Snow cover dynamics of four lake basins over Tibetan Plateau using time series MODIS data (2001–2010)

Authors

  • Guoqing Zhang,

    1. Key Laboratory of Tibetan Environmental Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences,Beijing,China
    2. Laboratory for Remote Sensing and Geoinformatics, University of Texas at San Antonio,San Antonio, Texas,USA
    3. State Key Laboratory Breeding Base of Nuclear Resources and Environment, East China Institute of Technology,Nanchang,China
    Search for more papers by this author
  • Hongjie Xie,

    Corresponding author
    1. Laboratory for Remote Sensing and Geoinformatics, University of Texas at San Antonio,San Antonio, Texas,USA
    Search for more papers by this author
  • Tandong Yao,

    1. Key Laboratory of Tibetan Environmental Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences,Beijing,China
    Search for more papers by this author
  • Tiangang Liang,

    1. State Key Laboratory of Grassland Agro-ecosystems, College of Pastoral Agriculture Science and Technology, Lanzhou University,Lanzhou,China
    Search for more papers by this author
  • Shichang Kang

    1. Key Laboratory of Tibetan Environmental Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences,Beijing,China
    Search for more papers by this author

Corresponding author: H. Xie, Laboratory for Remote Sensing and Geoinformatics, University of Texas at San Antonio, San Antonio, TX 78249, USA. (hongjie.xie@utsa.edu)

Abstract

[1] Snow over the Tibetan Plateau (TP) is an important water source of major Asian rivers and greatly influences water availability in the downstream areas. In this study, snow cover dynamics of the four characteristic lake basins, Cedo Caka, Selin Co, Nam Co, and Yamzhog Yumco during hydrological years 2001–2010 (September through August) are examined at the basin scale using the flexible multiday combined MODIS snow cover products. The time series of multiday, seasonal, and annual snow covered area (SCA), onset/disappearance dates of snow, snow covered days (SCD), peaks of maximum SCA, and snow cover index (SCI) for each hydrological year (HY) are examined. Results show there is no obvious trend of snow cover change in the examined period, although Nam Co basin has the greatest SCA in all four basins and in all years, and Cedo Caka and Selin Co basins show the smallest SCA in most of the years. Overall, the HY2007 shows a greater snow extent and HY2010 a smaller for the region, with exceptions for the Nam Co basin where the HY2003 is the greatest and for Cedo Caka basin where the HY2004 is the smallest. Statistical analysis between lake level changes and lake basin's SCA, precipitation and pan evaporation (ETpan) changes shows that (1) Cedo Caka's water level rise was highly correlated with the basin's SCA changes (r = 0.94, p = 0.063); (2) Selin Co's water level rise was significantly correlated with the basin's SCA, precipitation and ETpan changes (r = 0.99, p = 0.029); and (3) lake level changes of Nam Co and Yamzhog Yumco were correlated with their corresponding lake basin's SCA, precipitation and ETpan changes (r = 0.87 and r = 0.86, respectively), although insignificant at the 95% level. This could have been due to precipitation and ETpan data of a distant meteorological station for Nam Co lake basin and the complex hydrological processes in the Yamzhog Yumco basin. This study suggests that the examination of time series snow cover dynamics is important to evaluate the water budget of lake basins with snow as a major component of water balance.

1. Introduction

[2] Snow plays an important role in the energy and water balance of drainage basin in alpine regions. Contribution of snowmelt to runoff is one of the important water resources in mountainous regions in addition to rainfall and glaciers melting. Snow stores at least one-third of the water usage for irrigation and growth of crops worldwide [Stepphun, 1981]. For example, the snowmelt contributes up to 90% of annual runoff in the basins of Rocky Mountains with high elevation [Schmugge et al., 2002] and the agropastoral production in the Himalayan region relies heavily on snow cover dynamics [Paudel and Andersen, 2011; Singh and Bengtsson, 2005]. Snow over the Tibetan Plateau (TP) greatly influence water availability of several major Asian rivers such as Yellow, Yangtze, Indus, Ganges, Brahmaputra, Irrawaddy, Salween and Mekong [Barnett et al., 2005; Immerzeel et al., 2009]. Discharge from these rivers sustains the lives of more than 1 billion people living both in the region and downstream [Barnett et al., 2005].

