1. Introduction
Drought is a recurring, complex, and multidimensional natural phenomenon that occurs across a range of temporal and spatial extents and has significant impacts on regional hydrological cycles and water balance (Vicente-Serrano et al. 2012). Droughts can pose a serious threat to crop loss, ecosystem degradation, and economic development (Liu et al. 2020). In the twentieth century, 11 million people died from drought around the world, and more than 2 billion people were affected by the disaster (Etienne et al. 2016). Drought is one of the most devastating disasters in terms of crop yield reduction, sustainable development of river basins, and their dramatic influences on human societies (Konapala and Mishra 2020; Huo-Po and Jian-Qi 2015). In particular, most severe impacts of drought on population are thought to occur in the inland regions of China, where precipitation is low and evaporation is high; even under normal conditions available water resources are scarce and adaptation options for the population are limited (Ding et al. 2021; Y. Li et al. 2017; Xu et al. 2015). In northwest China, the areas affected and hit by droughts have been increasing over the past two decades (Y. Li et al. 2017; Zhang et al. 2012). Therefore, timely monitoring, analysis, and assessment of droughts in the region can help to improve the understanding of changes in drought risk under climate change.
Due to the different objectives in drought research and the complexity of drought formation mechanisms, drought can be divided into four categories: meteorological drought, hydrological drought, agricultural drought, and socioeconomic drought (Trenberth et al. 2014; Mishra and Singh 2011). Meteorological drought is defined as below-normal precipitation combined with increased potential evapotranspiration for a long period, known as the most basic and widespread drought. Hydrological drought, on the other hand, is associated with below-average surface water flow or subsurface water supply after a long period of precipitation and is considered as the most complete drought (Mishra and Singh 2010). Since atmospheric, surface, and subsurface water storages are closely linked in the hydrologic cycle, there is a propagation relationship from meteorological to hydrological drought (Eltahir and Yeh 1999; Wilhite 2000), which has become a hot topic in the hydrological community during the last two decades (Apurv et al. 2017; Haslinger et al. 2014; McEwen et al. 2021; K. Zhou et al. 2021; Zhu et al. 2019).
On one hand, Jehanzaib et al. (2020), Melo and Wendland (2016), Sattar et al. (2019), and Van Loon and Van Lanen (2012) found lagging and prolonged characteristics in the propagation of drought signals from meteorological drought to hydrological drought. On the other hand, some studies also found meteorological drought does not necessarily lead to hydrological drought. For example, Shin et al. (2018) calculated the propagation probability using conditional probability theory and found that 33%–48% of meteorological droughts propagated to hydrological drought instead of 100% probability. Liu et al. (2019) connected meteorological and hydrological drought events from a three-dimensional perspective and concluded that minor meteorological droughts generally were not prone to cause a hydrological drought, and that only when meteorological drought reached a certain critical value would the hydrological drought occur. The above studies suggested that there are two effects of drought propagation that cannot be ignored. One is interpreted as the cumulative time-lag effect of long-term precipitation deficit on hydrological drought, and the other is defined as the step-abrupt effect of the water deficit signal that occurs when a certain level of meteorological drought is reached, which are quantified by propagation time and propagation threshold, respectively. These response indicators determine the catchment resistance to hydrological droughts in the event of a meteorological droughts, and together they can enable water resource management agencies to mitigate hydrological droughts by taking measures such as water storage, increasing revenue, and reducing expenditure. In general, propagation time and propagation threshold are not only complementary and indispensable in the study of the two types of drought propagation processes, but are also an important bridge to understand the relationship between atmospheric and water resources, and an important basis for improving hydrological drought monitoring and early warning skills.
Previous studies have confirmed that the drought propagation time from meteorological to hydrological drought has a clear seasonal character. For example, Wu et al. (2016) concluded the lag time was 1 month on average in spring, summer, and autumn but about 3 months in winter in the Jinjiang River basin of China. Huang et al. (2017) found the propagation time in spring and summer was short, while it was long in autumn and winter in the Wei River basin of China. In terms of drought propagation threshold research, Guo et al. (2020) proposed a drought propagation threshold calculation model based on the Bayesian network from the perspective of drought events characteristics. Guo et al. (2021) then applied the above model to calculate the cumulative precipitation deficit thresholds that triggered different levels of hydrological drought to assess the impact of very large reservoirs with multiyear regulation capacity on the drought resilience of the basin. Such a Bayesian-network-based drought propagation threshold model provides a new way to calculate the propagation threshold. However, previous studies have not yet clarified one question: is the drought propagation threshold also seasonal as the drought propagation time? If so, what are the seasonal propagation thresholds of meteorological drought corresponding to different classes of hydrological drought (i.e., moderate, severe, extreme). To this end, the Bayesian network framework is proposed to calculate seasonal drought propagation thresholds triggering different classes of hydrological drought (i.e., moderate, severe, extreme) using the standard precipitation index (SPI) and to discuss the characteristics of different seasonal drought propagation in terms of time lags and step-abrupt effects in conjunction with seasonal changes in propagation times.
In addition, most studies related to drought propagation characteristics have been conducted along the southeast coast of China and along the Yellow River (Fang et al. 2020; Wu et al. 2021). Few studies have focused on the alpine region of Xinjiang. The Tarim River basin (TRB) in Xinjiang contains a purely dissipative inland river in an arid region of northwestern China, which has the features of low precipitation and high evaporation and arid climate (Chen et al. 2009, 2011). Most of the runoff from the mountains region is supplied to agricultural irrigation, and the rest mainly flows to the middle-lower reaches (Chen et al. 2009). Streamflow is a combined response to watershed climate, water transfer, water losses in the process of evapotranspiration and storage, and the effects of human activities on natural water flows. Prior to reaching the stream network, a large proportion of precipitation is stored in various hydrological subsystems (including snow cover, glacier, soil, moisture, groundwater reserves, reservoir storages, and so on), which respond to climatic conditions at different time scales (McGuire and McDonnell 2006; Miao et al. 2022; Vicente-Serrano and López-Moreno 2005; Zhang et al. 2021; Jiang et al. 2021). Previous studies have shown that the drought propagation under different regions varies depending on the meteorological conditions (evapotranspiration, snow cover, rainfall), catchments (geology, topography, soils) and man-made infrastructure (reservoir) (Tijdeman et al. 2018; Van Loon 2015; Y. Li et al. 2021; McEwen et al. 2021). In fact, the study area (viz., the headstreams of the TRB) itself has different climatic and catchment characteristics, which jointly develop the diverse (i.e., rapid or delayed) drought propagation patterns (Sun et al. 2017). Hence, it is necessary to explore the dynamics and the corresponding driving mechanism of propagation time and propagation threshold in a changing environment in order to better manage water resources and reduce the impacts of drought.
