Weather Regimes of Extreme Wind Speed Events in Xinjiang: A 10–30 Year Return Period Analysis

Abstract

Xinjiang is a critical wind energy region in China. This study characterizes extreme wind speed (EWS) events in Xinjiang by using ERA5 reanalysis (1979–2023) and station observations (2022–2023). Through k-means clustering and wind power density classification, four distinct regions and representative nodes were identified, aligned with the “Three Mountains and Two Basins” topography: Region #1 (eastern wind-rich corridor), Region #2 (Tarim Basin, west–east increasing wind power density), Region #3 (northern valleys), and Region #4 (mountainous areas with weakest wind power density). Peaks-over-threshold analysis revealed 10~30-year return levels varying regionally, with 10-year return level for Node #1 reaching Beaufort Scale 11 but only Scale 6 for Node #4. Since 2001, EWS occurrences increased, with Nodes #2–4 showing doubled 10-year event occurrences in 2012–2023. Events exhibit consistent afternoon peaks and spring dominance (except Node #2 with summer maxima). Such long-term trends and diurnal and seasonal preferences of EWS could be partly explained by diverging synoptic drivers: orographic effects and enhanced pressure gradients (Node #1 and #3) associated with Ural blocking and polar vortex shifts, both showing intensification trends; thermal lows in the Tarim Basin (Node #2) accounting for their summer prevalence; boundary-layer instability that leads to localized wind intensification (Node #4). The results suggest the necessity of region-specific forecasting strategies for wind energy resilience.

1. Introduction

Major economies worldwide have elevated climate change and carbon emissions as top policy priorities. Within this context, wind energy has emerged as a critical solution due to its zero greenhouse gas emissions, minimal ecological impact, and strong economic viability [1]. As a cornerstone of energy transition, wind power is projected to supply 22% of global electricity by 2030 [2], with its scalability directly impacting national energy security and climate governance efficacy [1,3]. Extreme wind speed (EWS) events, characterized by transient but high-intensity wind surges, pose severe risks to wind power infrastructure [4,5,6]. Turbines automatically enter cut-out mode when wind speeds exceed operational thresholds to prevent mechanical damage [5]. The core challenge for wind farm operations lies in balancing energy capture efficiency and EWS risk mitigation [7], compounded by climate variability-induced uncertainties [8,9]. Therefore, spatiotemporal patterns and predictive modeling of EWS have become focal points for both academia and policymakers.

While extensive research [10,11,12,13,14] has linked synoptic climatology and meteorological conditions to wind fields. EWS events typically originate from local/mesoscale windstorms, frontal system passages, and synoptic-scale deep low-pressure systems. The downward mixing of high-momentum air through turbulent vertical momentum transport in the boundary layer can also intensify near-surface winds [15,16,17]. Moreover, large orographic effects play a significant role [18]. However, research remains scarce on the specific relationships between more predictable recurrent atmospheric circulation patterns or weather regimes and EWS, with only a few studies limited to Europe [13], the Arctic [19], and Portugal [20], etc. Asian regions are markedly underrepresented.

Xinjiang Uygur Autonomous Region of China, with its unique “Three Mountains Encircling Two Basins” topography, hosts nine major wind resource zones—including the Alashankou, Dabancheng, and Hami Northern Gobi Wind corridors, Irtysh River Valley Wind Zone, Western Junggar Basin Wind Zone, Lop Nor Wind Zone, Tacheng Basin Wind Zone and Baili Wind Corridor [21,22]. Its theoretical wind energy potential reaches 957 GW, accounting for 37% of China’s total [1]. Recent advancements in low-wind-speed-optimized turbines with longer blades and higher capacity have expanded development beyond traditional zones [22]. However, significant challenges remain due to the pronounced seasonal variability and remarkable temporal fluctuations in wind resources [23,24,25,26]. Notably, Jiang et al. [27] highlighted that the frequency of EWS events tends to be higher in arid/semi-arid and coastal regions of China, including Xinjiang. Furthermore, Liu et al. [28] demonstrated that EWS events in these northern arid/semi-arid areas can further cause dust storms.

