Changes in precipitation amounts and extremes across Xinjiang (northwest China) and their connection to climate indices

Data and Selected indices

The meteorological data from 107 weather stations in Xinjiang were provided by the China Meteorological Data Network (http://data.cma.cn/data/cdcdetail/dataCode/SURF_CLI_CHN_MUL_DAY.html). After accounting for missing data and comparing the length of the recorded period, we then selected 89 stations for the period from 1961 to 2018. Missing values account for >1.5% of in the daily precipitation series at 4 stations and <1.5% at other stations. The missing data were completed using conventional statistical methods including: (1) if only one day has missing data, the missing data was replaced by the average value of its two nearest stations; (2) if consecutive two or more days have missing data, the missing data would be processed by simple linear correlation between its nearest stations. These missing data and interpolation have no effect on the results of our study. The selected stations were relatively evenly distributed, which enabled the regional precipitation changes in Xinjiang to be reflected (as shown in Fig. 1). Data quality control and calculation of the extreme precipitation index from the original datasets were undertaken using the R ClimDex software (http://etccdi.pacificclimate.org/software.shtml) (Zhang & Feng, 2004). Data quality control and the calculation process were conducted in strict accordance with the study requirements. Xinjiang is located in the westerly-dominated climatic regime, and the precipitation is impacted by the latitude wave propagation of the mid-latitude atmospheric circulation and the Atlantic Multidecadal Oscillation (AMO) (Huang et al., 2015; Chen et al., 2019), and the drought variability was closely related to AMO and Multivariate ENSO Index (ENSO3.4) events (Yao et al., 2019), Some studies have shown that changes in AMO can also influence climate change in northwest China (Wu, Zhou & Li, 2016; Dong & Dai, 2015; Meehl et al., 2013). The drought and flood events in Xinjiang had significant correlation with the annual North Atlantic Oscillation (NAO), for periods of less than 10 years (Ling et al., 2017). Guan & Yamagata (2003) have demonstrated that the Indian Ocean Dipole (IOD) is at least one possible cause of the abnormal East Asian summer climate, and influenced the precipitation over China. Correlations between various indices (the annual NAO, AMO, ENSO3.4 and IOD) and extreme precipitation events were analyzed. The IOD and NAO values between 1961 and 2018 were collected from the National Climate Center of China (https://cmdp.ncc-cma.net/cn/index.htm), the ENSO3.4 values between 1961 and 2018 were obtained from the monthly index of the National Weather Service Climate Prediction Center of USA (https://origin.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff). The AMO values between 1961 and 2018 were obtained from the National Center for Atmospheric Research (https://climatedataguide.ucar.edu/climate-data). The information used in this study was collected from the National Geographic Information Bureau of Surveying and Mapping Standard Map Website (http://bzdt.ch.mnr.gov.cn). Approval was received to download GS (2019) 3333 standard maps, which were reproduced without modification.

In our study, we used extreme precipitation indices to detect and analyze changes in precipitation. There are many definitions of extreme precipitation, and Nicholls & Murray (1999) presented a series of extreme precipitation indices. For the analysis of extreme precipitation indices, it is recommended to use the methods followed by the World Climate Research Programme project operated by the US Climate Variability and Predictability expert team on climate change detection (Easterling et al., 2003). These indices were initially used to study trends in the global climate by Frich et al. (2002), and were subsequently applied in other areas by various researchers (Jiang et al., 2011, 2013; Weili et al., 2015). The nine indices describe different aspects of the precipitation process, which were defined and used to analyze extremes and detect precipitation variations in our study (Table 1). All the selected indices can be roughly divided into four categories (Zhang et al., 2011b; Sillmann et al., 2013).

Annual precipitation and trends

The spatial-temporal dynamic characteristics and annual precipitation from 1961 to 2018 in Xinjiang are shown in Fig. 2. The annual precipitation in Xinjiang clearly changed over the 58-year period. The annual precipitation anomalies were based on the average (163 mm) from 1961 to 2018 in Xinjiang. The variance tendency over the 58-year period is shown in Fig. 2A. A number of key results were apparent: (1) The annual precipitation anomaly had surpluses exceeding 20 mm in 13 of these years, while in another 18 years, the annual precipitation anomaly had deficits exceeding 20 mm. Years with negative anomalies were mainly occur in the period from 1961 to 1987, and the positive anomalies mainly occurred after 1988. (2) These 58 years can be divided into two periods: a dry period (1961–1987) and a wet period (after 1988). (3) The driest years in the past 58 years were 1976 and 1997, while 2010 and 2016 were the wettest years. Figure 2B shows the changes in the annual precipitation over time and indicates that precipitation increases at a speed of 0.99 mm per year from 1961–2018.The low-pass filter also indicated that precipitation fluctuation increased, especially after the mid-1980s when the volatility became more intense and frequent. To determine whether the precipitation time series contained a mutation.the results of a Mann–Kendall (M–K) mutation test on the annual precipitation (Fig. 2C). It was found that the precipitation in Xinjiang changed abruptly in 1987, 1988 and 1989, with the results passing a significance test. Combined with the precipitation anomalies in Fig. 2A, it is found that 1987 was a transition year from negative anomalies to positive anomalies, and determined 1987 as the true mutation year.

