Spatiotemporal variations of snow characteristics in Xinjiang, China over 1961–2013

Snow cover is an important component of the hydrosphere and it reaches 46 million km² during winter in the northern hemisphere (Frei & Robinson 1999). Wintertime snow and spring snowmelt are critical to agriculture, ecosystem, and water resources management in cold regions. Snowpack is a special land cover type, and it is a key factor for local and global scales climate change in response to global warming (Liston 1999; Frei & Gong 2005; Bavay et al. 2009). Snow depth is an indispensable indicator for estimating snow water equivalent, calculating surface radiation and water budget, and simulating snowmelt runoff in spring. The high reflectivity and the low conductivity of snowpack affect the energy exchange within the land–atmosphere system (Lamb 1955; Namias 1964; Hahn & Shukla 1976; Dey & Kumar 1983; Foster et al. 1983).

Many studies have used ground observations and data to analyze long-term changes in snow depth, to reveal its variation and relationship with the other meteorological variables, and to predict future changes by different snow models (Pomeroy et al. 2007; Wu & Kirtman 2007; Jin & Wen 2012; Tang et al. 2013). In recent years, snow cover variations have not only been observed by the routine surface meteorological observation, but also with other advanced technologies, such as hyper-spectral, microwave radiometry, and remote sensing (Rango et al. 1989; Wang et al. 1992; Rott & Nagler 1995; Grody & Basist 1996; Tait & Armstrong 1996; Butt 2006; Butt & Kelly 2008). These technologies were applied successfully for snow monitoring, but the underlying surface is complex, and there is still a need for ground observation and data to retrieve snow properties. Moreover, snow observation and data are also vital for regional climate change, snowmelt modeling, and snowmelt floods. Stable snow cover areas in China are mainly located in the Tibet Plateau, the north Xinjiang Tianshan Mountains area, and the northeast three provinces Inner Mongolia region (Li & Mi 1983). The changes in snow depth affect snowmelt processes, which is important for water resources in the arid and semi-arid area of Xinjiang Autonomous region (Shen et al. 2013).

Spatiotemporal variability analysis involves the multivariate spatial analysis methods for hydrological time series, e.g., singular value decomposition (Wang et al. 2017), canonical correlation (Hardoon et al. 2004), cluster analysis (Nauditt et al. 2017), empirical mode decomposition (EMD), and empirical orthogonal functions (EOFs) (Hu et al. 2017), etc. EOF analysis is a widely applied statistical method for analyzing large multidimensional datasets (Perry & Niemann 2007). The EOF approach can condense the spatiotemporal information, study spatial and systematic structures, and identify dominant patterns of variations inherent at the measuring sites. To the present, the EOF method has been applied to analyze soil water content in the Canadian Prairie pothole region (Hu & Si 2016), terrestrial water storage in the Tarim River basin (Yang et al. 2017), water quality variation of Nakdong River, Korea (Sung et al. 2017), and many other water-related properties for different regions (Cerrone et al. 2017; Deepthi et al. 2017), but has not been widely used for spatiotemporal variability analysis of snow-related properties in Xinjiang, China. In addition, the EMD decomposes the overall spatial pattern into a finite and often small number of intrinsic modes, which are known as intrinsic mode functions (IMFs) that represent the characteristic scales of variability in the physical property.

This research aims to analyze the spatiotemporal variations of snow parameters, including snow cover starting date (D_s), ending date (D_e), duration (D_d), seasonal average snow depth (SD_avr), and maximum snow depth (SD_max). The snow depth data at 105 sites in Xinjiang, China, over the period 1961–2013 were collected. The seasonal trends, scales, and spatiotemporal variability of D_s, D_e, D_d, SD_avr, and SD_max have been investigated to illustrate the overall changes of snow cover.

