Transition from Slow Drought to Flash Drought Under Climate Change in Northern Xinjiang, Northwest China

Abstract

Flash drought (FD) is an extreme climate event that intensifies within days and exerts severe socio-environmental impacts. Its onset and evolution remain difficult to predict. Here, we quantify the spatio-temporal dynamics of FD across northern Xinjiang from 1961 to 2023 and identify the dominant driving factors. We apply linear trend detection, wavelet analysis, change-point detection, random forest (RF) modeling, and Pearson correlation. Results show that winter is becoming significantly wetter, whereas the annual signal and the other three seasons exhibit drying trends. After 1980, both FD frequency and FD duration increased; the longest single event lasted 40 days. Spatially, FD is concentrated in the Ili River Valley and the Altay region; the Akdala station recorded the highest count (nine events). Duration, rather than frequency, peaks on the northern slope of the Tianshan Mountains, where the maximum length reaches 40 days. RF importance ranks the Pacific Decadal Oscillation (PDO) as the leading driver (20.9%), followed by air temperature (17.8%); the sunspot index contributes only 6.1%.

1. Introduction

Drought is a natural hazard that exerts severe pressure on agriculture, sustainable development, and regional economies [1]. Traditionally, drought events have been viewed as long-term phenomena lasting several months or years. Recently, however, short-lived flash drought (FD) has become more frequent [2]. These events develop abruptly through anomalously high temperatures and below-average precipitation, depleting soil moisture within days and causing large crop losses [1,2]. The term “FD” describes droughts that intensify rapidly over a brief period [2]. Scientific interest emerged in 2002 [3] and expanded markedly after 2013 [4].

FD research now receives widespread attention in atmospheric science [5,6,7,8,9]. Seasonal-scale droughts are intensifying faster than before [10], and 74% of the regions assessed by the IPCC report rising FD frequency [11,12]. Hot spots include Brazil, India, the central United States, and northeastern China [13]. Qing et al. [14] identify humid-to-semi-humid areas in East Asia, southeast Asia, and eastern North America as the most prone, whereas Sreeparvathy et al. [15] locate FD primarily in arid and semi-arid regions. Both settings challenge climate change forecasting [16]. Prolonged soil moisture depletion [17,18,19,20], reduced precipitation [21], extreme heat, and strong wind [22] collectively enhance evaporation and evapotranspiration [23], triggering more frequent FD events. Without timely warning, these events amplify agricultural, water, and economic losses [24]. Researchers have found that the duration of flash drought in the northwest region is 10-20 days longer than in other regions, and the frequency of sudden drought in northern Xinjiang can reach four times/10a [25]. In addition, flash drought has been identified in the United States [26], western Russia [27], and Australia [28].

Xinjiang faces rising drought risk [29,30], which will cause larger economic losses than any other natural hazard in the region [31,32,33,34,35]. Between 1960 and 2007, the frequency of extreme dry and wet months in northern Xinjiang increased steadily [36], and the 2008 drought was the most severe in 30 years [37]. Long-lasting spring, summer, and autumn droughts regularly affect northern Xinjiang and trigger rodent and insect outbreaks on grasslands and in the Tianshan Mountains; only the 1974 drought was more damaging. Although drought eased slightly after the 1980s, extreme events remain frequent and continue to cause heavy social and economic losses. Using the Standardized Precipitation Evapotranspiration Index (SPEI), Zhang et al. [38] detected a weak downward trend in spring, summer, and autumn SPEI values across northern Xinjiang. While numerous studies have documented the occurrence of flash droughts in arid/semi-arid regions, there is a lack of systematic research exploring how combined climatic conditions modulate flash drought characteristics (duration, intensity, spatial extent), resulting in an incomplete understanding of their formation pathways under a changing climate. To compensate for this research gap, this study elaborates on the relationships between climate change, atmospheric circulation indices, and flash droughts through an in-depth analysis.

This paper clarifies the FD concept, highlights what is still poorly understood, and identifies deficiencies in current FD indices. Northern Xinjiang is selected as the study area to address the two following questions: (i) What are the spatio-temporal patterns of FD in northern Xinjiang? (ii) Which factors trigger FD outbreaks in the region?

2. Research Area and Data

2.1. Study Area

Northern Xinjiang lies north of the Tianshan Mountains and is bounded by the Altai Mountains in the north [39] (Figure 1). The region forms a trumpet-shaped basin that opens to the west; elevation decreases from the surrounding mountain rims toward the central plains that include the Gurbantunggut Desert [40]. Landforms are classified as hills, plains, mountains, the Gobi, and desert. More than 300 rivers drain the area, accounting for ~60% of all rivers in Xinjiang and sustaining a distinctive oasis irrigation agriculture system [41,42].

