A study is presented of the geographical distribution and spatial and temporal variabilities of the western China snow cover in the past 47 yr between 1951 and 1997. The data used consist of Scanning Multichannel Microwave Radiometer (SMMR) 6-day snow-depth charts, NOAA weekly snow extent charts, and the daily snow depth and number of snow cover days from 106 selected meteorological stations across western China. Empirical orthogonal function was performed on the SMMR dataset to better understand the spatial pattern and variability of the Qinghai–Xizang (Tibet) snow cover. A multiple linear regression analysis was conducted to show the association of interannual variations between snow cover and snow season temperature as well as precipitation. Further, the autoregressive moving average model was fitted to the snow and climate time series to test for their long-term trends. Results show that western China did not experience a continual decrease in snow cover during the great warming period of the 1980s and 1990s. It is of interest to note that no correlation was identified between temperature and precipitation in the snow cover season. However, year-to-year fluctuation of snow cover responds to both snowfall and snow season temperature. About one-half to two-thirds of the total variance in snow cover is explained by the linear variations of snowfall and snow season temperature. The long-term variability of western China snow cover is characterized by a large interannual variation superimposed on a small increase trend. The positive trend of the western China snow cover is consistent with increasing snowfall, but is in contradiction to regional warming. In addition, many constraints of the Qinghai–Xizang (Tibet) snow cover force the author’s challenge of Blanford’s hypothesis.
Snow cover is a vital water resource in western China. The largest rivers of China, such as the Yangtze River, Yellow River, etc., have their headwaters there. Agriculture and animal husbandry rely heavily on snowmelt water to be sustained. Crop failure and harvest have traditionally been tied strictly to the winter snow storage. Spring drought caused by snow scarcity represents potentially the most serious impact to agriculture and the ecosystem. Sometimes it even results in flow break off of the Yellow River. On the other hand, heavy snowstorms often bring disaster to animal husbandry in the Qinghai–Xizang (Tibet) Plateau, Xinjiang, and Inner Mongolia. In the context of global warming, changes in snow cover take on great significance and have clear economic impacts in western China.
The majority of the climatic community is convinced of a pronounced reduction in seasonal snow cover in response to CO2-induced global warming (Watson et al. 1996; Robinson and Dewey 1990; Groisman et al. 1994; Brown et al. 1996; Aizen et al. 1997). However, there are important regional exceptions (Moore et al. 2002; Ye et al. 1998; Davis et al. 1998; Mosley-Thompson et al. 1999; Vaughan et al. 1999; Ohmura et al. 1996; Li 1995). Many studies have shown that higher snowfall is a characteristic of a warming climate in cold regions (Houghton et al. 1996; Karl et al. 1993; Leathers et al. 1993). Up to now global snow cover monitoring has not found any convincing evidence of the trend in snow cover variations on global scale. How snow cover will react to global warming is presently a controversial issue.
South and East Asia experience the monsoon climate that undergoes high-amplitude variability (Webster et al. 1998). The effect of the Qinghai–Xizang (Tibet) snow cover on the Asian monsoon is one intriguing issue for climatologists (Blanford 1884; Walker 1910). Despite diagnostic and modeling investigations that have ascribed importance to the Qinghai–Xizang (Tibet) snow cover (Hahn and Shukla 1976; Barnett et al. 1988, 1989; Yasunari et al. 1991; Vernekar et al. 1995), the efforts to support Blanford’s hypothesis have left nothing but variant conclusions and sharp contrasts (Zwiers 1993) for lack of the ground truth of snow cover distribution and variability over the Qinghai–Xizang (Tibet) Plateau. So far, the open questions about whether an apparent or a weak correlation (Li 1994), a negative correlation or a positive correlation (Bamzai and Shukla 1999), is in existence between the Qinghai–Xizang (Tibet) snow cover and the Indian monsoon rainfall, and whether the albedo effect or hydrological effect of the Qinghai–Xizang (Tibet) snow cover is the key mechanism affecting Asian monsoon development, still elude us. Detailed and accurate information of Qinghai–Xizang (Tibet) snow cover is still essential to test Blanford’s hypothesis.