In this study, the propagation time was determined by the correlation coefficient between a single standardized runoff index (SRI) time scale and different SPI time scales. By constructing a Bayesian network framework, propagation thresholds for meteorological drought were calculated for different levels of hydrological drought in different seasons. Then, the seasonal dynamics and driving mechanism of drought propagation time and threshold in a changing environment were explored based on 25-yr moving window and Mann–Kendall (MK) test, and the most important factors affecting them were selected through random forest model and Pearson correlation. Results of the study may increase our understanding of the drought propagation mechanisms and help to improve drought monitoring and early warning systems.
2. Study area and datasets
a. Study area
The Tarim River basin is the largest inland river basin in China, mainly located in southern Xinjiang, with a drainage area of about 1.02 × 106 km2. With the intensification of the influence of modern human activities, especially the development of oasis agriculture, great changes have taken place in the TRB. At the same time, each river system has gradually disintegrated and separated from its connection with the main stream. At present, the Aksu River (including Kumarik and Toxkan River), Yarkant River, and Hotan River (including Yurungkax and Karakax River) are the only remaining headstreams that have direct surface water connection with the main stream of TRB, accounting for 73.2%, 3.6%, and 23.2% of the mainstream total water volume, respectively (Chen et al. 2007). These streams originate in the Tianshan, Karakoram, and Kunlun Mountains, which are surrounded by abundant glaciers and snow. Therefore, glacier snowmelt runoff and mountain precipitation are the main recharge sources of river runoff in the source area of TRB (Chen et al. 2011).
The TRB is surrounded by the Taklimakan Desert. It belongs to the continental warm zone with extreme arid climate, which has the characteristics of scarce precipitation, strong evaporation, and large climate difference between the four seasons (Sun et al. 2012). The annual average temperature is 2°–13°C, and the average annual precipitation is 37–110 mm in the headstreams of the TRB. Due to the special geographical location, the TRB is extremely short of water resources, with few forests and low vegetation coverage, the large Gobi Desert and bare land area, and extremely fragile ecology in the region.
According to the distribution of hydrological stations and the controlled area, the headstreams of the TRB are divided into five zones, namely, subbasins I, II, III, IV, and V (Fig. 1).
Overview map of the study area.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
b. Datasets
The datasets used in this study include hydrometeorological observations and GLDAS simulations. Hydrological data (monthly runoff) were collected from five hydrologic stations in the headstream mountain areas of the basin during 1960–2013 and were provided by the Tarim River Basin Management Bureau. Meteorological data (daily precipitation, air temperature, vapor pressure, sunshine hours, wind speeds, maximum and minimum air temperature data) from nine meteorological stations during 1960–2013, were provided by the National Climate Information Center of the China Meteorological Administration (Fig. 1). Furthermore, potential evapotranspiration (PET) was calculated through the Penman–Monteith equation (Allen et al. 1998).
The Global Land Data Assimilation System (GLDAS) provides valuable information of land surface water and energy fluxes for hydrometeorological investigation based on multiple models, ground-based and satellite-based observations, and data assimilation techniques (Rodell et al. 2004; Wang et al. 2016; K. Zhou et al. 2021). Previous studies have verified GLDAS performance in the Trim River (Chen et al. 2013). The GLDAS version 2/Noah land surface model product (GLDAS_Noah025_M) at 0.25° × 0.25° resolution is available via the Goddard Earth Sciences Data and Information Services Center website (http://disc.sci.gsfc.nasa.gov/hydrology/data-holdings). GLDAS soil moisture at four layers (including 0–10, 10–40, 40–100, 100–200 cm) was used in this study.
3. Methodology
a. Drought index
SPI was initially proposed by McKee et al. (1993) and recommended by the World Meteorological Organization as a reference drought index, which is widely used in many studies of different aspects of droughts such as drought propagation and impact assessment (Fang et al. 2020; Ding et al. 2021). Previous studies have also confirmed that SPI is applicable for TRB (Zhang et al. 2015; Yang et al. 2018). SPI is a multivariate meteorological drought index based on probability distribution of precipitation, which includes major merits of simplicity in calculation, low data requirements, and comparability over time and space as well as the multiscalar feature that allows us to monitor precipitation variability in different accumulation periods (McKee et al. 1993; Shi et al. 2020). Among them, the accumulation period is represented by SPIx. For example, SPI2 denotes a 2-month precipitation accumulation period. SRI (Shukla and Wood 2008) was adopted to characterize hydrological drought in this study because its calculation process and advantage is similar to those of the SPI (Aghakouchak and Nakhjiri 2012). SRI uses the monthly runoff time series to fit the distribution function. Therefore, SPI and SRI were used for identifying meteorological and hydrological droughts, respectively. The classification criterion for SPI and SRI are shown in Table 1 (Livada and Assimakopoulos 2007).
Drought classification schemes.
b. Seasonal drought propagation time calculation
Drought propagation time is a useful indicator for quantifying the time lag effect in the drought propagation (Fang et al. 2020; Gevaert et al. 2018). The long-term drought response is usually interpreted as strong drought resistance of the catchment to hydrological drought during the occurrence of meteorological drought. In contrast, rapid response is usually associated with low catchment resistance. The understanding of drought propagation time is of great significance for hydrological drought identification and early warning (Fang et al. 2020). The Pearson correlation analysis has been widely used to determine drought response time by previous studies (Ding et al. 2021; Hellwig et al. 2020; Huang et al. 2019; Z. Zhou et al. 2021). First, the SPI was calculated at different time scales ranging from 1 to 24 months (equivalent to a 2-yr timespan). The examination of multiple SPI time scales is beneficial to understand the response of runoff variability to precipitation anomalies and to clarify the time lag between two important variables. The SRI was investigated on a 1-month scale (SRI1) taking into account its description of the impact of runoff changes on the extent of drought in the short term. Based on the multiple time scale SPI and SRI1 sequences, the Pearson method was used to calculate the correlation coefficient between meteorological drought and hydrological drought. Second, the SPI time scale corresponding to the upper limit (viz., the first maximum value) of the correlation coefficient was chosen as the propagation time from meteorological to hydrological drought. Finally, the average propagation time of 3 months in the same season was denoted as the drought propagation time of that season (e.g., the summer drought propagation time was the average of the drought propagation time from June to August).
c. Seasonal drought propagation threshold framework based on Bayesian network
In the present study, the drought propagation framework based on the Bayesian network was proposed for calculating seasonal drought propagation threshold with the normalized index value of meteorological drought (SPI index value), which provides a clear visualization of the threshold in corresponding meteorological drought at given different classes of hydrological drought. The seasonal drought propagation framework is divided into three steps. First, the joint probability density function of SPIm (see section 3b for details) and SRI was calculated. Then, the conditional probability of hydrological drought under meteorological drought was determined. Finally, when we set a fixed conditional probability of 0.7, this means that when the conditional probability is greater than or equal to 0.7, the corresponding meteorological drought value (viz., SPI) will be considered as the propagation threshold for that level of hydrological drought in a given month. The average propagation threshold of 3 months in the same season is denoted as the drought propagation threshold of that season (Fig. 2).
