We used the definition of the geographical scope of Central Asia by Hu et al. (2018) as the spatial domain of this study area, namely the north-western regions of China (Xinjiang region) and five countries of the former Soviet Union: Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, and Uzbekistan. Located in the centre of Eurasia and far from the ocean, CA is a typical arid and semi-arid region, often experiencing long-term rainfall deficits and having a variable topography. Figure 1 shows that the topography descends from the eastern mountain ranges to the western steppe region. The Tianshan Mountains, the Pamir Plateau, and the Tarim Basin dominate the terrain and contribute to the region of the eastern alpine and arid desert climates (Xinjiang, Kyrgyzstan, and Tajikistan). The western and central areas are dominated by lower elevations (e.g. Caspian depression), characterized by extensive temperate grassland and shrubland (steppe) zones. This topography results in significant temperature and precipitation gradients from north to south and lowlands to mountains (Beck et al., 2018). The complex interplay between the variable topography and frequent precipitation deficits (Hu et al., 2018; Spinoni et al., 2019) leads to a fragile state within the terrestrial environment, creating conditions conducive to drought events (Guo et al., 2019).

In this study, precipitation (PCP) droughts, evapotranspiration (ET) droughts, runoff droughts, and root zone soil moisture (SM) droughts in the water cycle are investigated to identify and characterize the drought cascades across four types of systems, namely the atmosphere, hydrosphere, biosphere, and soil layer. The datasets used in this study cover the time window from 1 January 1980 to 31 December 2007. The datasets used in this study were standardized to a daily resolution and a spatial resolution of 0.25∘. The details of the dataset information are provided below.

The approach proposed here is summarized in three parts (Fig. 2). The three-dimensional (3D) single drought event is identified in Sect. 2.3.1 based on the continuously dynamic space–time motion. In Sect. 2.3.2, we integrated the different types of single drought events with similar space–time motion to form the cascading drought event. The features of single drought events and cascading drought events focusing on this study were introduced in Sect. 2.3.3. As precipitation deficits are generally considered as the initiation of droughts, PCP droughts are considered the commencement of a cascading drought event in this study.

This study identified 503 cascading drought events. Among these, the four-system cascading drought pattern (51.38 %) occurred more frequently than the three-system (37.99 %) and two-system (9.64 %) cascading drought pattern across the study region (Fig. 4). This higher prevalence of four-system cascading patterns suggests that PCP, ET, runoff, and SM droughts tend to occur in a consecutive manner instead of in isolation. Notably, the cascading drought patterns that occurred with a frequency greater than 10 % were PCP + ET + runoff (17.52 %), PCP + runoff + ET (16.93 %), and PCP + ET + runoff + SM (12.40 %). In contrast, the PCP + SM cascading pattern, which is associated with root zone soil moisture, was the least frequent cascading pattern, accounting for less than 0.4 % during the study period. ET droughts exhibited a shorter average time delay, with a higher concentration of data points in the lower quartile of the violin boxplot (Fig. S1 in the Supplement). In contrast, SM droughts have a longer average time delay, with a greater concentration of data points in the upper quartile of the violin boxplot. This evidence suggests that ET droughts occur predominantly as the initial system of the drought cascade following PCP droughts, while runoff and SM droughts tend to follow sequentially as the second and third positions, respectively.

Notable characteristic evolution in duration, area, and severity is observed among PCP, ET, runoff, and SM droughts during the cascading process. The duration of droughts was significantly extended when the droughts progressed from the atmosphere to the terrestrial system. Specifically, the duration of PCP droughts is the shortest among the four types of droughts. In contrast, SM droughts involving the root zone soil layer exhibit the most extended duration (Fig. 5a) in all cascading drought patterns. For the ET, runoff, and SM droughts occurring in the terrestrial system, the area of ET droughts is significantly greater than that of runoff droughts. The affected area of SM drought is consistently the smallest among all cascading patterns (Fig. 5b). Regarding severity (Fig. 5c), SM droughts exhibit the lowest average severity among the three types of cascading droughts occurring on the land surface in three-system and four-system cascading patterns. But several extremely severe SM droughts exceed the upper threshold of severity for ET and runoff droughts.