The study area extends over the latitude–longitude domain from 27° to 50°N, and from 70° to 105°E. Western China is physiographically divided into two regions: the Qinghai–Xizang (Tibet) Plateau and northwestern China. The former, with an average elevation exceeding 4500 m ASL and area of more than 2 × 106 km2, was acclaimed as the “Roof of the World.” The Himalayas, the world highest mountains, provide a natural screen in the southern frontier of the plateau. The latter is an arid region of China and is roughly encompassed by high mountains and large basins, such as the Altai, Tianshan, Pamirs, Karakoram, and Kunlun Mountains; and the Tarim and Junggar basins.
a. SMMR 6-day snow-depth data and their adjustment over western China
Assessment of spatial distribution and seasonal progress in snow mass requires reliable snow-depth data with a high spatial resolution and covering a sufficient length of time. The microwave snow estimates have been recognized as an efficient means of the large-scale mapping of snow depth and snow-water equivalent (SWE) with a high spatial density (Konig et al. 2001). The intensity of microwave radiation thermally emitted by snow cover is measured and expressed as brightness temperature. The Scanning Multichannel Microwave Radiometer (SMMR), on board the Nimbus-7, is a five-frequency dual-polarized microwave radiometer. The 37-GHz frequency with a 25-km spatial resolution was the sensor used for snow cover observation. The brightness temperature was spatially averaged for each 0.5° latitude × 0.5° longitude grid cell and was retrieved for snow depth by the snow parameter retrieval algorithm developed by Chang et al. (1987). The world SMMR snow-depth charts were compiled for 6-day periods from 1978 to 1987 on a continuous basis and in a consistent manner. However, employing a single global algorithm to extract snow depth from SMMR output significantly overestimates the snow cover area in the Qinghai–Xizang (Tibet) Plateau (Robinson et al. 1984) because snow cover is predominantly shallow and patchy, and is frequently of short duration. To calibrate SMMR estimates Chang et al. (1992) thoroughly compared SMMR data with Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS) shortwave images and daily snow depths reported by 175 weather stations over western China. Then, western China–specific retrieval algorithms were developed under the support of the Geographic Information System (GIS) to account for the effect of the atmospheric conditions and snow cover extent adjustment for shallow and patchy snow area. The algorithms were expressed as
for a plateau,
for high mountains, and
for rolling hills and basins. Here SD is the snow depth in centimeters; T18H and T37H are the horizontally polarized brightness temperature (K) for the SMMR 18- and 37-GHz radiometers, respectively.
In this study, National Aeronautics and Space Administration (NASA) 6-day SMMR snow-depth data during the period between 1978 and 1987 have been adjusted by using the western China algorithms.
b. NOAA weekly snow extent charts
Weekly snow extent charts produced by the National Oceanic and Atmospheric Administration (NOAA) from visible-band satellite imagery are the supplementary source of information for the Qinghai–Xizang (Tibet) Plateau snow cover investigation. The charting improved considerably in 1972 with deployment of the AVHRR sensor and was digitized on an 89 cell × 89 cell Northern Hemisphere grid with spatial resolution ranging from 16 000 to 42 000 km2. Presently, NOAA weekly snow charts constitute the longest satellite-derived snow cover dataset available on a continuous basis and produced operationally. However, they are limited by coarse resolution for the regional-scale study and by cloudiness, which frequently obscures portions of the plateau. Because the Tibetan Plateau has been one of the most difficult areas for snow cover monitoring, an efficient method by which to monitor the plateau snow cover is to utilize comparatively various satellite-derived snow datasets and station snow cover data. To minimize discrepancies between datasets, NOAA weekly snow extent charts covering the period from 1972 to 1989 were used to generate the plateau snow area time series for comparison with SMMR, as well as station-created snow time series.