A schematic diagram for identifying seasonal drought propagation characteristics (propagation time and threshold).
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
1) Copula theory
In addition, the fit test of the Clayton copula function was further validated by quantitative [squared Euclidean distance (SED)] and graphical fit (P–P plot).
2) Determination of seasonal drought propagation threshold
Based on the idea of conditional probability, with the increase of meteorological drought conditions, the probability of hydrological drought will be infinitely close to 1. In this study, a fixed conditional probability of 0.7 is set, which means while the conditional probability is equal to or greater than 0.7, the corresponding SPI index value is used as the propagation threshold of hydrological drought at this level. In general, the SPI index is iterated from −5 to 0 in the range of 0.01. The conditional probabilities of SRI less than or equal to −1, −1.5, and −2 (moderate, severe, and extreme hydrological drought) are estimated under different SPI intervals. When the conditional probability of each iteration is greater than or equal to 0.7, the SPI index value is taken as the threshold value of the propagation of meteorological drought to different levels of hydrological drought. The lower the meteorological drought propagation threshold that triggers hydrological drought, the greater the drought resilience of the catchment to a given hydrological drought. Therefore, in this study, SPI index values were used as triggers for specific hydrological drought propagation thresholds to represent the drought resilience of the catchment.
d. Reliability verification of the drought propagation threshold framework
e. Random forest model
Random forest (RF) model is a machine learning algorithm proposed by Breiman (2001) that integrates several relatively simple evaluators (decision trees) to form cumulative effects, which has the advantage of preventing overfitting, strong model stability and easy to deal with nonlinear regression (Chen and Liu 2005; Ward et al. 2006). At present, it has been widely used in various fields such as data mining and big data (Cutler et al. 2007). In particular, RF regression can evaluate the importance of each feature in classification.
In RF, ntree and mtry are two important parameters, where ntree is the the number of trees to grow (ntree = 500 in this study), mtry is the number of variables used to split each node (viz., the number of variables (mtry = 2 in this study). To make the model simulation result closer to the measured value, the Nash–Sutcliffe efficiency (NSE) and coefficient of determination (R2) were used to evaluate the RF model. Since variables (such as propagation time and threshold) were continuous data, the variable importance score (VIM) in RF regression was measured using increased mean-squared error (%IncMSE) and increase in node purity (IncNodePurity). Further details on the RF can be found in Z. Li et al. (2021).
In this paper, RF regression method was used to rank the importance of each influencing factor, combined with correlation analysis method, aiming to explain the key factors that affect drought propagation time and threshold dynamics from meteorological to hydrological drought and to provide a reference for understanding the driving mechanisms of propagation characteristics.
4. Results
a. The propagation time from meteorological to hydrological drought and its seasonality
Figure 3 shows the strength of the relationship between SPI (1–24-month time scales) and SRI1 of all months for the three headstreams on the TRB. Spatially, the upper limit (viz., the first maximum value) of the correlation coefficient decreases from subbasins I and II in the northern (0.63) to subbasins III–V in the south (0.26). Specifically, water vapor transport from zonal westerly circulations and dry and cold airflow from the Arctic Ocean are the atmospheric moisture sources for Xinjiang. The blockage of the Karakoram and Kunlun Mountains makes it difficult for airflow from westerly circulations and the southwest Arabian Sea reaching the southern mountains (including subbasins III–V). Therefore, there is more moisture in northern mountains (subbasins I–II) than in southern ones (subbasins III–V) (Bai et al. 2014; Ding et al. 2021). Ding et al. (2021) found that meteorological drought responded more strongly to hydrological drought in areas with high rainfall. In particular, there was negative correlation for every month in subbasin V, suggesting that meteorological drought in the area may not be the main factor inducing hydrological drought. Interestingly, most of the areas with longer propagation times (more than 9 months) were distributed across subbasins I, III, IV, and V, longer than in subbasin II (less than 9 months). This suggests that there are differences in the response of these watersheds in mediating streamflow to meteorological drought, which may be caused by differences in the sources of runoff supply. The runoff in subbasin II is derived predominantly from a mixture of glacier and snow meltwater and precipitation, whereas runoff in other subbasins is derived from glaciers and snow meltwater (Chen et al. 2011, 2009). Moreover, a large number of glaciers are distributed in the mountainous areas of the subbasins III–V, and when meteorological drought occurs, hydrological drought will not occur in a short period of time as that in the II subbasin due to the regulating effect of glacial meltwater.
Correlation coefficients between the monthly SRI and SPI series at 1–24-month scales in the headstreams of TRB. The Pearson correlation coefficients are shown in green to highlight the upper extremum value (viz., the first maximum value) in each month. (a)–(e) Subbasins I–V, respectively.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
The hydrological cycle shows different characteristics in different seasons; therefore, the hydrological responses to climate vary between seasons (Tallaksen 1995; Aryal and Zhu 2020; García-Ruiz et al. 2008). Figure 3 shows that there is a negative and weak correlation between SPI and SRI in summer. This phenomenon is discussed in detail in section 5b. Figure 4 shows the seasonal variability of drought propagation time. In general, the propagation time (more than 8 months) of autumn and winter are longer than those of spring. Glaciers and snow may be the key factors in explaining the variability of the hydrological propagation time of diverse subbasins within different seasons. This is a result of the storage of autumn and winter precipitation in the form of snow and the subsequent contribution to streamflow during the snowmelt period (in spring). Previous studies have shown that the recent evolution of spring runoff in alpine mountains (i.e., Pyrenees) is not determined by the climatic conditions in spring, but is clearly related to the temperature and precipitation in winter (López-Moreno and García-Ruiz 2004). In North America and northern Eurasia, where there is significant seasonal snow, peak river flows occur earlier in spring, while base flows increase in winter (IPCC 2007).
Seasonal variation of the drought propagation time from meteorological to hydrological drought in the headstreams of TRB (subbasins I–V). Note that the derived propagation time is in the unit of months.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
In general, the upper limit (viz., the first maximum value) of the correlation coefficient between SPI and SRI ranges from 0.26 to 0.63. The hydrological drought response to meteorological drought has a time lag of 2–21 months for the three headstreams of TRB. Previous studies have found that response of hydrological to meteorological drought tend to be achieved at shorter time scales, with an average of 2–6 months in the Yellow River basin (Huang et al. 2017) and the Pearl River basin (Z. Zhou et al. 2021). Given the effects of glacier melting and low temperature on the Tarim River, the propagation time form meteorological to hydrological drought of the TRB is relatively longer than the above basins, suggesting that the propagation process is significantly different in different climatic zones.
b. The propagation threshold from meteorological to hydrological drought and its seasonality
Significant correlation between meteorological drought and hydrologic drought is a prerequisite for the construction of copula function. Although the correlation between SPI and SRI was significant for summer in subbasin II (Fig. 3), to ensure seasonal consistency across all subbasins, different levels (moderate, severe, and extreme) of propagation thresholds are shown only for spring, autumn, and winter in Fig. 5.