[3] The TP has undergone warming in the past three decades [Liu and Chen, 2000]. Surface air temperature showed an increase of 1.8°C from 1961 to 2007 observed at over 90 meteorological stations [Wang et al., 2008], with a greater magnitude in winter [Liu and Chen, 2000; Wu and Zhang, 2008]. The glaciers over the TP show an accelerated retreating because of the warmer climate, except for in the Karakoram [Bolch et al., 2010, 2012; Ding et al., 2006; Gardelle et al., 2012; Yao et al., 2007, 2010]. This is further confirmed to be related also to increased black carbon deposition [Ming et al., 2009; Xu et al., 2009] and changes in the atmospheric circulation patterns [Yao et al., 2012]. However, the snow cover dynamics with detailed time series have received less attention than the glaciers. The area and pattern of snow cover change in dry and high mountains are significant to understand regional to global climate variability. The snow covered area (SCA) and snow covered days (SCD) are important parameters of hydrological models to predict the seasonal water supply, runoff and flooding risk in watersheds dominated by snowmelt [Hall et al., 1998; Jain and Lall, 2000; Tong et al., 2009; Yang et al., 2003].

[4] Detection of SCA from space has been available since 1966 from the National Oceanic and Atmospheric Administration (NOAA). Optical remote sensing, such as Landsat, Advanced Very High Resolution Radiometer (AVHRR), Moderate Resolution Imaging Spectroradiometer (MODIS), and Satellite Pour l'Observation de la Terre (SPOT) have been used to map snow at moderate to high spatial resolution [Dankers and De Jong, 2004; Hall et al., 2002; Hall and Riggs, 2007]. Clouds often obscure the observations, especially in mountainous terrain. The spaceborne passive microwave remote sensing can provide hemispheric or global-scale cloudless snow products by the Scanning Multichannel Microwave Radiometer (SMMR), Special Sensor Microwave/Imager (SSM/I), and the Advanced Microwave Scanning Radiometer-Earth Observation System (AMSR-E) [Pulliainen, 2006; Wulder et al., 2007]. However, the coarse spatial resolution (25 km or coarser) limits their advantages. Several approaches have been developed to interpolate between cloud covered pixels and retrieve snow-covered area from MODIS products, such as the MODSCAG algorithm (MODIS Snow-covered Area and Grain size) using spectral reflectance from MODIS [Dozier and Painter, 2004; Dozier et al., 2008; Painter et al., 2009], combination of MODIS Terra and Aqua data, spatial filtering, temporal filtering, zonal snowline and snow cycle [Gao et al., 2010a, 2010b, 2011; Parajka and Blöschl, 2008; Parajka et al., 2010; Paudel and Andersen, 2011; Xie et al., 2009]. A flexible multiday combined MODIS snow cover method developed by Xie et al. [2009] and Gao et al. [2010b, 2011] to decrease cloud contamination and retrieve snow cover is used in this study.

[5] The area changes of glaciers and associated lakes over the TP have been reported by many studies [Bian et al., 2009, 2010; Bolch et al., 2010; Chen et al., 2009; Ma et al., 2010; Wang et al., 2011; Wu and Zhu, 2008; Yao et al., 2007, 2010; Ye et al., 2007]. The precise lake level variations within snow/glacier fed basins during the period of 2003–2009 using ICESat (the ICE, Cloud, and land Elevation Satellite) altimetry data were first presented in recent studies [Zhang et al., 2011a, 2011b]. The results indicated that in the 74 lakes (56 salt lakes) examined, 62 (84%) of all lakes and 50 (89%) of salty lakes show lake level increase, a possible indication of accelerated melting of glacier/perennial snow cover in TP. However, as stated in Zhang et al. [2011b], it is currently difficult to quantify the contribution fractions between melting water from glaciers and perennial snow cover, precipitation, and evaporation for each lake basin. In other words, it is currently not possible to directly quantify the satellite-measured lake level increase to the water budget (i.e., difference between melting water of glacier, precipitation and evaporation), although the linkage between glacier/perennial snow cover melt and lake level increase appears strong.

