level of hazard intensity. The key to achieving low losses lies in low CEE, namely, proactively reducing the population, assets, infrastructure, and economic activities exposed to storm surge risks. In the short term, effective monitoring and early-warning systems can be used to organize the timely return of vessels to ports, temporarily close coastal production facilities, and transfer people and vulnerable assets to safe areas, thereby reducing the actual level of exposure when a disaster occurs. In the long term, risk-sensitive territorial spatial planning and coastal-zone land-use regulation should be used to restrict the excessive concentration of settlements, critical infrastructure, and high-value industries in areas exposed to high storm surge risks, thereby reducing potentially affected elements at the source of spatial development.

In summary, L1 reflects a defense strategy centered on proactive avoidance. This pathway indicates that, when storm hazard intensity has not yet reached an extreme level, reducing the scale of population and assets located in high-risk areas can effectively decrease the direct economic losses caused by storm surges. Its core does not lie in relying on rescue and recovery after a disaster occurs, but in reducing the possibility of losses at the source through pre-disaster prevention, spatial regulation, and exposure reduction.

2. Comprehensive Resilience-Driven Pathway (L2).

Configuration L2 takes low storm hazard intensity and high regional adaptive capacity as its core conditions, expressed as ~SHI and RAC. The consistency of this pathway is 0.9261, the highest among the two pathways, while its raw coverage and unique coverage are 0.4530 and 0.1369, respectively. The core of this pathway lies in a high level of regional adaptive capacity. This study uses per capita GDP as a proxy for regional adaptive capacity, mainly reflecting the financial, technological, and resource foundations provided by socioeconomic development for disaster-prevention infrastructure, public services, emergency management, and post-disaster recovery. It is worth noting that high regional adaptive capacity needs to operate jointly with low storm hazard intensity. This indicates that, even when a region possesses a strong resilience foundation, its loss-mitigation measures are more likely to function effectively within a range of hazard intensity that existing protection systems and emergency resources can manage. For extreme storm surges that exceed regional coping capacity, relying solely on adaptive capacity may not be sufficient to fully offset disaster impacts.

In this pathway, typical cases include Shanghai in 2014, 2016, 2017, and 2020, Jiangsu in 2017 and 2020, and Tianjin in 2018, 2020, and 2021, the comprehensive resilience supported by economic development is mainly reflected in three aspects. First, infrastructural resilience emphasizes ensuring that critical systems, including electricity, water supply, transportation, and communications, maintain basic operation during disasters. Its realization depends on higher engineering design standards, strict construction regulations, and continuous maintenance, upgrading, and protective investment. Second, socioeconomic resilience is mainly reflected in sound social security and economic recovery mechanisms. Through government emergency assistance, insurance protection, public services, and community mutual-aid networks, affected regions, households, and enterprises can restore production and daily life as quickly as possible. Third, ecological resilience emphasizes the protection and restoration of coastal ecosystems. By protecting natural barriers such as coastal wetlands and mangroves, the impacts of storm surges and waves on coastal areas can be weakened, thereby reducing direct economic losses (Menéndez et al., 2020).

In summary, L2 emphasizes reducing direct economic losses from storm surges by improving comprehensive regional resilience. Strengthening infrastructural support capacity, improving socioeconomic recovery mechanisms, and protecting coastal ecosystems can enhance the overall capacity of a region to withstand and recover from storm surge disasters. However, the presence of low SHI as a common core condition also indicates that the effectiveness of this pathway remains constrained by storm hazard intensity.

Overall, both pathways take low storm hazard intensity as a core condition, but their specific loss-mitigation mechanisms differ. L1 focuses on reducing coastal economic exposure through proactive avoidance and spatial regulation, thereby reducing the population, assets, and economic activities that may be affected by disasters at the source. L2 focuses on relying on socioeconomic development to improve the comprehensive resilience of infrastructure, public services, emergency management, and ecosystems, thereby reducing the extent to which disaster shocks are translated into actual economic losses. The two pathways reflect the equifinality characteristic of configurational analysis, indicating that low direct economic losses from storm surges can be achieved either by controlling exposure levels or by enhancing comprehensive regional adaptive capacity. However, the effective operation of both mitigation pathways is influenced by the level of disaster hazard intensity.

