Daily precipitation data from 77 stations in ARNC for the period 1968–2020 were obtained from the National Meteorological Science Data Center of the China Meteorological Administration (http://data.cma.cn/). Data quality was strictly controlled, including inspection of climate threshold values and extremes, and consistency checks (i.e., internal, spatial, and temporal consistency). Moreover, erroneous data were manually checked, validated, and corrected or removed. A quality control flag was provided in the dataset and the threshold for missing data was set at <20% for each site. In this study, valid data (flag = 0) were adopted, while data with other flags and records from sites with more than 20% missing data were excluded. Missing data over one day or 2 days were filled using average precipitation values of neighboring days, while if consecutive days had missing data, the missing values were replaced with the long-term averages of the same days (Zhang et al., 2013).

The National Centers for Environmental Prediction/National Center for Atmospheric Research reanalysis dataset was used in this study because it has been verified to have better applicability in representing the climatology and prominent interannual variations of the atmospheric circulation over ARNC (Li and Ma, 2018; Wang et al., 2018; Yin and Zhou, 2018). The specific reanalysis data comprised mean monthly wind fields, specific humidity, and geopotential height (GH) (https://psl.noaa.gov/), which were obtained on a 2.5° × 2.5° global grid with 17 pressure levels covering the period from January 1948 to the present.

This study adopted the percentile method (Zhai and Pan, 2003), which is one of the most widely used approaches for identification of climate/weather events (Zhai et al., 2005; Shi et al., 2021). Because of the huge differences among the different stations, extreme events are difficult to define using unified and fixed daily precipitation standards. Therefore, we used the 30-year mean value of the 95^th percentile value of the annual daily precipitation series during 1971–2000 to determine the threshold of extreme precipitation. When the daily precipitation exceeded the threshold, it was identified as an extreme precipitation event. The sum of all extreme precipitation amounts at a station in a given year was regarded as the annual extreme precipitation amount, which was further employed to calculate the EPCD and EPCP.

The temporal distribution of the extreme precipitation occurrence time (EPOT) during a year is described as the EPCD (Li et al., 2010), and the EPCP represents the month in which the maximum extreme precipitation occurred (Dong et al., 2020). Based on vector analysis, the extreme precipitation amount of each month in a year can be taken as a vector, which can then be decomposed into two components in the horizontal and vertical directions. The monthly extreme precipitation amount can be viewed as the length of the vector. The direction for a year can be seen as a cycle (360°), whereas the vector direction for each month is that of the corresponding month order (starting from 0) multiplied by 30° (Zhang and Qian, 2003). The sum of the vertical and horizontal components of extreme precipitation at a certain station (i.e., R _xi and R _yi , respectively) can be expressed as follows: R x i = ∑ j = 1 n r i j ⁡ sin ⁡ θ j , (1) R y i = ∑ j = 1 n r i j ⁡ cos ⁡ θ j , (2) Where, r _ij is the extreme precipitation during month j in year i, and θ _j is the azimuth of the month j. Thus, E P C D i = R x i 2 + R y i 2 R i (3)

And E P C P i = tan − 1 ( R x i R y i ) , (4) where R _i is the extreme precipitation during the year i.

If all extreme precipitation occurred only in one month of a year, the ratio of the synthetic component concerning the annual extreme precipitation would be 1, referred to as the maximum value. If the same extreme precipitation occurred every month of a year, the two summations of the separate additive components in the horizontal and vertical directions should be zero, referred to as the minimum value. Therefore, EPCD can reflect the concentration or dispersion degree of EPOT in a year. Additionally, EPCP indicates the month in which the maximal extreme precipitation during a year occurs, that is, the maximum extreme precipitation occurrence (MEPO). The corresponding relationship between EPCP and month is presented in Table 1.

EPCD can well reflect the inhomogeneity of extreme precipitation and thus improve the accuracy of drought and flood prediction (Li and Qian, 2006). However, the calculation of both EPCD and EPCP is complicated. Generally, larger time scales (month) can better describe the overall pattern, while smaller time scales (10-d and pentad) better describe the regional pattern (Wang et al., 2019). Moreover, EPCP reflects only the maximum extreme precipitation in a fixed period and it ignores the maximum extreme precipitation between periods (Wang et al., 2019).

The spatial heterogeneity of EPCD across ARNC is evident in Figure 2. Owing to the special structure of the terrain of “three mountains and two basins” in Xinjiang, notable differences in the spatial distribution of EPOT are evident. In southern Xinjiang, EPCD is largest in the central parts and decreases outward, which indicates that EPOT gradually disperses outward. This broadly circular distribution in southern Xinjiang is similar to the pattern of increasing precipitation from the desert toward the oasis, which might be attributable to interacting mechanisms between the local oasis cold-wet and water vapor transport (Zhou and Lei, 2018). In northern Xinjiang, the EPCD value is higher in mountainous areas than in the Junggar Basin and surrounding areas, which indicates EPOT of mountainous areas is concentrated. This regional discrepancy might be related to the fact that persistent extreme precipitation is more frequent in basins (except in summer) than in mountainous areas (Zhao et al., 2011), and it is considered that the future frequency of extreme precipitation will exhibit the same characteristics (Tang et al., 2020).