The Xinjiang Uygur Autonomous region is located in northwestern China with a longitude range of 73°40′E–96°23′E and altitude range of 34°22′N–49°10′N. The longest distance from east to west is 1,900 km, and from north to south is 1,500 km. The area of Xinjiang is about 1,660,000 km², about one-sixth of the total area of China. The border of Xinjiang region is 5,300 m, which is the longest border in the provincial regions in China. Xinjiang has a temperate continental climate and is a typical arid and semi-arid region, being characterized by ‘three mountains (i.e., the Kunlun Mountains in the south, the Tianshan Mountains in central Xinjiang, and the Altai Mountains in the northeast) and two basins (i.e., Zhungaer and Tulufan)’ (Li et al. 2017a, 2017b). There is a long, dry, and cold winter, and a short, dry, and hot summer. Annual mean precipitation is 145 mm, and evaporation capacity is about 200 mm. According to the climatic conditions, Xinjiang is divided into different sub-regions, i.e., north, south, and Tianshan Mountains area. The snowfall in north Xinjiang and Tianshan Mountains area accounts for about one-third of the annual precipitation, which is vital for water resources utilization in Xinjiang.

The daily snow data for the period 1961–2013 at the 105 stations in Xinjiang (Figure 1) were obtained from Xinjiang Meteorology Bureau, Urumqi, China. The elevations of these selected sites range between −197 and 8,494 m above sea level. The snow properties were measured and recorded manually. Snow depth was measured with a ruler, depth was measured three times to provide an average value. The snow data have undergone strict quality controls by Xinjiang Meteorological Observatory with 99.9% completeness. Missing data were replaced by the long-term averages for the neighboring days.

The snow cover parameters including D_s, D_e, D_d, SD_avr, and SD_max were analyzed to show snow variations. For easy statistical calculation, both D_s and D_e values were counted from July 1st, if daily snow depth is ≥1 cm, D_d increased by 1. If the snow lasted longer than 5 days, it belongs to a stable snow cover year (Wang et al. 2014). The D_d within the stable snow cover year is equal to D_e minus D_s.

EMD separates the variations in snow parameters according to their characteristic scales. EMD works directly in the temporal domain with the basis derived from the data. In a natural system, the overall variability is controlled by a number of processes occurring together at different intensities or scales (Goovaerts 1998). The processes with similar scales are separated into different IMFs. In defining the IMFs, they should satisfy the following conditions: (i) the mode might be linear and the number of minima or maxima and zero crossings must either be equal or differ at most by one, while zero crossing indicates the point where the function changes sign; (ii) the oscillation will be symmetric with respect to the local mean. According to these definitions, IMFs can be obtained after decomposing any function through a sifting process.

The trend test for annual time series cannot show seasonality, therefore a trend test for a seasonal dataset is also conducted for December, January, and February when snow cover lasts for the whole month. The detailed procedure for seasonal trend test is referred to in Helsel & Hirsch (2002).

Spatial distributions of the snow parameters, trends, and the EOFs were obtained by the inverse distance weight interpolation method in ArcMap 10.2 software.

There was a total of 105 sites. The Jimunai station in north Xinjiang was taken as an example. Figure 2 shows temporal variations of observed, IMFs, and variances in percentages for daily snow depth during 1961–2013 at Jimunai. Snow depth fluctuated periodically with different peaks for each year (Figure 2(a)). There was a SD_max of 75 cm in 2010 and a minimum of 0 cm each year. Snowfall was concentrated from October to February. The variances over different IMFs varied (Figure 2(b) and 2(c)). About 79.6% of the total variation of snow depth was separated in IMF9 (scale of 189 days), IMF10 (scale of 302 days), and IMF11 (scale of 437 days), respectively; the scale was about half (for IMF9) or one year (for IMF10 and IMF11). These scales corresponded to the period of climate variations. Therefore, variations of snow depth were determined by the vibrations of IMF9, IMF10, and IMF11. The residual line tended to increase from the 1980s, indicating a general increase of snow depth after the 1980s.

Daily snow depth at the other 104 stations in Xinjiang showed similar seasonal changes and scales but with different ranges. SD_avr was in the range of 6.3–21.3 cm for north Xinjiang, 4.2–10.8 cm for Tianshan Mountains area, 1.5–5.0 cm for south Xinjiang, and 4.8–12.8 cm for the whole of Xinjiang, respectively. SD_max was in the range of 13.1–42.5 cm for north Xinjiang, 11.8–23.6 cm for Tianshan Mountains area, 2.0–8.9 cm for south Xinjiang, and 10.3–25.4 cm for the whole of Xinjiang, respectively.

