An integrated assessment of tropical cyclone risks in mainland China by considering hazard, exposure, vulnerability and mitigation

Tropical cyclones (TCs), which are rapidly rotating weather systems associated with rainstorms, gales and storm surges, cause enormous property losses and casualties every year (Dong et al. 2010; Paerl et al. 2019; Yang et al. 2020). From 1912 to 2018, it is estimated that around the world, TCs have resulted in 175,500 deaths and 88.35 × 10⁷ United States dollars (USD) of economic losses (Debarati 2017). Located next to the Northwest Pacific Ocean, China is one of the most TC-prone nations in the world and suffers substantially from TCs (Ying et al. 2011; Yang et al. 2020). For instance, the super typhoons of Hato in 2017, Mangkhut in 2018 and Lekima in 2019 affected 14.02 million people and 1139.70 thousand hectares of cropland and led to a direct economic loss of 7.97 billion USD. To better cope with TCs, the spatial pattern of TC risks in the coastal and inland regions must be investigated (Nguyen et al. 2019; Sajjad et al. 2020; Mansour et al. 2021).

Generally, TC risks can be measured by TC hazard models, disaster loss functions and assessment indicators (Barcikowska et al. 2017; Knutson et al. 2020; Lyu et al. 2020). The different approaches pay attention to different aspects of TC risks: hazard models present explicit simulations of the TC intensity, frequency and track under different hydroclimatic scenarios (Barcikowska et al. 2017); disaster loss functions directly link attributes of TCs to socio-economic factors by fitting the joint probability distributions between TC destructiveness and economic losses, casualties and other damages (Bakkensen et al. 2018; Steptoe & Economou 2020); and a large number of indicators on TC hazard, exposure, vulnerability and mitigation capacity are proposed and then applied to the assessment of TC risks (e.g., Nakamura et al. 2015; Lyu et al. 2018).

According to the definitions proposed by the Intergovernmental Panel on Climate Change (IPCC) and the United Nations International Strategy for Disaster Reduction (UNISDR), the risk is composed of the probability of hazard and its negative consequences in hazard-formative environments. Peduzzi et al. (2009) incorporated the exposure into the framework of ‘Hazard–Vulnerability’ (HV) to identify the relationship between vulnerable elements and disaster losses. The framework of Hazard–Exposure–Vulnerability (HEV) was established (Cutter et al. 2013). In recent years, an increasing number of studies highlighted that the mitigation capacity plays an important part in alleviating disaster losses (Lyu et al. 2018, 2020; Nguyen et al. 2019). As a result, the framework of HEV–Mitigation (HEVM) was proposed for risk assessment (Hoque et al. 2018, 2019).

In this paper, the role of mitigation capacity in determining TC risks is investigated under the frameworks of HEV and HEVM (Frigerio & De Amicis 2016; Hoque et al. 2019). Specifically, attention is paid to four typical formulae of risk assessment: the multiplicative formula integrating the hazard, exposure and vulnerability attributes of risk from the perspective of joint probability of independent variables (Zhang & Chen 2019; Ali et al. 2020; Sajjad et al. 2020); the multiplicative/divisive formula incorporating the mitigation capacity as a divisor into the multiplicative formula (Hoque et al. 2019; Nguyen et al. 2019; Sajjad & Chan 2019); the additive formula adding up the attributes of hazard, exposure and vulnerability (Lummen & Yamada 2014; Lyu et al. 2018, 2020); and the additive/subtractive formula accounting for the mitigation as a subtractor in the additive formula (Yang et al. 2013; Ouma & Tateishi 2014; Lyu et al. 2020). Furthermore, the TC risks calculated by the above four formulae are validated by using the data of economic losses and casualties. Therefore, the objectives of this paper are three-fold: (1) to compare the TC risks obtained by the four typical formulae; (2) to examine the effects of hazard, exposure, vulnerability and mitigation in determining TC risks; and (3) to identify a suitable formula for the assessment of TC risks in China.

This paper focuses on mainland China and takes provinces that are affected by TCs into account. Specifically, the 6-h interval locations of TC centers, which are obtained from the 1949 to 2017 TC best track data provided by the China Meteorological Administration (CMA), are pooled to form TC tracks; a buffer distance of 250 km is then applied to each TC track to derive the average radius extent of TCs (Dong et al. 2010; Steptoe & Economou 2020). As shown in Figure 1, 28 provinces intersecting with the TC buffer zones are selected as the case study region in the analysis.

