Daily snow observation data from 672 stations in China, particularly the 296
stations with over 10 mean snow cover days (SCDs) in a year during the
period of 1952–2010, are used in this study. We first examine spatiotemporal
variations and trends of SCDs, snow cover onset date (SCOD), and snow cover
end date (SCED). We then investigate the relationships of SCDs with number of days
with temperature below 0
Snow has a profound impact on the surficial and atmospheric thermal conditions, and is very sensitive to climatic and environmental changes, because of its high reflectivity, low thermal conductivity, and hydrological effects via snowmelt (Barnett et al., 1989; Groisman et al., 1994). The extent of snow cover in the Northern Hemisphere has decreased significantly over the past decades because of global warming (Robinson and Dewey, 1990; Brown and Robinson, 2011). Snow cover showed the largest decrease in the spring, and the decrease rate increased for higher latitudes in response to larger albedo feedback (Déry and Brown, 2007). In North America, snow depth in central Canada showed the greatest decrease (Dyer and Mote, 2006), and snowpack in the Rocky Mountains in the United States declined (Pederson et al., 2013). However, in situ data showed a significant increase in snow accumulation in winter but a shorter snowmelt season over Eurasia (Bulygina et al., 2009). Decrease in snowpack has also been found in the European Alps in the last 20 years of the twentieth century (Scherrer et al., 2004), but a very long time series of snowpack suggests large decadal variability and overall weak long-term trends only (Scherrer et al., 2013). Meteorological data indicated that the snow cover over northwest China exhibited a weak upward trend in snow depth (Qin et al., 2006), with large spatiotemporal variations (Ke et al., 2009; Ma and Qin, 2012). Simulation experiments using climate models indicated that, with continuing global warming, the snow cover in China would show more variations in space and time than ever before (Shi et al., 2011; Ji and Kang, 2013). Spatiotemporal variations of snow cover are also manifested as snowstorms or blizzards, particularly excessive snowfall over a short time duration (Bolsenga and Norton, 1992; Liang et al., 2008; Gao, 2009; Wang et al., 2013; Llasat et al., 2014).
Total snow cover days in a year (SCDs hereafter) is an important index that represents the environmental features of climate (Ye and Ellison, 2003; Scherrer et al., 2004), and is directly related to the radiation and heat balance of the Earth–atmosphere system. The SCDs vary in space and time and contribute to climate change over short timescales (Zhang, 2005), especially in the Northern Hemisphere. Bulygina et al. (2009) investigated the linear trends of SCDs observed at 820 stations from 1966 to 2007, and indicated that the duration of snow cover decreased in the northern regions of European Russia and in the mountainous regions of southern Siberia, while it increased in Yakutia and the Far East. Peng et al. (2013) analysed trends in the snow cover onset date (SCOD) and snow cover end date (SCED) in relation to temperature over the past 27 years (1980–2006) from over 636 meteorological stations in the Northern Hemisphere. They found that the SCED remained stable over North America, whereas there was an early SCED over Eurasia. Satellite-derived snow data indicated that the average snow season duration over the Northern Hemisphere decreased at a rate of 5.3 days per decade between 1972/1973 and 2007/2008 (Choi et al., 2010). Their results also showed that a major change in the trend of snow duration occurred in the late 1980s, especially in western Europe, central and East Asia, and mountainous regions in western United States.
There are large spatiotemporal differences in the SCDs in China (Wang and Li, 2012). Analysis of 40 meteorological stations from 1971 to 2010 indicated that the SCDs had a significant decreasing trend in the western and south-eastern Tibetan Plateau, with the largest decline observed in Nielamu, reaching 9.2 days per decade (Tang et al., 2012). Data analysis also indicated that the SCDs had a linear decreasing trend at most stations in the Hetao region and its vicinity (Xi et al., 2009). However, analysis of meteorological station data in Xinjiang showed that the SCDs had a slight increasing trend, occurring mainly in 1960–1980 (Q. Wang et al., 2009). Li et al. (2009) analysed meteorological data from 80 stations in Heilongjiang province, Northeast China. Their results showed that the snow cover duration shortened, because of both the late SCOD (by 1.9 days per decade) and early SCED (by 1.6 days per decade), which took place mainly in the lower altitude plains.