This study focuses on EWS events in Xinjiang, with two primary objectives. One is to quantify extreme wind speeds at different return periods, which is a critical parameter for wind turbine structural design [13,29,30] and reveal their regional differences. The second is to diagnose the associated synoptic-scale weather systems and large-scale circulation patterns, which would provide foundational insights for improving the predictability of EWS events in this region.

2. Data and Methods

2.1. Data

This study used hourly ERA5 reanalysis data [31] from ECMWF, covering 1979–2023. We utilized single-level variables including 10 m zonal wind (u10m) and meridional wind speed (v10m), sea level pressure (SLP), boundary layer height (BLH), as well as geopotential height at 500 hPa pressure level (GHT500) on 0.5° × 0.5° grids. All data are publicly available via the Copernicus Climate Data Store (ERA5 pressure-level hourly data: https://cds.climate.copernicus.eu/datasets/reanalysis-era5-pressure-levels?tab=overview (accessed on 16 August 2025); single-level hourly data: https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels?tab=overview (accessed on 16 August 2025)). Surface height above sea level in Xinjiang was obtained via downloading the earth topography five-minute grid (ETOPO5), which is a gridded data base of worldwide elevations derived from several sources at a resolution of 5 min of latitude and longitude [32].

Existing studies support the validity of ERA5 wind speed data [5,33], with Vega-Bayo et al. [33] demonstrating high correlation coefficients (~0.7) between ERA5 outputs and in situ wind farm measurements while Larsen and Mann [34] argued an underestimation of the extreme events by approximately 15%. To evaluate this relationship in Xinjiang, we compared 2022–2023 ERA5 hourly instantaneous winds (bilinearly interpolated to station grids) against maximum wind speed observations from 105 national meteorological stations, provided by Xinjiang Meteorological Administration (https://pan.baidu.com/s/1SCUcfTcaA6S-V9TV8GJOCg?pwd=3si4 (accessed on 16 August 2025)). The station average correlation is 0.65. We also recognized that ERA5 underestimates peak magnitudes due to its instantaneous sampling, thus we adopted a tail-dependent regression approach [35,36] to address this bias for extreme wind analysis. First, we selected concurrent ERA5-observation pairs exceeding the 95th percentile threshold. Then, we derived a systematic scaling factor (k = 0.59) from linear modeling within this extreme subset to correct ERA5 values prior to extreme wind speed analysis.

2.2. Methods

2.2.1. Derivation of Related Variables

Total wind speed field at 10 m was derived from zonal and meridional components:

$$\left|U_{10m}\right|=\sqrt{{\left(u_{10m}\right)}^2+{\left(v_{10m}\right)}^2}$$

(1)

We followed the IEC 61400-12 standard for wind resource assessment to calculate the wind power density (WPD) as

$$WPD={\displaystyle \frac{1}{2}}\rho {\left|U_{10m}\right|}^3$$

(2)

Here, because the annual averages and standard deviations of wind speed is underestimated in ERA5, we applied mean and variance correction to obtain the corrected ERA5 wind speed data ($\left|U_{10m}\right|$) to eliminate systematic biases and prevent their amplification in WPD calculations. $\rho$ in Equation (2) is the air density field near earth’s surface, which was estimated from surface pressure and T_2m fields based on ideal gas law:

$$\rho ={\displaystyle \frac{SP}{R_d·T_{2m}}}$$

(3)

Note that $R_d$ is the gas constant for dry air, because given Xinjiang’s arid to semi-arid climate, humidity effects can be neglected for simplicity. The annual mean wind power density is the primary metric for evaluating wind resources. According to China’s National Wind Resource Assessment Technical Standards (National Energy Administration of China, 2014), there are five progressive tiers: Class V (≤50 W/m²), Class IV (50–100 W/m²), Class III (100–150 W/m²), Class II (150–200 W/m²), and Class I (≥200 W/m²).