The climate is continental, with a large diurnal temperature range, scarce precipitation that falls mainly in summer, and high inter-annual variability. Annual totals drop below 150 mm in the east and can be <100 mm in extreme years. Sunshine duration averages 2800 h yr⁻¹. The region is prone to drought, frost, gale, local flood, and rainstorm disasters [43]. Under recent climate change, northern Xinjiang exhibits a significant warming and moistening trend [44].

2.2. Data

2.2.1. Meteorological Data

Daily air temperature and precipitation records for 1961–2023 were obtained from 43 meteorological stations across northern Xinjiang. The dataset, provided by the Xinjiang Meteorological Administration, passed standard quality control procedures before release.

2.2.2. Atmospheric Circulation Index Data

Monthly indices of the North Atlantic Oscillation (NAO), Arctic Oscillation (AO), Pacific Decadal Oscillation (PDO), ENSO (Niño3), and the sunspot number were retrieved from NOAA Physical Sciences Laboratory (https://psl.noaa.gov/gcos_wgsp/Timeseries/SUNSPOT/, accessed on 31 December 2023). The NAO is defined as the normalized sea level pressure difference between the Azores and Iceland. The AO describes the leading mode of non-seasonal sea level pressure variability north of 20° N. The PDO reflects long-term sea surface temperature anomalies in the North Pacific. The ENSO denotes coupled ocean–atmosphere variations in the tropical Pacific; the Niño3 region (5° S–5° N, 150°–90° W) is routinely used to monitor its strength.

3. Method

3.1. Rules for Fast-Onset Drought (FOD) Identification

Following the flash drought definition in [45,46], we detect FOD on the pentad (5-day) scale using the Standardized Precipitation Evapotranspiration Index (SPEI). An event is flagged when the following occurs [47]:

(1)

The SPEI drops from ≥0 to <−1 within four consecutive pentads, signaling a rapid shift to at least moderate drought;

(2)

The dry spell persists for ≥3 pentads, ensuring agronomic impact;

(3)

The SPEI recovers to ≥−1, marking event termination.

3.2. Mann–Kendall Abrupt Change Detection

The Mann–Kendall (MK) abrupt change test [48] is applied to the SPEI series to identify the years when drought conditions shift suddenly.

3.3. Random Forest Model

The random forest (RF) model accommodates mixed data types and ranks predictors by permuting each variable across all decision trees and recording the resulting mean square error increase [49]. A total of 549 samples were split into 400 for the training set and 149 for the testing set, resulting in 400 trees. Variable importance was ranked by matlab R2023b. The resulting importance score indicates the relative contribution of each climate index to FD variability.

4. Results

4.1. The Spatio-Temporal Variation Characteristics of Drought Intensity

Figure 2 presents the annual and seasonal SPEI trends over 1961–2023. Annual spring and summer indices decline at −0.09, −0.103, and −0.124 yr⁻¹, respectively, whereas winter shows a significant increase of +0.213 yr⁻¹; autumn trends are negligible (−0.04 yr⁻¹). Only the annual and autumn slopes are statistically non-significant (α = 0.05). Extreme minima occur in 1997 (annual −1.91), spring 2020 (−1.96), summer 2023 (−1.83), autumn 1997 (−2.14), and winter 1967 (−1.80), indicating the most severe droughts of the 63-year record. In winter, a pronounced peak can be seen around 2009, reaching approximately 2.0. The main reason is that around 2009, temperatures rose (−8.9 °C) and precipitation decreased.

The winter moistening reflects a marked precipitation increase during the last five decades [50], accompanied by steadily rising winter air temperatures [51] that account for 57% of the regional annual mean temperature rise [52], thereby elevating the overall annual temperature in northern Xinjiang.

Complex terrain governs SPEI distribution in northern Xinjiang. The annual mean SPEI is high in the northeast and southeast but low on the northern Tianshan slope at Urumqi, Jinghe, and Hutubi (Figure 3); the extreme values are −0.0043 in Qiemo and +0.0084 in Wenquan. Seasonal patterns resemble the annual field (Figure 3). In spring, high values occur over the Altai and Ili River Valley, whereas the lowest value (−0.0045) is recorded at Xiaoquzi and the highest (0.0067) at Huoerguosi. The terrain of the western opening of the Ili River Valley has led to the infiltration of Atlantic water vapor. In recent years, the rising air temperature and frequent occurrence of extreme high temperatures have intensified evaporation. At the same time, after the snow melts in spring, the mountainous areas have fast water collection, while the plains have poor drainage, making it difficult to retain water. These reasons have led to frequent occurrences of flash droughts in the Ili River Valley. Summer warming intensifies drought on the northern Tianshan slope; the SPEI ranges from −0.0025 at Caijiahu to 0.0094 at Wenquan, and additional rainfall alleviates drought in the Ili River Valley. Autumn SPEI values vary from −0.0044 in Tekesi to 0.0091 in Fuyun; severe conditions are confined to the Ili River Valley and the northern Tianshan slope, while drought around Urumqi, the southern Altai slope, and Tacheng weakens. Winter SPEI values range from −0.0016 at Changji to 0.015 at Dabancheng, showing no clear spatial structure. Thus, topography, geography, and climate jointly create pronounced spatial heterogeneity of the SPEI across northern Xinjiang (Figure 4).