The upper limits regarding the duration, area, and severity of PCP, ET, runoff, and SM droughts are significantly higher in four-system cascading patterns compared to the same drought type in two-system and three-system cascading patterns (Fig. 5), indicating that the high-system cascading pattern consists of more pronounced drought events. For example, all droughts in two-system cascading patterns last less than 365 d, while in three-system and four-system cascading patterns, drought events persisting for more than 365 d are much more (Fig. 5a). These findings suggest that when single long-term droughts with markedly severe distribution over large areas are observed, the high-system cascading drought event will likely accompany.

Based on the total severity, the three most severe cascading drought events for different numbers of systems involved are identified and the corresponding drought characteristics are listed in Table 2. As shown, the total severities of three top cascading events in the two-system cascade are from -11.46 to -32.87 × 106 d km2, remarkably lower than that in three-system (-78.15 to -127.84 × 106 d km2) and four-system cascading events (from -2038 to -925.78 × 106 d km2). The higher total severity of cascading events with more systems involved suggests increased systematic risk when the cascading drought event involves more systems.

The most severe cascading drought event is the PCP + runoff + ET + SM (no. 7) in the four-system cascading pattern. This cascading event commenced from the PCP drought on 9 December 2006 and terminated by the end of PCP droughts on 29 December 2007, lasting for over 12 months and spanning 99.8 % of the study area. During the same period, the runoff drought emerged on 1 January 2007 and terminated on 13 May 2007, lasting 133 d. The ET drought followed the runoff drought and appeared in the summer lasting from 24 June to 29 July 2007. The SM drought occurred at the latest and persisted for 50 d with the smallest affected area. The severe drought event that occurred in 2007 coincided with the findings of Yoo et al. (2022). The Food and Agriculture Organization (FAO) has also documented substantial damage to the agriculture sector due to the drought event affecting Turkmenistan, Tajikistan, and Kyrgyzstan in 2007 (Patrick, 2017).

As the cascading connection in this study is identified by the spatial–temporal dynamic movement rather than fixed spatiotemporal overlapping thresholds, there is not always a temporal overlapping pattern in all cascading drought events. For instance, in the no. 2 cascading drought event, the PCP drought lingers from 19 July to 6 August 2002, while the ET drought commences on 26 July 2002, during the middle of the PCP's persistence period. In contrast, in the no. 5 cascading drought event, the PCP drought occurred during the spring of 1997, lasting from 22 March to 9 April 1997, and does not exhibit temporal overlap with the subsequent SM drought from 22 June to 17 July 1997.

A PCP + runoff + ET cascading drought event is taken as an example to elaborate the spatial–temporal continuum movement during cascading processes in detail (Fig. 6a–c). This cascading event lasted 5 months, from the onset of the PCP drought on 6 May 1981 to the end of the ET drought on 10 October 1981. The affected area spread around the northern region of Kazakhstan, where all single droughts spread east–westwards (Fig. 7). The PCP drought developed and decayed rapidly (Fig. 6d and e) within 30 d. The PCP drought began on 6 May 1981, extended quickly to the peak (88 125 km2) within 9 d, and then disappeared within 20 d. Runoff drought and ET drought develop relatively slower than PCP drought. The runoff drought lasted 105 d, starting on 13 June, peaking at 17 August (61 250 km2), and terminating on the 25 September. The ET drought began on 21 June, peaked on the 30 August (92 500 km2), and terminated on the 10 October.

