Analysis of the spatiotemporal patterns and propagation characteristics of drought risk in China

Abstract

Based on standardized precipitation index data, a systematic analysis was conducted of the spatiotemporal variations of drought events in China from 1978 to 2018. Drought events were identified using the run theory applied to the standardized precipitation index data set, and key variables such as drought frequency, duration, and intensity were quantified. Additionally, drought vulnerability, exposure, and resilience were calculated to comprehensively assess the regional drought risk. The spatiotemporal transmission characteristics and pathways of drought risk were further explored using the Markov chain model and its extended version based on spatial lag theory. The results revealed significant differences in the spatial and temporal distribution of drought events across China, with north-west China experiencing a particularly high frequency, duration, and intensity of droughts. Overall, the pattern of drought risk presented a gradient, being higher in the north-west and lower in the south-east. The risk was relatively stable from year to year, with few large fluctuations. Moreover, a strong spatial similarity in drought risk was observed among neighboring provinces, but there was no obvious spatial lag effect. This study provides a valuable scientific foundation for effective drought disaster risk management and the formulation of response measures.

Graphical abstract

Keywords

standardized precipitation index / drought risk / Markov chain / risk propagation

1 Introduction

As one of the most serious natural disasters in China, drought has a significant negative impact on agricultural production, water resource management, and the natural environment, and poses a potential threat to ecosystems and livelihoods (Zhai and Zou, 2005; Chiang et al., 2021; Jia et al., 2022). Due to global climate change and rapid socio-economic development, the risk of drought in China is gradually increasing (Zhang et al., 2020). Drought is one of the most difficult natural disasters to monitor and predict (Wang et al., 2012). To effectively mitigate the impact of drought, it is crucial to understand its spatial and temporal distributions, assess drought risks and their evolution, and provide a foundation for drought monitoring and early warning systems. This will help reduce the losses caused by drought disasters (Qian et al., 2001; Wang et al., 2012).

Drought indices are an important tool in drought research (Mishra and Singh, 2010). The standardized precipitation index (SPI) is an aridity index based on precipitation. It operates on the principle that, at specific temporal and spatial scales, precipitation deficits affect surface water, groundwater, soil moisture, snow cover, and streamflow (McKee et al., 1993). The SPI is sensitive to variations in drought conditions and can be used to predict droughts across different time scales (Hayes et al., 1999), making it one of the most widely used drought indicators. Therefore, in this study, an SPI data set for China from 1978 to 2018 was utilized to analyze the characteristics of drought risk and its transmission.

The identification of drought events and drought risk assessments based on drought indices are fundamental to analyzing and studying the characteristics of drought risk and its transmission processes (Yang et al., 2017; Wang et al., 2024). Drought risk is defined as the likelihood of a drought event occurring and is a natural characteristic of drought hazards. Drought hazard differs from drought risk because it encompasses both natural and social attributes (Yoo et al., 2013; Xiang et al., 2022). How to scientifically and comprehensively assess drought risk remains a critical issue in drought disaster research and represents an important scientific problem that requires urgent resolution. In previous studies, the Copula function has been widely used to construct multivariate joint distributions, which are frequently applied in drought risk analysis and assessment (Xu et al., 2015; Gu et al., 2021). However, most existing research has primarily focused on assessing the probability of drought events. In practice drought events are typically influenced by a variety of complex factors, and a comprehensive analysis is therefore required when evaluating drought risk (Xu et al., 2022). One novel approach is to identify drought events and assess drought risk by exploring the dynamic evolution of drought risk as a function of three key factors: vulnerability, exposure, and resilience (Mondal et al., 2023).

The risk transmission process of drought events is a key feature in the occurrence and development of drought disasters, and it remains a major research focus in the field of drought disaster studies. A Markov chain describes the process by which a system transitions from one state to another within a defined set, where the transition probability depends only on the current state and is independent of previous states. This property is known as “Markovianity” (Anderson and Goodman, 1957). Previous studies have attempted to predict the probability of drought state transitions using the Markov chain approach to uncover the evolutionary patterns and mechanisms of drought as a stochastic process. These methods have yielded promising results in drought monitoring and early warning systems. For example, (Paulo and Pereira, 2007) used a Markov model to analyze the transition probabilities between drought states to effectively assess the real-time drought risk. Alam et al. (2017) combined Markov chains with a three-dimensional log-linear model to explore the temporal evolution paths of different drought grades. Fadhil and Unami (2021) constructed a complex multi-state Markov chain model to predict the transfer probability of drought risk and explore potential strategies for mitigating increasing risk trends. Sun et al. (2014) used the Markov chain model to analyze two-dimensional variables in drought states and calculated transition probabilities between drought states in Xinjiang, providing a scientific basis for regional drought management. Xiao et al. (2012) combined the Copula function with the transition probability matrix of the Markov chain to study the duration of drought events and the average first arrival time of drought state transitions, based on the characteristics of drought in the Pearl River Basin. This study provides an important reference for water resource management and drought mitigation in the basin.

