Risk assessment of drought in Yun-Gui-Guang of China jointly using the Standardized Precipitation Index and vulnerability curves

Abstract Drought is one of the most serious natural disasters in the world and causes great economic losses in China every year, especially in its southwest region. Yet, few studies have reported the quantitative comprehensive risk of drought in the Yunnan, Guizhou, and Guangxi provinces of China. Taking these three provinces as the study area, we obtained annual precipitation, disaster loss, and agricultural planting data during 1964–2013. Following an optimal estimation of annual precipitation by the Bayesian maximum entropy method, we mapped the annual Standardized Precipitation Index. Based on the theory of information diffusion and exceeding probability, the hazard of drought was evaluated. We also fit the vulnerability curves using the drought loss data. As a basis, we constructed a multiplicative formula to calculate the comprehensive risk of drought, which integrates the hazard and the vulnerability and produces drought loss rate (DLR) maps. We found that the DLR caused by mild drought was about 3%, moderate drought 10%, severe drought 25%, and extreme drought 50%. We also created a risk zoning map to provide practical information, such as a scientific basis for optimization of regional allocation of resources for drought preparedness and response.


Introduction
Drought is one of the most serious natural disasters in the world. Its frequent occurrence on a global scale has a serious impact on the global ecological environment and people's daily lives. Drought causes an annual economic loss of 60-80 billion US dollars on an average every year around the world (Wilhite 2000). China is located in East Asia, where the instability of the monsoon climate causes the frequent occurrence of drought. The impact of drought on agriculture is the most serious. According to the effect of drought on agriculture during 1949-2014, the average annual drought area was about 13,544,300 hectares. China's southwestern region is vulnerable to drought. In 2010, the southwestern region experienced a once-in-a-century severe drought, resulting in nearly 10 billion RMB Yuan of agricultural losses and affecting the lives of 6 million people (Yang et al. 2011;Zhang et al. 2012;Li 2013). The affected agricultural area was greater than three million hectares in the province of Yunnan, and greater than one million hectares in the provinces of Guizhou and Guangxi. This event was the most serious drought since 1949. The study of drought risk assessment will help governments understand the characteristics of drought, provide guidance for dealing with drought, help government departments to develop drought prevention and mitigation measures, and provide a scientific basis for the preparation of contingency plans and the optimization of reserve resources configuration.
Risk assessment of natural disasters is a quantitative assessment and estimation of the intensity and form of risk. Common approaches include: the disaster risk index method, European multiple risk assessment methods, the disaster risk management index method, the United Nations Disaster Relief Organization (UNDRO) model, the forecastbased risk assessment model, and the exponential model (Ge et al. 2008;Zhang 2014).
Natural disaster risk zoning is the process of determining the geographical distribution of natural disaster risk. The common methods for natural disaster risk zoning include the risk zoning method with critical conditions of disaster-causing factors and historical series data, the risk division method based on information diffusion, the analytical method based on the physical/biological model, neighbourhood analogies, and empirical estimates (Zhang 2014). Like the risk assessment of natural disaster, the risk assessment of drought considers both disaster-causing factors and hazard-bearing bodies simultaneously. Drought risk involves the temporal and spatial distribution of drought hazard based on the variability of precipitation as well as the loss of hazard-bearing bodies such as agriculture. Since the main type of hazard-bearing body of drought is agriculture, the assessment of agricultural losses is particularly important for the assessment of drought risk.
The methods of assessment of agricultural drought loss mainly include index system methods (e.g. Analytical Hierarchy Process) and vulnerability curve methods (Zhang 2013;Zhang et al. 2015). The index system methods are considerably influenced by individual subjective factors. Many scholars have studied the vulnerability of drought from the perspective of crop yield by observing the crop production reduction under different severity of drought. Yamoah et al. (2000) conducted a statistical analysis of the Standardized Precipitation Indices (SPIs) to measure the hazard of drought and to assess the drought risk of maize production in Nebraska, USA. Some scholars have reported that the harm and loss caused by crop water shortage shows a logistic curve with prolonged water shortage from the mechanism of agricultural drought formation (Wang et al. 2010). Other scholars have used crop models to simulate and construct wheat, maize, and rice drought vulnerability curves (Wang et al. 2010;Wang et al. 2012;Jayanthi et al. 2013). Zhang et al. (2015) used the erosion-productivity impact calculator model to simulate the vulnerability of wheat drought and construct the wheat drought vulnerability curve to measure the impact of drought on the hazard-bearing bodies.
At present, drought risk assessment based on information diffusion theory and agricultural loss has also developed to a certain extent. Liu et al. (2013) used information diffusion theory and exceeding probability to assess multi-hazard natural disaster risk. Wang et al. (2013) assessed agricultural disaster risk in Gansu Province using information diffusion theory.
Based on the characteristics of drought, this paper proposes an integrated risk assessment framework, which seeks to improve the theoretical system of risk assessment of drought, especially from the beginning of the original data to the risk assessment of droughtthe complete risk assessment process. The paper is organized as follows: Section 2 describes the study area and the types of data that need to be used in this study, as well as the preprocessing methods and processes for the data. Section 3 describes the methodology, including the proposed risk assessment framework process, the SPI calculation, including the theory of information diffusion, and the risk assessment model. Section 4 provides a detailed analysis of the results and discussion. Finally, we arrive at our main findings in Section 5. This paper aims to provide guidance for the prevention and emergency preparedness of drought in the study area by evaluating the drought hazard of the region, analysing the impact of drought, and quantifying the comprehensive risk assessment of drought. For example, risk zoning can provide a scientific basis for the optimization of regional allocation of resources for drought preparedness and response.