The pressure gradient is a key metric for analyzing strong wind events, as it directly reflects the driving force behind atmospheric motion. Thus, we calculated the total pressure gradient magnitude as follows:

$$\left|\nabla SLP\right|=\sqrt{{\left({\displaystyle \frac{\partial SLP}{\partial x}}\right)}^2+{\left({\displaystyle \frac{\partial SLP}{\partial y}}\right)}^2}$$

(4)

where ${\displaystyle \frac{\partial }{\partial x}}$ and ${\displaystyle \frac{\partial }{\partial y}}$ are the gradient in zonal and meridional direction, respectively.

2.2.2. K-Means Clustering Method

The k-means clustering method [37] has been widely used in classification of weather elements including temperature and total cloud cover as well as extreme weather events such as thunderstorms and air pollutions [38,39,40]. Here, we applied the k-means clustering to daily maximum of hourly 10 m wind speed fields in the period 1979–2023 in Xinjiang. By minimizing the within-cluster sum of squares, k-means method groups grid points with similar spatiotemporal variations, revealing relatively coherent wind regimes. The optimal number of clusters is determined using a combination of the elbow method and the silhouette coefficient. Then, cluster labels are sorted by mean wind speed to establish a unified sequence from high to low wind speeds.

We further selected the representative node for each cluster by searching for a grid point whose time series has the minimum Euclidean distance to the cluster centroid:

$$p_j=\mathit{arg}min{‖x_i-{\mu }_j‖}_2,\hspace{1em} i\in C_j$$

(5)

where $x_i$ is the wind speed time series at grid point $i$ and ${\mu }_j$ is the centroid of cluster $j$. K-means clustering provides an efficient sample framework for extreme wind speed assessment in Xinjiang.

2.2.3. Calculation Scheme of Extreme Wind Speeds

To estimate the return levels of extreme wind speeds, we employed the peaks-over-threshold (POT) method from extreme value theory, widely used in climate-extreme studies [5]. In this approach, samples exceeding an appropriate threshold are assumed to follow a Generalized Pareto Distribution (GPD). Following Outten and Esau [41], we define the positive-anomaly threshold at each grid point as the lowest annual daily maximum 10 m wind speed, and the negative-anomaly threshold as the highest annual daily minimum wind speed. These thresholds ensure that a suitable number of extreme values are selected. The GPD is then fitted to the extreme values using maximum likelihood estimation, with its cumulative distribution function given by

$$H(y)=\left\{\begin{array}{c}1-{\left(1+{\displaystyle \frac{\xi y}{\sigma }}\right)}^{(-\frac{1}{\xi })}, \xi \ne 0\\ 1-e^{\left(-\frac{y}{\sigma }\right)}, \xi =0\end{array}\right.$$

(6)

where $\xi$ is the shape parameter and $\sigma$ is the scale parameter. The GPD corresponds to the exponential distribution $(\xi =0)$, the standard Pareto distribution $(\xi <0)$, or the Type II Pareto distribution $(\xi >0)$. For a suitable threshold, the number of exceedances approximately follows a Poisson distribution with parameter $\lambda$. Accordingly, the T-year return level $U_T$ is calculated as

$$U_T=\left\{\begin{array}{c}u+{\displaystyle \frac{\sigma }{\xi }}\left[{\left(\lambda T\right)}^{\xi }-1\right], \xi \ne 0\\ u+\sigma \mathit{ln}\left(\lambda T\right), \xi =0\end{array}\right.$$

(7)

In this study, extreme wind speeds are computed for return periods of 10, 20, 30, 50, and 100 years, with the 10–30-year return levels serving as the primary focus.