However, previous studies have paid relatively little attention to the geospatial characteristics of drought disasters (Yang et al., 2020). In some geospatial studies, spatial lag theory has been integrated with Markov chains to explore neighborhood correlations and spatial properties more thoroughly. For example, Pu et al. (2005) studied the influence of economic development in surrounding regions by combining the Markov chain with spatial lag theory in Jiangsu Province. Wang et al. (2015) applied the Markov chain approach and spatial lag theory to analyze spatial spillover effects in the regional development process in Guangdong. Few studies have combined a Markov chain with spatial lag theory for large-scale drought risk studies. Only a few researchers, such as Xu et al. (2022), have supplemented drought risk analysis with a spatial adjacency analysis using a Markov chain. After conducting a drought risk analysis in south-west China, they performed a spatial neighborhood analysis on the provinces, although they did not directly combine the Markov chain model with the spatial lag model. Therefore, integrating spatial analysis based on the Markov chain and spatial lag theory is of great significance for analyzing drought characteristics across China.

Combining the Markov chain model with the spatial lag model, the spatial and temporal distribution characteristics, drought risk factors and evolution process of drought events in China were systematically analyzed. The objective of this study is threefold: (i) achieve comprehensive analysis of drought events in China from the characteristic variables such as drought frequency, duration and intensity; (ii) reveal the dynamic evolution characteristics of drought risk factors (exposure, vulnerability, resilience); and (iii) explore the spatiotemporal transmission characteristics and pathways of drought risk. This study could provide a scientific basis for better understanding the evolution of risks and rational allocation of water resources.

2 Data and methods

2.1 Data

The SPI is an aridity index where precipitation is normalized after adjusting for its probability distribution (Yang et al., 2017). The ScienceDB data repository provides SPI data for China from 1978 to 2018, based on gamma, Gumbel, logistic, log-logistic, and normal distributions (Zhang et al., 2023). Previous studies have demonstrated that the gamma-distributed SPI generally offers broader adaptability and greater stability; therefore, the SPI data set based on the gamma distribution was used here (Stagge et al., 2015).

2.2 Drought event identification

The identification of drought events is the foundation of a drought event study. However, when selecting a threshold to identify drought characteristics, there may be limitations, particularly in cases where droughts are long-lasting and have relatively low intensity, but still exert significant cumulative impacts and cause substantial losses (Mondal et al., 2023). To address this issue, the run theory was used to set three SPI thresholds: R₀=0, R₁=−3, and R₂=−0.5 for identifying drought events. The specific method for determining these thresholds was as follows.

1) If the SPI value of a month was less than R₁, the month was preliminarily judged to be dry;

2) If the drought lasted for one month and the corresponding SPI value was greater than R₂, the month was recognized as a non-drought event and excluded;

3) If the interval between two consecutive droughts was one month and the SPI value for that month was less than R₀, the two droughts were combined into a single drought event. Otherwise, the two droughts were treated as separate events.

After identifying drought events using the run theory, the number of drought occurrences (N) and the drought duration, intensity, and peak intensity were analyzed. Drought duration (D) refers to the length of a drought event, drought intensity (S) represents the cumulative value of the difference between the SPI value and the drought threshold during the event, and peak intensity (P) denotes the maximum difference between the SPI value and the drought threshold within the event.

2.3 Drought risk assessment

Risk factors are indicators of drought evolution (Guo et al., 2019). Compared to evaluating drought risk based solely on the frequency of drought occurrences or their return periods, using risk factors can provide a more nuanced understanding of the underlying mechanisms driving drought events. In this study, the SPI sequence on a monthly scale was used to calculate the three drought risk factors of vulnerability (Vu), exposure (Ex), and resilience (Re), and to evaluate the regional comprehensive drought risk R. These calculation methods were based on the approaches proposed by Veettil et al., (2018) and Nie et al., (2020). The specific calculation was as follows.

1) Vulnerability (Vu)

(1)Vu=\frac{1}{N}{\sum }_{i=1}^t{S}_i,

(2)V{u}^{\prime }=\frac{Vu-V{u}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{V{u}_{\mathrm{m}\mathrm{a}\mathrm{x}}-V{u}_{\mathrm{m}\mathrm{i}\mathrm{n}}},

where N is the number of droughts; S_i is the intensity of the ith drought; and Vu' s the value after the normalized vulnerability treatment, with a range of [0,1].