Four common types of snow depth curve (Chen et al. 2015) were classified to show daily snow depth variations within the year at Jimunai station (Figure 3). The flat peak, multi-peak, sharp single-peak, and right-skewed types, corresponding to Figure 3(a), 3(b), 3(c) and 3(d), occurred 18, 15, 10, and 9 times at Jimunai, respectively.

Curve types for daily snow depth at the other 104 stations in Xinjiang also had different patterns. The number of sites in different sub-regions over 1961–2013 for each type are shown in Figure 4. The total number of sites in north Xinjiang, Tianshan Mountains area, and south Xinjiang were 45, 17, and 43, respectively. The number of sites for the flat peak type were 11, 3, 7, and 21 in north Xinjiang, Tianshan Mountains area, south, and all of Xinjiang over the study period, respectively. The number of sites for the multi-peak type were 14, 6, 11, and 32 in north Xinjiang, Tianshan Mountains area, south, and all of Xinjiang, respectively. For the sharp single-peak type, the number of sites were 7, 1, 2, and 10 in north Xinjiang, Tianshan Mountains area, south, and all of Xinjiang, respectively. The number of sites for the right-skewed type were 12, 5, 7, and 25 in north Xinjiang, Tianshan Mountains area, south, and all of Xinjiang, respectively. In general, the variations of snow depth curve types in each sub-region fluctuated with a curve type order of multi-peak > right-skewed > flat peak > sharp single-peak.

In addition, based on the MMK trend test, the number of sites with flat peak and multi-peak types in north Xinjiang (Figure 4(a) and 4(b)) had significant increasing trends. The number of sites with sharp single-peak in north and all of Xinjiang (Figure 4(c)) had significant decreasing trends. The number of sites with multi-peak in north, south, and all of Xinjiang (Figure 4(b)), sharp single-peak in Tianshan Mountains area and south Xinjiang (Figure 4(c)), right-skewed in Tianshan Mountains area (Figure 4(d)), and flat peak in all of Xinjiang (Figure 4(a)) had insignificant increasing trends. Meanwhile, the number of sites with multi-peak in Tianshan Mountains area, right-skewed in north, south, and all of Xinjiang, and flat peak in Tianshan Mountains area and south Xinjiang had insignificant decreasing trends. These snow depth variation patterns were closely connected with temperature conditions.

Values of D_s, D_e, and D_d in different sub-regions of Xinjiang over 1961–2013 are given in Table 1. The mean, maximal, and minimal D_s values generally decreased from north to south. The mean D_e decreased from north to south, and the maximal and minimal D_e values in north Xinjiang and Tianshan Mountains area were generally larger than in south Xinjiang. D_s and D_e values in north Xinjiang were close to Tianshan Mountains area. The D_d values ranged from 94 to 133 days for north Xinjiang, from 86 to 144 days for Tianshan Mountains area, and from 15 to 55 days for south Xinjiang, respectively. The mean D_d values in the whole of Xinjiang were 96 days. The spatial distribution of D_s, D_e, and D_d had a weak variability with C_v values <0.1.