4.2. SPEI Abrupt Change Analysis

The Mann–Kendall test applied to the annual and seasonal SPEI series is shown in Figure 5. On the annual scale, the UF curve fluctuates from 1961 to 2005 and declines thereafter; no abrupt change is detected. Spring, summer, and autumn exhibit similar behavior—fluctuation followed by a gentle decrease—without any significant jump. In winter, the UF curve rises with fluctuations and crosses the critical line in 1984, indicating an abrupt shift toward wetter conditions. Since the 1980s, autumn and winter have become progressively more humid, whereas spring and summer have drifted toward drier conditions.

4.3. Morlet Wavelet Analysis of SPEI

To investigate the change cycle of the SPEI in northern Xinjiang, we conducted Morlet wavelet analysis on the SPEI at different time scales over the past 63 years. The resulting wavelet real-part and variance maps for the different time scales are presented in Figure 6; higher values indicate more humid conditions, while lower values indicate drier conditions. The periodic characteristics are summarized in

Table 1. The SPEI change characteristics across various time series scales.

Time Scale	First Main Cycle (Year)	Number of Oscillations (First Main Cycle)	Second Main Cycle (Years)	Number of Oscillations (Second Main Cycle)	Third Main Cycle (Years)	Number of Oscillations (Third Main Cycle)
Annual	30	4	12	8	5	19
Summer	30	4	18	6	4	18
Autumn	10	9	20	5	40	2
Winter	22	5	10	10	45	3
Spring	10	9	27	4	50	3

. As observed in Figure 6 and Table 1, the wavelet coefficients exhibit a three-major-period variation pattern across seasons. In contour plots, positive SPEI values signify moisture, whereas negative SPEI values indicate dryness. This alternating oscillation between drought and wetness is consistent across spring, summer, autumn, winter, and the annual average scale, suggesting a global phenomenon. Furthermore, the different time scales also exhibit multiple small-period oscillations, albeit with higher turbulence. By examining the wavelet variance plot (Figure 6), we observed two peak periods in spring, with the highest and largest peak occurring in the 10th year, followed by the 30th year. Similarly, the highest and largest peak in summer occurs in the 30th year, while the largest peak in winter appears in the 22nd year. The annual SPEI exhibits the strongest variation in a period of 8 years in northern Xinjiang. Based on these findings, it can be concluded that the annual main cycle of the SPEI in northern Xinjiang is primarily influenced by the summer main cycle. Additionally, predictions about future drought conditions can be made using the results from the wavelet real-part contour. Consequently, it is likely that northern Xinjiang will experience shifts from drought to humid conditions at different time scales.

4.4. Identification and Feature Analysis of Flash Drought

Figure 7 presents the inter-annual variation in flash drought (FD) frequency and mean duration in northern Xinjiang. Frequency effectively indicates the regional coverage of FD, and higher frequencies are associated with more agricultural damage. The figure shows a steady increase in the number of FD events, signaling an expanding FD risk across the region. Between 1961 and 1980, only two FD events were recorded, but the frequency rose sharply after 1980, with notable peaks of 22 events in 2011 and 20 events in 2014; the highest count of 29 events occurred in 1967.

Mean duration serves as an effective indicator of how long FD disasters persist in a given area; longer durations imply a higher probability of local drought stress and more severe losses in agriculture and other sectors. The data in Figure 7 reveal a consistent upward trend in the average duration of FD events over the past 63 years, indicating an increasing impact of sudden drought on regional agriculture and related sectors. In 1981, the average duration reached its maximum of 40 days, representing the longest FD episode recorded during the 63-year period. In summary, the time series of both the occurrence and average duration of FD in northern Xinjiang exhibit coherent upward trends over the past 63 years, with 1981 marking the year of the most extensive coverage and longest duration of flash drought in almost six decades.