Furthermore, in northern Xinjiang, the EPCD value is less than that in southern Xinjiang, indicating that EPOT of northern Xinjiang is more dispersed, especially in Tacheng, which is similar to the distribution of water vapor supply. Specifically, abundant water vapor from the Arctic region and western Eurasia (Wang et al., 2020) is transported to northern Xinjiang and the Tianshan Mountains by Arctic air mass (Tian et al., 2007) and mid-latitude westerlies, resulting in most extreme precipitation occurring in months other than in summer. Additionally, the pattern in northern Xinjiang is in accord with the input pathway of the Arctic air mass. This might cause the dispersion of EPOT of northern Xinjiang and the Tianshan Mountains. Conversely, EPOT of southern Xinjiang is more concentrated because of the minimal source of moisture supply. Under the combined influence of the Iranian vortex and the equatorial vortex, water vapor from the Arabian Sea is transported by the South Asian summer monsoon to southern Xinjiang, which is a crucial input of water vapor at certain times in summer (Huang et al., 2015; Gao et al., 2020; Yao et al., 2020).

In eastern ARNC, including the Hexi Corridor, Qaidam Basin, and Alashan Plateau, EPCD gradually decreases from east to west, indicating that EPOT gradually disperses along the west–east direction but becomes concentrated over the eastern Alashan Plateau. This pattern of change is possibly caused by interaction between the westerlies and the summer monsoon. In the Qaidam Basin and the northern Hexi Corridor, water vapor is mainly transported by the westerlies. Further east, extreme precipitation is interacted by the westerlies and the summer monsoon (Li et al., 2015; Du et al., 2020). In summer, blocked by the Tianshan and Altai mountains, EASM provides abundant water vapor and causes precipitation over the eastern Alashan Plateau (Li et al., 2020).

Over the past 54 years, EPCD in ARNC has decreased slightly (Figure 3), reflecting that the extreme precipitation events have become more scattered over several months throughout the whole year, which is in accord with the advance of the start date and delay in the end date of extreme precipitation in ARNC (Lu et al., 2021a). EPOT of Xinjiang is concentrated from the north to the south, and matches well with the input pathway of a southward-moving air mass, indicating the potential influence of the Arctic air mass. Specifically, those stations that passed the significance test at the 95% level are mainly in northern Xinjiang, and the pattern is the same as the increase of number days of extreme precipitation (Wang et al., 2014), further reflecting the reliability of the pattern as well as the potential influence of the Arctic air mass. In eastern ARNC, EPOT gradually scatters from southeast to northwest, coincident with the input path of EASM, and it might be related to the prolonged duration of EASM (Ha et al., 2020).

The range of EPCP in ARNC is 160°–195°, revealing that MEPO occurs mainly in July (Figure 4; refer to Table 1), however, MEPO varies widely in different subregions. MEPO occurs in late July on the Alashan Plateau and in the eastern Hexi Corridor, coincident with the time of peak EASM (Wang and LinHo, 2002), indicating the potential influence of EASM. On the other hand, in Xinjiang, the MEPO is delayed from the south to the north; that is, it is in mid-July in the Altai Mountains and surrounding areas, in early July in the Junggar Basin and northern Tarim Basin, and in late June in southern Tarim Basin. The pattern of MEPO in Xinjiang is similar to the pathway of southward invasion of the Arctic air mass, indicating the potential influence of the Arctic air mass. Arctic air mass influences the mid-latitude weather/climate in early summer, owing to hemispheric-scale adjustments in the longwave structure of the tropospheric GH field (Hanna et al., 2013; Overland et al., 2016; Francis et al., 2017). It should be noted that the Gurbantunggut Desert in the Junggar Basin exhibits an anomaly; that is, MEPO is advanced in comparison with surrounding areas, which might be related to the regional topography and type of underlying surface, and it represents a subject that should be investigated further using model simulations.