Figure 5 illustrates the temporal variations of annual D_s, D_e, and D_d in different sub-regions of north Xinjiang, Tianshan Mountains area, south and all of Xinjiang. The annual variations of D_s in north Xinjiang and Tianshan Mountains area were similar and generally earlier than south Xinjiang. The annual mean D_s in south Xinjiang was November 30th, with the latest date on December 26th, 1973 and the earliest date on November 14th, 2000. The annual D_e in north Xinjiang and Tianshan Mountains area had a similar range, both being later than south Xinjiang. The annual mean D_e in south Xinjiang was March 5th, with the latest date on March 17th, 2010 and minimum date on February 17th, 1962. Consequently, the annual mean D_d in north Xinjiang and Tianshan Mountains area was much longer than in south Xinjiang. The annual mean D_d in south Xinjiang was 96 days, with a maximum of 120 days (in 1985), and a minimum of 71 days (in 1995). Annual mean D_s declined in north Xinjiang and south Xinjiang but increased in Tianshan Mountains area. In all of Xinjiang, annual mean D_s showed a decreasing trend with a linear slope of 0.014 day/year. Annual mean D_e declined in south Xinjiang and Tianshan Mountains area while it increased in north Xinjiang, and annual mean D_e in all of Xinjiang showed a decreasing trend with a linear slope of 0.014 day/year. Annual mean D_d increased in north Xinjiang and south Xinjiang while it declined in Tianshan Mountains area, and annual D_d in all of Xinjiang had an increasing trend with a linear slope of 0.002 day/year. In general, the fluctuations of D_s, D_e, and D_d were more abrupt in Tianshan Mountains area and south Xinjiang than in north Xinjiang, although their ranges were close, which showed that snow cover in north Xinjiang was more stable. Snow covered Xinjiang for about three months or longer. By the MMK trend test, D_s, D_e, and D_d had insignificant increasing or decreasing trends. The Sen's slope (b) values ranged from −0.03 to 0.098 for D_s, from −0.056 to 0.029 for D_e, and from −0.11 to 0.03 for D_d considering different sub-regions and all of Xinjiang. Different b values for D_s and D_e resulted in various trends in D_d, although all b values were small.

The temporal variations of annual SD_avr and SD_max in different sub-regions and four typical sites in Xinjiang are illustrated in Figure 6. The curves of SD_avr decreased from north to south over the study period (Figure 6(a)), and SD_avr showed an increase trend with a linear slope of 0.093, 0.054, 0.039, and 0.007 cm year⁻¹ in north Xinjiang the whole of Xinjiang, Tianshan Mountains area, and south Xinjiang, respectively. Among the four sites, Aletai and Jimunai belonged to north Xinjiang, Tianchi belonged to Tianshan Mountains area, and Luntai belonged to south Xinjiang. The variations of SD_avr not only had regional features, but also were site-specific. SD_avr decreased from north to south and ranking for sites was Aletai > Tianchi > Jimunai > Luntai (Figure 6(b)), and SD_avr showed an increased trend with a linear slope of 0.267, 0.112, 0.365, and 0.021 cm/year in the above four stations, respectively. The ranges of SD_avr varied from 0 to 53 cm for different sites across the region. SD_max curves decreased from north to south (Figure 6(c)) and ranked as Aletai > Tianchi > Jimunai > Luntai (Figure 6(d)), similarly to SD_avr variations. At either different sub-regions or the four typical sites, winter mean SD_max had increasing trends. The ranges for SD_max were much larger than SD_avr, which was reasonable. Other studies also showed that the SD_avr in north Xinjiang was deeper than in south Xinjiang, and SD_max in north Xinjiang (Fu et al. 2011). Our results agreed with theirs.

Further, the MMK trend test results showed that both SD_avr or SD_max had increasing trends, but the trends of SD_avr in north, south, all of Xinjiang and at the sites Tianchi and Jimunai, and the trends of SD_max in sub-regions of south and all of Xinjiang and at the sites Aletai, Tianchi, and Jimunai were insignificant. The b values ranged from 0.02 to 057 decreasing from north to south and ranked as Jinunai > Aletai > Tianchi > Luntai for either SD_avr or SD_max. This consistency in increasing trends of snow depth contributed to the overall increase of precipitation in Xinjiang, which has been reported by different studies (Shen et al. 2013; Li et al. 2017a).