The sources of the data on TC hazard, exposure, vulnerability and mitigation are listed in Table S1 of the supporting information for the sake of brevity. The best track dataset of the CMA has been widely used to characterize the central location, atmospheric pressure, maximum wind speed and frequency of the TC hazards in East Asia (Zhong et al. 2018; Lyu et al. 2020; Yang et al. 2020). The associated rainfall is also observed by the meteorological station in the ‘China Ground Climate Data Daily Value Dataset (1952–2017)’. The land use/land cover, road network, population, GDP, medical condition, household deposit and public fiscal revenue collected from different sources are used to characterize TC exposure and mitigation (Zhong et al. 2018; Zhang & Chen 2019; Lyu et al. 2020). Topographic and geomorphic datasets, in particular the digital elevation model (DEM), are adopted to extract vulnerability indicators of hazard-formative environment, such as slope, elevation, track proximity and river network density (Gao et al. 2014; Hoque et al. 2019; Zhang & Chen 2019). It is noted that all the data layers have been resampled to 1 × 1-km grid cells in ArcMap 10.5, considering that the original data are at different spatial resolutions.

The maximum wind speed has been identified as the primary indicator of TC intensity (Qi & Du 2018; Hoque et al. 2019). Besides, TC track frequency, process precipitation and duration are used to represent the spatial patterns of TC severity based on the historical observation data (Nakamura et al. 2015; Black & Gabriele 2019). This paper employs the indicators of maximum sustained wind speed, process rainfall, landfall frequency and average duration in hazard estimation (Qi & Du 2018; Fang & Zhang 2021; Yin et al. 2021). It is noted that storm surge height, which has a limited extent and only affects coastal areas, is not considered in this paper.

Population and properties, e.g., houses, croplands and public facilities, are highly likely to suffer from disaster-related losses and are selected as the exposure indicators (Peduzzi et al. 2009; Geiger et al. 2018). In this paper, gridded GDP density, population density and land use/land cover are used to represent exposure to TCs (Peduzzi et al. 2009; Zhang & Chen 2019). These datasets represent the sensitive objects affected by TCs from different aspects.

Vulnerability is the susceptible state of vulnerable elements, which involves economic, geographic, demographic and other factors in a broad sense (Yang et al. 2021). Previous studies highlighted the vulnerability that includes topography, vegetation, river networks and age structure (Peduzzi et al. 2009; Cutter et al. 2013; Aerts et al. 2018). In recent years, attention has also been paid to the social and environmental attributes of vulnerability (Frigerio & De Amicis 2016; Kilroy et al. 2016). Therefore, the ratio of primary industry, the ratio of tertiary industry, the ratio of vulnerable people, vegetation cover index, elevation, slope, river network density and coastal proximity toward TC disasters are considered in the analysis (Lummen & Yamada 2014; Chakraborty & Joshi 2016; Hoque et al. 2019; Ali et al. 2020; Mansour et al. 2021).

Mitigation capacity is associated with regional public disaster reduction facilities and economic conditions (Shreve & Kelman 2014; Frigerio & De Amicis 2016). For example, traffic accessibility and medical availability are the key factors in the treatment of wounded, transportation of materials and reconstruction after disasters (Cutter et al. 2013; Frigerio & De Amicis 2016; Hoque et al. 2019). Therefore, the medical levels of sickbeds/doctors per capita, public fiscal revenue and road network density are used as the indicators of mitigation capacity (Hoque et al. 2019). To consider the influence of resident assets, the per capita household deposit is also adopted (Sajjad & Chan 2019).

Coastal proximity, altitude, surface roughness, river network density, vegetation coverage index and mitigation capacity are categorized as negative indicators (Saaty 2008; Zhang et al. 2021).

The judgment matrix is constructed with the importance ranking of indicators in recent studies on TC risks (Hoque et al. 2019; Lyu et al. 2020). The weights of indicators and criteria are obtained by Equations (4)–(6) based on the importance ranking within and between indicators.

Both Equations (8) and (9) are formulated based on the HEV framework that has been widely applied to assess natural disaster risks (e.g. TC and flood) (Frigerio & De Amicis 2016; Aerts et al. 2018; Hoque et al. 2019; Mansour et al. 2021; Zhang et al. 2021).

The spatial analysis and reclassification tools in ArcMap 10.5 are used for the calculation of the TC hazard, exposure, vulnerability mitigation and risk. The results are classified into five levels, i.e., very low, low, moderate, high and very high, by using the natural break method (Jenks 1967).

The overall level of vulnerability is shown in Figure 5(i). It can be seen that regions with very high vulnerability generally do not coincide with regions exhibiting high hazards. The vulnerability tends to be more varied in inland areas than in coastal areas, with the high vulnerability mainly observed in settlement areas and the other levels observed in cropland, forest and grass areas. The moderate level of vulnerability covers about one-third of the study region, which is due to the proximity to the coastline, vegetation coverage index and terrain roughness. The very low and low vulnerability zones are distributed in inland areas next to the Qinghai–Tibet plateau.