4.3.3 Robustness test

To test the robustness of the configurational results for low direct economic losses from storm surges, this study follows Du et al. (2020) and conducts set-theoretic robustness tests by adjusting the fsQCA analytical parameters. Generally, when the new configurations obtained after parameter adjustment are subsets of the original configurations and their substantive interpretations remain unchanged, the analytical results can be considered robust.

Specifically, this study conducts three robustness tests. First, the PRI-consistency threshold is increased from 0.70 to 0.75. Second, the case-frequency threshold is increased from 1 to 2. Third, the case-frequency threshold is simultaneously increased to 2 and the PRI-consistency threshold to 0.75. Under all three parameter adjustments, the identified configurational pathways remain consistent with the baseline results, namely ~SHI and ~CEE and ~SHI and RAC. The overall solution consistency and overall solution coverage also remain unchanged at 0.8899 and 0.6751, respectively. Because the pathways obtained from all robustness tests remain unchanged, the corresponding results are not presented separately.

The above results indicate that the configurational pathways identified in this study for low direct economic losses from storm surges are insensitive to reasonable adjustments in the PRI-consistency threshold and case-frequency threshold and therefore exhibit good robustness.

5 Conclusion and policy implications

5.1 Conclusion

Based on panel data from 11 coastal provinces and municipalities in China from 2011 to 2023, this study comprehensively applies the Tobit model, the two-part model, the panel threshold model, and fsQCA to analyze the nonlinear characteristics and mitigation pathways of direct economic losses from storm surges from the three perspectives of baseline regression, threshold effects, and configurational pathways. The main conclusions are as follows.

First, storm hazard intensity and coastal economic exposure significantly increase direct economic losses from storm surges, whereas regional adaptive capacity has a significant loss-mitigation effect. The two-part model further shows that SHI and CEE simultaneously increase both the probability of loss occurrence and the magnitude of losses, whereas RAC reduces both. SDV is significantly negative in the Tobit model but insignificant in the two-part model. The interaction term between SDV and ISC is weakly significant and negative in the Tobit model, preliminarily indicating that information support capacity may alleviate losses associated with population vulnerability through warning dissemination, identification of vulnerable groups, and emergency coordination.

Second, direct economic losses from storm surges exhibit significant nonlinear characteristics. The SHI model identifies a threshold value of 33.1157 m/s, whereas the CEE, SDV, and RAC models identify a threshold value of 34.7958 m/s. Once SHI exceeds the threshold value, its loss-amplifying effect increases significantly; CEE also exhibits clear regime changes before and after the threshold value; SDV shifts from a negative relationship to a significantly positive relationship, indicating that population vulnerability is more likely to be translated into actual losses under high-intensity disasters; RAC has a significant loss-mitigation effect in the low-intensity regime but is no longer significant in the high-intensity regime, indicating that the protective effect of regional adaptive capacity may fail under extreme disasters.

Third, low direct economic losses from storm surges exhibit causal complexity and equifinality. The necessity analysis shows that no antecedent condition alone constitutes a necessary condition for the low-loss outcome. The configurational analysis identifies two mitigation pathways. The first is the Defense Strategy-Driven Pathway, jointly constituted by low SHI and low CEE, which emphasizes reducing losses through proactive avoidance, spatial regulation, and exposure reduction. The second is the Comprehensive Resilience-Driven Pathway, jointly constituted by low SHI and high RAC, which emphasizes mitigating disaster impacts by relying on the resilience of infrastructure, public services, emergency management, and ecosystems. Both pathways pass the robustness tests and jointly indicate that the effectiveness of socioeconomic loss-mitigation measures is constrained by extreme physical hazard intensity.

