Dynamic risk assessment of drought disaster: a case study of Jiangxi Province, China

The dynamic risk assessment of drought is crucial in the transition from the crisis management model to the risk management model, which can reveal the evolution mechanism of drought disasters. Due to a lack of data and research perspectives, most current studies are still based on static risk assessment. This study proposes a conceptual model for the dynamic risk assessment of droughts based on the probability of their occurrence and potential impacts. The developed dynamic risk index considers the hazard, exposure, vulnerability, and capacity for drought mitigation. The analytic hierarchy process (AHP) method was used to determine the weight coefficient of each indicator in the model. The novelty of the proposed model lies in the integration of four elements of drought disasters with spatiotemporal characteristics. Jiangxi Province, which is frequently affected by drought, was selected as the study area to validate the proposed model. Experimental results demonstrate that the proposed model rapidly reflects the degree of drought disaster risk caused by drought events and the influencing factors at monthly and annual scales. Moreover, the datasets based on the influencing factors of drought disasters in different regions have a good commonality in the proposed model.


INTRODUCTION
Drought is probably the most complex and severe natural disaster, which affects more people than any other type of natural hazard (Wilhite ). Globally, it has the widest range of influence, the longest duration of prevalence, and causes huge economic loss (Sheffield et al. ).
The risk of drought disasters is their possible impact on social, economic, and natural environmental systems (Zhang ). Owing broadly to a lack of data and varying research perspectives, most of the drought disaster risk research does not involve the essential characteristics of the risk of loss caused by drought disaster and only focuses on drought risk. Researchers use the theory of risk formation based on natural disasters for drought risk assessment. A common tool for drought risk assessment is to estimate the drought indices using observable meteorological and hydrological data, which can be used as an individual index or as a combination of other indices. A variety of drought indices have been proposed, such as the Palmer drought severity index (PDSI) (Vasiliades et al. ), the standardized precipitation index (SPI) (Raziei et al. ), the standardized precipitation evapotranspiration index (SPEI) (Hernandez & Uddameri ), the soil moisture deficit index (SMDI) (Narasimhan & Srinivasan ), and the normalized difference vegetation index (NDVI) (Choudhury et al. ). Integrated analysis of the frequency, recurrence period, exposure, and vulnerability are the variables that are used to characterize the risk posed by droughts (Birkmann ). Studies dynamically assessing the risk of drought disasters by considering all four factors (i.e., hazard, exposure, vulnerability, and capacity for drought mitigation) of drought disasters are rare. Therefore, with increased data availability and improved computation techniques, a dynamic risk assessment of drought disasters based on the spatiotemporal change in hazards, exposure, vulnerability, and capacity for drought mitigation is essential.
Research on the dynamic assessment of natural disaster risks has not progressed as desired owing to the negligence of time-dependent changes in risks, resulting in non-technical conceptual studies (Huang ). In 1988, the Office of () analyzed the relationship between water shortage and drought risk in three stages (before, during, and after the drought disaster) and showed that the contributions of the hazard factors to the risk varied at the different stages.
The current trend in drought disaster research is the transformation from drought crisis management to drought risk management (Wilhite ). Dynamic monitoring and assessment of drought disasters at different scales is the most vital component of drought risk management. A dynamic risk assessment of drought disaster can reveal their spatiotemporal evolution mechanisms and provide essential inputs for the scientific formulation of disaster reduction plans. The main purpose of this study is to propose a conceptual model for the dynamic risk assessment of drought disasters through the construction of a mathematical model based on the interactions between the factors of drought at different times. To achieve this, the analytic hierarchy process (AHP) (Saaty ) method that uses weight coefficients of the hazard, exposure, vulnerability, and capacity for drought mitigation indicators was developed.
This model was applied to the study area to dynamically analyze the integrated risk of drought disaster.