Study area
The study area covers the Yunnan, Guizhou, and Guangxi provinces of China. The longitude range is 97.31 E-112.04 E, the latitude range is 20.54 N-29.16 N, as shown in Figure 1. The study area is located in the southwest of China, in the Yunnan-Guizhou Plateau, near China's borders with the South China Sea, Burma, Laos, and Vietnam. There are 39 cities in the study area, including three provincial cities (Kunming, Yunnan, and Nanning), Lijiang, Dali, Guilin, and so on. The Yunnan-Guizhou region is one of the most at-risk areas for severe drought and is also an important agricultural area. The area has large areas of rugged terrain, the average altitude is 1,000-2,000 m, the total area is 795,300 square kilometres, and the resident population is about 130 million, accounting for 9.5% of the total population of China. According to the 2014 China Statistical Yearbook, the region's GDP accounts for only 6% of the country's total GDP.
The area has a subtropical monsoon climate, with less precipitation in the spring and winter. There is strong transpiration, so the winter and spring seasons are prone to drought. Agriculture is susceptible to drought in the study area. The agricultural losses caused by drought in the region account for 10.6% of the total drought losses in the country. More seriously, drought losses in the region account for 43.6% of the agricultural losses caused by all kinds of meteorological disasters. The Yun-Gui-Guang area is also one of China's four distinct drought centres. Considering that the region is susceptible to drought, the population density is relatively high, the agricultural sector in the economic structure is relatively heavy, and the climate change is very diverse, it is of great significance to study the drought in this area.

Dataset
Two categories of data were used in this study, precipitation data for drought hazard assessment, and data on the disaster loss and agricultural acreage for the comprehensive risk assessment of drought.
The Chinese annual surface precipitation data, which was collected by 90 meteorological stations in and around the study area during 1964-2013, were obtained from the China Meteorological Data Sharing Service Network (http://cdc.nmic.cn). The basic format of the meteorological data is shown in Table 1. The annual data are produced from raw observations according to specific algorithms and organized into CSV format (for details refer to http://cdc. nmic.cn). Seventy-three of the 90 meteorological stations were located within the study area. Since the Bayesian maximum entropy estimation method used in our study considers the spatial association in the estimation process, 17 stations outside the study area were selected.
Historical disaster loss data were obtained from the historical natural disaster database of the Ministry of Agriculture Planting Management Division of China. We  extracted the agriculture area affected by drought during 1964-2013 within the Yunnan, Guizhou, and Guangxi provinces, which was taken as a measure of the annual agricultural losses caused by the drought to the study area. The data on the area of agricultural acreage were obtained from the crop database of the Ministry of Agriculture of the People's Republic of China. The basic data format is shown in Table 2 for the annual crop planting area data of Yunnan, Guizhou, and Guangxi in 1964-2013.