5.2 Policy implications

Based on the above findings, this study proposes several policy implications to support coastal disaster risk governance and enhance resilience against storm surge impacts.

First, hazard-intensity thresholds should be incorporated as an important reference for tiered storm surge response. Storm surge early-warning systems may use the estimated thresholds within the range of 33–35 m/s as a reference for risk assessment and response escalation, while calibrating them to local coastal types, protection standards, and the spatial distribution of population and assets. When forecast wind speeds approach the thresholds, the focus of risk management should shift in advance from routine preparedness to the evacuation of vulnerable groups, the shutdown of critical facilities, the return of vessels to port, and the pre-positioning of emergency resources. Once the thresholds are exceeded, priority should be given to maintaining the operation of lifeline infrastructure, coordinating emergency resources across regions, and accelerating post-disaster recovery.

Second, differentiated mitigation pathways should be adopted according to regional hazard conditions and development characteristics. In coastal regions with relatively low development intensity and without large-scale asset concentration, priority should be given to the Defense Strategy-Driven Pathway. Coastal land-use regulation, risk-sensitive spatial planning, and approval controls for new development projects should be strengthened to prevent the formation of new high-exposure areas. In coastal cities where populations and industries are highly concentrated and existing exposure is difficult to reduce substantially, priority should be given to the Comprehensive Resilience-Driven Pathway by strengthening critical infrastructure, public services, emergency management, and post-disaster recovery capacity. In regions that are persistently exposed to high-intensity storm surge risks, exposure reduction and resilience enhancement should be implemented simultaneously.

Third, regional adaptive-capacity building should shift from general resource investment toward ensuring functional continuity under extreme scenarios. The threshold results show that conventional adaptive capacity can effectively reduce losses under relatively low hazard intensity, but its protective effect becomes less stable under high-intensity conditions. Disaster-prevention systems should therefore be designed with greater redundancy and a stronger capacity to maintain essential functions under extreme shocks. Specific measures include deploying backup power supplies and communication systems, establishing cross-regional mechanisms for rescue operations and emergency-material allocation, and improving the substitutability and operational continuity of transportation and energy networks.

Fourth, a coordinated mechanism combining information support with in-person assistance for vulnerable groups should be established. Information support capacity may help reduce losses associated with population vulnerability, but its effectiveness cannot depend solely on the scale of network infrastructure. Coastal regions should establish dynamically updated risk registries for older adults, children, persons with disabilities, and other mobility-impaired groups, while assigning responsibility for warning delivery, evacuation routes, shelter arrangements, and emergency assistance to local communities. A multi-channel warning system integrating mobile networks, broadcasting, satellite communications, community-based grid management, and door-to-door notification should also be developed to prevent information gaps caused by the digital divide. Information support should be implemented in coordination with evacuation, medical assistance, and emergency resource allocation rather than serving as a substitute for physical protection and in-person services.

5.3 Limitations and future research

This study has several limitations that provide directions for future research. First, the use of provincial-level panel data may obscure intra-regional heterogeneity, and future studies could employ finer spatial scales or geospatial data to capture localized disaster dynamics more accurately. Second, the analysis focuses on direct economic losses and does not account for indirect and long-term impacts, such as supply chain disruptions and macroeconomic spillovers, which warrant further investigation using network-based or input–output approaches. Third, key variables are measured by proxy indicators, which may not fully reflect the multidimensional nature of vulnerability and resilience; more comprehensive indicators could be incorporated in future research. Finally, while this study identifies nonlinear threshold effects and configurational mitigation pathways, the dynamic evolution of disaster systems and the generalizability of findings beyond China remain to be explored through longitudinal and cross-regional comparative analyses.