The SCDs are sensitive to local winter temperature and precipitation, latitude (Hantel et al., 2000; C. Wang et al., 2009; Serquet et al., 2011; Morán-Tejeda et al., 2013), and altitudinal gradient and terrain roughness (Lehning et al., 2011; Ke and Liu, 2014). Essentially, the variation in SCDs is mainly attributed to large-scale atmospheric circulation or climatic forcing (Beniston, 1997; Scherrer and Appenzeller, 2006; Ma and Qin, 2012; Birsan and Dumitrescu, 2014), such as monsoons, the El Niño–Southern Oscillation, the North Atlantic Oscillation, and the Arctic Oscillation (AO). Xu et al. (2010) investigated the relationship between the SCDs and monsoon index in the Tibetan Plateau and their results indicated that there were great spatial differences. As an index of the dominant pattern of non-seasonal sea-level pressure variations, the AO shows a large impact on the winter weather patterns of the Northern Hemisphere (Thompson and Wallace, 1998; Thompson et al., 2000; Gong et al., 2001; Wu and Wang, 2002; Jeong and Ho, 2005). The inter-annual variation of winter extreme cold days in the northern part of eastern China is closely linked to the AO (Chen et al., 2013). Certainly, the AO plays an important role in the variation of SCDs. An increase in the SCDs before 1990 and a decrease after 1990 have been reported in the Tibetan Plateau, and snow duration has positive correlations with the winter AO index (You et al., 2011), and a significant correlation between the AO and snowfall over the Tibetan Plateau on an inter-decadal timescale was also reported by Lü et al. (2008).
The focus of this study is the variability in the snow cover phenology in
China. A longer time series of daily observations of snow cover is used for
these spatial and temporal analyses. We first characterise the spatial
patterns of change in the SCDs, SCOD, and SCED in different regions of
China; we then examine the sensitivity of SCDs to the number of days with
temperature below 0
Locations of weather stations and major basins, mountains, and plains mentioned in the paper, overlying the digital elevation model for China.
We use daily snow cover and temperature data in China from the 1 September 1951 to the 31 August 2010, provided by the National Meteorological Information Centre of China Meteorological Administration (CMA). According to the Specifications for Surface Meteorological Observations (China Meteorological Administration, 2003), an SCD is defined as a day when the snow cover in the area meets the following requirement: at least half of the observation field is covered by snow. For any day with at least half of the observation field covered by snow, snow depth is recorded as a rounded-up integer. For example, a normal SCD is recorded if the snow depth is equal to or more than 1.0 cm (measured with a ruler), or a thin SCD if the snow depth is less than 1.0 cm. A snow year is defined as the time period from 1 September of the previous year to 31 August of the current year. For instance, September, October, and November 2009 are treated as the autumn season of snow year 2010, December 2009 and January and February 2010 as the winter season of snow year 2010, and March, April, and May 2010 as the spring season of snow year 2010.
Station density is high in eastern China, where the observational data for most stations are complete, with relatively long histories (as long as 59 years), while station density is low in western China, and the observation history is relatively short, although two of the three major snow regions are located in western China. If all stations with short time series are eliminated, the spatial representativeness of the data set would be a problem. Therefore, a time series of at least 30 years is included in this study.
Because of topography and climate conditions, the discontinuous nature of snowfall is obvious in western China, especially in the Tibetan Plateau, with patchy snow cover, and there are many thin SCD records (Ke and Li, 1998). However, in order to enhance data reliability, according to the previous studies (An et al., 2009; Wang and Li, 2012), thin SCDs in the original data set are not taken into account in this paper.