Study area
The study area, Jiangxi Province (Figure 1), is located in southeastern China (from 113 34 0 36″ E to 118 28 0 58″ E and from 24 29 0 14″ N to 30 04 0 41″ N). It spans about 620 km from north to south and is 490 km wide from east to west. The total land area of the province is 166,948 km 2 , accounting for 1.74% of the total land area of China. It is located near the Tropic of Cancer and belongs to a subtropical warm and humid monsoon climate. It is warm and rainy in spring and hot and humid in summer.
The annual average temperature ranges from 16.3 to 19.5 C, increasing from north to south, and the province experiences 240-307 days of a frost-free period per annum.
The average annual temperature of northeast Jiangxi, the northwestern mountainous region of Jiangxi, and the Poyang Lake Plain is 16.3-17.5 C. The average annual temperature of Gannan Basin is 19.0-19.5 C, and the extreme maximum temperature is above 40 C at one of the hottest areas in the middle reaches of the Yangtze River. The daily average temperature is stable over 10 C for a duration of 240-270 days, and the active accumulated temperature is 5,000-6,000 C within this period. The  region-wise ratios represent the scale of the annual disastertolerant component. The capacity for drought mitigation was calculated using capacities of the beneficial reservoirs and the regions serviced by them. Table 1 provides detailed information on the data used in this study for the dynamic risk assessment of drought disasters in Jiangxi Province.

Methodology
Conceptual model for the dynamic risk assessment of drought disasters Drought disaster risk has both natural and social attributes (Martin et al. ). The objective of a dynamic risk assessment of drought disasters is to construct a dynamic model that can represent the non-linear changes in the risks over time and space. The model presented here not only determines the form of representation of drought disaster risks but also expresses the method of dynamic risk analysis.
Scholars who study the theory of disaster systems believe that disasters are the result of interactions between hazards and the potential impact of disasters (Smith ). Therefore, the concept of risk in the context of drought disasters is defined as follows: where P(Disasters) is the probability of the occurrence of drought disasters, and Effect is the potential impact of drought disasters.
The dynamic risk assessment of drought disaster with four influencing factors is a more comprehensive consideration in current drought disaster assessment research. Accordingly, the integrated risk (R t ) of a drought disaster at time t can be expressed as follows: where R t is the integrated risk of the drought disaster at time t, P(Disasters) t is the probability of the drought disaster at time t, and Effect t is the potential impact of the corresponding drought disaster at time t. The potential impact of drought disasters is a result of the combined effects of prevalent disaster environmental factors, disaster-bearing factors,

Hazard assessment of drought disasters
The hazard of drought disaster is the natural variation factor and abnormal degree of drought disaster, which includes the dynamic change of drought disaster in time, space, and intensity. Based on the characteristics of the drought disaster, the hazard of the drought disaster can be identified and analyzed. The SPEI has the characteristics of multiple timescales, which can represent various types of droughts and better reflect the variations in drought characteristics.
The SPEI was utilized as the drought event identification index, and a drought event was identified when SPEI was less than À1. The risk of the drought disaster hazard is expressed by the absolute value of the difference between the probabilities of SPEI of the drought event and that of SPEI being À1, as follows: where h x is the hazard risk when SPEI is x, p x is the probability when SPEI is x, and p À1 is the probability when SPEI is À1 (Zhang et al. ).
There are two types of non-drought moments during a drought event: (1) the time when the SPEI value is greater than À1 occurs in the middle of the drought event and (2) the time when the SPEI is greater than À1 occurs at the beginning and end of the drought event. When the SPEI value is greater than À1, p x À p À1 is greater than zero.
Months with an SPEI greater than À1 occur at different times during the drought event, and the associated risks differ. In such situations, h x cannot be calculated using formula (3). For the first case, the risk index for a particular month is calculated by adding the average hazard risk value of the month (when the SPEI value of the preceding and following months was less than À1) and the p x À p À1 value of the month. In the second case, p x À p À1 is the risk index.

Exposure assessment of drought disasters
Exposure to drought disasters is a measure of the degree of contact between various objects affected by the drought. In China, statistics on losses caused by drought disasters are available for different administrative divisions. Based on the area of each city or region and the area of each land-use type present within each city or region, the exposure ratios for each land-use type within each city or region were computed (Chen et al. ). Moreover, this land-use information, reflecting various types of surface features, is spatiotemporal. Usually, when the timescale considered is small, many types of features do not change significantly, and when it is large, the changes are ignored. The main landuse type in the study area is natural vegetation, and its growing period extends far more than 1 year. Furthermore, in this study, the annual scale was selected as the unit of time for the extraction of land-use information. Therefore, the drought disaster exposure index based on the surface-cover type in a certain year is as follows: where E(mn) t is the drought disaster exposure index of the mth city or region and nth feature at time t; a(mn) t is the area of the mth city or region and nth feature at time t.