2) Exposure (Ex)

(3)Ex=\frac{1}{T}\sum _{i=1}^N{D}_i,

(4)E{x}^{\prime }=\frac{Ex-E{x}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{E{x}_{\mathrm{m}\mathrm{a}\mathrm{x}}-E{x}_{\mathrm{m}\mathrm{i}\mathrm{n}}},

where D_i is the duration of the ith drought; T is that the total period; and Ex' is the value after exposure normalization, with a range of [0,1].

3) Resilience (Re)

(5)Re=\frac{N}{{{\displaystyle \sum }}_{i=1}^N{D}_i},

(6)R{e}^{\prime }=\frac{Re-R{e}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{R{e}_{\mathrm{m}\mathrm{a}\mathrm{x}}-R{e}_{\mathrm{m}\mathrm{i}\mathrm{n}}},

where Re' is the normalized value of the restoring force, with a range of [0,1].

4) Combined drought risk (R)

(7)R=w_1 V{u}^{\prime }+w_{2}E x^{\prime }+ w_3(1-R{e}^{\prime }),

where R is the drought risk index; w₁, w₂, and w₃ are the weights of the three risk factors in the drought risk, with w₁=w₂=w₃=1/3.

2.4 Markov transfer matrix

The Markov transition matrix is a method of Markov analysis characterized by its discrete nature in both time and state. Once the transition reaches stability, the probability of transition in the current year becomes independent of the initial state (Hu et al., 2021). In this study, the Markov chain was used to calculate the drought transition matrix for each province in China. First, based on the comprehensive R value calculated from the SPI, drought risk was categorized into four levels: low, medium-low, medium-high, and high, using the natural breakpoints method. Next, for each study area, a Markov state transition probability matrix M was constructed. Here, P_ij denotes the probability of transitioning from state i in year t to state j in year t + 1. A transfer frequency approximation was then used to estimate the state transition probability P_ij:

(8)P_{ij} (n,n +k)= P\{X_{n+k}=j|X_n= i\},

(9)P_{ij} =\frac{n_{ij}}{{{\displaystyle \sum }}_{i=1}^s{n}_{ij}},

where i,j = 1, 2, …, m; P_ij = n_ij/n_i; n_ij denotes time n to n + 1 time, state i shifts to state j; and S is the number of possible states of the Markov chain. The drought risk status of the next year can be predicted by evaluating the drought risk status of the unit in the previous year.

2.5 Risk transfer model based on Markov chain

The spatial Markov chain approach integrates the traditional Markov method with the concept of spatial autocorrelation or spatial lag, thereby addressing the limitations of traditional Markov chains in accounting for spatial transfer and spatial lag effects. This approach enables an in-depth analysis of the spatiotemporal coupling relationships of drought risk, and facilitates the exploration and identification of the internal transmission characteristics and pathways of regional drought risk. The spatial Markov matrix is constructed based on the spatial lag of drought risk types in the initial year. The Markov risk transfer probability matrix is then decomposed into K × K Markov probability matrices, forming the Markov spatial probability matrix for drought risk transfer. This matrix can be used to evaluate the probability of changes in drought risk levels within a spatial unit across different drought risk levels. Additionally, it enables an assessment of how the drought risk levels of other geographic units impact the drought risk transfer within a spatial unit.

In this study, the spatial lag value was calculated by multiplying the risk attribute value of the drought evaluation unit and the spatial weight matrix (Hu et al., 2021). The calculation was as follows:

(10)La{g}_a=\sum Y_b{W}_{ab} ,

where Lag_a s the spatial lag value of evaluation unit a; and Y_b is the attribute value of evaluation unit b; The spatial weight matrix W_ab evaluates the spatial relationship between units a and b, i.e., if spatial units a and b are adjacent, W is 1, but if they are not, W is 0. By comparing the elements in the spatial Markov transfer probability matrix, the influence of regional drought risk state background on neighborhood risk state transfer could be determined.

3 Results and analysis

3.1 Drought characteristics in China

Fig.1 shows the average annual drought duration, annual average drought frequency, average annual drought intensity, and peak drought intensity over a 40-year period in China based on SPI data. Fig.1(a) shows that there was significant variation in the distribution of average annual drought duration across China, with the maximum duration reaching 6.23 months. The regions with longer average annual drought durations were primarily concentrated in the north-west, and some central and south-western areas of China. Additionally, the distribution of the average annual drought duration was uneven, reflecting the strong influence of geographical spatial differentiation.

Fig.1(b) shows that, in most parts of China, the average annual drought frequency was approximately one, with a maximum of two drought events per year recorded. High incidence areas were predominantly identified in the north-west, with more sporadic occurrences in other regions. Some areas experienced an average of more than one drought event annually. Conversely, coastal areas and some central regions of China had an average annual drought frequency of less than one, indicating that drought events in these regions were relatively rare and episodic.