Data processing
We divided the precipitation data into two types: hard data (which was recorded completely) and soft data (which was recorded incompletely). The corresponding sites were called hard data sites and soft data sites, respectively. There were a total of 72 hard data sites and 18 soft data sites in the study area. Since the soft data could not be directly used for the analysis of the risk of droughts, we used the gamma distribution to fit the historical precipitation data in soft data sites, and finally obtained the complete hard and soft dataset to be used for mapping the spatial distribution of precipitation with BME (Bayesian Maximum Entropy) (Watterson and Dix 2003;Li et al. 2012;Zhang et al. 2016;Wang et al. 2017). The spatial and temporal covariance analysis of precipitation was carried out using hard data. The covariances of annual precipitation with different spatial and temporal lags were calculated in MATLAB software (MathWorks, Inc. Natick, Massachusetts, USA). Then, based on the calculated results, the spatial and temporal experimental covariance models of the precipitation data were described and fitted.
The temporal and spatial estimates of precipitation were carried out to obtain a continuous regionalized annual precipitation distribution. According to the Bayesian maximum entropy method, the complete dataset of hard data and soft data, the spatiotemporal covariance model, and the parameters of the spatiotemporal covariance model were used as the external inputs of the estimation process. The annual precipitation estimates for each point (with a certain location of time and space, and a size of 0.2 Â 0.2 ) were calculated with the BMELib tool and mapped in ArcGIS 10.2.2 (Esri, Redlands, California, USA) as shown in Figure 2. Our previous publication Wang et al. 2017) gives the detailed procedure.

Workflow
In this study, we developed a framework for comprehensive risk assessment of drought. Figure 3 shows the process and utilization of this framework. The main steps were as follows.
1. We mapped the continuous distribution of precipitation according to the observations from meteorological stations. In this step, we utilized the latest developed BME method by Christakos (1990) in the regional variable theory to interpolate the precipitation observations, which is intended to obtain optimal estimates. 2. Based on the interpolated precipitation, the SPI was calculated as the index to quantify the drought hazard. An adapted SPI proposed by Zhang et al. (2006) was used in this step; SPI values less than À0.5 reflect drought hazard. 3. Because agriculture was the main affected object by drought, the vulnerability curves of drought in the study area were constructed according to the statistical data of historical agricultural disaster. The curves were used to model the relationship between drought hazard and agricultural loss. In this step, information diffusion theory was selected to deal with small sample, and exceeding probability was used to depict hazard possibility. 4. The risk assessment of drought was achieved through the integration of the above drought hazard and vulnerabilities. A function of the two components was formulated to implement the integration and quantify the comprehensive risk of drought.

Standardized precipitation indicator
The main influencing factor of drought hazard is precipitation. In this paper, SPI (Mckee et al. 1993) was used to characterize the hazard of drought. The SPI is a measure of the drought and flood conditions in a region based on the amount of precipitation. It can be used to analyse the drought at different time scales. The most common time scales include 1 month, 3 months, 6 months, 12 months, and 24 months. According to the calculated SPI values, drought hazard can be divided into different grades and the specific classification criteria are shown in Table 3. SPI values greater than 0.5, which are used for flood analysis, were not considered in the study.  Meteorological drought level (GB/T20481-2006) is a Chinese national standardized method for calculating the precipitation index. The calculation method in this standard is more suitable for the case of China, and some of the calculation parameters have been fitted using Chinese history data. The calculation method in this standard makes a modification to the original SPI calculation method considering the actual situation of China, and it also provides some of the calculation parameters by China's historical precipitation data fitting. These parameters were used directly in our study. The annual SPI of each point (defined in the 'Data Processing' section) was estimated by spatiotemporal estimations of the annual precipitation data obtained from space-time interpolation of BME.