Figure 3 shows that there was an obvious spatial difference in the annual precipitation of Xinjiang. It was much higher in the mountains compared to the other areas, much higher in the west compared to the east, and much lower in the south compared to the north. The area with the most abundant precipitation was the Tianshan Mountains, with an annual precipitation of more than 600 mm. The precipitation patterns in Xinjiang varied with topography.

Changes of annual precipitation extremes

Abrupt changes in precipitation extremes

To detect whether there was a mutation in the precipitation extreme time series from 1961 to 2018, an M–K mutation test was performed on the regional average extreme precipitation indices, as shown in Fig. 4. Nine extreme precipitation indices were selected for use in this study and all of them indicated that there were abrupt changes from 1961 to 2018. However, there were some differences in the timings of the mutation points among the indices. The mutation time, based on the consecutive dry days (CDD) appeared in 1987 and 1989, and the consecutive wet days (CWD) appeared in 1987 (shown in Figs. 4A and 4B). The threshold indices, that is, number of heavy precipitation days (R10) and annual total wet-day precipitation (PRCPTOT) mutated in 1990 (shown in Figs. 4D and 4C). The percentile-based threshold indices, that is, R95p and R99p, changed abruptly in 1990 (R95p appeared in 1989–1990, shown in Figs. 4E and 4F). The mutations based on the other indices, maximum 1-day precipitation amount (RX1day), maximum 5-day precipitation amount (RX5day), and simple daily intensity index (SDII), occurred in 1987, 1990 and 1992, respectively (shown in Figs. 4G, 4H and 4I). The mutations of these indices passed a significance test. Although the mutation times of these indices were different, they were concentrated in the period from 1986 to 1992.

Variation of the trends in precipitation extremes

The statistical results of the trends of the extreme precipitation indices for the 89 meteorological stations in Xinjiang from 1961 to 2018 are shown in Fig. 5. Figure 5B shows the percentage of the meteorological stations with zero, positive, and negative trends for each extreme precipitation indices. The percentage of meteorological stations passing the significance level is shown in Fig. 5A. Except for the CDD, which was dominated by a negative trend, all other indices were dominated by positive trends over the 1961–2018 period. The percentage of zero and negative trends was low. It should be noted that the percentage of results that passed a significance test (0.05> P ≥ 0.01 and P < 0.01) was not high, as shown in Fig. 5A. This indicates that the significant increases of extreme precipitation in Xinjiang were concentrated in some local areas and did not extend across the entire region.

To evaluate the changes in the annual precipitation extreme indices of Xinjiang from 1961 to 2018 based on the mutation test results for each index, a time sequence of each index was constructed (Fig. 6), and the probability distribution functions (PDFs) of the periods before and after the abrupt change in each annual precipitation index were evaluated. The PDFs of the annual precipitation extremes across the two periods are shown in Fig. 7. The CDD and CWD had different trends from 1961 to 2018 (Fig. 6A and 6B). Over the 1961–2018 period, the average CDD was 105.5 days, with a decreasing trend of −3.6 days/10a, but approximately 31.5% (28) of the stations passed a significance test. The slope of the trend was different before and after the mutation (Fig. 6A), with values of −1.26 days/10a for the period from 1961 to 1987 and −1.44 days/10a for the period from 1988 to 2018. The average CWD was 2.96 days and it presented a gradually increasing trend of 0.08 days/10a, but only 22.7% (20) of the stations passed a significance test. The slope of the trends was also noticeably different before and after the mutation, with values of 0.009 days/10a (1961–1987) and 0.03 days/10a (1988–2018). The aforementioned decreasing trend in CDD and increasing trend in CWD could also be determined from the variation in their PDFs (as shown in Figs. 7A and 7B). The PDF curves of CDD were negative from 1988 to 2018, but the CWD moved in the opposite direction and shifted positively. The shifts of the PDF curves clearly indicated a decrease in CDD and increase in CWD. The CDD values corresponding to the maximum probability and maximum values were 113 and 165 days from 1961 to 1987. These values decreased to 95 and 150 days, respectively, during 1988–2018 (Fig. 7A). The CWD values corresponding to the maximum probability and maximum values were approximately 2.7 and 3.3 days from 1961 to 1987, but they increased to 3.2 and 4.2 days, respectively, over 1991–2018 (Fig. 7D). The CDD and CWD trends show that the climate in Xinjiang is becoming wetter.