Propagation threshold from meteorological to different levels (moderate, severe, and extreme) of hydrological droughts for each month of spring, autumn, and winter in the headstreams of TRB (subbasins I–IV).
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
In general, the evaluation of prolonged and severe hydrological drought can directly influence irrigation/residential water supply and hydropower generation, and ultimately affect the country’s agriculture and economy. Therefore, in this study, moderate, severe, and extreme hydrological droughts were selected as the specific hydrological droughts in drought propagation threshold model. Based on the drought propagation threshold model introduced in section 3c, drought propagation thresholds were calculated at a probability level of 0.70 for the SPI index values that trigger various hydrological droughts (i.e., moderate, severe, and extreme). Figure 5 shows the propagation threshold from meteorological to different levels of hydrological drought in each month of spring, autumn, and winter. When the color turns blue, the propagation threshold is lower (i.e., higher meteorological drought level), which also means that the hydrological system is more resistant to drought during the month. Conversely, red indicates a higher propagation threshold (i.e., lower meteorological drought levels), which means that the hydrological system is more vulnerable to meteorological drought and should be taken into account by the relevant authorities.
Spatially, the propagation thresholds for the different levels of hydrological drought scenarios in subbasin III are dominated by severe meteorological drought, ranging from −4.12 to −2.09, while water resources in subbasin II are less resistant to meteorological drought, with thresholds ranging from −3.18 to −1.19. From a seasonal perspective, the propagation thresholds from meteorological to hydrological drought are higher in spring and winter in most subbasins, showing a lower resistance to drought, but the opposite is true in autumn. However, there is one exception in subbasin III. Interestingly, there is no corresponding propagation threshold in September for subbasins III and IV. It might be caused by the negative or weak correlation coefficient. This indicates that these elements have a probability of less than 70% for triggering hydrological drought. In general, as the intensity of hydrological drought increases, the propagation threshold for meteorological drought triggering hydrological drought is expected to decrease.
c. Propagation time and threshold dynamics of drought and the corresponding driving mechanisms
To further explore the variation of drought propagation process and investigate the drought mechanism, the 25-yr moving window (Villalba et al. 2019) and MK trend test (Kendall 1957; Mann 1945) were used to detect the trend dynamics in drought propagation time (or threshold) between meteorological and hydrological drought and driving factors (precipitation P, temperature T, potential evapotranspiration PET, and soil moisture SM) during 1960–2013. Furthermore, correlation analysis and RF were utilized to evaluate the most important factors between propagation time (or propagation threshold) and driving factors. The factors were smoothed using the 25-yr moving window. This will result in better investigation the dependency structure between meteorological and hydrological drought propagation over the last half century. It is noted that trend detection and attribution analysis are focused on spring, autumn, and winter.
The SPI–SRI series under the 25-yr sliding window were used to calculate 28 sets of drought propagation time (or propagation threshold) according to the proposed framework as depicted in Fig. 2, which was used as dependent variables of RF model. Since this paper only focuses on the importance of independent variables to dependent variables, the whole sequence is used for RF simulation, and its performance is evaluated through NSE and R2. As shown in Table 2, it can be found that the NSE and R2 of propagation time and threshold are generally higher than 0.8, which suggests that the RF model has good performance for various seasons.
Performance evaluation of propagation time and threshold during the whole sequence of the RF model.
1) Temporal dynamics of drought propagation time and preliminary attribution analysis
As shown in Table 3, drought propagation time in the Aksu River basin (subbasins I–II) mostly becomes longer in most seasons over the period 1960–2013, with significant upward trends. The exception is subbasin II, which has an exception in autumn. Subbasins I–II are from the southern slope of the Tianshan Mountains and have similar drought propagation characteristics. The results in Fig. 6 (VIM by %IncMSE) and Table 4 (VIM by IncNodePurity) are generally consistent with those in Fig. 7, indicating that the faster response of hydrological drought to meteorological drought in the Aksu River basin (subbasins I–II) is inferred to be related to the propagation of changes in P and PET into the surface hydrological cycle, especially in spring, when the absolute value of the correlation coefficient is greater than 0.5. Trend detection using the MK test reveals a descending trend in PET (Fig. 8). From a land surface water balance perspective, PET is mainly controlled by soil moisture and energy availability (Seneviratne et al. 2010). Less PET in the catchment can better replenish soil moisture deficits, thereby accelerating the time to infiltration into the hydrological system, reducing the resistance of the catchment to precipitation deficits and slowing the spread of meteorological drought through the terrestrial system. At the same time, Fig. 8 clearly shows that P show a significant upward trend, which to some extent mitigates the severity of meteorological drought in both basins. As a result, the catchment is able to withstand sustained longer periods of precipitation deficit, delaying the hydrological response.
Importance scores (%IncMSE) of drivers (P, T, PET, and SM) vs propagation time for (a) spring, (b) autumn, and (c) winter in the headstreams of TRB (subbasins I–V). All factors were processed by a moving window of 25 years.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
Correlation coefficients between propagation times and drivers (P, T, PET, and SM) for each month of spring, autumn, and winter in the headstreams of TRB (subbasins I–V). All factors were processed by a moving window for 25 years. The circles represent the magnitude of the correlation coefficients.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
The MK test statistics of driving factors (P, T, PET, and SM) for (a) spring, (b) autumn, and winter (c) in the headstreams of TRB (subbasins I–V). All factors were processed by a moving window for 25 years.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
The MK test statistics of drought propagation time in the headstreams of TRB. The MK test statistics are marked with a single asterisk (*) to inform that the detected trend is significant at the 5% level of significance.
Importance scores by IncNodePurity for propagation time. The IncNodePurity value is marked in bold to inform that the corresponding driver is the most important.
The Hotan and Yarkant River basins (subbasins III–V), both from the northern slopes of the Karakorum and Kunlun Mountains, have similar dynamic characteristics in terms of propagation time, with the exception of subbasins I–II, which exhibit opposite characteristics (Sun et al. 2017). Figures 6 and 7 and Table 4 show that T plays a key role in streamflow changes in subbasins III–V, with regional heterogeneity in other factors (P, PET, and SM), due to temperature changes that affect the rate of glaciers and snowmelt (Li et al. 2020; Xiang et al. 2020), resulting in different seasonal variations in drought propagation times. As shown in Table 4, the drought propagation time under the moving window shows a decreasing trend in most seasons, which means that the time to monitor and prevent drought is shortened and should be given more attention. However, the drought propagation time shows a clear upward trend in autumn, suggesting that meteorological and ground-based factors in autumn may not be the main factors triggering drought propagation and that the role of diverse anthropogenic interventions, such as the construction of reservoirs, should be emphasized.