[6] In this study, we select four specific lakes to examine the snow cover dynamics over their individual basins and to see if there is any linkage between SCA, precipitation, evaporation, and lake level changes at the temporal scale of the Earth Observation System (EOS) record. The lake Cedo Caka had the largest lake level increase (+0.80 m yr−1) among all examined lakes in Zhang et al. [2011b]. The lake Selin Co had the second largest lake level increase (+0.69 m yr−1). The lake Nam Co, the largest lake in the area, with an area of 2015 km2 in 2004 [Zhu et al., 2010], had a moderate lake level increase (+0.25 m yr−1). The lake Yamzhog Yumco, however, had the fastest drop in water level (−0.40 m yr−1) among all examined lakes in Zhang et al. [2011b]. We examine data for the four lake basins for 10 hydrological years (2001–2010). We define a hydrological year (HY) from 1 September to 31 August; for example, HY2003 referring to hydrological year 2003, from 1 September 2002 to 31 August 2003.

2. Study Area

[7] The selected four lakes, Cedo Caka, Selin Co, Nam Co, Yamzhog Yumco, and their respective drainage basins are located in the central TP and extend from 28.0°N to 34°N and 87.5°E to 92.5°E (Figure 1), with rainfall and melting glaciers/snow contributing to the lakes. Many studies have analyzed observation data from around 70 meteorological stations in the eastern and central TP. For example, the linear trends of mean daily minimum and maximum temperature increased 0.41°C/decade and 0.18°C/decade during 1961–2003, respectively, and annual number of warm days also increased [Liu et al., 2006]. Precipitation exhibited an increasing trend in most regions over the period of 1961–2001 [Xu et al., 2008]. The annual mean sunshine duration series shows a significant decrease from 1983 to 2005 with a rate of −65.1 h/decade in the central TP [You et al., 2010].

Figure 1.

Location of lakes Cedo Caka, Selin Co, Nam Co, and Yamzhog Yum Co (also in the inset map) and their corresponding drainage basins over Tibetan Plateau. Streams and boundaries of lake basins are delineated from SRTM DEM and glacier coverage from GLIMS at www.glims.org. Four available meteorological stations are also denoted.

[8] The Nam Co basin is predominantly covered by grass Kobresia pygmaea [Cong et al., 2009]. Nyainqentanglha Mountains, south of Nam Co, represent a SW-NE high-mountain range of around 230 km in length with elevation of 5000–7000 m [Bolch et al., 2010]. Nam Co is a closed basin with main water discharge from melting glaciers/snow over the Nyainqentanglha Mountains, as well as rainfall and rainfall discharge to the lake.

[9] In the drainage basin of Selin Co, several major rivers, such as Zajia Zangbo, the longest inland river (409 km) of Tibet Autonomous Region, and Zagen Zangbo, the largest inland river of the drainage basin with 355 km in length, flow into the lake [Bian et al., 2010; Li et al., 2009]. There are two lakes, Yamzhog Yumco and Puma Yumco, in the Yumzhog Yumco basin. Yamzhog Yumco is the largest inland lake on the northern foot of the Himalayan Mts. with an area of 600 km2 in 2010 [Chu et al., 2012]. The Puma Yumco is a fresh lake with an area of ∼290 km2, receiving glacial melt water and discharging water into Yamzhog Yumco. Precipitation is the dominant water supply for Yumzhog Yumco, while glaciers melt accounts for 16% of total water supply [Chu et al., 2012]. The lake area of Yamzhog Yumco indicated a decreasing trend over the 1972–2010 period, especially since 2004 due to negative balance between precipitation and evaporation [Bian et al., 2009; Chu et al., 2012]. The lake Cedo Caka was rarely studied prior to the recent finding with the fastest lake level increase [Zhang et al., 2011b].