c. In situ snow cover
To assess snow mass variability, particularly in light of long-term variations and trends, the satellite data are still far from sufficient length. Ground station data could provide for longer time series generation. The meteorological network of western China consists of more than 200 synoptic stations. They report snow depth daily and the number of snow cover days monthly. Snow density is measured at primary stations only every 5 days when snow depth ≥ 5cm. The data span the period from 1951 to the present. Considerable quality problems are inadequate spatial coverage and varying length. The former case does not refer to the irregularity of station spatial distribution, but to high mountainous areas where snow cover is heavy and is affected by large interannual variability, and the station is geographically sparse or virtually absent. For instance, over the Qinghai–Xizang (Tibet) Plateau the limited stations tend to be in an inhabited river valley over the eastern plateau. In addition, it was only after 1956 that the Qinghai–Xizang (Tibet) network became the very least bit dense to ensure any adequate spatial coverage, except for western Tibet (Fig. 1). To minimize the defects of station data two subsets of the network were created in northwestern China and the Qinghai–Xizang (Tibet) Plateau, respectively. The former consists of 46 stations, of which 38 were selected in such a way that only 1 station was chosen from each grid cell of 2° latitude × 2° longitude. An additional eight high-elevation (2000–4000 m ASL) stations were added to account for orographic effects on snow cover distribution and to improve data coverage in high mountains. The station selection criteria included availability of a longest time series, with few missing records and without site relocation. The station records were used without further adjustments, and no attempt was made to fill in a few missing data (Balling and Idso 2002). After the strict quality control the station point records were integrated over the snow cover year (from September to the following August) and space to derive regional time series. With regard to large-scale area averages, biases and errors associated with specific point data were further minimized to such an extent that a homogeneous and meaningful signal could be extracted. Over the Qinghai–Xizang (Tibet) Plateau the station network consisted of 60 primary stations, of which 35 are located in Qinghai Province and only 25 are in the Xizang (Tibet) Autonomous Region.
d. Climate data
Monthly temperature and precipitation from 1957 to 1992 for 60 stations over the Qinghai–Xizang Plateau, and from 1951 to 1997 for 46 stations in northwestern China, were obtained from the Central Meteorological Bureau of China.
The principal objective of this paper was to investigate the spatial distribution and temporal variability of snow cover over western China, and particularly over the Qinghai–Xizang (Tibet) Plateau. The chief approaches used are briefly given below.
a. EOF analysis
Empirical orthogonal function (EOF) analysis is used for compressing an initial huge quantity of information and extracting the main dominant spatiotemporal modes that capture the maximum proportion of initial variance (Richman 1986). The decomposition in EOFs follows:
where fi and gi are two sets of orthogonal functions in space and time, obtained by diagonalizing the covariance matrix. The corresponding eigenvalues represent the portion of the variance. In this study we did not use rotation in the EOF analysis. The EOFs are based on unnormalized SMMR snow-depth data that are not converted to anomalies. It was performed on 10-yr SMMR 6-day snow-depth field during the winter snow maxima (January and February) over the Qinghai–Xizang Plateau. Compressing the original 90 × 875 matrix into a 90 × 9 matrix, the first two EOFs take into account 60.4% of the total variance. The spatial distribution of the loadings corresponding to the EOFs represents the spatial pattern of snow-depth distribution and variability.
b. Multiple linear regressions
The linear regression used in this study is an ordinary regression approach. Because we are interested in diagnosing snow cover sensitivity to winter temperature and snowfall, a two-variable regression is described as the following:
where S, P, and T denote snow cover, snowfall, and temperature, respectively. Other notations, such as means (S0, P0, T0), standard deviations (σS, σP, σT), and regression coefficients (rSP, rST, rPT) are easy to understand. Their determinations can be found in any standard statistics book. For each regression coefficients a standard F test was performed with degrees of freedom that were two less than the number of years were used (Pollard 1981).
c. Trend test
At the core of trend testing is the ability of the model to distinguish whether an observed trend in time series is a random trend or a deterministic trend. In this study, a statistical model consisting of a possible trend plus correlated noise is fitted to the snow and climate time series,
where yt represents snow and climate parameters in year t; Et is the deviation from a straight line, and is assumed to be a stationary zero mean process.