Fig.1(c) and Fig.1(d) show the characteristics of drought intensity in China. The peak drought intensity reached 24.74, while the maximum mean drought intensity was 5.32. Areas with a high average drought intensity were mainly located in north-west China. Compared to the average drought intensity, regions with high peak drought intensity were more dispersed, although they were still concentrated in the north-west. This distribution also extended to other parts of the country, suggesting that extreme drought events have occurred throughout China over the past 40 years.

Overall, there were significant regional variations in the occurrence of drought events in China over the past four decades. The most severely affected areas were primarily in the north-west, where droughts were long-term and frequent, although some other regions also experienced drought disasters. Additionally, the peak drought intensity indicated that extreme drought events were more sporadic compared to general drought events.

3.2 Drought risk characteristics in China

Risk factors (vulnerability, exposure, resilience) are key indicators of drought risk and its evolution (Guo et al., 2019). Based on SPI data, the results of drought vulnerability, exposure, and resilience in China are shown in Fig.2. Drought vulnerability reflects the degree to which various social, economic, and natural elements of the human-environment system are disrupted by drought, thereby affecting the system’s ability to manage drought impacts (Shi et al., 2017). The greater the intensity of a single drought event, the lower the ability of the human-environment system to withstand the disturbances caused by drought. Fig.2(a) shows the drought vulnerability across China. The maximum recorded value of drought vulnerability in China was 0.16. The areas with high drought vulnerability were primarily concentrated in the arid and semi-arid regions of north-west China, such as Nei Mongol and Xinjiang, indicating that these regions experienced more severe disruptions to the human-environment system than the other areas of the country. Additionally, some regions in the central, south-western, and coastal areas of China also exhibited a high vulnerability to drought and were prone to extreme drought events.

Drought exposure (Jia et al., 2020) refers to the proportion of time a region is exposed to drought; thus, a higher drought exposure indicates that the region has experienced drought conditions for a longer duration. Fig.2(b) shows the overall drought exposure across China. The maximum value of drought exposure in China was 0.52. Regions with a high drought exposure were primarily located in the north-west and south-west of China, as well as some northern and central areas, indicating that these regions were more susceptible to prolonged drought events. Conversely, in the rest of the country, drought events were relatively rare, particularly prolonged drought.

Drought resilience (Nie et al., 2021) refers to a region’s ability to recover from water scarcity to a state of water availability. The higher the value of a region’s drought resilience, the greater its ability to withstand and recover from drought events, meaning that the region can rebound more quickly after being affected by drought. The regions with strong drought resilience in China were primarily located in coastal, central, and some northern areas, indicating that these regions can recover from drought impacts in a relatively short time. In contrast, the regions with poor drought resilience were mainly concentrated in the arid and semi-arid areas of north-west China, suggesting that recovery from drought damage would be more challenging in these regions.

The drought risk index provides a comprehensive evaluation of drought risk based on three key assessment indicators: drought vulnerability, drought exposure, and drought resilience. The distribution of drought risk across China is shown in Fig.3. The spatial pattern of drought risk distribution in China was aligned with the spatial distribution patterns of vulnerability, exposure. The overall drought risk was highest in the north-west and lowest in the south-east, and was significantly influenced by the distribution of precipitation across the country. The maximum drought risk in China was 0.43. This risk was influenced by non-zonal variations in topography and land cover, with high drought risk areas also scattered throughout other regions beyond the north-west.

3.3 Characteristics of the drought-risk state-transition in different provinces of China

Based on the drought risk index results for China and using the natural breakpoints method, the drought risk index across China was categorized into four levels: low risk, medium−low risk, medium−high risk, and high risk. The classification criteria are detailed in Tab.1 below.

Based on the calculated drought risk index, the drought risk in China was classified into different levels using the natural breakpoints method. Using the Markov chain model, the drought risk state transition was assessed for each point in 34 provincial-level administrative regions in China from 1978 to 2018. The probability of drought-risk state-transitions between consecutive years was calculated for each region, resulting in a drought-risk state-transition matrix, as shown in Fig.3. The x-axis represents the drought risk level after a transition, while the y-axis represents the initial drought risk level. The diagonal of the matrix indicates cases where the drought risk level remained unchanged between two consecutive years. The top-right portion reflects the probability of a transition to a higher drought risk, while the bottom-left diagonal shows the probability of a transition to a lower drought risk. Blue indicates a lower probability of transition, red indicates a higher probability of transition, and the darker the color of the matrix, the greater the difference in transition probabilities between the four levels. This highlights distinct regional characteristics in drought risk transfer.