3. Data Used and Methods

[10] The precipitation measurements at Nam Co station, Multisphere Observation and Research of the Institute of Tibetan Plateau Research, Chinese Academy of Sciences started on 14 July 2005. To match the SCA records (HY2001–2010) used for the study, the station Baingoin with full records of precipitation and pan evaporation (ETpan), the closest station to Nam Co drainage basin (Figure 1), is collected. In addition, the precipitation and ETpan data at stations Xainza and Nagarze are also collected for Selin Co and Yamzhog Yumco drainage basins, respectively (Figure 1). The precipitation and ETpan at these three stations are obtained from China Meteorological Data Sharing Service System (see http://cdc.cma.gov.cn/). The ETpan is measured in a 20-cm diameter pan, which is much higher than actual evapotranspiration (ETa) in TP. There are no meteorological stations in or near the Cedo Cake drainage basin.

[11] The Shuttle Radar Topography Mission (SRTM) DEM is used to derive drainage basins of the four lakes, and multiday combined MODIS snow cover products are used to derive snow cover time series in the study. The ICESat lake level and changes over 2003–2009 period from Zhang et al. [2011b] are also used for comparison purposes. The ICESat mission, part of NASA's Earth Observing System (EOS), was launched in January 2003 and ended in February 2010 [Zwally et al., 2002]. ICESat provides global measurements of surface elevation with unprecedented accuracy [Zwally et al., 2002]. The precision of ICESat measurements of mean flat surfaces is ∼2 cm within 70 m footprints spaced at 172 m along track [Kwok et al., 2004; Zwally et al., 2008]. Although the ICESat data set is not temporally continuous, it fully covers both winter (February to March) and autumn (October to November) seasons. Snowfall starts in autumn and usually ends spring in TP. Therefore, in this study, the relationship between the lake level changes from autumn of present year to autumn of next year, and corresponding SCA, precipitation and ETpan changes between present hydrological year (September through August) and the next hydrological year is examined using multiple-variant linear regression model. There are 5 year data (4 differences) except 2005, 6 year data (5 differences) of lake level changes from 2003–2008 for Cedo Caka and Selin Co, 7 year data (6 differences) from 2003–2009 for Nam Co and Yamzhog Yumco, respectively.

3.1. SRTM DEM and Watershed Delineation

[12] The SRTM successfully collected elevation data over 80% of the earth land surface between 60°N and 57°S during an 11-day mission in February 2000. The high-quality DEM acquired in C-and X-bands with a resolution of 1 arc sec (∼30 m) over the U.S. and its territories and 3 arc sec (∼90 m) over non-U.S. territory are freely available from the U.S. Geological Survey (USGS) (see http://seamless.usgs.gov/). Vertical errors of the DEM are ±16 m and ±6 m for absolute and relative accuracy, respectively; the horizontal positional accuracy is ±20 m at a 90% confidence level [Farr et al., 2007; Rabus et al., 2003]. In mountainous terrain, the SRTM DEM shows sections with data gaps, generally due to radar shadow, layover and insufficient interferometric coherence [Kääb, 2005]. A recent study of using ICESat to validate the SRTM DEM in the TP shows that the vertical error of SRTM is within 16 m [Huang et al., 2011].

[13] In this study, the SRTM DEM is used to derive drainage basins of the four lakes using the Watershed Modeling System (WMS v8.3, see http://www.aquaveo.com/wms) (Figure 2). The derived basin areas of Cedo Caka, Selin Co, Nam Co and Yamzhog Yumco (Figure 1) are 6016, 49,895, 10,857 and 10,199 km2, respectively.

Figure 2.

Flowchart showing the processes of delineating watersheds, producing flexible multiday combination of MODIS snow cover images and SCD maps.

3.2. MODIS Snow Cover and Flexible Multiday Combination

[14] MODIS instruments on board the Terra and Aqua satellites as part of NASA's EOS program were launched in December 1999 and May 2002, respectively. The Terra satellite crosses the equator about 10:30 am with a descending node, and Aqua crosses the equator about 1:30 pm with an ascending node [Savtchenko et al., 2004]. Both Terra and Aqua have a sunsynchronous, near-polar, circular orbit. MODIS provides imagery of the Earth's surface and clouds in 36 discrete, narrow spectral bands from approximately 0.4 to 14.0 μm with a spatial resolution of 250 m, 500 m and 1 000 m [Hall et al., 2002]. The standard MODIS snow cover products are based on the normalized difference snow index (NDSI) and a set of thresholds and decision rules to provide global snow-mapping with a binary algorithm [Hall et al., 2002]. The accuracy of MODIS snow cover data has been validated against in situ observations and show a very high agreement in clear sky conditions [Gao et al., 2010a, 2010b; Liang et al., 2008; Parajka and Blöschl, 2006; Pu et al., 2007; Wang and Xie, 2009; Wang et al., 2009; Xie et al., 2009; Zhou et al., 2005], although the standard MODIS snow cover product misses snow in the transitional periods during accumulation and melt [Dong and Peters-Lidard, 2010; Rittger et al., 2012] and sparse snow conditions [Painter et al., 2009].