Drought vulnerability curve
Drought vulnerability curves of crops generally have a logistic shape (Wang et al. 2008;Zhang et al. 2015). The SPI was used as a measure of the intensity of drought hazard. The drought vulnerability curve was fitted using the drought loss rate (DLR) and the SPI. The focus of this paper is drought, but when the SPI is greater than 0.5, the precipitation is excessive for crop growth and there is a possibility of flood. Therefore, only the complete annual record data and the SPI values less than 0.5 were used to fit the curves. And we assumed that when SPI values vary from À1 to 0.5, the precipitation deficit has decreasing effect on crop.
Using the SPI and the sample of DLR fhSPI; lrig, the drought vulnerability curve of each province was obtained by fitting the function as follows (Zhang et al. 2015;Guo et al. 2016): where lr refers to DLR, the value range of SPI is ðÀ1; 0:5Þ, and a; b; c; d are fitted parameters. Lr is monotonic decreasing functions in the SPI range.

Information diffusion theory
As the precipitation data distribution was relatively scattered and the sample size was small, the traditional statistical analysis methods for large sample were not applicable for our study. For such small sample data, Huang proposed the theory of information diffusion (Huang 1997), which mainly carries out fuzzy diffusion and obtains the extended dataset which is supposed to be close to the actual cases (Xie et al. 2015). The calculation procedure of information diffusion is as follows (Huang 2000;Huang 2002aHuang , 2002b: Assuming that t denotes the disaster, the sample observation data of the disaster is T i ¼ ft 1 ; t 2 ; t 3 ; :::t m ; g, and U t is the domain of discourse of the risk assessment index of disaster t, then each risk assessment indicator set T i can spread the carried information to each of the indicators on the domain. The diffusion equation is where the value of the information diffusion coefficient h is determined by the minimum a and maximum b values of the risk assessment indicator set and the total number of sample observations m. Some researchers have recommended h and the The information distribution l i ðu jt Þ can be obtained by the ratio of f i ðu jt Þ and C i : where C i ¼ P n j¼1 f i ðu jt Þ: Let pðu jt Þ is the probability distribution when the domain of discourse of the risk assessment index is located at u j , and Q is the total number of observed samples within each u jt . pðu jt Þ is calculated as: Finally, the calculated frequency value can be treated as a probability value, then the probability exceeding u jt is: where pðu t > u jt Þ denotes the comprehensive risk of drought that is determined according to the exceeding probability.

Establishment of drought comprehensive risk assessment model
Assuming the comprehensive risk of drought in a region is R s , the factors that affect the comprehensive risk of drought in the region are the hazard and vulnerability of drought. The comprehensive risk can be expressed as (Guo et al. 2016): Comprehensive risk of drought ¼ hazard of droughtÂvulnerability of drought; (7) i.e.
The hazard of drought is the intensity of disaster-causing factors of drought (here shortage of precipitation), and the risk of drought is the synthesis of the probability of the occurrence of drought at different hazard levels and the corresponding consequences. Using the SPI to measure the intensity of the hazard of drought, the hazard of drought can be expressed as: Considering the situation of the study area and the validity of the data, we used historical drought data to construct the vulnerability curves. The vulnerability of drought is measured by the loss of drought, and the main factor affecting the loss is the hazard of drought. The vulnerability of drought can be expressed as: According to Equations (9) and (10), the comprehensive risk of drought can be expressed as: It can be seen that the comprehensive risk of drought is a function of the hazard of drought. Since the evaluated results of the drought hazard are the probability density form, the risk of drought is also the probability density form, that is, the possibility of different DLRs.