Beijing, Tianjin, Shanghai, Fujian, Henan, Taiwan, Hong Kong, Macao, Xizang, Qinghai, and Xinjiang showed significant differences in drought-risk state-transition probabilities, demonstrating clear regional features. For Xinjiang, Qinghai, Xizang, and Taiwan, the state transition probabilities in states 1-1 and 4-4 were particularly high, while the probabilities of a transition from state 1 to state 4 and vice versa were very low. This suggests that the drought risk in these regions was relatively stable, with little change when regions were in either high or low drought risk states. The relative stability of the drought risk for Xinjiang, Qinghai, and Xizang may be due to their location in China’s arid and semi-arid inland areas, while Taiwan’s generally humid coastal climate made it less susceptible to short-term meteorological changes, which contributed to its stable drought risk level.

For Beijing, Tianjin, Shanghai, Fujian, Henan, Taiwan, Hong Kong, and Macao, the drought risk levels mainly fell into categories 2 and 3, with a high probability of a shift between these two levels. This suggests that these regions are prone to level 2 to level 3 drought risks over extended periods. Overall, there was no clear upward or downward trend in drought risk across Chinese provinces, and drought risk transitions mostly occurred between adjacent levels. The probability of transitioning between distant drought risk levels was low, indicating that drought risk remained relatively stable over consecutive years, with few large shifts.

3.4 Characteristics of the spatial transfer of drought risk in China

The spatial Markov transition probability matrix was derived under varying spatial lag conditions within the region. The annual drought risk level for each province was determined using the average drought risk of each province in China and the classification of drought risk levels. Neighboring provinces were identified following the principle of adjacency, and a Markov transition probability matrix for different spatial lag types was calculated, as shown in Fig.4. This matrix enabled an analysis of the spatial transmission patterns of drought risk across regions in China.

The results showed a strong spatial similarity in the drought risk levels of neighboring provinces within the same year, reflecting the regional nature of drought risk in China. When the initial state was characterized by a level 1 drought risk, the likelihood of neighboring provinces experiencing a level 1 or level 2 drought risk was significantly higher than that of other risk categories. Conversely, when the initial drought state was a level 3 risk, the surrounding provinces tended to have either the same or lower drought risk levels. When the initial drought risk state was level 2 the probability of adjacent areas also being in a low drought risk state was considerably higher than for other risk states, suggesting that the level 2 drought risk state exhibited a stronger regional coherence than the other initial states.

However, the spatial transfer probability over two consecutive years indicated that drought risk transfer in China was not strongly influenced by neighboring provinces, with significant impacts observed only in specific cases. For example, when the drought risk level shifted from level 2 to level 3, the likelihood of neighboring provinces also experiencing a level 3 drought risk was higher than their transitioning to other drought levels. This suggests that, in certain cases, the drought risk level of neighboring provinces may influence post-transfer drought risk. Overall, the lag relationship between the drought risk changes within a province and the risk conditions in adjacent regions was not pronounced because drought risk is more strongly influenced by broader regional conditions.

4 Discussion

This study used standardized precipitation index (SPI) data to systematically examine drought events across China from 1978 to 2018, evaluating both the drought risk and its transmission dynamics. The study revealed the key characteristics of drought risk and its spatial transfer within China, but there were several limitations that warrant further investigation.

4.1 Uncertainty of risk assessment results

First, the run theory was employed to identify drought events, and three key drought risk indicators were used for risk assessment. However, due to China’s vast geographical expanse and significant regional variability, determining an appropriate threshold for event identification and the weight distribution of risk indicators remains challenging. Addressing this is crucial for accurately reflecting regional drought risk conditions in future studies.

Second, SPI data on a monthly scale were used to analyze drought risk, but given the substantial variation in drought conditions across the different regions of China, future studies should consider incorporating SPI data at different time scales or considering alternative drought risk indicators to refine the analysis.

Finally, while the spatial Markov transfer probability matrix analysis revealed the spatial similarity of drought risk among neighboring provinces, the spatial lag effect was found to be minimal. This could be attributed to China’s complex geographical and climatic conditions. It would therefore be beneficial to explore the spatial attributes of drought risk by adjusting the spatial lag model or conducting analyses at different spatial scales.

4.2 Intercomparison with previous studies

In the research on drought risk assessment in China, by comparing the results of different studies, the majority of studies have emphasized the exacerbating effects of climate change and human activities on drought risk (Xiang et al., 2022). They all point out that reduced precipitation, rising temperatures, and excessive exploitation of water resources are the main factors leading to frequent droughts (Zhang et al., 2019). Additionally, these studies generally employ meteorological data (Chai et al., 2019), remote sensing information (Sun et al., 2015), and numerical models (Otkin et al., 2016) to conduct drought risk assessments, ensuring the reliability of the assessment results.