When Et is serially correlated, in order to detect a deterministic trend for time series resulting from random trend presence, an autoregressive moving average (ARMA) model is appropriate for adoption (Woodward and Gray 1993). The statistical significance of the trends is evaluated by using the Student’s t test with following significance parameter:
where n is the total number of years, and r is correlation coefficient.
4. Spatial pattern of snow depth over western China
Figure 2 shows the spatial pattern of average snow depth (cm) during the winter snow maxima (January and February) between 1978 and 1987, estimated by SMMR, covering 2500 cells of a 0.5° latitude × 0.5° longitude grid over western China. It is characterized by uneven geographical distribution. Altitudinal variation is surprisingly pronounced. Snow depths vary spatially between large mountains and basins in northwestern China. The highest snow depth is seen in the Altay Mountains, the second highest occurs in the Tianshan, Pamirs, Karakorum, and Kunlun Mountains. In addition, an appreciable snow cover is also noted in the Elgis and the Ili Valleys. In contrast, snow cover is rare in the Tarim basin, Lop Nur basin, and Badain Jaran Desert. In the Junggar basin snow is predominantly thin, and frequently is of short duration. Aside from the great distance to moisture sources, the blocking mountains keep the basins very dry in snow cover.
Snow cover is far from a pervasive feature over the Qinghai–Xizang (Tibet) Plateau. It was about 59% snow covered in winter. Only in the peripheries, including the Himalayas, Pamirs, Nyainqentanglha, and eastern Tanggula Mountains, is there heavy snow cover present. In the vast interior, such as the Qaidam basin, Yarlung Zangbo Valley, snow cover is rare, and in the north of the Qinghai–Xizang (Tibet) Plateau snow cover is thin and of a short duration.
To better understand and interpret the structures of the SMMR data, Fig. 3 shows the first EOF pattern of 10-yr SMMR 6-day snow-depth data during the winter snow maxima from 1978 to 1987. It contains 44.7% of the total variance and has uniform negative loadings over the Tibetan Plateau, which is beyond the domain of China territory. The two largest loadings occurred along the eastern periphery and the western periphery of the plateau, which represent the two heavy snow cover areas. An apparent similarity between Figs. 3 and 4 indicates that the first EOFs represent an average of the original 10-yr SMMR snow depth during the winter snow maxima. Only in the peripheries over the Tibetan Plateau, particularly in the eastern and the western peripheries, was a heavy snow cover noted. Unfortunately, many diagnostic and modeling investigations ascribed the importance to the Qinghai–Xizang (Tibet) snow cover acting as a huge elevated cool source on atmosphere, but paid no attention to this key feature of snow cover distribution.
5. Annual cycle of snow cover
The regularity of annual cycle of snow cover seems to be ideal for agricultural practices. Large changes in timing, for instance, late or early spring snow cover dissipation and ill-timed snow peak, would have the potential for significant societal consequences.
The normal annual cycle of snow-water equivalent in northwestern China derived from the SMMR 6-day snow-depth averages during the 10-yr observation and pentad snow density area averages based on 36 primary weather stations in northwestern China (Fig. 5a) demonstrated that snow begins accumulating in the mid-November, increases to a late February or early March peak, followed by a rapid decline until early April. The late peak and short ablation duration are beneficial to humans for overcoming spring droughts. In some winters the peak was delayed by about 1 month, and sometimes it appeared earlier by half a month. The broad peak lasted for 72 days, while the sharp peak lasted for only 30 days. The peak amounts of snow storage (snow water equivalent) between the heavy snow winter and the light snow winter have a difference of 70 × 108 m3.
Over the Qinghai–Xizang (Tibet) Plateau snow season normally begins in mid-September (Fig. 5b). Snow cover growth is rapid in the first half of winter, with the maximum occurring in January. This is followed by a slow decline until June. The long snow season, early snow peak, rapid snow growth, and slow snow decay are evident. Of all snow seasons, winter (December–January–February) has the greatest snow storage. The winter, spring (March–April–May), and autumn (September–October–November) are represented by 45.2%, 28.0%, and 21.2% of the annual snow storage, respectively. The largest variability of peak snow amount is the most striking feature with great differences of as much as 300 × 108 m3 between the heavy snow winter and the light snow winter.