Totally, there are 722 stations in the original data set. Since station relocation and changes in the ambient environment could cause inconsistencies in the recorded data, we implement strict quality controls (such as inspection for logic, consistency, and uniformity) on the observational data sets in order to reduce errors (Ren et al., 2005). The standard normal homogeneity test (Alexandersson and Moberg, 1997) at the 95 % confidence level is applied to the SCDs and temperature series data in order to identify possible breakpoints. Time series gap filling is performed after all inhomogeneities are eliminated, using nearest neighbour interpolation. After being processed as mentioned above, the 672 stations with annual mean SCDs greater than 1 (day) are finally selected for subsequent investigation (Fig. 1).
The observation period for each station is different, varying between 59 years (1951/1952–2009/2010) and 30 years (1980/1981–2009/2010). Overall, 588 stations have observation records between 50 and 59 years, 47 stations between 40 and 49 years, and 37 stations between 30 and 39 years (Fig. 2). Most of the stations with observation records of less than 50 years are located in remote or high-elevation areas. All 672 stations are used to analyse the spatiotemporal distribution of SCDs in China, while only 296 stations with more than 10 annual mean SCDs are used to study the changes of the relationships of SCOD, SCED, and SCDs with TBZD, MAT, and the AO index.
The daily AO index constructed by projecting the daily (00Z) 1000 mb height
anomalies poleward of 20
A digital elevation model from the Shuttle Radar Topographic Mission (SRTM,
Percentage of weather stations with different measurement lengths.
We apply a Mann–Kendall (MK) test to analyse the trends of SCDs, SCOD, and SCED. The MK test is an effective tool to extract the trends of time series, and is widely applied to the analysis of climate series (Marty, 2008). The MK test is characterised as being more objective, since it is a non-parametric test. A positive standardised MK statistic value indicates an upward or increasing trend, while a negative value demonstrates a downward or decreasing trend. Confidence levels of 90 and 95 % are taken as thresholds to classify the significance of positive and negative trends of SCDs, SCOD, and SCED.
At the same time, if SCDs, SCOD, or SCED at one climate station has a significant MK trend (above 90 %), their linear regression analyses are
performed against time, respectively. The slopes of the regressions represent
the changing trends and are expressed in days per decade. The statistical
significance of the slope for each of the linear regressions is assessed by
the Student's
Correlation analysis is used to examine the SCDs relationships with the
TBZD, MAT, and the AO index, and the Pearson product-moment correlation
coefficients (PPMCCs) have been calculated. The PPMCC is a widely used
estimator for describing the spatial dependence of rainfall processes, and
it indicates the strength of the linear covariance between two variables
(Habib et al., 2001; Ciach and Krajewski, 2006). The statistical
significance of the correlation coefficients is calculated using the
Student's
The spatial distribution of SCDs, SCOD, and SCED, and their calculated results, are spatially interpolated by applying the ordinary Kriging method.
Prediction errors of cross-validation for the spatial interpolation with the ordinary Kriging method (the unit is day for snow cover days (SCDs), snow cover onset day (SCOD), and snow cover end day (SCED); there is no unit for the coefficient of variation (CV)).
Annual mean snow cover days (SCDs) from 1980/1981 to
2009/2010
Seasonal variation of SCDs; the number in the centre denotes annual mean SCDs, the blue colour in the circle the SCDs represents the winter season, the green colour spring, and the red colour autumn.
SCDs anomalies in 1957
All mean errors are near zero, all average standard errors are close to the corresponding root mean squared errors, and all root mean squared standardised errors are close to 1 (Table 1). Prediction errors are unbiased and valid, except for slightly overestimated coefficients of variation (CVs) and slightly underestimated SCDs in 2002. Overall, the interpolation results have small errors and are acceptable.