Vulnerability assessment of drought disaster
Vulnerability to a disaster is common (Blaikie et al. ) and Thus, the vulnerability index V ij of the affected component in year i and month j was obtained as follows: where V ij is the vulnerability index of year i and month j, F i is the ratio of the non-agricultural to the agricultural population in year i, and N j is the dimensionless vegetation health coefficient in the jth month of the year.
The dimensionless vegetation health coefficient N j was obtained from the normalization of the monthly average NDVI values.
where NDVI j is the multi-year monthly average NDVI value for the jth month.

Disaster reduction capacity assessment
Disaster reduction capacity is the ability to resist and recover from drought disasters in risk areas, and the reservoir is an important engineering measure to deal with the losses caused by drought disasters. While considering the storage and water supply capacity of the reservoirs, the upstream and downstream relationships of the relevant reservoirs and the data on water supply and water transfer were found to be lacking. Therefore, the disaster reduction process of the reservoirs was simplified. It was assumed that the reservoirs provide water supply to their respective captive cities or regions only, and hence, the reservoir storage capacity at a certain time was divided by the city/region area to derive the disaster reduction capacity of the city/region (Equation (7)) (Ehsani et al. ).
where a m,t is the disaster reduction capacity of the mth city/ region at time t, (c i ) m,t is the reservoir storage capacity of the ith reservoir in the mth city/region at time t, s m is the total area of the mth city/region, and n is the number of all the reservoirs in the mth city/region.

Disaster reduction capacity varies for different cities/
regions, and these can be ranked by a difference of 1 between them. The normalized disaster reduction capacity index of each city/region was obtained as follows: where A m,t is the disaster reduction capacity index of the mth city at time t and a m,t is the disaster reduction capacity of the mth city at time t. There were 11 cities in the study area.

Determination of index weight of the model
The risk assessment based on the four factors of disaster theory is in a primary stage, and there are still significant differences and uncertainties in the judgment of the importance of each factor by different experts. AHP and fuzzy analytical hierarchy process (FAHP) (Chang ) are commonly used methods to determine the weight coefficients of each index in the model. The AHP method makes use of the characteristics of behavioral science to quantify the empirical judgment of decision-makers. When the factor structure is complex and there is a lack of necessary data, this method is more practical. FAHP is a highly complex process compared with AHP. Therefore, in this study, AHP was adopted to calculate the weight coefficient of each factor. In the selection of the scale, compared with the original one-to-nine scale, the four-factor scale makes it easier to judge the importance of each index, and the consistency check process is omitted. This method is simple, fast, and the accuracy conforms to the calculation requirements. Through this method, the four-factor scale of drought disasters was constructed by approximately replacing the results of direct expert investigations.  (Table 3).
Based on the dynamic risk assessment model and the weight coefficients of the various indicators in Table 3, the calculation formula for the dynamic risk assessment of drought disasters in the model is: where R t is the integrated risk of a drought disaster at time t; h t is the risk index of hazard factors at time t; e t is the risk of exposure at time t, including forest, wetland, dryland (i.e., arable land, grassland, and others), and artificial surfaces; v t is the risk of vulnerability at time t; and a t is the capacity for drought mitigation at time t. The SPEI was utilized as the drought identification index. As the study area is located in the subtropical monsoon climate region, the water cycle is rapid, such that the monthly scale is more sensitive, which can reflect the development trend of drought events (Duan ).
Simultaneously, the monthly dynamic risk assessment can conform to the requirements of disaster risk management for the early warning of drought disasters and achieve the purpose of effectively mitigating the impact of drought disasters (Grasso & Singh ). Therefore, the monthly scale was selected for drought event identification and drought disaster risk analysis. Finally, the annual scale was selected to analyze the total dynamic change process in drought disaster risk from 2003 to 2013.