The highly variable nature of annual cycles is principally responsible for the anomalies in spring runoff as well as seasonality of river flow in western China.
6. Interannual variability of snow cover
a. Station-derived snow cover time series verification
Over northwestern China snow cover time series were created from 46 of the best stations over the past 47 yr from 1951 to 1997. First, daily snow depth and the monthly number of snow cover days are summed over the snow cover season (September–August), respectively, at a single station. Then, northwestern China snow cover time series were generated by averaging the annual snow cover data of 46 stations. To verify station-derived snow cover time series we compared time series of the annual number of snow cover days developed from the 46-selected-station network with annual cumulative 6-day snow storage estimated by SMMR in northwestern China during the overlap period between 1978 and 1987 (Fig. 6).
It is of interest to note that any year-to-year fluctuations experienced by the snow cover duration time series derived from station data showed up in the SMMR snow storage time series as well. A strong correlation (r = 0.59) exists between the two. The same also holds true for the two station-generated time series of the annual number of snow cover days and annual cumulative daily snow depth, with a correlation coefficient of 0.72. It argues that long-term snow cover time series constructed from station data have ability to represent the ground truth of the interannual variation of snow cover in northwestern China. Over the Qinghai–Xizang (Tibet) Plateau three snow cover time series were generated from area-averaged annual cumulative daily snow depths based on 60 primary stations, the SMMR 6-day snow-depth charts, and NOAA weekly snow cover area charts, respectively (Fig. 7). It can be seen that Tibetan station data do not have the ability to perfectly represent the ground truth of snow cover. Despite the major similarities between the station- and satellite-derived time series, for instance, heavy snow cover in 1977/78 and 1988/89, light snow cover in 1984/85, etc., some discrepancies could be found. An apparent example was 1985/86. While both SMMR and NOAA recorded a very heavy snow winter, the stations failed to report it.
b. Characteristics of spatial and temporal variations of snow cover
From the 10-yr SMMR 6-day snow-depth data, we computed the anomaly, which is the difference in snow depth during the snow pick period between the maximum (1985/86) and the minimum (1984/85) snow cover years for every grid cell.
The spatial pattern of SMMR snow-depth ranges showed that large interannual variability of snow depth is the most striking feature over the Qinghai–Xizang (Tibet) Plateau. The maximum and minimum area-averaged snow depths were 21.3 (1985/86) and 10.4 (1984/85) cm, respectively, during the period of SMMR operation. In fact, only the eastern part of the Qinghai–Xizang (Tibet) Plateau is affected by the most substantial year-to-year fluctuation in snow depth (Fig. 8). It is here that turns out to be one of the largest variation areas of China snow cover, as well as Eurasian snow cover (Vernekar et al. 1995). Most of the anomaly is about 30 cm in snow depth and in a relatively small region over the eastern Tibetan Plateau it is deeper than 50 cm. In Fig. 9 we display the second EOFs of the 10-yr SMMR 6-day snow-depth data during the winter snow maxima (January and February), which account for 15.7% of the total variance, with a broad maximum negative loading extended over eastern Tibet and a weaker positive loading located over western Tibet. The striking analog between Fig. 8 and Fig. 9 suggests that the second EOFs of SMMR snow data represent the east–west difference of snow variability over the Tibetan Plateau (Moron 1997). From Fig. 8 and Fig. 9 it is well established that the plateau snow cover fluctuation is almost always exhibited in such a regional pattern. Snow-depth fluctuation in the eastern part of the Qinghai–Xizang (Tibet) Plateau not only dominates interannual variability of snow cover over the entire Qinghai–Xizang (Tibet) Plateau, but it also is out of phase with that in western Tibet. In contrast with the Qinghai–Xizang (Tibet) Plateau, snowy areas affected by substantial year-to-year variation in snow depth are scattered in six large mountain systems, and the Ili, Elgis, and Junggar basins over northwestern China. The variability of snow depth is not as large as in the Qinghai–Xizang (Tibet) Plateau. It would be remiss not to mention another notable feature—that snow depths display the largest variability in the coldest months (Table 1), and they could dominate annual variation over the Qinghai–Xizang (Tibet) Plateau.