Reservoirs are mega-man-made hydraulic structures that distribute, regulate, and store water resources and reduce variability in natural river flows (Wang et al. 2019; Shiklomanov et al. 2000; Peisert and Sternfeld 2005). In 2007, the Xiabandi Reservoir with the storage capacity of 8.67 million m3 came into operation at the middle and lower reaches of Tashkurgan River, the branch of Yarkant River (subbasin III) in the Tarim River system. Since 2003, the Wuluwati reservoir has been the key control project of the Karakax River basin (the V subbasin), with a total storage capacity of 3.47 billion m3 and a regulating storage capacity of 2.24 billion m3. Previous studies have found that reservoirs have a significant impact on the drought propagation (Guo et al. 2021; López-Moreno et al. 2013; J. Wu et al. 2018). Specifically, regulating rivers through dams reduces hydrological sensitivity to changes in precipitation over shorter time scales and leads to delayed responses, which coincides well with subbasins III and V in autumn.
2) Dynamics of drought propagation thresholds and preliminary attribution analysis
It can be seen from Table 5 that the dynamics of trigger thresholds vary little from meteorological drought to different levels of hydrological drought (moderate drought, severe drought, and extreme drought). Taking moderate drought as an example, RF importance assessment (Fig. 9 and Table 6) and correlation analysis (Fig. 10) between the trigger thresholds of moderate drought and driving factors are explored.
Importance score (%IncMSE) of driving factors (P, T, PET, and SM) for (a) spring, (b) autumn, and (c) winter on trigger threshold of moderate hydrological drought in the headstreams of TRB (subbasins I–IV). All factors were processed by a moving window for 25 years.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
The correlation coefficients between propagation threshold and driving factors (P, T, PET, and SM) for each month of spring, autumn, and winter in the headstreams of TRB (subbasins I–IV). All factors were processed by a moving window for 25 years. The circles represent the magnitude of the correlation coefficients.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
The MK test statistics of drought propagation threshold in the headstreams of TRB. The MK test statistics are marked with a single asterisk (*) to inform that the detected trend is significant at the 5% level of significance.
Importance scores by IncNodePurity for propagation threshold. The IncNodePurity value is marked in bold to inform that the corresponding driver is the most important.
The drought propagation threshold shows an increasing trend in autumn and winter, indicating that the drought resilience of the basin is weakened for subbasins I–II. Moreover, correlation analysis and RF between each driving factor and propagation threshold reveal that PET and T show higher importance score and significant correlations, indicating that these factors have high sensitivity to the drought propagation threshold, especially in subbasin II. In addition, the impact of SM on subbasin I cannot be ignored (Figs. 9 and 10 and Table 6). However, unlike subbasins I–II, the season with weak resilience in subbasins III–IV mainly occurs in spring. The P and PET are the most sensitive to drought trigger threshold in subbasins III–IV, respectively, with a correlation of around 0.5 in spring. Because meteorological drought can affect hydrological drought through soil water infiltration processes, when a meteorological drought occurs, it will cause a continuous deficit of soil water, with potential amplification of soil water driven by the increase in PET (Fig. 8). At the same time, insufficient precipitation will deplete soil water, further exacerbating drought conditions and reducing the resistance of hydrological system, resulting in higher propagation threshold (the lower the SPI level) described above.
In addition to the aforementioned controls on drought propagation thresholds, the role of diverse human interventions should be highlighted. Table 4 shows that river damming can delay the drought propagation from the atmosphere into the hydrological system in subbasin III, and the corresponding drought thresholds show a decreasing trend in Table 5.
5. Discussions
a. Verifying the reliability of the seasonal drought propagation threshold framework
The proposed drought propagation threshold framework is verified to accommodate the existence of complex hydrometeorological relationship by comparing the estimated SRI conditional density distributions and corresponding paired SRI–SPI observations for different SPI values over the period 1960–2013. Prior to this, the fit of the Clayton copula function should be further verified. Figure 11a shows the results corresponding to the SED and P–P plot during January (I), September (II), December (III), and April (IV), which is selected randomly from the corresponding regions. As observed, based on the evaluation indexes of P–P plot and SED, the Clayton copula in this study is reliable for studying the seasonal drought propagation threshold.
Randomly selected January (I), September (II), December (III), and April (IV) in the headstreams of TRB. Goodness of fit tests: (a) P–P plots of empirical and theoretical distribution of Clayton copula and SED value. (b) Conditional probability of occurrence of moderate, severe, and extreme hydrological drought under the condition of various meteorological droughts. (c) Conditional density distributions of SRI given the diverse SPI value.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
The simulation of SRI density distributions conditioned given a specific SPI value are computed using Eq. (6). Figure 11c shows the results corresponding to the scatterplots of SPI and SRI. Each panel indicates where a pair of SRI–SPI observations (black dots) lie in the conditional PDF of SRI estimates given a particular SPI value (colored portion). The conditional PDFs are scaled between 0 and 1 for visualization purposes. The results show that most of the pairwise SRI–SPI observations fall well within the high-density region.
The conditional probability of occurrence of hydrological drought with different levels under the diverse SPI value is shown in Fig. 11b. As can be seen from the figure, the probabilities of occurrence of moderate, severe, and extreme hydrological drought all decrease under the same meteorological drought conditions and curve up with increasing meteorological drought severity, eventually converging to 1. The drought propagation threshold was set at 0.7, which is due to factors such as glacier melting and reservoir construction that interfere with the response of the watershed hydrological system to meteorological drought in headstreams of TRB. In conclusion, drought propagation threshold framework developed in this study is reliable for investigating hydrologic drought responses to various meteorological drought severity.
b. Possible causes for the weak correlation between meteorological and hydrological drought in summer
In arid and semiarid areas, the response of river runoff to climate and glacier change is complex. They are caused not only by the decrease of precipitation, but also by other factors affecting the water balance of these basins (Z. Li et al. 2017; Lan et al. 2020). Climate change is likely to significantly impact river basins, but the specific impacts would vary according to the hydrological characteristics of the basins. The precipitation–runoff relationship partly reflects the correlation coefficient between meteorological and hydrological drought (Gu et al. 2020). In contrast to most regions, which show negative and low correlation between runoff and precipitation, subbasin II exhibits positive and higher correlation in summer for the past five decades (Fig. 12). As shown in Fig. 3, the correlation between meteorological drought and hydrological droughts in almost all subbasins are not high in summer. The difference between the source region was mainly caused by the different recharge sources of runoff in these basins. The runoff in subbasin II is derived predominantly from a mixture of glacier and snow meltwater and precipitation, while runoff in other subbasins is derived from glaciers and snow meltwater (Chen et al. 2019; Xu et al. 2010). The proportion of runoff resulting from glacier melt is mainly determined by the relationship between the runoff and the mountain’s precipitation or temperature (Z. Li et al. 2017). Y. Wu et al. (2018) found that snowmelt reduced hydrological drought and disturbed the drought propagation in snow regulated area. As shown in Fig. 12, the correlations between temperature and runoff for summer are 0.46, 0.72, 0.39, and 0.38 over the recent half-century in subbasins I, III, IV, and V, respectively. Higher temperatures accelerate the movement and melting of glaciers and snowpack in summer, but have a negative impact on runoff due to the intensification of PET in summer, which subsequently leads to a decrease in water content in the catchment.