The analysis of observations from 672 stations indicates that there are three major stable snow regions with more than 60 annual mean SCDs (Li, 1990): Northeast China, north Xinjiang, and the Tibetan Plateau, with Northeast China being the largest of the three (Fig. 3a). In the Daxinganling, Xiaoxingganling, and Changbai Mountains of Northeast China, there are more than 90 annual mean SCDs, corresponding to a relatively long snow season. The longest annual mean SCDs, 163 days, is at Arxan Station (in the Daxinganling Mountains) in Inner Mongolia. In north Xinjiang, the SCDs are relatively long in the Tianshan and Altun Mountains, followed by the Junggar Basin. The annual mean SCDs in the Himalayas, Nyainqentanglha, Tanggula Mountains, Bayan Har Mountains, Anemaqen Mountains, and Qilian Mountains of the Tibetan Plateau are relatively long, although most of these regions have fewer than 60 annual SCDs. The Tibetan Plateau has a high elevation, a cold climate, and many glaciers, but its mean SCDs are not as large as those of the other two stable snow regions.
Areas with SCDs of 10–60 per year are called unstable snow regions with
annual periodicity (definite snow cover every winter) (Li, 1990). It includes
the peripheral parts of the three major stable snow regions, Loess Plateau,
Northeast Plain, North China Plain, Shandong peninsula, and regions north of
the Qinling–Huaihe line (along the Qinling Mountains and Huaihe River to the
east). Areas with SCDs of 1–10 per year are called unstable snow regions
without annual periodicity (the mountainous regions are excluded) (Li, 1990).
It includes the Qaidam Basin, the Badain Jaran desert, the peripheral parts
of Sichuan Basin, the northeast part of the Yungui Plateau, and the middle
and lower Yangtze River Plain. Areas with occasional snow and mean annual
SCDs of less than 1.0 (day) are distributed north of the Sichuan Basin and in
the belt along Kunming, Nanling Mountains, and Fuzhou (approximate latitude
of 25
The spatial distribution pattern of SCDs based on climate data with longer time series is similar to previous studies (Li and Mi, 1983; Li, 1990; Liu et al., 2012; C. Wang et al., 2009; Wang and Li, 2012). Snow distribution is closely linked to latitude and elevation, and is generally consistent with the climate zones (Lehning et al., 2011; Ke and Liu, 2014). There are relatively more SCDs in Northeast China and north Xinjiang, and fewer SCDs to the south (Fig. 3a). In the Tibetan Plateau, located in south-western China, the elevation is higher than eastern areas at the same latitude, and the SCDs are greater than in eastern China (Tang et al., 2012). The amount of precipitation also plays a critical role in determining the SCDs (Hantel et al., 2000). In the north-eastern coastal areas of China, which are affected considerably by the ocean, there is much precipitation. In north Xinjiang, which has a typical continental (inland) climate, the precipitation is less than in Northeast China, and there are more SCDs in the north of Northeast China than in north Xinjiang (Dong et al., 2004; Q. Wang et al., 2009). Moreover, the local topography has a relatively large impact on the SCDs (Lehning et al., 2011). The Tarim Basin is located inland, with relatively little precipitation, thus snowfall there is extremely rare except in the surrounding mountains (Li, 1993). The Sichuan Basin is surrounded by high mountains, therefore situated in the precipitation shadow in winter, resulting in fewer SCDs (Li and Mi, 1983; Li, 1990).
The three major stable snow regions, Northeast China, north Xinjiang, and the eastern Tibetan Plateau, have smaller CVs in the SCDs (Fig. 3b). Nevertheless, the SCDs in arid or semi-arid regions, such as South Xinjiang, the northern and south-western Tibetan Plateau, and central and western Inner Mongolia, show large fluctuations because there is little precipitation during the cold seasons, and certainly little snowfall and large CVs of SCDs. In particular, the Taklimakan Desert in the Tarim Basin is an extremely arid region, with only occasional snowfall. Therefore, it has a very large range of fluctuations of SCDs. Additionally, the middle and lower Yangtze River Plain also has large SCDs fluctuations because of warm-temperate or subtropic climate with a short winter and little snowfall. Generally, the fewer the SCDs, the larger the CV (C. Wang et al., 2009). This is consistent with other climate variables, such as precipitation (Yang et al., 2015).
Seasonal variation of SCDs is primarily controlled by temperature and
precipitation (Hantel et al., 2000; Scherrer et al., 2004; Liu et al., 2012).