[15] The available MODIS daily snow cover products (only binary SCA data used), MOD10A1 and MYD10A1 (MODIS Terra/Aqua Snow Cover Daily L3 Global 500 m SIN GRID V005) during HY2001–2010 (spanning from 1 September 2000 to 31 August 2010) are used in this study. Four MODIS tiles H25V05, H25V06, H26V05 and H26V06 are needed to cover the entire study area. The MODIS Reprojection Tool (User's manual, release 4.0, 2008, https://lpdaac.usgs.gov/lpdaac/tools/modis_reprojection_tool) is used to reproject into Albers Equal-Area Conic Projection with datum of WGS84 and resample from original 463.3 m pixel size to 500 m. Meanwhile, the four MODIS tiles (images) are mosaicked and converted to GEOTIFF. Daily snow cover products for each lake basin (derived from section 3.1) are then extracted from daily Terra MODIS and Aqua MODIS, separately.

[16] However, cloud contamination in MOD10A1 and MYD10A1 limits mapping of daily snow cover. A flexible multiday combination method [Gao et al., 2010b, 2011] is used in the paper to derive cloudless snow cover products (Figure 2). The daily combined Terra-Aqua can reduce cloud contamination where there is not persistent, homogenous cloud cover, which is the case for the study area. The 3 h difference of Terra and Aqua make it possible to reduce cloud cover and reveal more snow cover, i.e., cloud covered pixels in the Terra MODIS map would be replaced with the Aqua/MODIS values assuming cloud-free conditions for Aqua, and vice versa. Therefore, the combination of the two can often remove the cloud cover and reveal more snow cover if pixels covered by cloud are snow.

[17] The method includes two steps: a daily combination of Terra and Aqua is executed first, which can reduce cloud pixels by ∼10% [Xie et al., 2009]. Then, the combined daily snow cover images are further combined into multiday combination images, based on two predefined threshold values: less than 10% in fractional cloud cover and maximum composite days of 8. Once any of the two threshold values reaches, the combination process stops. Thus, this method is called flexible multiday combination. Note that for the period before the Aqua MODIS was available, the daily combination step is not possible, while the flexible multiday combination is directly performed based on daily Terra/MODIS snow cover products only, with the same two predefined threshold values to control the combination process.

3.3. SCD and SCI Derivations

[18] The flexible multiday combination of MODIS images are also used to derive the snow covered days (SCD) map and snow cover index (SCI) for each hydrological year. Snow covered days (SCD) is defined as the total number of days that a pixel covered with snow in a hydrological year. The maximum SCD of a pixel could be less than 365 days due to cloud cover from those flexible multiday combination images, since there are still pixels (less than 10% in one combined image) with cloud cover. In addition, the daily combined products could produce data gaps and errors during combination due to different viewing geometry of MODIS Terra/Aqua [Dozier et al., 2008].

[19] The snow cover index (SCI, unit km2 d−1), proposed by Wang and Xie [2009], is used to examine quantitative variation of snow cover condition for a basin in 1 hydrological year, and is defined as below:

display math

where A is the area of a pixel (in km2; for MODIS, it is 0.25 km2); SCDi is the days of SCD at pixel i within a hydrological year; N is the total pixels in the basin.

[20] To intercompare snow cover condition for the four basins, this paper introduces a normalized SCI (NSCI, unit km2 d−1 km−2). The NSCI is defined as the SCI divided by the corresponding basin area, which indicates the mean snow days in the basin. The NSCI provides useful information for the snow cover condition, one number per hydrological year per basin, as compared with the SCD map that is one map for a hydrological year per basin.