The correlation matrix between driving factors (P, T, PET, SM, and runoff R) for summer in the headstreams of TRB. (a)–(e) Subbasins I–V, respectively.
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
As a result, increased runoff in most sub basins is not evident during the summer months (Fig. 13). Runoff in Central Asia heavily relies on glacier melting. It has been observed that glacier retreat will accelerate with global warming (Chen et al. 2016; Z. Li et al. 2017). In the Tianshan Mountains (including the I and II subbasins), the area and mass of glaciers have decreased in the last half century (Farinotti et al. 2015). Interestingly, this trend accelerated in the 1990s, but has remained stable or even declined slightly since 2000. Chen et al. (2016) found this is consistent with the change of runoff in subbasins I–II (Fig. 13). Then, Xiang et al. (2020) found that the area of glacial coverage in the Yarkant River basin (subbasin III) initially decreased and then increased, which is consistent with the trend of runoff for subbasin III in Fig. 13. However, glaciers have been stable and even gaining in the Hotan River basin (the IV and V subbasins) over the past 50 years (Bolch et al. 2012; Kapnick et al. 2014). In addition, the runoff of Hotan River (the IV and V subbasins) in summer responded significantly to the change of 0°C isotherm height (Qin et al. 2019). The above analysis during summer explains why low correlations are found between meteorological drought and hydrological drought in the headstreams of TRB.
The trend of runoff (R) for summer in the headstreams of TRB (subbasins I–IV).
Citation: Journal of Hydrometeorology 23, 9; 10.1175/JHM-D-21-0250.1
6. Conclusions
To date, little is known about the dynamics of seasonal propagation characteristics (i.e., propagation time and threshold) and the driving mechanisms that determine the propagation from meteorological to hydrological drought. To this end, the time lag in drought propagation from meteorological to hydrological drought was investigated. Then, the drought propagation threshold (characterized by step abrupt effect) was estimated for moderate, severe, and extreme hydrological droughts, where SPI and SRI dependence patterns were modeled by a Bayesian network-based seasonal drought propagation threshold framework. Paired observations (SPI–SRI) were compared with the simulations to validate the proposed model framework. Finally, dynamics of drought propagation times and propagation thresholds over the past half-century were detected, which were followed by a preliminary attribution analysis at various seasons.
A case study was conducted on the largest inland river in China (viz., the headstreams of the Tarim River basin). The results show that there is a time lag of 2–21 months in the response of hydrological drought to meteorological drought. The noticeable spatial variability was found to be determined by the contributions of precipitation, glaciers, and snowmelt to runoff in different subbasins, which largely represents the interrelation between meteorological drought and hydrological drought. In watersheds dominated by glacial snowmelt runoff (i.e., subbasins I, III, IV, and V), relatively long propagation time (more than 9 months) and smaller propagation threshold (from −3.18 to −1.19) can be expected, suggesting that these subbasins are more drought resistant. Apart from the spatial variability, drought propagation time also presents a clear seasonality, with short-term and long-term hydrological response being observed in spring and autumn (or winter). This is a consequence of the storage of autumn and winter precipitation as snow, and its subsequent contribution (in spring) to streamflow during the snowmelt period.
Additionally, in a changing environment, hydrological drought in most seasons tends to respond to meteorological drought at increasingly longer time scales for subbasins I–II, while decreasing propagation time is seen in subbasins III–V. The drought threshold shows an increasing trend in autumn and winter for subbasins I–II, indicating a weakening of the basin’s drought resilience. However, unlike subbasins I–II, the season with weak drought resistance in subbasins III–IV mainly occur in spring. Such changes are preliminarily attributed to variations in catchment climate factors and catchment characteristics over the last half-century, which have a significant control on the spatiotemporal variability in the process of drought propagation. The impacts of precipitation, temperature, potential evapotranspiration, and soil moisture on drought propagation dynamics exhibit regional and seasonal heterogeneity. In particular, higher temperatures accelerate the movement and melting of glaciers and snowpack in summer, leading to a low or even negative correlation between precipitation and runoff, which inhibits the propagation of meteorological to hydrological drought and thus prevents the calculation of drought propagation thresholds.
Moreover, the reservoir serves as a buffer to regulate the propagation from meteorological to hydrological drought. On one hand, it is suggested that potential evapotranspiration can be regulated through the rational allocation of land cover types and thus the rational allocation of water resources. On the other hand, water conservancy facilities such as reservoirs are built to enhance the drought resilience. These findings can provide guidance for drought management in water resource systems, improve our understanding of drought propagation mechanisms, and support the development of practical drought monitoring systems. Given the complexity of factors affecting drought propagation mechanism in different subbasins analyzed above, a combined drought index accounting for various factors such as of glacial melting as well as temperature and evapotranspiration should be used in future studies. In general, the time-delayed and step-abrupt effects of meteorological to hydrological drought propagation should be characterized by propagation time and propagation threshold, respectively, and their dynamics and the corresponding driving mechanisms in a changing environment can enable water resources management to improve the understanding of disaster prevention. The results of the study can help water use authorities to provide appropriate water allocation plans in advance. For example, water managers can reduce water resources consumption during the noncritical water demand periods and reserve sufficient water in critical water demand periods. It should also be noted that the framework of seasonal propagation thresholds based on the Bayesian network in this study can be transferred and used to evaluate seasonal regulation performance in other regions. However, the propagation characteristics with only a single meteorological to hydrological drought does not account for other types of drought (such as agriculture and vegetation drought, etc.). Therefore, a long chain on drought propagation is still needed to address these challenges.
Acknowledgments.