The MKK statistic (Z_m) and Sen's slope (b) values for D_s, D_e, and D_d were obtained for 105 sites throughout Xinjiang. The spatial distribution of multi-year mean D_s, D_e, and D_d and their trend test results are shown in Figure 7. In Figure 7(a), D_s showed clear regional differences between north and south Xinjiang with Tianshan Mountains area as a division, indicating much earlier snowfall in north than in south Xinjiang. For example, at Daxigou in Tianshan Mountains area, the earliest D_s was October 1st, while at Baluntai in south Xinjiang, the latest D_s was February 2nd. D_s ranged from October 5th to November 29th in north Xinjiang and Tianshan Mountains area, while in south Xinjiang it ranged from December 12th to February 5th, respectively. This is due to the warm temperature or less snowfall in the dry winter there. In Figure 7(b), D_s at 45 out of the total 105 sites had increasing trends, with D_s being significant at seven sites. D_s at the other 60 sites had decreasing trends, with D_s at two sites being significant. Values of b for D_s varied from −1.5 to 3.0 across Xinjiang. In Figure 7(c) and 7(e), D_e and D_d also showed clear regional differences between north and south Xinjiang. D_e ranged from February 18th to April 15th in north Xinjiang and Tianshan Mountains area while in south Xinjiang it ranged from January 9th to February 4th, respectively. Since snow started earlier and ended later in north Xinjiang than in south Xinjiang, as a consequence the D_d values in north Xinjiang were much larger than in south Xinjiang. In Figure 7(d), D_e at 34 sites had increasing trends, with D_e at six sites being significant. D_e at the other 71 sites had decreased trends, and only at one site was its trend significant. D_d ranged from 74 to 199 in north Xinjiang and Tianshan Mountains area while in south Xinjiang it ranged from 14 to 46, respectively. In Figure 7(f), trends in D_d at 53 sites increased, of which, trends at two sites were significant. Among the other 52 sites which had decreasing trends in D_d, trends at four sites were significant. Values of b varied from −1.55 to 4.0 for D_e and from −1.3 to 2.0 for D_d, respectively. D_s at the 24 sites in north Xinjiang, eight sites in Tianshan Mountains area, and 13 sites in south Xinjiang had increasing trends, while D_e at the 36 sites in north Xinjiang, eight sites in Tianshan Mountains area and 24 sites in south Xinjiang had decreasing trends. D_d at the 27 sites in north Xinjiang, seven sites in Tianshan Mountains area and 16 sites in south Xinjiang had increasing trends. Therefore, more sites in north Xinjiang had increased trends of D_s and decreased trends of D_e than in south Xinjiang. The number of sites with different trends in D_s, D_e, and D_d for different sub-regions are given in Table 2. More sites had decreasing trends (e.g., D_s in Tianshan Mountains area, south and all of Xinjiang, D_e in north, south Xinjiang, and all of Xinjiang, and D_d in north Xinjiang) than increasing trends.

The increasing D_d may induce larger snow depth for the study sites. This may produce higher snow water equivalent and high discharge in Xinjiang. In the meantime, it may cause higher snowmelt flood risks when temperature increases rapidly within a very short period in spring. Therefore, the increase of D_d in Xinjiang could be either beneficial or a threat for society and human safety.

Figure 8 shows the spatial distribution of multiyear mean monthly (in December, January, and February), annual mean SD_avr and their trends. SD_avr in all of the three months varied within 9.0 cm for south Xinjiang, but reached 17.7 cm in December, 29.3 cm in January, and 34.4 cm February for north Xinjiang, respectively. Similarly, the b values had large variations in north Xinjiang and Tianshan Mountains area but much smaller variations in south Xinjiang, especially for the monthly data. The annual SD_avr ranged between 5 and 25 cm in north Xinjiang and below 6 cm in south Xinjiang. Eighty-six sites had increasing trends in annual SD_avr, of which trends at 16 sites were significant; among the other 19 sites with decreasing trends, trends at two sites were significant. Annual SD_avr at 41 sites in north Xinjiang, 16 sites in Tianshan Mountains area and 29 sites in south Xinjiang had increasing trends. Therefore, most sites in all of Xinjiang had increasing trends in annual SD_avr (Table 2).

Figure 9 shows the spatial distribution of multiyear mean monthly SD_max, annual mean SD_max and their trends. Also similar to SD_avr, SD_max varied little in the south but greatly in north Xinjiang and Tianshan Mountains area. The values of b varied in December, January, and February, and the annual scale in all of Xinjiang. The annual SD_max ranged between 24 and 95 cm in north Xinjiang and below 33 cm in south Xinjiang, respectively. The smallest SD_avr and SD_max were in the Taklamakan desert zone. Ninety sites had increasing trends in annual SD_max, of which, trends at 18 sites were significant, while the other 15 sites had insignificant decreasing trends. SD_max at the 43 sites in north Xinjiang, 16 sites in Tianshan Mountains area and 31 sites in south Xinjiang had increasing trends. Similar to annual SD_avr, most sites in all of Xinjiang had increasing trends in annual SD_max (Table 2).