Cascading drought events, which present a systematic drought risk induced by drought evolution, deserve more attention than the traditional focus on individual droughts due to their higher severity (Fig. 8) and the more dynamic spatial and temporal scope. Traditional approaches to drought assessment have focused primarily on analysing single systems, neglecting the potential for cascading linkages between different systems, potentially leading to an underestimation of overall drought risk. Some isolated droughts may not be noticeable. However, when multiple hazards are aggregated and occur in conjunction, their combined severity becomes more apparent, leading to more pronounced consequences on the socio-economic system. For example, a single PCP drought of only 30 d, such as the cascading event shown in Sect. 3.4 (Figs. 6–7), may not be considered a significant drought event. However, when considering the subsequent runoff and ET drought, the cumulative systematic risk of all the droughts included in the cascading drought event must still be considered. Figure 8 shows an apparent increasing trend in total severity as the number of systems increases. CA is dominated by the four-system cascading drought pattern (Fig. 4). This suggests that once the PCP droughts occur, the high systematic drought risk in Central Asia, spreading across the atmosphere, biosphere, hydrosphere, and root zone, is likely to emerge. To this end, it is imperative to broaden the scope of drought risk assessments to include cascading hazards to enable the development of comprehensive drought management plans that take into account the magnitude and characteristics of impending disasters (de Brito, 2021).

The droughts in terrestrial system identified in cascading drought events, including ET droughts in the biosphere, runoff droughts in the hydrosphere, and SM droughts in the root zone soil layer, exhibit a prolonged duration compared to PCP droughts in the atmosphere (Fig. 5). This phenomenon is consistent with the findings of Jiang et al. (2023), where ecological droughts defined by the drought index including soil moisture and normalized difference vegetation index (NDVI) persist longer than meteorological droughts defined by SPI. The persistence of ET droughts, runoff droughts, and SM droughts may be attributed to the fragile ecosystem resilience in Central Asia, which is originally shaped by the arid environment and sparse vegetation cover, leading to desertification risks and poor recovery capability (Jiang et al., 2019). SM droughts have the longest duration, slightest severity, and smallest area relative runoff droughts and ET droughts (Fig. 5), suggesting that the fluctuations in drought signal become smoother and broader when they proceed from meteorology through depth to the root zone soil. This result is consistent with the findings of existing research (Farahmand et al., 2021; Van Loon, 2015). This phenomenon may be driven by the complex trade-offs between vegetation, groundwater, and human activities that make droughts in terrestrial systems slower, stealthier, and more persistent than droughts in the atmosphere.

As the commencement of the cascading drought event, precipitation has the more readily available data compared to ET, runoff, and SM. To examine whether the PCP single drought events could signify the severity of systematic drought risk, features of PCP drought in the cascading events involving the different amounts of systems were investigated in Fig. 9. Interestingly, the upper threshold of PCP droughts duration in four-system, three-system, two-system, and no cascading drought patterns corresponds to the types of yearly (420 d), seasonal (101 d), monthly (55 d), and sub-monthly (22 d), respectively. These data imply that there would be a four-system cascading drought event after seasonal PCP drought events lasting longer than 101 d occur. The severity and area of magnitude in PCP droughts increase as the number of systems increases (Fig. 5), further proved in Fig. 9c and d. The upper threshold of severity (Fig. 9c) in four cascading droughts patterns could reach up to -2022.9, -29.7, -4.7, and -0.76 106 d km2, whereas that of area percentage is 99 %, 30 %, 6.5 %, and 1.5 %, respectively (Fig. 9d). The intensity of PCP droughts in four-system cascading patterns is higher than in the other three patterns (Fig. 9b). However, the intensity of PCP droughts in the no-cascading drought pattern is higher than in the three-system and two-system cascading drought pattern and in some cases even higher than in the four-system cascading pattern. However, it may be limited by their short duration and small areas. PCP droughts in the no cascading drought pattern did not form the cascading drought event.

The above result highlights that the characteristics of PCP droughts could serve as indicators of systematic drought risk. Long-term PCP droughts with greater severity and intensity in larger areas are more likely to form extensive cascading droughts across more systems and cause greater systematic risk. Further research is needed to understand the characteristics and underlying mechanisms of moderate and small PCP droughts that trigger cascading drought events with more systems involved.