3. Results

3.1. Climatic Characteristics and Long-Term Trends of Wind Speed in Xinjiang

As displayed in Figure 1a, the climatological pattern of annual mean wind speed across Xinjiang exhibits pronounced regional heterogeneity, broadly reflecting the influence of the region’s topography. The maximum wind speeds (>4 m/s) are observed mainly in the eastern Xinjiang along mountain slope areas, which could be related to localized topographic channeling and accelerating effects. Particularly, they are along the southern slopes of the Altay Mountains (including the Irtysh River Valley Wind Zone and Tacheng Basin Wind Zone), both the northern and southern slopes of the eastern Tian Shan Mountains (including the Hami Northern and Southern Gobi Wind Zones and Baili Wind Zone), and the northern slopes of the Altun Mountains (including the Lop Nor Wind Zone). In contrast, the lowest wind speeds are concentrated in the western mountainous regions, including the ridge areas of central Tian Shan Mountains, the Kunlun-Altun Mountains in the southwest, and the northern Altay Mountains.

While topographic effects remain constant, Xinjiang’s near-surface winds exhibit pronounced seasonal variability and long-term trends. This indicates significant influences from climate change and evolving atmospheric circulation patterns. Climatological patterns of monthly mean wind speeds in January, April, July and October show that wind speeds are lowest in winter across the entire region (Figure 2a) and rise to maximums in spring (Figure 2b). Wind speeds remain strong in summer but are slightly weaker than in spring (cp. Figure 2c with Figure 2b). A noticeable decline is observed in October (Figure 2d). From the long-term trends in the period from 1979 to 2023, as shown in Figure 1b, we can see that the annual mean wind speed exhibits a significant increasing trend in the transitional zone between the central-southern Altay Mountains and the Junggar Basin, and the central Tarim Basin. These accelerated trends highlight how the two primary wind speed maxima belts in the northernmost and southernmost wind corridors of Xinjiang have undergone persistent intensification throughout the study period. The intensification in the annual mean is mainly contributed to summer and spring months (Figure 2f,g), while there is no significant trend in winter winds (Figure 2e).

By applying k-means clustering to the spatiotemporal variability of daily maximum 10 m wind speeds, Xinjiang can be classified into four distinct regions (Figure 3a). When combined with hierarchical wind energy resource classification based on annual mean wind power density (Figure 3b) and topographic distribution (Figure 3c), each region exhibits unique characteristics and representative features. Particularly, Region #1 represents the most resource-abundant and spatially homogeneous zone, covering the eastern Xinjiang areas including the Hami Northern and Southern Gobi Wind Zones and the Lop Nor Wind Zone, with its representative node located in the latter. This region forms a wind corridor between three major mountain systems in eastern Xinjiang. It exhibits exceptionally rich wind energy resources with power density consistently exceeding 200 W/m², falling into Class I based on China’s National Wind Resource Assessment Technical Standards. Region #2 covers the Tarim Basin, where the representative node is situated in the central area. This region displays a distinct west-to-east stepwise increasing distribution in wind power density (from Class III to I), inversely corresponding to the terrain elevation gradient. Thus, this region is characterized by significant spatial heterogeneity in wind resources. Region #3 spans northern Xinjiang’s low-lying areas, incorporating the Irtysh River Valley Wind Zone, Tacheng Basin Wind Zone, and Dzungarian Gate Wind Corridor, with wind power densities ranging 100–200 W/m² (Class I–II). Finally, Region #4 comprises the remaining high-altitude mountainous areas, including the Tian Shan Mountains periphery where Node #4 is located, characterized by relatively scarce wind resources (below 100 W/m²).

3.2. Identification and Characteristics of Extreme Wind Speed Events

Return Wind Speed at 10–30 Year Levels

We next assess the extreme winds at a single grid point in the historical period from 1979 to 2023. We begin with the four nodes selected for representing the four regions as detected in previous section. As seen from Figure 4, the curve of the GPD is fitted to the exceedances over the threshold and the return wind speeds at various levels are detected. We can observe minimal variability in extreme wind speeds across increasing return periods within individual nodes but pronounced spatial disparities in intensity between nodes. For instance, for node #1 (wind-rich eastern Gobi corridor, Figure 4a), the 10-year and 30-year return events are 29.78 m/s and 31.66 m/s, respectively, a difference of only 1.88 m/s, both exceeding the 99% confidence level using bootstrapping and falling into Beaufort Scale 11 (violent storm). For node #4 located in Tian Shan highlands (Figure 4d), the 10-year and 30-year return events are only 11.42 m/s and 12.49 m/s, respectively, exceeding the 99% confidence level and belonging to Beaufort Scale 6 (strong breeze). The difference between them can be as large as 18~19 m/s.