In north Xinjiang and Northeast China, snow is primarily concentrated in the
winter (Fig. 4). In these regions, the SCDs exhibit a single-peak
distribution. In the Tibetan Plateau, however, the seasonal variation of SCDs
is slightly different, i.e. more snow in the spring and autumn combined than
in the winter. The mean temperature and precipitation at Dangxiong station
(30
Percentage (%) of stations with anomalies (P for
positive and N for negative) of snow cover days (SCDs) in a year, snow cover
onset date (SCOD), and snow cover end date (SCED). Percentage (%) of
stations with anomalies of SCDs, SCOD, and SCED larger (smaller) than the
mean
The temporal variation of SCDs shows very large differences from 1 year to
another. We define a year with a positive (negative) anomaly of SCDs in the
following way: for a given year, if 70 % of the stations have a positive
(negative) anomaly and 30 % of the stations have SCDs larger (smaller)
than the mean
Years with a negative anomaly of SCDs include 1953, 1965, 1999, 2002, and 2009 (Table 2). If there is too little snowfall in a specific year, a drought is possible. Drought resulting from little snowfall in the cold season is a slow process and can sometimes cause serious damages. For example, East China displayed an apparent negative anomaly of SCDs in 2002 (Fig. 5b), and had very little snowfall, leading to an extreme winter drought in Northeast China, where snowfall is the primary form of winter precipitation (Fang et al., 2014).
Because of different atmospheric circulation backgrounds, vapour sources, and topographic conditions in different regions of China, there are great differences in the SCDs, even in 1 year. For example, in 2008, there were more SCDs and longer snow duration in the Yangtze River Basin, North China, and the Tianshan Mountains in Xinjiang (Fig. 5c), especially in the Yangtze River Basin, where large snowfall was normally not observed. However, four episodes of severe and persistent snow, extreme low temperatures, and freezing weather occurred in 2008 and led to a large-scale snowstorm in this region (Gao, 2009). As reported by the Ministry of Civil Affairs of China, the 2008 snowstorm killed 107 people and caused losses of USD 15.45 billion. Both the SCDs and scale of economic damage broke records from the past 5 decades (Wang et al., 2008). On the contrary, there was no snowstorm in north Xinjiang, the Tibetan Plateau, and Pan-Bohai Bay region in 2008. Moreover, Northeast China had an apparent negative anomaly of SCDs (Fig. 5c).
There are great differences in the temporal variations of SCDs, even in the
three major stable snow regions. If we redefine a year with a positive
(negative) anomaly of SCDs using a much higher standard (i.e. 80 % of stations
have a positive (negative) anomaly and 40 % of stations have an SCDs larger
(smaller) than the mean
Years with a positive anomaly of SCDs in the Tibetan Plateau include 1983 and 1990, whereas years with a negative anomaly of SCDs include 1965, 1969, and 2010 (Table 3). The climate in the Tibetan Plateau is affected by the Indian monsoon from the south, westerlies from the west, and the East Asian monsoon from the east (Yao et al., 2012). Therefore, there is a spatial difference in the SCDs within the Tibetan Plateau, and a difference in the spatiotemporal distribution of snowstorms (Wang et al., 2013). Our results differ from the conclusions drawn by Dong et al. (2001), as they only used data from 26 stations, covering only a short period (1967–1996).
The same as Table 2, but only for the years with a positive (negative) anomaly of SCDs and only for the three major stable snow regions: Northeast China (78 stations), north Xinjiang (21 stations), and the Tibetan Plateau (63 stations).
Significance of trends according to the Mann–Kendall test of SCDs, SCOD, and SCED, significance of relationships among SCDs, SCOD, SCED, respectively, with TBZD, significance of relationship between SCDs and MAT, and significance of relationship between SCDs and AO (296 stations in total). All of them have two significance levels, 90 and 95 %.