This research was jointly funded by the National Key R&D Program of China (Grant 2021YFC3000203), the Shaanxi University Science and Technology Association Youth Talent Promotion Project (20190413), and the Strategic Priority Research Program of the Chinese Academy of Sciences (Grant XDA28060100). Additionally, authors would like to extend sincere appreciation to the editor and anonymous reviewers for their constructive comments, which help to improve the quality of the manuscript substantially. The authors declare that they have no conflicts of interest to this work. We declare that we do not have any commercial or associative interest that represents a conflict of interest in connection with the work submitted.
Data availability statement.
Hydrological data (monthly runoff) are available from by the Tarim River Basin Management Bureau. Meteorological data (daily precipitation, air temperature, vapor pressure, sunshine hours, wind speeds, maximum and minimum air temperature data) are available from the National Climate Information Center of the China Meteorological Administration (https://data.cma.cn/). The Global Land Data Assimilation System (GLDAS) is available via the Goddard Earth Sciences Data and Information Services Center website (http://disc.sci.gsfc.nasa.gov/hydrology/data-holdings).
REFERENCES
Aas, K., C. Czado, A. Frigessi, and H. Bakken, 2009: Pair-copula constructions of multiple dependence. Insur. Math. Econ., 44, 182–198, https://doi.org/10.1016/j.insmatheco.2007.02.001.
AghaKouchak, A., and N. Nakhjiri, 2012: A near real-time satellite-based global drought climate data record. Environ. Res. Lett., 7, 044037, https://doi.org/10.1088/1748-9326/7/4/044037.
Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop evapotranspiration: Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56, 300 pp., www.fao.org/docrep/X0490E/X0490E00.htm.
Apurv, T., M. Sivapalan, and X. Cai, 2017: Understanding the role of climate characteristics in drought propagation. Water Resour. Res., 53, 9304–9329, https://doi.org/10.1002/2017WR021445.
Aryal, Y., and J. Zhu, 2020: Effect of watershed disturbance on seasonal hydrological drought: An improved double mass curve (IDMC) technique. J. Hydrol., 585, 124746, https://doi.org/10.1016/j.jhydrol.2020.124746.
Bai, Y., H. Xu, and H. Ling, 2014: Drought-flood variation and its correlation with runoff in three headstreams of Tarim River, Xinjiang, China. Environ. Earth Sci., 71, 1297–1309, https://doi.org/10.1007/s12665-013-2534-5.
Bolch, T., and Coauthors, 2012: The state and fate of Himalayan glaciers. Science, 336, 310–314, https://doi.org/10.1126/science.1215828.
Breiman, L., 2001: Random forests. Mach. Learn., 45, 5–32, https://doi.org/10.1023/A:1010933404324.
Bushra, N., J. C. Trepanier, and R. V. Rohli, 2019: Joint probability risk modelling of storm surge and cyclone wind along the coast of Bay of Bengal using a statistical copula. Int. J. Climatol., 39, 4206–4217, https://doi.org/10.1002/joc.6068.
Chen, H., Y. Chen, W. Li, and Z. Li, 2019: Quantifying the contributions of snow/glacier meltwater to river runoff in the Tianshan Mountains, central Asia. Global Planet. Change, 174, 47–57, https://doi.org/10.1016/j.gloplacha.2019.01.002.
Chen, X. W., and M. Liu, 2005: Prediction of protein-protein interactions using random decision forest framework. Bioinformatics, 21, 4394–4400, https://doi.org/10.1093/bioinformatics/bti721.
Chen, Y., W. Li, C. Xu, and X. Hao, 2007: Effects of climate change on water resources in Tarim River Basin, Northwest China. J. Environ. Sci., 19, 488–493, https://doi.org/10.1016/S1001-0742(07)60082-5.
Chen, Y., C. Xu, X. Hao, W. Li, Y. Chen, C. Zhu, and Z. Ye, 2009: Fifty-year climate change and its effect on annual runoff in the Tarim River Basin, China. Quat. Int., 208, 53–61, https://doi.org/10.1016/j.quaint.2008.11.011.
Chen, Y., Z. Ye, and Y. Shen, 2011: Desiccation of the Tarim River, Xinjiang, China, and mitigation strategy. Quat. Int., 244, 264–271, https://doi.org/10.1016/j.quaint.2011.01.039.
Chen, Y., K. Yang, J. Qin, L. Zhao, W. Tang, and M. Han, 2013: Evaluation of AMSR-E retrievals and GLDAS simulations against observations of a soil moisture network on the central Tibetan Plateau. J. Geophys. Res. Atmos., 118, 4466–4475, https://doi.org/10.1002/jgrd.50301.
Chen, Y., W. Li, H. Deng, G. Fang, and Z. Li, 2016: Changes in central Asia’s water tower: Past, present and future. Sci. Rep., 6, 39364, https://doi.org/10.1038/srep39364.
Cutler, D. R., T. C. Edwards Jr., K. H. Beard, A. Cutler, K. T. Hess, J. Gibson, and J. J. Lawler, 2007: Random forests for classification in ecology. Ecology, 88, 2783–2792, https://doi.org/10.1890/07-0539.1.
Dehghani, M., B. Saghafian, and M. Zargar, 2019: Probabilistic hydrological drought index forecasting based on meteorological drought index using Archimedean copulas. Hydrol. Res., 50, 1230–1250, https://doi.org/10.2166/nh.2019.051.
Ding, Y., J. Xu, X. Wang, H. Cai, Z. Zhou, Y. Sun, and H. Shi, 2021: Propagation of meteorological to hydrological drought for different climate regions in China. J. Environ. Manage., 283, 111980, https://doi.org/10.1016/j.jenvman.2021.111980.
Eltahir, E. A. B., and P. J. F. Yeh, 1999: On the asymmetric response of aquifer water level to floods and droughts in Illinois. Water Resour. Res., 35, 1199–1217, https://doi.org/10.1029/1998WR900071.
Etienne, E., N. Devineni, R. Khanbilvardi, and U. Lall, 2016: Development of a demand sensitive drought index and its application for agriculture over the conterminous United States. J. Hydrol., 534, 219–229, https://doi.org/10.1016/j.jhydrol.2015.12.060.
Fang, W., S. Huang, Q. Huang, G. Huang, H. Wang, G. Leng, and L. Wang, 2020: Identifying drought propagation by simultaneously considering linear and nonlinear dependence in the Wei River basin of the Loess Plateau, China. J. Hydrol., 591, 125287, https://doi.org/10.1016/j.jhydrol.2020.125287.
Farinotti, D., L. Longuevergne, G. Moholdt, D. Duethmann, T. Mölg, T. Bolch, S. Vorogushyn, and A. Güntner, 2015: Substantial glacier mass loss in the Tien Shan over the past 50 years. Nat. Geosci., 8, 716–722, https://doi.org/10.1038/ngeo2513.