The spatial patterns of 10-year, 20-year, and 30-year return level wind speed for all grid points in the domain are further shown in Figure 5. We can see their strong consistency with both the climatological mean 10 m wind speed pattern (Figure 1a) and topographic features (Figure 3c), with regional difference up to ~30 m/s. This confirms the significant regional differences in the intensity of extreme winds within Xinjiang. The extreme winds show a gradual intensification from 10- to 30-year return periods, with localized hotspots exceeding 37 m/s (Beaufort Scale 12) for 30-year events. These hotspots are located in four of the nine most notorious wind corridors, namely the Tacheng Lao Fengkou (Dzungarian Gate), Baili Wind Corridor, Hami Southern Gobi, and Lop Nor Wind Zones, indicating that extreme wind intensities in these regions, are exceptionally high, which would pose significant challenges for wind turbine survivability and infrastructure resilience.

Based on the 10-year, 20-year, and 30-year return wind speed at representative nodes, we have identified EWS events, as listed in

Table 1. Time table of extreme wind speed events at 10-year return level at four representative nodes. Times are provided in Coordinated Universal Time (UTC) and local time (LT; UTC+8).

Number	Node #1 40.75° N, 91.75° E	Node #2 39.75° N, 83.25° E	Node #3 47.00° N, 89.00° E	Node #4 43.00° N, 80.50° E
10-year return level
1	1996-08-30 5:00–9:00 UTC [13:00–17:00 LT]	1982-05-10 3:00 UTC [11:00 LT]	1984-04-24 9:00–10:00 UTC [17:00–18:00 LT]	1982-5-10 7:00–8:00 UTC, 10:00 UTC [15:00–16:00 LT, 18:00 LT]
2	2001-04-07 14:00–16:00 UTC [22:00–24:00 LT]	1999-07-19 12:00 UTC [20:00 LT]	2001-04-05 11:00 UTC [19:00 LT]	1994-10-07 05:00–09:00 UTC [13:00–17:00 LT]
3	2010-03-28 10:00 UTC [18:00 LT]	2008-05-01 20:00 UTC [05-02 4:00 L]	2015-04-27 3:00–4:00 UTC [11:00–12:00 LT]	2007-04-21 06:00–10:00 UTC [14:00–18:00 LT]
4	2023-04-18 11:00–12:00 UTC [19:00–20:00 LT]	2014-07-16 8:00 UTC [16:00 LT]	2018-11-30 22:00 UTC [12-01 6:00 LT]	2012-04-22 06:00–09:00 UTC [14:00–17:00 LT]
5	-	2020-06-28 12:00 UTC [20:00 LT]		2014-05-16 05:00–10:00 UTC [13:00–18:00 LT]
20-year return level
1	1996-08-30 5:00–7:00 UTC [13:00–15:00 LT]	1999-07-19 12:00 UTC [20:00 LT]	2015-04-27 3:00–4:00 UTC [11:00–12:00 LT]	1994-10-07 06:00–7:00 UTC [14:00–15:00 LT]
2	2001-04-07 15:00 UTC [23:00 LT]	2008-05-01 20:00 UTC [05-02 4:00 LT]	2018-11-30 22:00 UTC [12-01 6:00 LT]	2007-04-21 06:00–10:00 UTC [14:00–18:00 LT]
3	2023-04-18 11:00 UTC [19:00 LT]	2014-7-16 8:00 UTC [16:00 LT]		2014-05-16 06:00–08:00 UTC [14:00–16:00 LT]
30-year return level
1	1996-08-30 6:00–7:00 UTC [14:00–15:00 LT]	2014-07-16 8:00 UTC [16:00 LT]	2015-04-27 3:00–4:00 UTC [11:00–12:00 LT]	1994-10-07 07:00 UTC [15:00 LT]
2			2018-11-30 22:00 UTC [12-01 6:00 LT]	2007-04-021 07:00–09:00 UTC [15:00–17:00 LT]