Note: I
Changing trends of annual SCDs are examined, as shown in Fig. 6a, and summarised in Table 4. Among the 296 stations, there are 35 stations (12 %) with a significant negative trend, and 37 stations (13 %) with a significant positive trend (both at the 90 % level), while 75 % of stations show no significant trends. The SCDs exhibit a significant downward trend in the Xiaoxingganling, the Changbai Mountains, the Shandong peninsula, the Qilian Mountains, the North Tianshan Mountains, and the peripheral zones in the south and eastern Tibetan Plateau (Fig. 6a). For example, the SCDs decreased by 50 days from 1955 to 2010 at the Kuandian station in Northeast China, 28 days from 1954 to 2010 at the Hongliuhe station in Xinjiang, and 10 days from 1958 to 2010 at the Gangcha station on the Tibetan Plateau (Fig. 7a–c).
Significance of trends according to the Mann–Kendall test of
SCDs
Variations in SCDs at Kuandian (40
Spatial distribution of SCOD
The SCDs in the Bayan Har Mountains, the Anemaqen Mountains, the Inner Mongolia Plateau, and the Northeast Plain, exhibit a significant upward trend (Fig. 6a). For example, at the Shiqu station on the eastern border of the Tibetan Plateau, the SCDs increased 26 days from 1960 to 2010 (Fig. 7d). The coexistence of negative and positive trends in the change of SCDs was also reported by Bulygina et al. (2009) and Wang and Li (2012).
The SCOD is closely related to both latitude and elevation (Fig. 8a). For
example, snowfall begins in September on the Tibetan Plateau, in early or
middle October on the Daxinganling, and in middle or late October on the
Altai Mountains in Xinjiang. The SCOD also varies from one year to another
(Table 2). Using the definition of a year with a positive (negative)
anomaly of SCDs, as introduced before (i.e. 70 % stations with positive (negative)
SCOD anomaly and 30 % stations with SCOD larger (smaller) than the mean
There are 196 stations (66 %) with a significant trend of late SCOD, and eight stations (3 %) with a significant trend of early SCOD (both at the 90 % level), while 31 % of the stations show no significant trends (Table 4). The SCOD in the major snow regions in China exhibits a significant trend towards late SCOD (Fig. 6b). These significantly late trends dominate the major snow regions in China. In particular, the late SCOD in Northeast China is consistent with a previous study (Li et al., 2009). Only the SCOD in the east Liaoning Bay region exhibits a significant trend towards early SCOD. For example, the SCOD at the Pingliang station in Gansu province shows a late rate of 5.2 days per decade from 1952 to 2010, but the SCOD at the Weichang station in Hebei province shows an early rate of 5.2 days per decade from 1952 to 2010 (Fig. 7e–f).
SCDs relationships with TBZD at Chengshantou
(37
The pattern of SCED is similar to that of SCOD (Fig. 8b), i.e. places with early snowfall normally show late snowmelt, while places with late snowfall normally show early snowmelt. Like the SCOD, temporal variations of SCED are large (Table 2). Using the same standard for defining the SCOD anomaly, we judge a given year as a late (early) SCED year. Three years, 1957, 1976 and 1979, can be considered as late SCED years on a large scale (Table 2). It is evident that 1957 was a typical year whose SCED was late, which was also the reason for the great SCDs (Fig. 5a and e). The SCED in 1997 was early for almost all of China except for the Tibetan Plateau, western Tianshan Mountains, and western Liaoning (Fig. 5f).
For the SCED, there are 103 stations (35 %) with a significantly early trend (at the 90 % level), while 64 % of stations show no significant trends (Table 4). The major snow regions in China all show early SCED, significant for Northeast China, north Xinjiang, and the Tibetan Plateau (Fig. 6c). The tendency of late SCED is limited, with only three stations (1 %) showing a significant trend. For example, the SCED at the Jixi station in Northeast China shows an early rate of 3.5 days per decade from 1952 to 2010, while the SCED at the Maerkang station in Sichuan province shows a late rate of 4.2 days per decade from 1954 to 2010 (Fig. 7g–h).