Figure 10 depicted time series of annual and spring ablation season (March and April) numbers of snow cover days and annual cumulative daily snow cover depth over northwestern China from 1951 through 1997. As will be readily seen from visual inspection, it demonstrated that long-term variability of snow cover is characterized by normal oscillation. Snow cover fluctuated around the mean. Heavy and light snow winters occurred alternatively. Neither abrupt changes nor continuation of snow minima from the late 1980s and early disappearance of spring snow cover were found. Only from the end of 1980s was a longer-lived decrease in snow cover evidenced, but it was not so great as the three previous snow deficits. In the 1960s and early 1970s snow cover was the lowest in second half of the twentieth century.
Over the Qinghai–Xizang Plateau long-term variability of snow cover is characterized by the largest interannual variability superimposing on a continuous increase trend. Furthermore, the annual amplitude of snow cover variability has increased significantly since the 1980s (Fig. 11). Both an extremely heavy snow cover year and light snow cover year occurred more frequently. The anomalies did not appear to be outside the range of natural variability.
7. Response of snow cover to climate change
a. Association between interannual variabilities of snow cover, temperature, and precipitation in the snow cover season
Snowfall and low temperature are full conditions to meet the formation of snow cover. To find a clue to the response of snow cover to climate change, it is necessary to understand the linkage of variations between snow cover, snow season temperature, and snowfall.
In China, the meteorological observation practices have not used an independent method for snowfall measurement. We have to use snow season precipitation data. It should not have a qualitative effect on the conclusions because the snow season temperature is well below the freezing over western China (see Fig. 12). Monthly average temperature and monthly total precipitation were used from 46 selected stations and from 60 primary stations to develop area-averaged time series of snow season temperature and precipitation over northwestern China and over the Qinghai–Xizang (Tibet) Plateau, respectively. What interested us about Fig. 12 is that in northwestern China the snow season temperature fluctuated in a fairly similar manner as that of the global temperature (Jones et al. 1999) over the past half century. The warming trend is most apparent. The snow season temperature rose by 1.7°C over the past 47 yr, which is one of the strongest warming regions all over the world. However, temperature increase has not been monotonic. Most of the warming occurred after 1976. It increased by 4.1°C during the five winters between 1976 and 1981. The 1990s exhibited the warmest decade during the past 47 yr. The warmest winters were 1996/97, 1994/95, 1988/89, and 1980/81. The first three occurred on a global basis, and the last one occurred in the Arctic (Przybylak 2000). Moreover, the variation is characterized by an alternating occurrence of warm and cold periods. The same is true for snow season precipitation. Here we lay special emphasis on the fact that the precipitation variability exhibits little relationship to temperature (r = +0.008) in snow season from 1951 to 1997. It is evident that the cause and effect relation between them is not in existence. Different atmospheric circulations control both of them. A major control on interdecadal variation in winter temperature over northwestern China is the North Atlantic Oscillation (NAO) (Clark et al. 1999; Kushnir 1999). From the 1960s until the early 1970s when the NAO index exhibited a downward trend to the minimum, the wintertime temperature was lower than normal and cold winters lasted for the longest period over northwestern China. The sharp warming has occurred with unprecedented strongly positive NAO index values since the end of the 1970s (Hurrell 1995). In contrast, the NAO’s impact on winter precipitation is limited for the precipitation that is primarily determined by availability of moisture brought by the southwestern flows.
To diagnose the climate influences on western China snow cover, we conducted a multiple linear regression analysis. The area-averaged time series of annual snow duration and annual cumulative daily snow depth were related to area-averaged snow season precipitation and temperature time series. The resulting regression equations are given by
where Sn and Sd are the annual number of snow cover days and annual cumulative daily snow depth (cm), respectively; Ps represents the total precipitation (mm) in snow cover season; and Ts denotes the surface air temperature (°C) during snow cover season. The first two equations are for northwestern China, and the third equation is for the Qinghai–Xizang (Tibet) Plateau. The multiple regression coefficients are 0.70, 0.81, and 0.67, respectively, and are significant at the 95% confidence level (Table 2). The results highlight the important influences of both temperature and precipitation on snow cover. The year-to-year fluctuation of snow cover is fundamentally tied to the snowfall and snow season temperature variabilities, while positive snowfall and negative temperature relationships were found. About one-half to two-thirds of the total variation in the snow cover could be explained by the linear relationship of corresponding precipitation and temperature variabilities.