. If threshold exceedances are consecutive (allowing single-hour interruptions) were counted as a single event. Notably, almost all events exhibited regional specificity, with only the 10 May 1982 event simultaneously affecting both Node #2 (Tarim Basin) and Node #4 (Tian Shan foothills). A closer investigation on the decadal, seasonal, and diurnal occurrences of EWS events (Figure 6) yields that 10-year EWS events show higher decadal occurrence after 2001 at all nodes, with Node #2–4 showing doubled occurrences in the latest decade (2012–2023) (Figure 6a). The similar increasing tendency is found for 30-year events that occurred at Node #2–3 (Figure 6d). Seasonally, 10-year events occurred more frequently in spring months (March–May), except Node #2 with summer peaks (Figure 6b). This is consistent with previous studies reporting [27,28] that strong winds in China tend to occur mainly in spring. Despite no clear seasonal preference found for 30-year events, complete absence of winter EWS events exist for both 10-year and 30-year return periods (Figure 6e). The diurnal distribution of EWS events (Figure 6c,f) exhibits a pronounced afternoon peak (13:00–18:00 LT, UTC+8) for both 10- and 30-year return periods across all nodes except Node #3. Such strong preference suggests that thermal forcing might be one of the primary drivers.

3.3. Local Synoptic Systems and Large-Scale Circulation Regimes

From the composite mean wind and SLP fields averaged over all time steps within EWS events for each node (Figure 7 and Figure 8a,c,e,g), we observe distinct circulation patterns related to the intensification of wind speed for different nodes.

It is seen from Figure 7a that compared to EWS events at other nodes, Node #1 events exhibited the most extensive wind acceleration, characterized by a bifurcated flow originating along the southern slopes of Altai Mountains. One is an eastward branch turning southeastward toward the Turpan-Hami Basin, and the other one is a southward branch traversing the Tian Shan Mountains and finally invading into the Tarim Basin. The winds are obstructed by the Altun Mountains and converged near the Lop Nor region, contributing to their intensification. Such an anticyclonic wind shear is dominated by a strong high-pressure system spanning southern East Europe–West Siberia, as manifested by composite SLP field (Figure 8a). This high-pressure system exhibited a center over Lake Balkhash, with composite center values exceeding 1035 hPa. Due to the high’s zonal elongation and extension into Mongolia, pressure gradients along the high’s eastern periphery are enhanced (shadings in Figure 8a), particularly over northeastern Xinjiang, generating sustained and strengthened anticyclonic wind flow.

EWS events at Node #3 are also associated with a strong high-pressure system, but its center intensity is slightly weaker (~1030 hPa) with its center displaced over 10° westward, compared to Node #1 events. This anticyclone is flanked by an eastern low-pressure system, confining its influence to western Xinjiang (Figure 8c). The resulting pressure gradients peaked over northern Xinjiang, driving strong northwesterlies that are further amplified by orographic channeling due to Tian Shan Mountains. Meanwhile, the Tian Shan Mountains act as a dynamic barrier, preventing significant wind speed enhancements south of the range. It can be inferred that, while both Node #1 and Node #3 EWS events are linked to high-pressure systems, the exact position (particularly eastward protrusion) of these systems determines the spatial distribution of wind intensification.

Node #2 EWS events exhibit weaker but distinct wind patterns, associated with the strongest cyclonic circulation around a deep low-pressure system over Tarim Basin (Figure 7b and Figure 8c). This low generates steep pressure gradients within the low-pressure system. Considering that this region overlays with deserts and there were no remarkable pressure systems upstream or downstream, this strong low-pressure system is local and thermally driven. Because the Tarim Basin thermal low is strongest in summertime, Node #2 events tend to occur more frequently in summer, rather than spring as other nodes.