Drought hazards
The distribution map of the drought hazard in each year of the study area was obtained based on the calculated SPI and drought hazards classification method. Figure 4 shows the distribution of drought hazard levels in the study areas of 1965, 1975, 1985, 1995, 2005, and 2013.
Through a visual inspection in Figure 4, we see that the distribution of drought hazards in the study area was uneven, extreme droughts occurred occasionally, and there was a great deal of uncertainty in the locations of drought occurrences. In general, the frequency of extreme drought in Guizhou was relatively high, while western Yunnan and southern Guangxi had few extreme droughts. In addition, the frequency of severe drought and moderate drought was relatively high in most of the prefectures in Guizhou, the central and western part of Yunnan, and northern part of Guangxi. Moreover, the impact of near-normal drought was small, and most of the study area experienced near-normal drought often. Through comparing these three provinces, it was found that Guizhou was a drought-prone area, especially for extreme and severe drought, and the number of droughts in Yunnan and Guangxi were relatively small. The occurrence of drought in different years was quite distinctive, mainly due to the impact of precipitation in various regions.
Using the historical data, the estimated results were only the distribution of SPI of some regions in 1964-2013, in which case the distribution of SPI was scattered and the sample data points were relatively few. In order to make the evaluation results more accurate, the information diffusion theory was applied to spread the sample data of time series estimation of the SPI, by which we obtained the probability of different SPI, i.e. the probability density function of the possibility of the drought occurrence. Since the SPI had a wide range, to reduce calculation, we set the range of the SPI as SPI 2 ½À5; 5 and obtained the probability density of the drought hazard (H s ) in ½À5; 5 according to the drought hazard risk analysis above. There were 39 prefectures in the three provinces of the study area. The average value of SPI was calculated, and the risk assessment of drought was carried out in every prefecture. Using the SPI as an indicator of the hazard of drought, the evaluation index set was: where s represents prefectures, the numbers from 1 to 39 denote Nanning, Liuzhou, Guilin, Wuzhou, Beihai, Fangchenggang, Qinzhou, Guigang, Yulin, Baise, Hezhou, Hechi, Laibin, Chongzuo, Guiyang, Liupanshui, Zunyi, Anshun, Bijie, Tongren, Qianxinan, Qiandongnan, Qiannan, Kunming, Qujing, Yuxi, Baoshan, Shantong, Lijiang, Pu'er, Lincang, Chuxiong, Honghe, Wenshan, Xishuangbanna, Dali, Dehong, Nujiang, and Diqing orderly, and i is the year. The domain of discourse of the risk index of drought was set as U t ¼ fÀ5; À4:98:::; À0:02; 0; 0:2; :::; 4:98; 5g (there are 501 indicators in total). Then the risk assessment index set of each prefecture was diffused to the domain of discourse of the drought risk index to calculate the probability density function of drought hazard. The probability density distributions of the drought hazard in Guangxi, Guizhou, and Yunnan are shown in Figures 5-7, respectively. The exceeding probabilities of drought under different SPI values and risk of drought in various cities of the study area were calculated in the paper. The exceedance probability distribution maps of the drought risk in Guangxi, Guizhou, and Yunnan provinces are shown in Figure 8-10, respectively. The black line, which is perpendicular to the SPI axis, indicates an SPI of 0.5. The left side of the line represents no drought or drought, and the right side denotes flood. The curve is the exceedance probability of drought hazard, and the horizontal dash lines from up to down denote the exceedance probability of near-normal drought, moderate drought, severe drought, and extreme drought respectively.
The probabilities of the hazard of no drought, near-normal drought, moderate drought, severe drought, and extreme drought were calculated in each prefecture of the study area, as shown in Figure 11. It can be seen that the types of droughts in Guangxi and Yunnan were generally near-drought and severe drought, while those of Guizhou were usually severe and extreme drought. Twenty out of 39 (i.e. 51.3%) prefectures had a probability of drought over 30%. Therefore, the study area had a high Figure 11. Probability distribution of the hazard of drought (no drought, near-normal drought, moderate drought, severe drought, extreme drought).
incidence area of drought. In terms of the probabilities of the hazard of different drought levels, we calculated the corresponding exceedance probabilities.
As previously stated, we took the precipitation deficit to be the hazard factor of crop drought in our study. Many studies have taken a similar approach by assuming that crop growth conditions vary year-to-year due to prevailing weather conditions (e.g. the variability of precipitation). Nonetheless, it is worth noting that the precipitation deficit is just one of the factors that may reduce crop losses. Other causes, including infestation, crop rotation, and other weather conditions also amount to some proportion in certain conditions. In a fine evaluation these factors should be considered. For example, the rotation between agricultural crops (e.g. corn and soybeans) implies that the assumption may not hold, as each crop type may have distinct phenological variability across the growing season (Yagci et al. 2015).