In the context of global warming, 196 stations (66 %) show significantly late SCOD, and 103 stations (35 %) show significantly early SCED, all at the 90 % confidence level. It is not necessary for one station to show both significantly late SCOD and early SCED. This explains why only 12 % of stations show a significantly negative SCDs trend, while 75 % of stations show no significant change in the trends of SCDs. The latter is inconsistent with the overall shortening of the snow period in the Northern Hemisphere reported by Choi et al. (2010). One reason could be the different time periods used in the two studies, 1972–2007 in Choi et al. (2010) as compared to 1952–2010 in this study. Below, we discuss the possible connections between the spatiotemporal variations of snow cover and the warming climate and changing AO.
The number of days with temperature below 0
For the SCOD, there are 245 stations with negative correlations with TBZD, accounting for 83 % of 296 stations, whereas only 51 stations (17 %) show positive correlations (Table 4). This means that for smaller TBZD, the SCOD is later. For the SCED, there are 269 stations with positive correlations, accounting for 91 % of 296 stations, whereas only 27 stations (9 %) have negative correlations. This means that for smaller TBZD, the SCED is earlier.
Very similar results are found for the MAT (Table 4, Fig. 6e), and Fig. 9b shows an example (the Tieli station).
Although the AO index has shown a strong positive trend in the past decades (Thompson et al., 2000), its impact on the SCDs in China is spatially distinctive. Positive correlations (46 % of 296 stations) are found in the eastern Tibetan Plateau and the Loess Plateau (Table 4, Fig. 6f), and Fig. 9c shows an example (the Huajialing station). Negative correlations (54 % of 296 stations) exist in north Xinjiang, Northeast China, and the Shandong peninsula, and Fig. 9d shows an example (the Tonghua station).
This study examines the snow cover change based on 672 stations in 1952–2010 in China. Specifically, the 296 stations with more than 10 annual mean SCDs are used to study the changing trends of SCDs, SCOD, and SCED, and SCD relationships with TBZD, MAT, and AO index during snow seasons. Some important results are summarised below.
Northeast China, north Xinjiang, and the Tibetan Plateau are the three major snow regions. The overall inter-annual variability of SCDs is large in China. The years with a positive anomaly of SCDs in China include 1955, 1957, 1964, and 2010, while the years with a negative anomaly of SCDs are 1953, 1965, 1999, 2002, and 2009. Only 12 % of stations show a significantly negative SCDs trend, while 75 % of stations show no significant SCDs trends. Our analyses indicate that the distribution pattern and trends of SCDs in China are very complex and are not controlled by any single climate variable examined (i.e. TBZD, MAT, or AO), but by a combination of multiple variables.
It is found that significantly late SCOD occurs in nearly the whole of China except for the east Liaoning Bay region; significantly early SCED occurs in nearly all major snow regions in China. Both the SCOD and SCED are closely related to the TBZD and MAT, and are mostly controlled by local latitude and elevation. Owing to global warming since the 1950s, the reduced TBZD and increased MAT are the main reasons for overall late SCOD and early SCED, although it is not necessary for one station to experience both significantly late SCOD and early SCED. This explains why only 12 % of stations show significantly negative trends in SCDs, while 75 % of stations show no significant SCDs trends.
Long-duration, consistent records of snow cover and depth are rare in China because of many challenges associated with taking accurate and representative measurements, especially in western China; the station density and metric choice also vary with time and locality. Therefore, more accurate and reliable observation data are needed to further analyse the spatiotemporal distribution and features of snow cover phenology. Atmospheric circulation causes variability in the snow cover phenology, and its effect requires deeper investigations.
This work is financially supported by the Program for National Nature Science Foundation of China (no. 41371391), and the Program for the Specialized Research Fund for the Doctoral Program of Higher Education of China (no. 20120091110017). This work is also partially supported by the Collaborative Innovation Center of Novel Software Technology and Industrialization. We would like to thank the National Climate Center of China (NCC) in Beijing for providing valuable climate data sets. We thank the three anonymous reviewers and the editor for valuable comments and suggestions that greatly improved the quality of this paper.Edited by: H.-J. Hendricks Franssen