Node #4 events are featured by highly localized stronger winds (Figure 7d), and no evident features can be found in the SLP field except a commonly seenn low-pressure system in southern Xinjiang (Figure 8g). However, analysis of BLH reveals significant large values over vast areas of southern half of Xinjiang only for Node #4 EWS (Figure 8b,d,f,h). BLH exceeded 2 km in the neighborhood of Node #4. This indicates enhanced atmospheric instability, which can strengthen turbulent mixing within the boundary layer, facilitating vigorous vertical momentum transport that effectively couples upper-level winds to the surface, according to Dai and Deser [15]. The dominance of exceptional boundary-layer instability that is caused by thermal forcing over synoptic-scale drivers at Node #4 EWS, explaining its exclusive afternoon occurrence (Figure 6c), as well as the localized nature of wind acceleration.

In sum, large-scale and local circulation forcing and boundary-layer processes play different roles in wind acceleration in the four nodes and regions they represent. Nodes #1 and #3 are primarily forced by synoptic-scale dynamics, featuring strong high-pressure systems west of Xinjiang that interact with orographic effects from surrounding mountains. In contrast, Node #2 events are dominated by a thermally driven low-pressure system over the Tarim Basin. Node #4 presents a unique case where no significant sea-level pressure changes occur; instead, wind acceleration results from afternoon thermal boundary-layer instability, demonstrating the region’s dependence on localized surface-atmosphere interactions rather than synoptic forcing.

Interestingly, the dominant synoptic systems during Node #1 and Node #2 EWS events exhibit significant baroclinic development signatures throughout the troposphere. As illustrated in Figure 9, vertically stacked high–low pressure anomaly pairs are observed at both middle-tropospheric and sea levels, displaying a characteristic westward tilt with centers displaced northward. This is a classic signature of baroclinic instability. Notably, in Node #1 events, the high-pressure anomaly (centered near the Ural Mountains) dominates over its eastern low-pressure counterpart, suggesting reinforcement by the Ural-blocking high. In contrast, Node #2 events feature a pronounced trough extending from the polar region, likely associated with southward shifts in the polar vortex toward Asia. Note that both the Ural-blocking high and the polar vortex represent recurrent weather regimes over Asia during winter and spring months. Our analysis of their annual intensities (Figure 10) further reveals significant strengthening trends from 1979 to 2023, which may partially explain the increased occurrences of EWS events at Nodes #1 and #3 (Figure 6a).

4. Discussion and Conclusions

In this study, we used hourly ERA5 Reanalysis in the period 1979–2023 and two-year (2022–2023) national weather station observational data to characterize regional-scale extreme wind speed (EWS) events in Xinjiang, which is a critical area for wind energy development in China. Using k-means clustering of hourly 10 m wind speed fields combined with wind power density (WPD) classification, we objectively divided Xinjiang into four distinct regions, each represented by a key node. POT and GPD analysis further revealed different thresholds of wind speed for defining the EWS events at 10-, 20- and 30-year return level for specific regions. The regional difference corresponds well to Xinjiang’s unique “Three Mountains and Two Basins” topography, which is consistent with studies in other regions [13].

In particular, Region #1 is the eastern wind-rich zone (Hami Northern/Southern Gobi and Lop Nor areas), which is characterized by the highest wind speeds with exceptionally consistent WPD exceeding 200 W/m². The return wind speeds at 10- to 30-year levels correspond to violent storm scale. Region #2 is the Tarim Basin region, with a clear west-to-east increase in WPD inversely correlated with elevation. Region #3 is Northern Xinjiang’s lowland areas (Irtysh Valley, Tacheng Basin, and Dzungarian Gate), with moderate-to-rich wind resources. Region #4 is mountainous regions, particularly around the Tian Shan Mountains, exhibit the poorest wind resources (<100 W/m²). The difference between regions in the 10-/30-year return level values up to 18~19 m/s.