References

• 1

  BalanP.NordenL. (2024). Are measures of corporate financial constraints universal? Evidence from Brazil. Finance Res. Lett.70, 106353. doi: 10.1016/j.frl.2024.106353

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  CalafatF. M.WahlT.TadesseM. G.SparrowS. N. (2022). Trends in Europe storm surge extremes match the rate of sea-level rise. Nature603, 841–845. doi: 10.1038/s41586-022-04426-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  DebD.UddinM. M.Mahbub-E-KibriaA. S.DasM. K.HasanM. (2024). Coastal vulnerability assessment to multi hazards in the exposed coast of Southeastern Coastal Region of Bangladesh. Reg. Stud. Mar. Sci.73, 103484. doi: 10.1016/j.rsma.2024.103484

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  DuY.LiuQ.ChengJ. (2020). What kind of ecosystem for doing business will contribute to city-level high entrepreneurial activity? A research based on institutional configurations. J. Manage. World36, 141–155. doi: 10.19744/j.cnki.11-1235/f.2020.0143

  • CrossRef
  • Google Scholar

• 5

  FissP. C. (2011). Building better causal theories: A fuzzy set approach to typologies in organization research. Acad. Manage. J.54, 393–420. doi: 10.5465/amj.2011.60263120

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  FuX.HouJ.LiuQ.LiM.LiangS. (2023). Evaluation of surge hazard based on a storm surge hazard indicator along the mainland coast of China. Nat. Hazard.116, 3481–3493. doi: 10.1007/s11069-023-05820-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  GhoraiD.SenH. S. (2015). Role of climate change in increasing occurrences of oceanic hazards as a potential threat to coastal ecology. Nat. Hazard.75, 1223–1245. doi: 10.1007/s11069-014-1368-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  GuoH.YinK.HuangC. (2022). Modeling of direct economic losses of storm surge disasters based on a novel hybrid forecasting system. Front. Mar. Sci.8, 804541. doi: 10.3389/fmars.2021.804541

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  GuoT.LiG. (2020). Study on methods to identify the impact factors of economic losses due to typhoon storm surge based on confirmatory factor analysis. Nat. Hazard.100, 515–534. doi: 10.1007/s11069-019-03823-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  GuoY.van de LindtJ. (2019). Simulation of hurricane wind fields for community resilience applications: A data-driven approach using integrated asymmetric Holland models for inner and outer core regions. J. Struct. Eng.145, 04019089. doi: 10.1061/(asce)st.1943-541x.0002366

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  HanR.YinK.DaiX.LiH.ZhouS. (2025). Economic footprint assessment of storm surge disasters in China based on disastrously-extended input-output analysis. Front. Mar. Sci.12, 1673928. doi: 10.3389/fmars.2025.1673928

  • CrossRef
  • Google Scholar

• 12

  HansenB. E. (1999). Threshold effects in non-dynamic panels: Estimation, testing, and inference. J. Econometrics93, 345–368.

  • Google Scholar
  • View reference in article

• 13

  LiL.LiZ.HeZ.YuZ.RenY. (2022). Investigation of storm tides induced by super typhoon in macro-tidal Hangzhou Bay. Front. Mar. Sci.9, 890285. doi: 10.3389/fmars.2022.890285

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  LiY.ZhangX.ZhaoX.MaS.CaoH.CaoL. (2016). Assessing spatial vulnerability from rapid urbanization to inform coastal urban regional planning. Ocean Coast. Manage.123, 53–65. doi: 10.1016/j.ocecoaman.2016.01.010

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  LiY.ZhouW. H.ShenP. (2023). Flood risk assessment of loss of life for a coastal city under the compound effect of storm surge and rainfall. Urban Clim.47, 101396. doi: 10.1016/j.uclim.2022.101396

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  LinJ.WangS.ZhaoY.XueX. (2024). The inequality of economic development increases the direct economic loss of storm surge disaster in China. Ocean Coast. Manage.258, 107427. doi: 10.1016/j.ocecoaman.2024.107427