Multiple cascading drought patterns with mixed temporal order were identified in this study. The result likely reflects the integrated driving effect of the water-limited and energy-limited phases. When at the energy-limited phase, the soil moisture is sufficient and does not regulate the evaporation. During this period, the precipitation deficit might enhance the potential and actual evaporation demand. The cascading drought pattern might present the PCP + runoff, PCP + runoff + ET, or PCP + ET + SM. When the water-limited phase begins, PET could continue to increase. However, the increase in AET is no longer sustained and subsequently decreases as soil moisture deficits are limited, during which the evaporative deficit gradually increases and evaporative droughts are exacerbated (Pendergrass et al., 2020). During the water-limited phase, the cascading drought pattern possibly presented as the PCP + SM + ET.

The three most common cascading drought patterns identified in this study are PCP + ET + runoff, PCP + runoff + ET, and PC P+ ET + runoff + SM. This result could be explained from two aspects. (1) The drought signal in soil moisture is only sometimes detected. (2) Given the more significant evaporative deficits in the water-limited phase, the significant ET drought signal after PCP droughts shown in the cascading drought pattern indicates the occurrence of cascading droughts under the water-limited phase. This result could be for three possible reasons. First, the SM data adopted in this study are from the soil layer of 40–100 cm. The shrubland and grassland with short roots are the dominant vegetation in the study area (Klein et al., 2012), and their root density is typically concentrated on the 0–40 cm soil layer (Kurc and Small, 2007). This could result in lower vegetation water requirements and less fluctuation of the space–time signal of soil moisture in the deep layer.

Second, the coupling of SM and evaporation raw data might not be reflected by their drought index. The fluctuation in deep layer soil moisture is most stable relative to the precipitation and runoff (Van Loon, 2015). The SM drought index might not recognize declines in SM in the constantly dry region that fluctuates within a small magnitude. Moreover, the spatiotemporal progression of SM and ET droughts is different. The spatiotemporal progression of SM droughts is the slowest compared to the other three droughts, which is reflected by the fact that SM droughts are the longest in duration and the smallest in the area (Fig. 5).

Third, the effect of energy-limited regimes could be underestimated in Central Asia. Hsu and Dirmeyer (2023) recognized most Central Asia as the transitional and energy-limited regime. Yuan et al. (2023) found that around 30 %–50 % of soil moisture drought events in CA are flash droughts possibly induced by the precipitation deficit and increased evapotranspiration at the energy-limited phase. This could be presented as the PCP + ET + SM or PCP + ET + runoff + SM cascading drought pattern. Central Asia is typically characterized by significant warming in the past years compared to the average worldwide (Mirzabaev, 2013). Under significantly warming conditions, annual evapotranspiration (ET) significantly increased during low precipitation and sequentially amplified the runoff deficit in alpine regions (Mastrotheodoros et al., 2020). Zhao et al. (2022) found that previous studies overestimate the reduction in evaporation from bare soil during droughts in arid climates under water-limited regimes. They found that ET increases during almost half of the drought months worldwide (drought was defined by the deficits in precipitation and total water storage) as a result of increased atmospheric water demand outweighing the limiting effect of soil moisture supply on ET. Possible reasons may be that traditional modelling of the ET response to soil moisture stress downplays the deep soil moisture supply to ET, could better constrain plant drought response traits, and ignores the effects of soil structure on soil hydraulic conductivity.