The number of sites in Table 2 corresponding to Figures 6–8 show that most sites in different sub-regions had increasing trends in the studied snow properties; many less sites had significant decreasing trends. As mentioned above, the increases in SD_avr and SD_max reflect the increase in precipitation, which is consistent with Li et al. (2017b) who showed generally increasing precipitation over 1961–2013 in Xinjiang, China. Both SD_avr and SD_max had seasonality.

The variance percentage for each EOF was obtained. EOF1, EOF2, EOF3, and EOF4 was 70.0%, 6.7%, 4.0%, and 2.4%, accounting for 83.1% of the total spatial variability, of which, only EOF1 was significant at the 95% confidence level. Although EOF2 through EOF10 were insignificant, the total variation explained by these patterns was about 20%, which implied that about one-fifth of the spatial variation in snow depth was random in time and did not belong to a temporary correlated spatial pattern. Therefore, Figure 10 shows the EOF1 and EC1 that explain the most variation. As shown in Figure 10(a), the values of EOF1 were negative for south Xinjiang but positive for north Xinjiang, indicating the differences of snow depth in space, which could be caused by the climatic conditions. Absolute values of EOF1 in north Xinjiang were larger than in south Xinjiang, which indicated that snow depth in north Xinjiang varied more. In Figure 10(b), the EC1 values fluctuated with the years and had obvious periods but no typical trends.

Spatiotemporal characteristics of snow properties, including starting and ending date of snow cover, snow depth, and snow cover duration in Xinjiang, China, were investigated. For the temporal variations, the annual mean starting date ranged from November 15th to December 27th. The annual mean ending date ranged from February 17th to March 16th. The annual snow cover duration lasted for 71 to 120 days. Both the average and maximum snow depths fluctuated and decreased from north to south. Four typical types of daily snow depth within the snow-cover year, i.e., flat peak, multi-peak, sharp single-peak, and right-skewed, were generalized. Among the 17 IMFs and residual generated by the empirical mode decomposition (EMD), about half- and one-year scales were demonstrated by IMF10, IMF11, and IMF12.

For the spatial distributions, starting (ending) date values delayed (advanced) but snow depth values decreased from north to south Xinjiang. Average and winter maximum snow depth were much larger in Tianshan Mountains area and north Xinjiang than in south Xinjiang. Starting date increased from October 5th to February 5th, while ending date, snow cover duration, average and winter maximum snow depths decreased and ranged from January to April, 14–199 days, 1.5–24.9 cm, and 4.0–95.0 cm, respectively. The MMK test showed that, starting date at 60 sites had decreasing trends, of which, trends at two sites were significant; moreover, seven out of the other 45 sites for starting date had significant increasing trends. Trends of ending date at 71 and 34 sites decreased and increased, respectively, of which, trends at one and six sites were significant, respectively. Fifty-two and 53 sites had decreasing and increasing trends in snow cover duration, of which, trends at four and two sites were significant, respectively. Nineteen (86) sites had decreasing (increasing) trends in annual average snow depth, of which, trends at three and 22 sites were significant, respectively. As to winter maximum snow depth, 15 and 90 sites had decreasing and increasing trends, of which, trends at none and 25 sites were significant, respectively. The first empirical orthogonal function (EOF1) accounted for 79% of the spatial variability in snow depth, of which, larger variability was shown for north than south Xinjiang.

Seasonality of the monthly average and maximum snow depths were shown by comparing data series in December, January, and February. Further studies for snow density are needed in future research, to clearly reveal how snow-water equivalent varies in Xinjiang, China.

This research was financially supported by China National Science Foundations (No. U1203182 and 51579213), and the China 111 project (B12007). We thank the three anonymous reviewers who provided many helpful comments which helped improve the manuscript.