In terms of differences, there are significant variations among different studies in terms of assessment methods and focuses. For example, some studies emphasize the statistical analysis of historical drought events (Dai, 2013), assessing the severity and frequency of droughts by constructing drought indices; while other studies focus more on predicting future drought scenarios (Jia et al., 2022), using climate models to simulate the impact of future climate change on drought risk. Additionally, some studies concentrate on specific regions, such as the North China Plain (Zuo et al., 2020) or the Yangtze River Basin (Zhang et al., 2023), conducting in-depth analyses of the drought characteristics and causes in these areas, whereas other studies carry out comprehensive assessments at the national scale (Jia et al., 2023), exploring the spatial and temporal distribution patterns of drought risk.

While there’s a general consensus on the increasing drought risk trend in China, discrepancies in assessment methodologies, scales, scopes, indicator choices, model selections, consideration of socioeconomic factors, timeframes, and scenario settings across various studies have led to uncertainties and variations in the findings. Future research should focus on enhancing the integration of different scales and methods, as well as incorporating socioeconomic factors more comprehensively, to achieve more accurate drought risk assessments in China.

4.3 Driving forces behind risk transition

The continuous evolution of drought, characterized by the gradual intensification or mitigation of its severity, has been demonstrated in many studies (Sherwood and Fu, 2014; Kang et al., 2023). This finding is consistent with our observation that transitions between adjacent drought categories are more probable. As an example, research utilizing Markov chain models has shown higher probabilities for shifts from mild to moderate or moderate to severe drought conditions compared to direct transitions from mild to severe. This inherent continuity reflects drought’s nature as a natural hazard, influenced by the interplay of atmospheric circulation, hydrological processes, and other factors, ultimately resulting in a cumulative impact on its development. Due to differences in climatic conditions, geographical environments, and water resource management levels, the patterns of drought risk transition may vary among different regions. Some studies have found that in areas with severe climate fluctuations or vulnerable ecological environments, the probability of cross-level transitions might be relatively higher (Merabti et al., 2018; Xu et al., 2022). Therefore, future research needs to further focus on the driving mechanisms of drought risk transition and consider the influences of various factors, in order to gain a more comprehensive understanding of the laws of drought evolution.

5 Conclusions

Utilizing standardized precipitation index (SPI) data, this study systematically investigated the spatiotemporal distribution patterns, drought risk factors, and evolution of drought events across China from 1978 to 2018. Through the application of the Markov chain approach and a spatial expansion model, the study also identified the factors governing the spatiotemporal transfer of drought risk. The main conclusions were as follows.

1) There were significant spatial and temporal disparities in the distribution of drought events across China, with prolonged and frequent droughts primarily concentrated in the north-west, whereas the south-east coastal areas experienced relatively few droughts.

2) The regions with long average annual drought durations were predominantly located in the north-west, central, and south-western parts of China. While the average annual drought frequency was approximately one event per year in most areas, drought frequency in the north-west was notably higher than in other regions. Both the mean and peak values of drought intensity highlighted the severity of droughts in the north-west, although extreme drought events with varying intensities also occurred across other regions. Collectively, these features highlighted the regionality and complexity of drought events in China.

3) In terms of drought risk assessment, after evaluating the three key risk factors of drought vulnerability, exposure, and resilience, it was found that the regions with high vulnerability were largely situated in the arid and semi-arid areas of north-west China, such as Nei Mongol and Xinjiang, where the human-environment system has a weaker capacity to respond to drought events. Regions with high exposure to drought were similarly concentrated in the north-west and south-west, indicating that they face long-term drought threats. In contrast, the areas with higher drought resilience were located in the coastal and central regions, and had the ability to rapidly recover from the effects of drought. The comprehensive drought risk assessment showed that drought risk was higher in the north-west and lower in the south-east, and was closely linked to regional precipitation patterns.

4) The Markov chain analysis revealed that the drought risk level in each province remained relatively stable over consecutive years, with few large changes. Most transitions in drought risk occurred between adjacent risk levels, with a low probability of crossing between non-adjacent levels. It was also observed that drought risk in inland arid and semi-arid regions, such as Xinjiang, Qinghai, and Xizang, tended to remain stable, while in coastal and central regions, there was a greater tendency for risk levels to shift toward the level 2 and level 3 risk categories.

5) The spatial Markov transition matrix analysis further indicated a strong spatial similarity in drought risk levels among neighboring provinces within the same year. However, the probability of drought risk transfer over two consecutive years was not significantly influenced by neighboring provinces, suggesting that changes in drought risk across China were more influenced by regional drought conditions than by the drought status of adjacent provinces.