4. Results and Discussion

4.1. Cloud Reduced MODIS Snow Cover Products

[21] The 3601 images of MOD10A1 and 2965 images of MYD10A1 are available for daily combination, resulting in 3585 daily Terra-Aqua combined images (including Terra/MODIS images only before the Aqua/MODIS images were available) (Table 1). The number of missing dates (images) for MOD10A1 and MYD10A1 account for 1.39% and 0.54% of total images, respectively. Table 2 shows the ranges and means of cloud and snow percentages of MODIS snow cover products in the four basins during HY2001–2010, respectively. The original daily MODIS Terra or Aqua products have the minimum cloud cover of 33.1% (CT) and 41.7% (CA). It is also true that the Aqua products always map higher cloud cover (CA) and lower snow cover (SA) than those of Terra products (CT, ST), with a overall mean of 49.5% (CA) and 2.4% (SA) against 40.0% (CT) and 3.4% (ST).

Table 1. MOD10A1 and MYD10A1 Daily Snow Cover Data Used Over the Period of 1 September 2000 to 31 August 2010a
Hydrological YearTotal Number of Days With MOD10A1Total Number of Days With MYD10A1Daily Combined Images
  • a

    MOD10A1 is Terra. MYD10A1 is Aqua. Period of 1 September 2000 to 31 August 2010 is HY2001–2010.

20013450345
200235450345
2003364362361
2004357366357
2005364365364
2006364362361
2007365365365
2008366365365
2009361365361
2010361365361
Total360129653585
Table 2. Ranges and Means of Cloud and Snow Percentagesa
 CTSTCASACDCSDCCMCSMCMean Days
  • a

    Shown are ranges and means of cloud and snow percentages of daily MODIS Terra (CT, ST) and Aqua (CA, SA), daily MODIS Terra-Aqua combination (CDC, SDC), flexible multiday combined products (CMC, SMC), and mean days of multiday combination for the Cedo Caka (CC), Selin Co, Nam Co and Yamzhog Yumco (YY) basins between HY2001 and 2010. Overall means of the four basins are shown.

CC         
 range37.2–45.00.9–5.850.8–56.50.5–3.828.0–40.01.2–6.93.1–4.72.9–12.42.6–3.6
 mean40.72.453.41.432.82.93.86.83.0
Selin Co         
 range36.7–44.21.5–4.445.4–53.81.1–2.827.0–38.71.9–5.54.4–5.43.7–10.82.6–3.7
 mean39.52.648.61.831.53.24.97.03.0
Nam Co         
 range38.7–46.22.5–8.547.8–58.91.7–6.429.5–43.03.4–10.54.3–6.37.4–19.22.7–4.1
 mean42.55.251.33.834.76.35.012.93.3
YY         
 range33.1–41.51.8–7.041.7–49.71.2–5.726.7–37.02.4–8.34.0–6.75.0–12.02.7–3.8
 mean37.43.444.92.531.74.24.98.53.2
Overall         
 mean40.03.449.52.432.74.24.78.83.1

[22] Daily Terra-Aqua composites reduce cloud cover and expose more snow cover with respect to original Terra and Aqua as shown in Table 2. The combination of Terra and Aqua perform well when the cloud cover patterns in both products differ due to 3 h of time shift. However, in most years, the cloud cover percentage in the daily combination (CDC) is still high, with the minimum of 26.7% and overall mean of 32.7% in the four basins.

[23] Flexible multiday combinations, with predefined threshold values of maximum cloud percentage (10%) and maximum composite days (8), show that the cloud cover percentage (CMC) is greatly reduced and snow cover percentage (SMC) is increased (Table 2). The overall mean cloud cover at all four basins is 4.7%, while the snow cover is 8.8%, with a mean composite of 3.1 days. Compared to the other three basins, Nam Co basin shows the greatest snow cover with a mean of 12.9% and a maximum of 19.2%. Figure 3 shows an example of cloud cover decrease and snow cover increase for daily Terra-Aqua combination on 10 October 2002, and flexible multiday combination from 7–10 October 2002 in the Nam Co drainage basin. The following analyses are all based on the flexible multiday composite snow cover images.

Figure 3.