García-Ruiz, J. M., and Coauthors, 2008: Flood generation and sediment transport in experimental catchments affected by land use changes in the central Pyrenees. J. Hydrol., 356, 245–260, https://doi.org/10.1016/j.jhydrol.2008.04.013.
Gevaert, A. I., T. I. E. Veldkamp, and P. J. Ward, 2018: The effect of climate type on timescales of drought propagation in an ensemble of global hydrological models. Hydrol. Earth Syst. Sci., 22, 4649–4665, https://doi.org/10.5194/hess-22-4649-2018.
Gu, L., J. Chen, J. Yin, C.-Y. Xu, and H. Chen, 2020: Drought hazard transferability from meteorological to hydrological propagation. J. Hydrol., 585, 124761, https://doi.org/10.1016/j.jhydrol.2020.124761.
Guo, Y., S. Huang, Q. Huang, G. Leng, W. Fang, L. Wang, and H. Wang, 2020: Propagation thresholds of meteorological drought for triggering hydrological drought at various levels. Sci. Total Environ., 712, 136502, https://doi.org/10.1016/j.scitotenv.2020.136502.
Guo, Y., Q. Huang, S. Huang, G. Leng, X. Zheng, W. Fang, M. Deng, and S. Song, 2021: Elucidating the effects of mega reservoir on watershed drought tolerance based on a drought propagation analytical method. J. Hydrol., 598, 125738, https://doi.org/10.1016/j.jhydrol.2020.125738.
Hao, Z., F. Hao, V. P. Singh, and W. Ouyang, 2017: Quantitative risk assessment of the effects of drought on extreme temperature in eastern China. J. Geophys. Res. Atmos., 122, 9050–9059, https://doi.org/10.1002/2017JD027030.
Haslinger, K., D. Koffler, W. Schöner, and G. Laaha, 2014: Exploring the link between meteorological drought and streamflow: Effects of climate-catchment interaction. Water Resour. Res., 50, 2468–2487, https://doi.org/10.1002/2013WR015051.
Hellwig, J., I. E. M. de Graaf, M. Weiler, and K. Stahl, 2020: Large-scale assessment of delayed groundwater responses to drought. Water Resour. Res., 56, e2019WR025441, https://doi.org/10.1029/2019WR025441.
Huang, S., P. Li, Q. Huang, G. Leng, B. Hou, and L. Ma, 2017: The propagation from meteorological to hydrological drought and its potential influence factors. J. Hydrol., 547, 184–195, https://doi.org/10.1016/j.jhydrol.2017.01.041.
Huang, S., L. Wang, H. Wang, Q. Huang, G. Leng, W. Fang, and Y. Zhang, 2019: Spatio-temporal characteristics of drought structure across China using an integrated drought index. Agric. Water Manage., 218, 182–192, https://doi.org/10.1016/j.agwat.2019.03.053.
Huo-Po, C., and S. Jian-Qi, 2015: Drought response to air temperature change over China on the centennial scale. Atmos. Ocean. Sci. Lett., 8, 113–119, https://doi.org/10.1080/16742834.2015.11447247.
IPCC, 2007: Climatic Change 2007: Impacts, Adaptation and Vulnerability. Cambridge University Press, 976 pp., https://www.ipcc.ch/pdf/assessment-report/ar4/wg2/ar4_wg2_full_report.pdf.
Jehanzaib, M., M. N. Sattar, J.-H. Lee, and T.-W. Kim, 2020: Investigating effect of climate change on drought propagation from meteorological to hydrological drought using multi-model ensemble projections. Stochastic Environ. Res. Risk Assess., 34, 7–21, https://doi.org/10.1007/s00477-019-01760-5.
Jiang, J., B. Ming, Q. Huang, J. Chang, P. Liu, W. Zhang, and K. Ren, 2021: Hybrid generation of renewables increases the energy system’s robustness in a changing climate. J. Cleaner Prod., 324, 129205, https://doi.org/10.1016/j.jclepro.2021.129205.
Kapnick, S. B., T. L. Delworth, M. Ashfaq, S. Malyshev, and P. C. D. Milly, 2014: Snowfall less sensitive to warming in Karakoram than in Himalayas due to a unique seasonal cycle. Nat. Geosci., 7, 834–840, https://doi.org/10.1038/ngeo2269.
Kendall, M. G., 1957: Rank correlation methods. Biometrika, 44, 298, https://doi.org/10.2307/2333282.
Konapala, G., and A. Mishra, 2020: Quantifying climate and catchment control on hydrological drought in the continental United States. Water Resour. Res., 56, e2018WR024620, https://doi.org/10.1029/2018WR024620.
Lan, T., and H. Zhang, C-y Xu, V. P. Singh, and K. Lin, 2020: Detection and attribution of abrupt shift in minor periods in human-impacted streamflow. J. Hydrol., 584, 124637, https://doi.org/10.1016/j.jhydrol.2020.124637.
Li, Y., N. Yao, S. Sahin, and W. M. Appels, 2017: Spatiotemporal variability of four precipitation-based drought indices in Xinjiang, China. Theor. Appl. Climatol., 129, 1017–1034, https://doi.org/10.1007/s00704-016-1827-5.
Li, Y.-L., Y. Gong, and C. Huang, 2021: Construction of combined drought index based on bivariate joint distribution. Alexandria. Eng. J., 60, 2825–2833, https://doi.org/10.1016/j.aej.2021.01.006.
Li, Z., Y. Chen, G. Fang, and Y. Li, 2017: Multivariate assessment and attribution of droughts in Central Asia. Sci. Rep., 7, 1316, https://doi.org/10.1038/s41598-017-01473-1.
Li, Z.-H., X. Shi, Q. Tang, Y. Zhang, H. Gao, X. Pan, S. J. Déry, and P. Zhou, 2020: Partitioning the contributions of glacier melt and precipitation to the 1971–2010 runoff increases in a headwater basin of the Tarim River. J. Hydrol., 583, 124579, https://doi.org/10.1016/j.jhydrol.2020.124579.
Li, Z.-Y., and Coauthors, 2021: Clarifying the propagation dynamics from meteorological to hydrological drought induced by climate change and direct human activities. J. Hydrometeor., 22, 2359–2378, https://doi.org/10.1175/JHM-D-21-0033.1.
Liu, P., Z. Fang, C. Lv, and A. Ruan, 2020: China’s agricultural water-use efficiency and its influencing factors under the constraint of pollution emission. Int. J. Des. Nat. Ecodyn., 15, 579–585, https://doi.org/10.18280/ijdne.150416.
Liu, Y., Y. Zhu, L. Ren, V. P. Singh, B. Yong, S. Jiang, F. Yuan, and X. Yang, 2019: Understanding the spatiotemporal links between meteorological and hydrological droughts from a three-dimensional perspective. J. Geophys. Res. Atmos., 124, 3090–3109, https://doi.org/10.1029/2018JD028947.