Calculation of DLR
Considering that agriculture is the pillar industry in the study area (a relatively underdeveloped part of China) and crops are sensitive and susceptible to drought (Bi 1993), we took the crops as the hazard-bearing bodies of drought. In terms of the validity and accessibility of data, the drought vulnerability curves were constructed with the province as the unit by using the historical disaster data, the crop planting data, and the SPI data of the provinces in the study area.  In order to obtain a better fitting curve, we defined the index of DLR, which refers to the ratio of crop area loss due to drought and planting area of farm crops in the study area in the corresponding year. In this paper, the crop area loss due to drought was used to measure the drought loss of crop. The formula to calculate the DLR was DLR ¼ The crop area loss due to drought Planting area of farm crops in the study area : Using the crop area loss due to drought data and planting area of farm crops data, the DLRs for the 1964-2013 years in the three provinces of the study area were calculated, as shown in Table 4. In the case of drought vulnerability curve fitting, the degree of drought hazard corresponding to DLR was measured using SPI. Based on the estimation results of the previous section, the SPI of the three provinces in the study area was also calculated, as shown in Table 4.

Fitting of drought vulnerability curves
The drought vulnerability curve fitting process was done in MATLAB, and we obtained the scatter plots and the fitting curves of the sample data, as shown in Figure 12. The curves in Figure 12 are the vulnerability curves obtained by fitting the sample data. The dotted lines are the trends of the vulnerability curves drawn from the theoretical function model (Equation (1)). It can be seen that the basic shapes of the three vulnerability curves were roughly the same, except that there were some differences in the slopes. The absolute value of the slope of the vulnerability curve of Guizhou Province was the largest, so the sensitivity to drought hazard (i.e. SPI) was greater than that of the other two provinces. This means that if a drought occurred in this area it would cause relatively large economic losses. The absolute value of the slope of vulnerability curve in Guangxi was the smallest, so the sensitivity to drought hazard was less than that of the other two provinces.
According to the theoretical function model of vulnerability curve, the obtained parameters and the goodness-of-fit coefficients are shown in Table 5. We can see that the sample data were relatively small and scattered, the sum of squares due to error (SSE) was about 0.19, and the result was acceptable. The root mean squared error (RMSE) was below 0.1, close to zero, indicating that the selected drought vulnerability curve model and the fitting result were relatively accurate. R-square and adjusted Rsquare of the fitting results were greater than 0.9, very close to 1, indicating that the results of the fitting matched well with sample data. The vulnerability of drought was measured by DLR which was calculated using the obtained vulnerability curves. The different degrees of drought hazards (i.e. the SPI) are associated with different DLR values.