Northern Xinjiang exhibits distinct spatial distributions of flash drought that reflect its complex and diverse terrain. To reveal regional FD characteristics, we analyzed the frequency and duration of events at each station; the spatial patterns are shown in Figure 8. Overall, higher FD frequencies are observed in the northeast and southeast, whereas lower values occur on the northern slope of the Tianshan Mountains, including Urumqi, Jinghe, and Hutubi. Topography exerts clear control: the Tianshan and Altay Mountains modulate the occurrence of FD, with stations such as Caichahu and Qitai recording no events, while Akedala experienced nine events during the past 63 years. The spatial pattern of FD duration closely resembles that of frequency, although some differences exist. Notably, fewer FD events are detected on the northern Tianshan slope, but the events that do occur tend to last longer, with durations ranging from 3 to 5.5 pentads. Consequently, the spatial distribution of FD duration exhibits significant variation across northern Xinjiang under the combined influences of topography, geography, climate, and other local factors.

4.5. Influencing Factors of Flash Drought

Although northern Xinjiang is geographically distant from the ocean, sea surface conditions can still induce noticeable weather changes across the region. In this study, we calculated FD frequency and examined its consistency with eight climate factors and atmospheric circulation indices, namely air temperature, precipitation, the SPEI, the PDO, the NAO, the AO, the sunspot index, and the ENSO. Pearson correlation coefficients between FD frequency and each index were all below 0.3, indicating no significant linear relationship. To further explore non-linear influences, we employed a random forest model that explains 81% of the variance in annual FD frequency. The simulation results identify the PDO as the dominant driver, accounting for 20.9% of the variability, followed by air temperature at 17.8%. By contrast, the sunspot index exerts the smallest influence, contributing only 6.1% to FD variability (Figure 9). The PDO (Pacific Decadal Oscillation) is the dominant factor (contributing 20.9%), and its impact mechanism is closely related to the long-term oscillation characteristics of air–sea coupling. The cold/warm phase indirectly regulates regional water vapor transport and energy balance by changing the intensity of the Walker circulation. For example, the PDO cold phase will enhance the La Niña phenomenon, leading to a decrease in precipitation in northern Xinjiang, consistent with the observed drought trend. Regional warming exacerbates surface evaporation, reduces soil moisture, and amplifies drought intensity.

5. Discussion

With global warming, the climate in northern Xinjiang has also shown a clear trend of change. As shown in Figure 10, from 1961 to 2023, the annual average temperature and precipitation in the northern Xinjiang region have shown an increasing trend, at 0.36 °C/10a and 8.4 mm/10a, respectively. Especially since 1980, the increasing trend has been significant, and the occurrence of high-temperature weather has accelerated the frequency of flash drought.

Climate change in northwest China is a complex, multi-scale phenomenon that is still poorly quantified. Rising temperature, increased evaporation, and soil moisture loss have collectively amplified drought risk across Xinjiang [53], yet long-term precipitation has also increased, producing a warmer and moister environment in northern Xinjiang.

Since 2000, the Tianshan Mountains and northern Xinjiang have experienced more frequent extreme drought events [54], although recent observations suggest a slight moderation in drought severity. These conflicting signals arise because most assessments rely solely on precipitation and air temperature records and ignore zero-precipitation days, soil moisture deficits, and antecedent wetness. Consequently, projections of future drought remain highly uncertain. Future work must integrate the full water cycle response to climate change—including atmospheric moisture transport, soil moisture dynamics, and evaporative demand—to improve confidence in dry/wet forecasts for the region.

6. Conclusions

Winter has become significantly wetter, whereas the annual mean and the other three seasons show a drying trend. The SPEI varies spatially: summer and autumn drought is more severe on the northern Tianshan slope and in the Ili River Valley than elsewhere.

An abrupt wetting shift occurred in winter in 1984. Wavelet analysis reveals a ~30 yr dominant cycle superimposed on shorter, region-specific periodicities.

After 1980, both the frequency and duration of flash drought (FD) increased; the longest events persisted for 40 days. FD is concentrated in the Ili River Valley and the Altay region, with Akdala recording the highest count (nine events). On the northern Tianshan slope, FD is less frequent but lasts up to 40 days.

Linear correlations between FD and individual climate indices are weak (r < 0.3). Random forest attribution nevertheless identifies the Pacific Decadal Oscillation as the primary driver (20.9%), followed by air temperature (17.8%); the sunspot index contributes only 6.1%.

This study analyzed mountain drought in northern Xinjiang, and future work should integrate multi-source data, optimize models with machine learning, expand applications like agricultural warnings, and strengthen data quality and international collaboration to enhance its academic and practical value.