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  LinJ.XuY.HouY.XueX. (2023). Spatio-temporal distribution, composition and influencing factors of economic losses from storm surge disasters: An empirical study from China (2007-2016). Int. J. Disaster Risk Reduct.90, 103669. doi: 10.1016/j.ijdrr.2023.103669

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  LockwoodJ. W.LinN.OppenheimerM.LaiC. (2022). Using neural networks to predict hurricane storm surge and to assess the sensitivity of surge to storm characteristics. J. Geophysical Research: Atmospheres127, e2022JD037617. doi: 10.1029/2022jd037617

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  MenendezP.LosadaI. J.Torres-OrtegaS.NarayanS.BeckM. W. (2020). The global flood protection benefits of mangroves. Sci. Rep.10, 4404. doi: 10.1038/s41598-020-61136-6

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  MengH.XuH.CuiY.LiuG. (2024). International container intermodal competitiveness: An empirical study from Chinese hub ports. Ocean Coast. Manage.254, 107178. doi: 10.1016/j.ocecoaman.2024.107178

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  NeedhamH. F.KeimB. D.SathiarajD. (2015). A review of tropical cyclone-generated storm surges: Global data sources, observations, and impacts. Rev. Geophys.53, 545–591. doi: 10.1002/2014rg000477

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  NohrstedtD.HilemanJ.MazzoleniM.BaldassarreG. D.ParkerC. F. (2022). Exploring disaster impacts on adaptation actions in 549 cities worldwide. Nat. Commun.13, 3360. doi: 10.1038/s41467-022-31059-z

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  Pineau-GuillouL.BouinM. N.ArdhuinF.LyardF.BidlotJ. R.ChapronB. (2020). Impact of wave-dependent stress on storm surge simulations in the North Sea: Ocean model evaluation against in situ and satellite observations. Ocean Modell.154, 101694. doi: 10.1016/j.ocemod.2020.101694

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  RaginC. C. (2008). Redesigning Social Inquiry: Fuzzy Sets and Beyond (Chicago: University of Chicago Press).

  • Google Scholar
  • View reference in article

• 25

  ShiX.LvY.DongD.JiaN.GeJ.YinJ. (2024). Quantitative assessment of building risks and loss ratios caused by storm surge disasters: A case study of Xiamen, China. Appl. Ocean Res.145, 103934. doi: 10.1016/j.apor.2024.103934

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  ShiX.TanJ.GuoZ.LiuQ. (2013). A review of risk assessment of storm surge disaster. Adv. Earth Sci.28, 866–878. doi: 10.11867/j.issn.1001-8166.2013.08.0866

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  SuiX.HuM.WangH.ZhaoL. (2023). Improved elasticity estimation model for typhoon storm surge losses in China. Nat. Hazard.116, 2363–2381. doi: 10.1007/s11069-022-05768-z

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  SunZ.HuangS.NieH.JiaoJ.HuangS.ZhuL.et al. (2015). Risk analysis of seawall overflowed by storm surge during super typhoon. Ocean Eng.107, 178–185. doi: 10.1016/j.oceaneng.2015.07.041

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  SunH.WangJ.YeW. (2021). A data augmentation-based evaluation system for regional direct economic losses of storm surge disasters. Int. J. Environ. Res. Public Health18, 2918. doi: 10.3390/ijerph18062918

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 30

  TienV. P.ThuyN. B.KimS.CuongN. K.NgocP. K.KhiemM. V.et al. (2025). Impact of the interaction of surge, wave, and tide on the surge and wave on the northern coast of Vietnam for a marine storm surge and wave forecast system. Reg. Stud. Mar. Sci.87, 104234. doi: 10.1016/j.rsma.2025.104234

  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  WangT.TanX.TianY.HuangX.ChenX.XiaoP.et al. (2025). Risk assessment based on Bayesian network for the typhoon-storm surge-flood-dike burst disaster chain: A case study of Guangdong, China. J. Hydrology: Regional Stud.58, 102251. doi: 10.1016/j.ejrh.2025.102251