Cloud and snow cover changes of MODIS images on 10 October 2002 in the basin Nam Co: (a) MODIS Terra; (b) MODIS Aqua; (c) Daily Terra-Aqua combination; (d) Flexible multiday combination of 7–10 October 2002.

4.2. Snow Cover Dynamics in the Four Basins

[24] Snow cover parameters within a hydrological year include the timing of snow cover onset/disappearance in a snow cycle, SCA, and maximum snow cover. Figures 4, 6, 8, and 10 present the snow cover time series for the four lake basins. Figures 5, 7, 9, and 11 present the seasonal snow cover changes, ICESat-derived lake elevation change from Zhang et al. [2011b], and available precipitation and ETpan data for each lake basin. It appears that, from the time series snow cover plots, the snowpack accumulates and melts several times in a hydrological year. The minimum SCA values could show the glaciers and perennial snow cover of the basin. Nam Co basin (Figure 8) shows more continuous snow cover before summer melting, such as in HY2003, 05, 07, 09, than any of the three other basins. Table 3 summarizes the numbers of SCA peaks (40% or above) and percentages in each season for each basin.

Table 3. Number of Peaks With Snow Cover Percentages of More Than 40% in Spring, Summer, Autumn, and Winter During Hydrological Years 2001–2010
Hydrological YearCedo CakaSelin CoNam CoYamzhog Yumco
  • a

    Spring, spr; summer, sum; autumn, aut; and winter, win.

20012 (40%, spr)a<30%2 (∼50%, spr),1 (∼40%, aut)
   1 (∼40%, spr-sum) 
20022 (50–60%, win)2 (∼40%, ∼50%, win)1 (∼60%, aut),1 (40%, aut),
   1 (∼55%, win),3 (40–50%, spr)
   1 (∼50%, spr) 
20032 (∼40%, aut),2 (∼40%, aut)2 (∼65%, 75%, aut),2 (∼55%, ∼65%, aut)
 1 (∼65%, win) 1 (50%, win), 
   2 (∼40%, spr) 
2004<30%2 (40–50%, win)2 (∼40%, win),3 (45–50%, 1 win, 2 spr)
   1 (∼50%, spr), 
   1 (∼40%, sum) 
20051 (∼75%, aut),3 (40–50%, aut),3 (50–65%, aut),1 (∼55%, spr)
 1 (∼50%, win),1 (∼70%, win)1 (∼45%, win) 
 1 (∼45%, spr)   
20061 (∼50%, aut),2 (∼40%, 60%, aut),1 (65%, aut),1 (∼40%, aut),
 1 (∼45%, spr)1 (∼60%, spr)2 (∼50%, spr)1 (∼50%, win),
    1 (∼65%, spr)
20071 (∼70%, aut),2 (∼60%, 40%, aut),1 (40–65%, aut),1 (∼75%, aut),
 2 (40–50%, win),2 (∼50%, ∼65%, win)3 (50%, 70%, 50%, win)1 (∼60%, win),
 1 (∼40%, win–spr),  1 (∼40%, spr),
 1 (95%, spr)  1 (∼40%, sum)
20081 (∼60%, aut),2(80%, 50%, win), 2(40–50%, spr)1 (∼45%, aut)7 (40–55%, 1 aut, 2 win, 4 spr)1 (∼40%, spr)
20092 (∼60%, ∼40%, aut), 1 (∼55%, win), 2(∼55%, 70%, spr)1 (∼65%, aut),2 (55%, ∼80%, aut),1 (∼80%, aut),
 1 (∼40%, spr)2 (∼40%, ∼50%, spr),1 (∼40%, win)
20101 (∼50%, aut), 1 (∼70%, win)<30%4 (∼40%, 1 aut, 2 win, 1 spr)1 (40%, spr)

4.2.1. In Cedo Caka Basin

[25] Figure 4 shows the time series of snow cover and mean seasonal snow cover for the Cedo Caka basin for each hydrological year, with details also summarized in Table 3. The HY2001 and 2004 show the smallest snow extent with peaks of less than 40% and 30%, respectively. The HY2002, 2003 and 2010 show greater extent than HY2001 and 2004, but smaller than HY2005–2009. The HY2007 shows the greatest areal extent of snow during the 10 year period with the highest peak of ∼95% in spring, which is the greatest SCA in the basin during 2001–2010. Overall, the maximum SCA peak of a year is usually in two periods: November–December or May for the basin.