By definition, the existence of a positive change rate of EPCP corresponds to delay of MEPO, and vice versa. As shown in Figure 5, the change of MEPO presents a pattern of advance–delay–advance–delay from east to west over ARNC, for which only 11.7% (9) of the stations passed the significance test at the 95% level. MEPO has become delayed by < 4 d/10a in the western margins of the Qaidam Basin and the Qilian Mountains, while it has advanced at a rate of between 2 and 8 d/10a from west to east in eastern ARNC (only three of the stations passing the significance test at the 95% level), which might be associated with the fact that the change of heavy precipitation in summer has been less intense than that in spring and autumn (Wang et al., 2015). The change of MEPO in eastern ARNC is also consistent with the input pathway of the EASM. Moreover, the advance of MEPO has gradually decreased at a rate of between 2 and 8 d/10a from south to north in eastern Xinjiang (only two of the stations passing the significance test at the 95% level), except for a delay of <4 d/10a in the northern margins of Xinjiang, reflecting the influence of the invasion of the Arctic air mass. Among these changes, MEPO delay in northern Xinjiang is similar to the increases in precipitation and extreme precipitation intensity during winter time in northern Xinjiang (He et al., 2020; Tang et al., 2020). In contrast, MEPO has become delayed by < 4 d/10a in western Xinjiang, for which only two of the stations passed the significance test at the 95% level. This regional difference is similar to the changes in the intensity of extreme precipitation in Xinjiang, with strong precipitation extreme in western Xinjiang and weak precipitation extreme in eastern Xinjiang (Zhang et al., 2012).

Based on the result of the abrupt point of MEPO in 1993, the differences in the atmospheric parameters between 1968–1992 and 1993–2021 were analyzed. As shown in Figure 7A, a positive anomaly of GH appears in the high latitudes of the Northern Hemisphere, which weakens the polar vortex and drives southward the input of the Arctic air mass. Meanwhile, the northeasterly airflow caused by the cyclone over the Japan Sea transports water vapor that converges with the Arctic air mass at 120°E (Figure 7B). Furthermore, the enhanced anticyclone with a positive anomaly of GH centered over Mongolia strengthens the transport of water vapor toward the study region. Moreover, the northwesterly airflow associated with the Iranian Plateau anticyclone and the westerly airflow associated with the Siberia cyclone also strengthens the transport of water vapor toward the study region. Consequently, abundant water vapor converges over the study region (Figures 7C,D). These two main pathways of water vapor transport provide abundant water vapor for triggering extreme precipitation events in the study region.

The highest EPCD indicates the most concentrated EPOT, which reflects scarce water vapor sources. As shown in Figure 8A, a negative anomaly is evident in the high latitudes of the Northern Hemisphere, which means a deeper polar vortex. An anticyclone with a positive anomaly of GH over the Okhotsk Sea, together with a cyclone with a negative anomaly of GH over the Barents Sea, suppresses the southward input of the Arctic air mass (Figure 8B). Moreover, the negative anomaly of GH over the Indian subcontinent strengthens the southwesterly airflow. However, owing to the northward input of the southwesterly airflow being impeded by the negative anomaly of GH over the northwestern ARNC, it ultimately only reaches eastern ARNC. Consequently, water vapor is scarce and it diverges within the study region (Figures 8C,D), which is unfavorable for the occurrence of precipitation. We infer that extreme precipitation might be related to recycled moisture over local regions of ARNC (Wu et al., 2012; Peng and Zhou, 2017; Zhang et al., 2021).

The lowest EPCD indicates the most scattered EPOT, which reflects abundant water vapor sources. As shown in Figures 9A,B, the zonal wave pattern is positive–negative–positive with anticyclones with positive anomalies GH over the Eastern Europe Plain and northwestern Pacific, as well as a cyclone with anomalous negative GH over western Siberian in between, bringing water vapor to ARNC (Wu et al., 2017). Therefore, the northwesterly airflow caused by the anticyclone over the Eastern Europe Plain and the northeasterly airflow caused by the cyclone over western Siberian act together to strengthen the transport of water vapor toward the study region, consistent with the transport pathway of summer extreme precipitation events in Xinjiang (Ning et al., 2021). Abundant water vapor is transported toward the study region (Figure 9C) and converges over western Xinjiang (Figure 9D), which is favorable for the occurrence of precipitation. Consequently, the occurrence of extreme precipitation events is scattered in this year.

The highest EPCP indicates that MEPO is later than in other years and occurs in August. As shown in Figure 10A, a negative anomaly of GH develops over the high latitudes in the Northern Hemisphere, but a positive anomaly centered over western Siberia extends northwestward to 70°N, which strengthens the southward input of the Arctic air mass (Figure 10B). The southern flank of an anticyclone with a positive anomaly of GH over Northeast China causes an easterly wind that transports water vapor toward the study region. Moreover, the cyclone with a negative anomaly of GH over the Iranian Plateau strengthens the southwesterly airflow transporting water vapor to ARNC. These main two water vapor channels provide water vapor to the northern margins of Xinjiang and the southern margins of ARNC (Figure 10C). In particular, water vapor converges over the southern margins of ARNC (Figure 10D), enhancing the occurrence of precipitation.