The driving mechanism of large-scale drought evolution is highly complex involving modulators (atmospheric oscillations), climate drivers (precipitation and temperature), feedback of land surface phenology, and anthropogenic activities. All these factors and their interactions construct the initial condition and running situation for drought evolution and trigger the variation of cascading drought patterns. In this study, the diversity of topography and land surfaces complicates the regional climate regimes of Central Asia. The introduction of spatiotemporal movement in the definition of cascading drought events has resulted in that a cascading drought event may be involved in multiple climatology–land-cover regimes, further adding to the complexity of identifying the driving mechanism. Therefore, the simple comparison or common correlation method with single factors, often applied at gauge/catchment scale, may not be appropriate for exploring the driving mechanism of the large-scale cascading phenomenon. More information and an appropriate attribution framework need to be explored in the next stage.

Many studies are customized to link meteorological drought to hydrological drought by point-to-point time series correlation based on the drought index (Geyaert et al., 2018; Guo et al., 2020; Van Loon et al., 2012). These studies identified their lag, attenuation, prolongation, and pooling based on the time series between two drought types. However, these studies are often conducted at a fixed spatial scale, such as at a gauge station or in a catchment, and the spatial variation in drought progression is not considered.

More recently, researchers have begun to define three-dimensional single drought events by introducing continuous spatial–temporal motion and, based on this, have adopted temporal and spatial overlap thresholds to determine whether there is a relationship between two drought events (Jiang et al., 2023; Liu et al., 2019a). The method proposed in this study not only introduces spatiotemporal motion but also integrates the time series correlation method. The daily timescale used in this study provides more detailed time series of drought area changes and more accurate spatial trajectories (Fig. 7). The high-resolution time series at daily timescale provides more input data for analysing temporal and spatial movement correlations (Fig. 6). Based on the above, the method can detect the relationship without temporal overlap (Table 2 and Fig. 6), identify the linkage among more than two types of droughts, and address their temporal orders. It is more suitable to the actual situation of droughts cascade in the real world as the drought evolving along the water circle may span a long period and interrelated droughts do not necessarily present overlap in time. The similarity rule of the space–time motion could break the lots of temporal scale and space pre-restrictions in the conventional methodology and reduce the spatiotemporal incomparability, which is helpful to reduce subjective pre-determination in exploring drought cascades and proceed in recognizing drought cascades more flexibly. Moreover, beyond the four types of droughts discussed in this study, the dynamic spatiotemporal movement can be observed within droughts occurring in different systems, such as the socio-economic system, and other related hazards, such as heatwaves. Consequently, the framework presented in this study offers valuable prospects for identifying multidimensional cascading linkages between different types of hazards.

Farahmand et al. (2021) investigated the cascading drought phenomenon involving precipitation, runoff, soil moisture, and streamflow to explore their temporal structure. This study differs from our work in three aspects. (1) This work did not involve identifying the relationship between different drought events, as they directly linked the most significant drought events in a given region to form a cascading event. (2) Moderate and small drought events were also not considered (e.g. flash droughts). (3) They do not take into account area variations and spatial motions during drought progression. The framework proposed in this paper based on the dynamic space–time motion at the daily timescale enables us to capture more sub-seasonal droughts at moderate and small spatial scales (Table 2 and Fig. 6), such as flash drought. It will provide a better understanding of the evolution of sub-seasonal drought across multiple systems, which is critical to design more scientific adaptation policies in the face of increased flash drought and related compound extreme events under human-induced climate change.

Cascading events, in general, refer to the sequential occurrence of different hazards over time. These events can exhibit correlations from shared drivers, causal mechanisms, or simply by chance. However, distinguishing between these cases poses a challenge due to limited sample sizes and an incomplete understanding of the intricate system comprising myriad influencing factors (Zscheischler et al., 2020). This challenge is also present in this study, given the limited knowledge of external environmental parameters (e.g. temperature, vegetation, and human activities) in the evolution of droughts. In addition, the cascading connection has proceeded in the spatial area of PCP droughts, and information outside the PCP drought area is not discussed in this work. Furthermore, only one ET/runoff/SM drought event is identified in each, which may lead to an underestimation of the severity of cascading droughts. Consequently, further research is warranted to explore these aspects in future studies.