References

[1] Alam N M, Sharma G C, Moreira E, Jana C, Mishra P K, Sharma N K, Mandal D (2017). Evaluation of drought using SPEI drought class transitions and log-linear models for different agro-ecological regions of India.Phys Chem Earth, 100: 31–43

[2] Anderson T W, Goodman L A (1957). Statistical inference about Markov chains.Ann Math Stat, 28(1): 89–110

[3] Chai Y F, Li Y T, Yang Y P, Li S X, Zhang W, Ren J Q, Xiong H B (2019). Water level variation characteristics under the impacts of extreme drought and the operation of the Three Gorges Dam.Front Earth Sci, 13(3): 510–522

[4] Chiang F, Mazdiyasni O, AghaKouchak A (2021). Evidence of anthropogenic impacts on global drought frequency, duration, and intensity.Nat Commun, 12(1): 2754

[5] Dai A G (2013). Increasing drought under global warming in observations and models.Nat Clim Chang, 3(1): 52–58

[6] Fadhil R M, Unami K (2021). A multi-state Markov chain model to assess drought risks in rainfed agriculture: a case study in the Nineveh Plains of Northern Iraq.Stochastic Environ Res Risk Assess, 35(9): 1931–1951

[7] Gu L, Chen J, Yin J B, Guo Q, Wang H M, Zhou J Z (2021). Risk propagation from meteorological to hydrological droughts in a changing climate for main catchments in China.Adv Water Sci, 32(3): 321–333

[8] Guo Y, Huang S Z, Huang Q, Wang H, Wang L, Fang W (2019). Copulas-based bivariate socioeconomic drought dynamic risk assessment in a changing environment.J Hydrol (Amst), 575: 1052–1064

[9] Hayes M J, Svoboda M D, Wilhite D A, Vanyarkho O V (1999). Monitoring the 1996 drought using the standardized precipitation index.Bull Am Meteorol Soc, 80(3): 429–438

[10] Hu S L, Jiao S T, Zhang X Q (2021). Spatio-temporal evolution and influencing factors of China’s tourism development: based on the non-static spatial Markov chain model.J Nat Resourc, 36(4): 854–865

[11] Jia H C, Chen F, Du E Y, Wang L (2023). Drought vulnerability curves based on remote sensing and historical disaster dataset.Remote Sens (Basel), 15(3): 858

[12] Jia H C, Chen F, Zhang C R, Dong J W, Du E Y, Wang L (2022). High emissions could increase the future risk of maize drought in China by 60–70%.Sci Total Environ, 852: 158474

[13] Jia Y T, Cui X Y, Liu Y X, Liu Y L, Xu C, Li T, Ran Q W, Wang Y F (2020). Drought vulnerability assessment in inner Mongolia.Acta Ecol Sin, 40(24): 9070–9082

[14] Kang L, Wen Y L, Zhou L W, Chen H, Ye J W (2023). Drought driving mechanism and risk situation prediction based on machine learning models in the Yellow River Basin, China.Geomatics Nat Hazards Risk, 14(1): 2279493

[15] McKeeT B, Doesken N J, KleistJ (1993). The relationship of drought frequency and duration to time scales. In: Proceedings of Vulnerability. Cambridge: Cambridge University Press

[16] Merabti A, Martins D S, Meddi M, Pereira L S (2018). Spatial and time variability of drought based on SPI and RDI with various time scales.Water Resour Manage, 32(3): 1087–1100

[17] MishraK A, SinghP V (2010). A review of drought concepts. J Hydrol (Amst), 391(1−2): 204–216

[18] Mondal S, Mishra A K, Leung R, Cook B (2023). Global droughts connected by linkages between drought hubs.Nat Commun, 14(1): 144

[19] Nie M Q, Huang S Z, Leng G Y, Zhou Y L, Huang Q, Dai M (2021). Bayesian-based time-varying multivariate drought risk and its dynamics in a changing environment.Catena, 204: 105429

[20] Nie MQ, Huang SZ, Huang Q, Wang L, Zhang Y, Guo Y (2020). Meteorological-hydrological drought risk assessment and dynamic evolution based on nonparametric method.J Nat Disasters, 29(2): 149–160

[21] Otkin J A, Anderson M C, Hain C, Svoboda M, Johnson D, Mueller R, Tadesse T, Wardlow B, Brown J (2016). Assessing the evolution of soil moisture and vegetation conditions during the 2012 United States flash drought.Agric For Meteorol, 218-219: 230–242

[22] Paulo A, Pereira L S (2007). Prediction of SPI drought class transitions using Markov chains.Water Resour Manage, 21(10): 1813–1827

[23] Pu Y X, Ma R H, Ge Y, Huang X Y (2005). Spatial-temporal dynamics of Jiangsu regional convergence with spatial Markov chains approach.Acta Geograph Sin, 60(5): 817–826

[24] Qian Z A, Wu T W, Song M H, Ma X B, Cai Y, Liang X Y (2001). Arid disaster and advances in arid climate researches over Northwest China.Adv Earth Sci, 16(1): 28–38

[25] Sherwood S, Fu Q (2014). A drier future.Science, 343(6172): 737–739

[26] Shi Y Z, Wang J, Wang Z Q, Lu D M, Yang X J (2017). Rural household vulnerability to drought and adaptation mechanism on the Loess Plateau.Prog Geogr, 36(10): 1281–1293