Figure 13 shows a comparison of calculated time series of the annual number of snow cover days from snow season temperature and precipitation by using the multiple linear regression [Eq. (8)] with the measured result. The similarity between the two is striking, except for the difference in trends. While observed time series exhibit a positive trend, the calculated one turned out to be a negative trend. The clear message of this discrepancy is that while much of the year-to-year fluctuation can be explained as snow cover response to precipitation and temperature variations in snow season, the long-term trend of snow cover is by no means predictable by the regression [Eq. (8)]. It is obvious that important limitations may affect the validity of Eq. (8). Three possible explanations are presented. First, the standard rain gauge currently being used worldwide at meteorological networks undercatch snowfall because of wind-induced turbulence. For instance, a number of snow cover days, snow season temperature, and precipitation rose by 9 days, 1.7°C, and 5.3 mm, respectively, in northwestern China over the past 47 yr. However, Eq. (8) predicts that the precipitation should rise by 12.3 mm. In other words, the observed precipitation is much smaller than it should be. Another reason is that all regression models, except those with r = 1, underestimate the variance in the observed values. In the case of a time series with a monotonic trend, this feature can cause the model to underestimate the magnitude of the trend. In addition, snow cover variability exhibits a very small trend, which is embedded in a large year-to-year fluctuation that makes the long-term trend more difficult to detect. Confirmation of the inability of the statistical models to accurately represent the true trend is vital for projecting the future behavior of snow cover. This comparison addresses the question of where we now stand with respect to prediction of snow cover based on snowfall data. At present, it is not possible to make a confident statement about a reliable trend of snow cover by using the regression equation because the actual relationship between snow cover and temperature, as well as precipitation, in snow season could not be obtained. One of the main reasons is the large bias in snowfall measurements (Goodison et al. 1992)
b. Testing for trends in the time series of snow cover, temperature, and precipitation in snow season
Searching for long-term trends in snow cover variation forced by climate change may provide a starting point for understanding the future behavior of snow cover. Results of trend estimates for standardized time series of snow cover, snow season temperature, and precipitation are listed in Table 3. They reveal a general and uniform positive trend of snow cover over western China. An increase in snow cover is a systematic development as evidenced by the presence of deterministic trends. Both snow depth and snow cover duration exhibited a gradual increase. Although increasing rates are small, they are statistically different from zero. The increase in snow cover is more evident over the Qinghai–Xizang (Tibet) Plateau. The annual cumulative daily snow depth increased by 2.3% yr−1 during the period between 1957 and 1998. The long-term trends of western China snow cover are in good agreement with the snowfall trend, but are in contradiction to regional warming. It is an unexpected result that the apparent unprecedented warming of the 1990s and 1980s was accompanied by an increase in snow cover over western China. A persistent and misleading assertion is that global warming would decrease snow cover because snow cover has a strong negative relationship to temperature. Our study suggests that the global warming may have various impacts on snow cover depending on the different correlations between local winter temperature and snowfall. In a warming world, negative correlation corresponds to snow cover decrease, positive correlation corresponds to snow cover increase, and zero correlation means snow cover is more vulnerable to changes in snowfall. The snow cover trend is dependent on the snowfall trend rather than on the temperature trend, if winter temperature is well below freezing. In other words, even with global warming there is no reason for snow cover to decrease when snowfall is increasing and temperature is still to remain well below freezing in the region.
8. Summary and concluding remarks
Western China snow cover has been a long-standing gap in our knowledge despite many diagnostic and modeling studies ascribing importance to it. Accurate monitoring of snow cover still remained a tough question because the Qinghai–Xizang (Tibet) Plateau is marked by the highest terrain and complex mountain ranges.