The RX1day and RX5day values increased over time (Figs. 6G and 6H). For the period from 1961 to 2018, the average RX1day was 18.96 mm, with an increasing trend of 0.84 mm/10a, but only approximately 27% (24) of the stations passed a significance test. The slope of the trend was different before and after the mutation (Fig. 6G), with values of 0.56 mm/10a for the period from 1961 to 1987 and 0.9 mm/10a from 1988 to 2018. In contrast, the average RX5day value was 27.2 mm and it presented a gradually increasing trend of 1.26 mm/10a, but only 25% (22) of the stations had significant growth trends. The slope of the trends, which were also noticeably different before and after the mutation, was 0.8 mm/10a (1961–1990) and 0.98 mm/10a (1991–2018). The increasing trend of the RX1day and RX5day values could also be determined from the variation in their PDFs (as shown in Figs. 7G and 7H). The PDF curves of the RX1day and RX5day values shifted positively. The RX1day values corresponding to the maximum probability and maximum values were approximately 17 and 24.5 mm, respectively, from 1961 to 1987. These figures increased to approximately 19.5 and 28 mm, respectively, during 1988–2018 (Fig. 7G). The maximum probability and maximum values of the RX5day were approximately 25 and 37 mm, respectively, from 1961 to 1987, but they increased to 27 and 43 mm, respectively, over 1991–2018 (Fig. 7H). Similar to the trends of RX1day and RX5day, the PRCPTOT and SDII displayed increasing trends (Figs. 6C and 6I). The average PRCPTOT was 153.5 mm and it displayed an increasing trend of 9.9 mm/10a, with approximately 69% (61) of the stations passing a significance test. The slope of the trend was different before and after the mutation (Fig. 6C), with a value of 8.1 mm/10a for the period from 1961 to 1990 and 0.97 mm/10a from 1991 to 2018, respectively. In contrast, the average SDII was 4.64 mm/d and it presented a gradually increasing trend of 0.09 (mm/d)/10a, but only 22.7% (20) of the stations passed a significance test. The slopes of the trends, which were noticeably different before and after the mutation, were 0.12 (mm/d)/10a for the period from 1961 to 1992 and 0.15 (mm/d)/10a from 1993 to 2018. The PDF curves of PRCPTOT and SDII also moved in a positive direction (Figs. 7C and 7I). The PRCPTOT values corresponding to the maximum probability and maximum values were approximately 135 and 210 mm, respectively, from 1961 to 1990. These values increased to around 165 and 250 mm, respectively, over 1991 to 2018 (Fig. 7C). The SDII value corresponding to the maximum probability and maximum values were approximately 4.5 and 6.1 mm/d, respectively, from 1961 to 1992, but changed to 4.75 and 5.75 mm during 1993–2018. The changes in the four indices indicated that there were increases in extreme precipitation intensity in Xinjiang. The R95p and R99p values were 33.3 and 11.1 mm, respectively, with both indices displaying upward trends at rates of 3.96 and 1.81 mm/10a, in the same order (Figs. 6E and 6F). For the two indices, 28.4% (26) and 18.2% (16) of the stations passed a significance test, with the slope of the trends being 2.45 and 0.93 mm/10a, respectively, for the period from 1961 to 1990 and 3.65 and 1.85 mm/10a, respectively, from 1991 to 2018, that is, before and after the mutation. The R95p values corresponding to the maximum probability and maximum values were approximately 25 and 58 mm, respectively, from 1961 to 1990. These values increased to approximately 40 and 85 mm, respectively, over the latter period (Fig. 7E). Similarly, the maximum probability and maximum values of R99p were around 8 and 20 mm, respectively, from 1961 to 1990. These values increased to approximately 11 and 42 mm, respectively, in the latter period (Fig. 7F). The average R10 value was 3.6 days, and it also exhibited an increasing trend of 0.31 d/10a (Fig. 6D), with 44.3% (39) of the stations displaying statistically significant trends. The R10 values corresponding to the maximum probability and maximum values were around 3 and 5.6 days, respectively, from 1961 to 1990, which increased to approximately 4 and 6.5 days, respectively, over the latter period (Fig. 7D). This indicates that the frequency of extreme precipitation in Xinjiang has increased.

Spatial variation of extreme precipitation

The spatial changes in each index value are shown in Fig. 8. It can be seen that the indices indicated a spatial heterogeneity and diversity in extreme precipitation across Xinjiang. Generally, the CDD value was lower in mountainous areas than in other regions. In northern Xinjiang, the CDD value was lower than that of southern Xinjiang, and in the west, it was lower than that of the east. Compared with the CDD, the spatial changes of the other eight indices had the opposite pattern. In a similar manner to the annual precipitation amounts, the value of the other eight indices were greater in mountainous areas than in the other regions. In northern Xinjiang, the values were greater than that in southern Xinjiang, and they were also greater in the west than in the east.