Dynamic risk assessment of the drought disaster in 2003
The risk presented by drought disaster events, calculated based on the SPEI probability of each meteorological station, was used to obtain the spatial distribution of the risk using the inverse distance weighted (IDW) spatial interpolation method (Figure 3). The study area was dominated by a summer drought in 2003, which occurred in the central and northeast regions of the study area in June and July. This drought spread throughout the entire province between August and September, and gradually eased from north to south from October to November.
The disaster reduction capacity of each city/region before the occurrence of the drought was calculated based on the cumulative rainfall of the cities/regions in the study area 3 months before the drought. It was assumed that each reservoir only has a 4-month drought resistance capacity, and the disaster reduction capacity was considered the same for the months in between. Finally, the integrated risk for each city/region was calculated. As shown in Figures 3 and 4, the drought phenomenon first appeared in the central area of the study area in June.
Since agriculture is mainly distributed in the central region of Jiangxi Province, the scope and intensity of the integrated drought disaster risk are higher than the risk of drought events. At the same time, Nanchang had the largest population, and its inhabitants were more vulnerable. Therefore, the overall risk for Nanchang was higher than that of the surrounding Jiujiang and Shangrao areas.
The drought occurred throughout the province in July, and Nanchang was affected by the drought, and the integrated risk continued to increase. At this time, the reservoirs in Ji'an, Ganzhou, and Fuzhou began to supply water, and the integrated risk of the three places was reduced. In August and September, droughts occurred in all cities in the study area. Owing to the lack of disaster reduction capabilities of the reservoirs in Nanchang, its integrated risk was highest among all cities. With the easing of the drought conditions in the northern part of the study area in October and November, the integrated risk of drought disasters in Nanchang was reduced. However, the exposure of Xinyu City (mainly drylands and paddy fields) was higher than that of the surrounding cities such that when the drought continued in the central region, its integrated risk was significantly higher than that of the other areas.

Dynamic risk assessment of the drought disaster in 2013
Similarly, analyses were performed for 2013. Figure 5 shows that the study area suffered a summer drought in 2013. In July, the drought mainly occurred in the southern Ganzhou region, and it gradually spread from the south to the northeast. In October, in the majority of the areas, its intensity was at a maximum, as evidenced by the risk values. By December, the drought showed a decreasing trend from the north, and the south got relieved. The northern regions of the spatial patterns, the integrated drought disaster risk intensity in the northern region was higher than that in the southern region, which, hence, requires more attention.
The results of dynamic drought disaster risk assessment accord with the actual situation.

Dynamic risk assessment of the drought disaster from 2003 to 2013
For a typical year, the monthly scale was used to characterize the changes in drought and the drought disaster risk during the year. In this section, the annual scale data were  Although the monthly and annual scales of the model were selected to identify drought events and assess the drought disaster risk in the study area, it is limited by the existing observation and data collection conditions. In future studies, by integrating the influencing factors collected at a smaller time scale or live data, such as realtime monitoring of rainfall, evapotranspiration, soil water content, reservoir scheduling information, or night light remote sensing data, at a unified scale, we will be able to receive more timely dynamic information on drought disaster risks to provide decision support for drought disaster risk managers. Meanwhile, the selection of drought disaster indicators remains in an exploratory stage. For example, in the evaluation of the vulnerability indicators, only two influencing factors, namely the ecological vulnerability coefficient and the ratio of non-agricultural to agricultural population, were selected. However, the vulnerability response caused by drought is in all aspects of human social activities, and it is difficult to quantify the vulnerability to drought disasters through only two factors. The lack of other indicators is due to insufficient research on socioeconomic attributes, and it is difficult to collect data on the corresponding time scale.
In addition, the model should be verified in more study areas with different climatic conditions to improve the common applicability of the model.

CONCLUSIONS
Based on a conceptual model of the dynamic risk assess-

CONFLICTS OF INTEREST
No conflict of interest exits in the submission of this manuscript, and the manuscript is approved by all authors for publication.

DATA AVAILABILITY STATEMENT
All relevant data are available from an online repository or repositories.