The average risk values at the provincial scale are plotted against the density of direct economic losses for the four formulae in Figure 8. To better illustrate the relationship between TC risks and economic losses, a linear relationship is fitted for the two variables. The popular goodness of fit (R²) is selected to indicate the strength of the dependence relationship (Zhang et al. 2021; Huang & Zhao 2022). It can be seen that HEV-2 and HEVM-2, respectively, lead to higher R² than HEV-1 and HEVM-1. Specifically, the R² value is 0.40 for HEV-1 and 0.29 for HEVM-1; in the meantime, the R² value is 0.55 for HEV-2 and 0.56 for HEVM-2. Therefore, in comparison to the multiplicative/divisive formula, the additive/subtractive formula can improve the goodness of fit between the risk value and economic loss density. Furthermore, it is noted that the comparison of R² for HEV-2 and HEVM-2 suggests that the consideration of mitigation capacity can contribute to the goodness of fit.

The relationship between TC risk values and the density of casualties is shown in Figure 9. Similar to the results in Figure 8, the goodness of fit is the highest for HEVM-2, followed by that for HEV-2. This result confirms the effectiveness of the additive/subtractive formula and the importance of mitigation capacity. Compared to Figure 8, the goodness of fit for casualties in Figure 9 is overall lower but remains significant. This result is owing to the fact that TC casualties are caused not only directly by TCs but also by TC-related secondary disasters, including floods, landslides and debris flows (Wang & Chen 2016; Kundzewicz et al. 2018; Yin et al. 2021). It is noted that besides coastal provinces of Zhejiang, Fujian, Guangdong, Guangxi and Hainan, and inland provinces with moderate TC risk, including Jiangxi and Hunan, can also suffer from casualties.

Following TC indicators used in peer studies (Yang et al. 2016; Liu et al. 2020; Lyu et al. 2020), this paper has illustrated the spatial patterns of TC hazard, exposure, vulnerability and mitigation for TC-affected coastal and inland areas in China. The analysis is built upon the datasets of TC tracks, disaster losses, socio-economic census and surface underlying conditions. The results show that TC hazard levels tend to attenuate from the southeast coastal area to the northwest inland area owing to the landfall frequency and duration of TCs (Fang & Zhang 2021). Through the analysis of disaster losses (direct economic loss and casualties) during the period from 2004 to 2017, it is found that inland provinces, such as Yunnan, Hunan, Hubei and Henan, can also suffer from TCs (Figure 8). Although the TC frequency of these provinces is low, the extreme TC events contribute to plenty of disaster losses due to high vulnerability and low mitigation capacity (Shreve & Kelman 2014). Therefore, besides coastal provinces, attention can also be paid to inland provinces to build the capacity to cope with TC disasters (Ying et al. 2011).

The TC risks estimated by the multiplicative, multiplicative/divisive, additive and additive/subtractive formulae can serve as references for the assessment of TC risks in China. Besides the different formulae, TC risks can be calculated by using process-based models. For example, Peduzzi et al. (2012) employed ocean and climate models to forecast the future TC intensity, exposure and empirical loss functions under certain climate scenarios for TC risks of global coastal countries. Nakamura et al. (2015) projected the migration of TC tracks with an ensemble of numerical models from phase 5 of the Coupled Model Inter comparison Project (CMIP5). Miao et al. (2018) used the Inter-criteria Correlation (CRITIC) and gray relational analysis (GRA) methods to evaluate the population's vulnerability to disasters. These models account for the atmospheric processes and describe how TCs lead to the disasters of floods, landslides and debris flows (Shreve & Kelman 2014; Bakkensen et al. 2018; Sajjad et al. 2019). In the future, more efforts can be devoted to performing process-based analysis of TC hazards, exposure, vulnerability and mitigation to yield insights into TC risks.

This paper has implemented four typical formulae to assess TC risks in China based on the datasets of TC tracks, socio-economic census, disaster losses and surface underlying conditions. The indicators of hazard, exposure and vulnerability are integrated into the multiplicative formula HEV-1 and the additive formula HEV-2; the mitigation capacity is furthermore considered by the multiplicative/divisive formula HEVM-1 and the additive/subtractive formula HEVM-2. The results show that TC risks under HEV-1 and HEVM-1 tend to concentrate in large cities; by contrast, TC risks under HEV-2 and HEVM-2 are more dispersed and show an attenuating tendency from coastal areas to inland areas. For validation, disaster losses and casualties collected from the Yearbooks of Meteorological Disasters are applied to examine the estimated TC risks. The results suggest the use of the additive/subtractive formula and highlight the importance of mitigation capacity. Overall, the estimated TC risks and the spatial plots of TC hazard, exposure, vulnerability and mitigation serve as a useful tool for TC risk zoning and preparation.

This research is supported by the National Natural Science Foundation of China (U1911204, 51979295, 51861125203 and 52109046), the National Key Research and Development Program of China (2021YFC3001000) and the Guangdong Provincial Department of Science and Technology (2019ZT08G090).

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

The authors declare there is no conflict.