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  WangF.WuM.WangJ. (2023). Can increasing economic complexity improve China's green development efficiency? Energy Econ.117, 106443. doi: 10.2139/ssrn.3985233

  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  WangK.YangY.ReniersG.LiJ.HuangQ. (2022). Predicting the spatial distribution of direct economic losses from typhoon storm surge disasters using case-based reasoning. Int. J. Disaster Risk Reduct.68, 102704. doi: 10.1016/j.ijdrr.2021.102704

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  WangX.ZhengY.XiaoX.WangX.WuJ.NieM.et al. (2026). China's coastal zone shifts from land-dependent to green and sustainable development. Geogr. Sustainability, 100434. doi: 10.1016/j.geosus.2026.100434

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  WuS.ZhangR.WangC.FengD. (2022). The impact of natural disasters on rural household wealth: Micro evidence from China. Front. Environ. Sci.10, 993722. doi: 10.3389/fenvs.2022.993722

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  XiongD.LiW. (2025). Digital economy and urban economic resilience: Evidence from China's national big data policy. Environ. Dev. Sustainability, 1–27. doi: 10.1007/s10668-025-07151-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 37

  YangX.YaoY.TianK.JiangW.XingQ.YangJ.et al. (2023). Disaster response strategies of governments and social organizations: From the perspective of infrastructure damage and asymmetric resource dependence. Heliyon9, e20349. doi: 10.1016/j.heliyon.2023.e20432

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  YangY.ZhangY.JiangL.XuH. (2025). Assessing ecological efficiency, spatial spillovers, and configurational drivers of recreational fisheries in the United States. Ocean Coast. Manage.269, 107803. doi: 10.1016/j.ocecoaman.2025.107803

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  YiX.ShengK.WangY.WangS. (2021). Can economic development alleviate storm surge disaster losses in coastal areas of China? Mar. Policy129, 104531. doi: 10.1016/j.marpol.2021.104531

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  YinK.HanR.HuangC. (2023). Effect on high-quality economic development of foreign direct investment in China from the triple perspectives of financial development. J. Cleaner Prod.427, 139251. doi: 10.1016/j.jclepro.2023.139251

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 41

  YinK.ZhaoY.ZhouS.LiX. (2024). How do storm surge disaster losses affect economic development? Perspectives from disaster prevention and mitigation capacity. Sci. Total Environ.951, 175526. doi: 10.1016/j.scitotenv.2024.175526

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 42

  YuanH.GaoX.QiW. (2019). Fine-scale spatiotemporal analysis of population vulnerability to earthquake disasters: Theoretical models and application to cities. Sustainability11, 2149. doi: 10.3390/su11072149

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 43

  ZhaiA. R.JiangJ. H. (2014). Dependence of U.S. hurricane economic loss on maximum wind speed and storm size. Environ. Res. Lett.9, 64019. doi: 10.1088/1748-9326/9/6/064019

  • CrossRef
  • Google Scholar
  • View reference in article

• 44

  ZhaiW.ZengQ.WangL. (2025). Configurational pathways to port environmental efficiency: An fsQCA-NCA analysis of CO2 reduction drivers in port management. Reg. Stud. Mar. Sci.92, 104618. doi: 10.1016/j.rsma.2025.104618

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 45

  ZhangC.YinK.ShiX.YanX. (2021). Risk assessment for typhoon storm surges using geospatial techniques for the coastal areas of Guangdong, China. Ocean Coast. Manage.213, 105880. doi: 10.1016/j.ocecoaman.2021.105880

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 46

  ZhangM.ZhouC.ZhangJ.ZhangX.TangZ. (2022). Numerical simulation and analysis of storm surges under different extreme weather event and typhoon experiments in the South Yellow Sea. J. Ocean Univ. China21, 1–14. doi: 10.1007/s11802-022-4750-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article