Comprehensive risk of drought
Based on the assessment of drought hazard and the fitting of the drought vulnerability curve, a comprehensive risk assessment model of drought was established, and then the comprehensive risk of drought was evaluated and zoned. In this study, the possibility of the DLR was used to measure the comprehensive risk level of drought.
The comprehensive risks of drought of the provincial capital cities Nanning, Guiyang, and Kunming of Guangxi, Guizhou, and Yunnan provinces are shown in Figure 13. The DLR caused by the near-normal drought, moderate drought, severe drought, and extreme drought was about 3%, 10%, 25%, and 50%, respectively. It can be seen from Figure 13 that when DLR was less than 60%, the order of comprehensive risk index in three cities was Guiyang > Kunming > Nanning. When DLR was greater than 60%, the order of comprehensive risk index in three cities was Nanning > Kunming > Guiyang. Under normal circumstances, DLR was below 60%.
The comprehensive risk of drought of all prefectures in Guangxi, Guizhou, and Yunnan is shown in Figures 14-16. It can be seen from these figures that the comprehensive risk trend between different cities was the same, but the risk level was different. At the same DLRs, the risk levels of the cities in Guizhou were higher than those in the other two provinces. Liupanshui and Zunyi (both cities in Guizhou) had the highest levels of comprehensive risk of drought.
The comprehensive risk of drought was calculated when DLR was 0%, 3% (Nearnormal), 5%, 10% (Moderate), 20%, 25% (Severe), 50% (Extreme), and 60%, as shown in Table 6 (the expected DLRthe averaged DLR with probability distributionis also given). From Table 6, it can be seen that the comprehensive risk index of drought was not 1 when DLR >0 due to the fact that a region may not have any droughts in some years and may have floods. When DLR was 3%, which is near-normal drought, the comprehensive risk index of drought decreased by 0.5 or so.
There was about 3% loss when there was no drought. The range of expected DLR caused by drought was 3.7-17.1%, and the fluctuation was large. Qianxinan of Guizhou was the region with the largest expected DLR, and Hechi of Guangxi is the region with the lowest. The expected DLR in the whole study area was 8.6%. The expected DLR in Guizhou was larger than those in Guangxi and Yunnan, which shows that the comprehensive risk of drought in Guizhou was larger than those in the other two provinces. Next, we divided the comprehensive risk of drought into five gradeslow, medium-low, medium, medium-high, and highaccording to Table 7, and the zoning results were visualized in the ArcGIS software. Figure 17 shows the comprehensive risk zoning result of expected DLR. It can be seen that high risk and mediumhigh risk areas were concentrated in Guizhou Province, and the comprehensive risk of drought grades of all the cities in Guizhou Province was above medium risk grade.  The comprehensive risk of drought grades in Yunnan and Guizhou was relatively low. Figure 18 shows the risk zoning results of the study area when DLR was appointed to 3%, 10%, 25%, and 50%. According to the comparative analysis of the comprehensive risk of drought in different places, Guizhou was a high-risk area, especially in its southwest region. The relatively high-risk area in Yunnan was located in its central region, and the high-risk area of Guangxi was evenly distributed. On the whole, in the three provinces of the study area, the comprehensive risk of drought in Guizhou was the highest, followed by Yunnan and then Guangxi.  Some possible reasons for the high comprehensive risk in Guizhou is that Guizhou has large areas of rugged terrain and extends deep inside the mainland, which results in less precipitation, and more proneness to drought, especially severe and extreme drought. The hazard-bearing bodies of drought in Guizhou are various and easily influenced by terrain. The disaster prevention and mitigation measures such as irrigation are relatively undeveloped, which make the hazard-bearing bodies more sensitive and susceptible to drought. Therefore, the same level of drought in Yunnan, Guangxi, and Guizhou would result in more losses in Guizhou, making the risk of drought greater in Guizhou.

Conclusions
In this paper, we proposed a comprehensive risk assessment framework of drought which jointly considered the intensity of disaster-causing factors of drought and the vulnerability of hazard-bearing bodies. After mapping the spatial distribution of precipitation, an SPI-based evaluation of drought hazard was implemented. Drought vulnerability curves were developed using historical disaster data. The comprehensive risk assessment of drought was carried out according to the hazard assessment results and the vulnerability curves. Finally, for each prefecture in the study area, we obtained the risk of drought under different DLR values and the expected DLR, which allowed us to perform risk zoning. The main conclusions and findings of this study are as follows.
1. The probability of occurrence of drought was the largest in Chuxiong, whereas Yunnan was mildly drought prone. Anshun and Guizhou had the smallest possibilities of occurrence of drought. The types of droughts in Guangxi and Yunnan were generally mild drought and moderate drought, while severe and extreme drought occurred in Guizhou. 2. The vulnerability curves of the three provinces were similar, but the absolute value of the vulnerability curve in Guizhou Province was relatively large, which indicates that the disaster-bearing bodies in this area were sensitive to drought. The absolute value of the slope of the vulnerability curve was relatively small for Guangxi, which means it had low sensitivity to drought. 3. When DLR was less than 60% (which is observed under general circumstances), the order of comprehensive risk of drought in three cities was Guiyang > Kunming > Nanning. The comprehensive risk trend of drought between different cities was the same, but the risk levels were quite different. The risk levels of the cities in Guizhou were higher than those of the cities in the other two provinces under the same loss rate. In Guizhou Province, Liupanshui City and Zunyi City had the highest levels of comprehensive risk of drought. According to the resultant risk zoning map, Guizhou had the highest risk of drought, especially in the southwest region, followed by Yunnan and then Guangxi with the lowest risk.

Disclosure statement
No potential conflict of interest was reported by the authors.