In this paper the geographical distribution and spatial and temporal variability of western China snow cover has been investigated for the past 47 yr between 1951 and 1997. The data used consist of 10-yr SMMR 6-day snow-depth charts, which are revised by the regional retrieval algorithms, NOAA weekly snow extent charts, daily snow depth, and number of snow cover days recorded at 106 selected meteorological stations. An empirical orthogonal function was performed on the SMMR dataset. A multiple linear regression was conducted between snow cover, snow season temperature, and snow season precipitation. An autoregressive moving average model was fitted to the snow and climate time series to test for trends. The major findings are summarized below.
Snow cover distribution is far from a pervasive feature over the Qinghai–Xizang (Tibet) Plateau. Only in the peripheral mountains is any appreciable snow cover noted. In the vast interior snow cover is rare or very thin, patchy, and of a short duration. The blocking mountains keep the interior of the Qinghai–Xizang (Tibet) Plateau very dry in snowfall. The annual cycle of Qinghai–Xizang (Tibet) snow cover is characterized by an early peak occurring in January, a very slow snow decay, and a long snow dissipation progress from February to June. More than half of the snow mass was lost by sublimation in winter. Although the Qinghai–Xizang (Tibet) Plateau is one of the largest year-to-year variation areas of Eurasian snow cover, only the eastern plateau, a one-quarter part of the plateau in area, is affected by substantial interannual variability in snow depth, and it is out of phase with the western plateau. The above-mentioned characteristics constrain the Qinghai–Xizang (Tibet) snow cover to be a key variable influencing the Asian monsoon, and challenge Blanford’s hypothesis that a simple inverse relationship between Tibetan snow cover and Indian monsoon rainfall could provide a plausible explanation for the monsoon variations.
Western China did not experience a continual decrease in annual snow storage and early disappearance of spring snow cover, even during the great warming periods of the 1980s and 1990s. Over northwestern China the long-term variability of snow cover is marked by a stochastic oscillation superimposed on a small increasing trend over the past 47 yr. No abrupt change in snow cover was found. Over the Qinghai–Xizang (Tibet) Plateau, large interannual variability of snow depth is the most striking feature, and annual amplitude has increased significantly since the 1980s. An increase in snow depth is more evident over the Qinghai–Xizang (Tibet) Plateau than in northwestern China. The annual cumulative daily snow depth increased by 2.3% yr−1 over the Qinghai–Xizang (Tibet) Plateau during the period between 1957 and 1998. The increasing trend of western China snow cover is in good agreement with the snowfall positive trend, but is in contradiction to the regional warming.
This study highlights the importance of both the low temperature and snowfall influence on snow cover. The year-to-year fluctuation of western China snow cover is fundamentally tied to the precipitation (snowfall) and temperature in snow season. About one-half to two-thirds of the interannual variation in snow cover could be explained by the linear relationship of the precipitation and temperature variabilities. In contrast, the long-term trend of snow cover is by no means predictable by the multiple linear regression equation between snow cover, temperature, and precipitation—the main reason for which are the large biases in snowfall measurement. Further, the authors argued that the precipitation variability exhibits little relationship to the temperature (r = +0.008) in snow season over northwestern China. This is evidence that the cause and effect relation between the two is not in existence. The results address the question of where we now stand with respect to prediction of the future behavior of snow cover based on snowfall data.
The authors are indebted to Dr. Francis Zwiers of the Canadian Centre for Climate Modelling and Analysis for carefully reading and helpful comments on the original manuscript. Thanks are also extended to the two anonymous reviewers and to the Journal of Climate scientific editor Dr. Scott Denning for their excellent reviews and constructive comments. This project is funded by the Nation Natural Science Foundation (NSFC) (Grants 90202013 and 40121101) and the Chinese Academy of Sciences (Grants KZCX3-SW-345 and KZCX3-SW-339).
Corresponding author address: Dr. Liu Shiyin, Key Laboratory of Cryosphere and Environment, Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences, Lanzhou 730000, China. Email: firstname.lastname@example.org