[27] Stagge J H, Tallaksen L M, Gudmundsson L, Van Loon A F, Stahl K (2015). Candidate distributions for climatological drought indices (SPI and SPEI).Int J Climatol, 35(13): 4027–4040

[28] Sun P, Zhang Q, Bai Y G, Zhang J H, Deng X Y, Liu J Y (2014). Transitional behaviors of hydrometeorological droughts in Xinjiang using the Markov chain model.Geogr Res, 33(9): 1647–1657

[29] Sun Y, Fu R, Dickinson R, Joiner J, Frankenberg C, Gu L H, Xia Y L, Fernando N (2015). Drought onset mechanisms revealed by satellite solar-induced chlorophyll fluorescence: insights from two contrasting extreme events.J Geophys Res Biogeosci, 120(11): 2427–2440

[30] Veettil A V, Konapala G, Mishra A K, Li H Y (2018). Sensitivity of drought resilience vulnerability- exposure to hydrologic ratios in contiguous United States.J Hydrol (Amst), 564: 294–306

[31] Wang F, Lai H X, Men R Y, Wang Z P, Li Y B, Qu Y P, Feng K, Guo WX, Jiang Y Z (2024). Dynamic variations of agricultural drought and its response to meteorological drought: a drought event-based perspective.J Geophys Res: Atmosph, 129(12): e2024JD041044

[32] Wang J S, Li Y H, Wang R N, Feng J Y, Zhao Y X (2012). Preliminary analysis on the demand and review of progress in the field of meteorological drought research.J Arid Meteor, 30(4): 497–508

[33] Wang S J, Wang Y, Zhao Y B (2015). Spatial spillover effects and multi-mechanism for regional development in Guangdong province since 1990s.Acta Geogr Sin, 70(6): 965–979

[34] Xiang Y Y, Wang Y, Chen Y N, Zhang Q F, Li H W (2022). Comprehensive evaluation of hydrological drought characteristics and their relationship to meteorological droughts in the upper Tarim River Basin, central Asia.Front Earth Sci, 16(4): 890–905

[35] Xiao M Z, Zhang Q, Chen X H (2012). Spatial-temporal Patterns of Drought Risk across the Pearl River Basin.Acta Geograph Sin, 67(1): 83–92

[36] Xu K, Yang D W, Xu X Y, Lei H M (2015). Copula based drought frequency analysis considering the spatio-temporal variability in Southwest China.J Hydrol (Amst), 527: 630–640

[37] Xu X L, Peng T, Lin Q X, Dong X H, Wang Y X, Liu J, Chang W J, Wang G X (2022). Analysis on characteristics and driving factors of drought risk transmission in southwest China.J Hydroelect Eng, 41(12): 69–79

[38] Yang Q, Li M X, Zheng Z Y, Ma Z G (2017). Regional applicability of seven meteorological drought indices in China.Sci China Earth Sci, 60(4): 745–760

[39] Yang W T, Deng M, Tang J B, Jin R (2020). On the use of Markov chain models for drought class transition analysis while considering spatial effects.Nat Hazards, 103(3): 2945–2959

[40] Yoo J Y, Kim U, Kim T W (2013). Bivariate drought frequency curves and confidence intervals: a case study using monthly rainfall generation.Stochastic Environ Res Risk Assess, 27(1): 285–295

[41] Zhai P M, Zou X K (2005). Changes in temperature and precipitation and their impacts on drought in China during 1951−2003.Adv Clim Chang Res, 1(1): 16–18

[42] Zhang Q, Yao Y B, Li Y H, Huang J P, Ma Z G, Wang Z L, Wang S P, Wang Y, Zhang Y (2020). Progress and prospect on the study of causes and variation regularity of droughts in China.Acta Meteorol Sin, 78(3): 500–521

[43] Zhang Q, Yao Y B, Wang Y, Wang S P, Wang J S, Yang J H, Wang J, Li Y P, Shang J L, Li W J (2019). Characteristics of drought in Southern China under climatic warming, the risk, and countermeasures for prevention and control.Theor Appl Climatol, 136(3−4): 1157–1173

[44] Zhang Y, Hu X, Zhang Z X, Kong R, Peng Z H, Zhang Q, Chen X (2023). The increasing risk of future simultaneous droughts over the Yangtze River basin based on CMIP6 models.Stochastic Environ Res Risk Assess, 37(7): 2577–2601

[45] Zuo Q, Jiang L, Feng Y, Diao Y (2020). Spatiotemporal variation of ecological footprint of water resources in the provinces in the Yellow River basin.J Irrigatd Drain, 39(10): 1–8

RIGHTS & PERMISSIONS

Higher Education Press