Figure 8 also shows the spatial changes in the trends of the annual series of precipitation indices. There was a significant difference in the spatial distributions of the CDD and CWD trends. The largest areas with a decreasing CDD were in northern Xinjiang and the Tianshan Mountains, with the stations that passed a significance test concentrated in northern Xinjiang. Only six of these stations displayed an increasing trend and they were scattered across the region (Fig. 8A). In contrast to the CDD, for the CWD, the largest area with an increasing trend was western Xinjiang. Only nine of the stations displayed an increasing trend and they were scattered across the western region (Fig. 8B). The spatial distribution of the changing trends of RX1day was very similar to that of RX5day (Figs. 8G and 8H). There were mainly increasing trends for RX1day and RX5day, and the stations that passed a significance test were mainly located in the Tianshan Mountains. However, there were 18 and 19 stations for which a decreasing trend of RX1day and RX5day were observed, respectively, and they were mainly scattered across the eastern Tianshan Mountains and northern Xinjiang. There were certain differences among the spatial distributions of the changing trends of PRCPTOT, R10, and SDII (Figs. 8E, 8D and 8I). Most locations displayed increasing trends and the stations with the largest increases were concentrated in mountainous regions. However, there were clear differences in the spatial distributions of stations that passed a significance test. For PRCPTOT, a large number of stations passed the significance test and they were evenly distributed across the study area. In contrast, for R10 and SDII, only a few stations passed the significance test and they were concentrated in the Tianshan Mountains. However, six stations (PRCPTOT), three stations (R10) and 15 stations (SDII) had a decreasing trend, and they were scattered across the study region. The R95p and R99p mainly increased, and the stations with the largest increase that passed a significance test were mainly located in the Tianshan Mountains (Figs. 8E and 8F). However, there were also 16 stations (R95p) and 25 stations (R99p) in which a decreasing trend was observed. These stations were distributed throughout Xinjiang.

Period analysis and correlation with climate indices

The time series of annual AMO, ENSO, IOD, and NAO from 1961 to 2018 are shown in Fig. 9, with the four indices showing fluctuations in the different trends. The periodic changes of the extreme precipitation indices (R95p and R99p) and climate indices (AMO, ENSO, IOD, and NAO) in the time series are shown in Figs. 10 and 11. The XWT correlation between R95p and AMO is shown in Fig. 10A. There were two significant power bands: a 1–3-year period from 1992 to 2000 (band (1)) and another from 2005 to 2018 (band (2)), although the direction of the arrows was different. In band (1) the arrows pointed down, while in band (2) the arrows pointed to the right, which indicated that the phases of the two time series differed by 90° and there was a positive correlation between R95p and AMO, respectively. The WTC showed a significant 3–7-year band from 2005 to 2018 and a 1–4-year band around 1962–1975, as shown in Fig. 10B. The arrows pointed down (phases differed by 90°) or to the right (positive phases).

The XWT correlation between R95p and ENSO (Fig. 10C) revealed three significant power bands: a 3–6-year period from 1985 to 1993 (band (1)), a 2–3-year period in the mid-1990s (band (2)), and another from around 2007–2016 (band (3)). The arrows were different in the three bands, with anti-phase conditions in the mid-1990s (band (2)). The WTC between R95p and ENSO showed a 4–7-year period around 1982–2000 (Fig. 10D). It also showed a 2–4-year band from 1961 to 1975. The arrows indicated that the phase of two time series differed by 45° or were in a positive phase. As shown in Fig. 10E, there was one band with a good correlation between IOD and R95p, which revealed a 1–3-year period from 1990 to 1998. The WTC results for the two time series showed that their periodicity and periodic occurrence time were consistent with the XWT results, and the two series were basically in anti-phase (Fig. 10F).

The XWT correlation between R95p and NAO (Fig. 10G) revealed three significant power bands: a 5-year period around 1990, a 2–3-year period in the late 1970s, and a 1–3-year period from 2000 to 2010. There were two significant bands in the WTC between the R95p and NAO (Fig. 10H), which indicated a 5–7-year period from 1982 to 1998 and a 2–3-year period from 1992 to 1998. The direction of the arrows indicated that the time series were neither in phase nor in anti-phase.

Through the analysis of XWT and WTC, the correlation and periodicity of R99p and climate indices were found to be similar to the results of the XWT correlation between the R95p and climate indices (Fig. 11). There were specific correlations between the extreme precipitation events (R95p and R99p) and the four climate indices (AMO, ENSO, IOD and NAO). It was neither a simple linear relationship nor a simple positive or negative relationship, but was more of an advance or delay in phase. This also indicates that the extreme precipitation events in Xinjiang were caused by multiple factors.