Statistics of the occurrence of EWS events reveal distinct decadal, seasonal, and diurnal patterns across four nodes. Since 2001, the frequency of 10-year return period events has increased at all nodes, with Nodes #2–4 experiencing a doubling of occurrences during 2012–2023. A similar upward trend is observed for 30-year events at Nodes #2 and #3. Annual mean wind speed also exhibited increasing trends in Regions #2 and #3, attributable to summer months and spring months for the former and autumn months for the latter. Seasonally, 10-year events predominantly occur in spring, except at Node #2, where summer peaks dominate. Diurnally, an afternoon peak characterizes EWS events across all nodes except Node #3, implicating thermal forcing as one of the key factors. The upward trend and distinct seasonality of EWS shows consistency with changes in daily maximum wind speed in China, with negative trends in winter and autumn, and positive trends in spring and summer [42].

Further analysis yields distinct dominant synoptic drivers for 10-year return level EWS events. EWS events that occurred at Nodes #1 and #3 are primarily driven by synoptic-scale dynamics, with a strong high-pressure system to the west enhancing pressure gradients along its periphery. This high-pressure system structurally resembles the Mongolia–Siberian High, whose seasonal migration is a dominant driver of seasonal wind speed variability across Asia [43]. Altai, Tian Shan, and Altun Mountains further amplify winds through orographic channeling, while also restricting the spatial extent of extreme winds via blocking effects, thus making the location of ESW sensitive to the spatial extent and location of high-pressure. Node #2 features a thermally driven deep cyclonic low-pressure system there, strongest in summer due to desert heating. ESW at these three nodes are essentially driven by lower-tropospheric pressure-gradient force, which has been argued as the primary source of the wind speed change in China by Zhang et al. [44]. Distinctly, Node #4 is primarily forced by afternoon thermal boundary-layer instability, via boundary-layer turbulence coupling upper-level winds to the surface, as mentioned as one of the mechanisms of extreme winds by previous studies [16,17]. This also explains the localized nature of wind acceleration in this node.

The dominant synoptic systems during EWS events that Nodes #1 and #3 exhibit are linked to high–low pressure anomaly pairs in the middle troposphere, resembling two recurrent weather regimes: Ural-blocking highs for Node #1 events and southward-shifted Arctic polar vortex toward Asia for Node #3. These two regimes exhibited an intensifying trend, which possibly make contributions to higher occurrence of EWS events in recent decades. Such trends have also been documented by Zhang et al. [45] and Tyrlis et al. [46] but they focused mainly on wintertime. Current forecast systems demonstrate high predictive skill for large-scale regimes, notably Ural blocking beyond a one-week lead time and the Arctic polar vortex up to 10–15 days in advance [47,48,49,50]. This extended predictability of synoptic-scale patterns can provide accurate initial and boundary conditions for regional models, thereby potentially improving the forecast lead time for extreme wind events.

Note that this study has certain limitations that present valuable opportunities for future research. Firstly, quantifying the statistical relationship between regime indices and EWS frequency is necessary, but requires longer-term reliable data to diagnose their individual and synergistic effects. Therefore, large-ensemble numerical modeling experiments are essential to verify these relationships, as in some extreme event studies including Bevacqua et al. [51] and Sippel et al. [52]. Secondly, this analysis employs the GPD; however, it is important to note that return level estimates can be sensitive to the choice of extreme value distribution, particularly when working with relatively short data records. Other common choices, such as the Generalized Extreme Value (GEV) or Pearson Type III distributions [53,54,55], may yield different results. Other future work should focus on projecting EWS trends under climate change with explicit attribution to these evolving regimes, and operationalizing synoptic (e.g., Ural blocking and polar vortex for Nodes #1 and #3) and boundary-layer indicators (e.g., BLH for Node #4) to enhance region-specific EWS predictability in Xinjiang.