Figure 4.

Time series of snow cover area over the basin area (%) from flexible multiday combined MODIS images, with seasonal means (unit %) (left to right: autumn, winter, spring, and summer) on top of each hydrological year, in the Cedo Caka basin during HY2001–2010.

[26] Figure 5 (bottom) shows annual and seasonal means of SCA in each hydrological year. There is a clear seasonal variation and there is no dominant season with consistently great or small SCA, while there seems an overall annual SCA increase from 2001 to 2007 and then decrease from 2007 to 2010, except the fluctuations of HY2004 and HY2006. ICESat derived lake level for the Cedo Caka lake shows a clear increase trend (0.8 m yr−1) from 2003 to 2009 (Figure 5, top).

Figure 5.

(bottom) Mean seasonal SCA (%) in the Cedo Caka drainage basin during HY2001–2010 and (top) lake level changes of Cedo Caka derived from ICESat data arranged by hydrological year of 2003–2009.

[27] The meteorological data is not available in the Cedo Caka drainage basin. The correlation coefficient between lake level changes and SCA changes is examined directly. The result shows that SCA changes highly correlated with lake level changes (r = 0.94) with a p value of 0.063, which might be attributed to limited available data in this basin. The lake level increased rapidly from 2004 to 2007 but flattened out from 2007 to 2009, corresponding well with continuous increase in annual SCA from 2001 to 2007 and then decrease from 2007 to 2009. The increase in SCA means more melt water in summer to increase lake level, the reverse is also true, although SCA is not the same as the snow water equivalent (SWE) [Martinec, 1980]. The HY2007 has the greatest SCA among all years for the basin, which indicates relatively more water that could have contributed to the larger lake level increase in October 2007 (or early HY2008). The same case is also seen for the HY2005, the relatively great SCA might have also contributed to the larger lake level increase from spring 2005 to fall 2005 (or early HY2006).

[28] However, summer precipitation and evaporation also affect the lake level. For example, the lake level did not show a decrease after 2007, although the SCA shows a clear decrease; a reasonable guess (since we do not have precipitation and evaporation data for the basin) is that precipitation increased in those years to largely balance out the water lose due to both evaporation and less melt water from snow.

4.2.2. In Selin Co Basin

[29] The time series of snow cover in the Selin Co basin is shown in Figure 6 and Table 3. The HY2001 and 2010 shows low snow cover (<30%). The overall SCA throughout the HY2002 was similar or even smaller than HY2001, although there are two snow cover peaks (40% or above) in HY2002. Snow cover conditions of HY2003 and 2004 show similar patterns. The HY2005 indicates a greater extent of snow cover, especially in autumn and winter. The HY2007 shows the greatest SCA. Since HY2008, the snow cover indicates a decrease with respect to HY2005, 2006 and 2007. Overall, the maximum SCA peak of a year is usually in months between October and February of the following year for the basin.

Figure 6.

Time series of snow cover area over the basin area (%) from flexible multiday combined MODIS images, with seasonal means (unit %) (left to right: autumn, winter, spring, and summer) on top of each hydrological year, in the Selin Co basin during HY2001–2010.

[30] Similar to the Cedo Caka drainage basin (in Figure 5), Figure 7 (bottom) shows large variation of seasonal SCA and an overall increase of annual SCA from 2001 to 2007, then decrease from 2007 to 2010. The figure also shows a large variation of mean annual precipitation ranges from the lowest 200 mm in HY2010 to the highest ∼500 mm in HY2008. The annual ETpan indicates obvious interannual fluctuations and is much larger than the precipitation. The continuous lake level increase (0.69 m yr−1) based on ICESat from HY2003 through HY2009 (Figure 7, top) can be well explained from SCA, precipitation and ETpan together (r = 0.99, p = 0.029). For example, the large precipitation in HY2003 and 2004 mostly contributed to the lake level increase from 2003–2004, as well as the large precipitation in HY2007–2009 contributed to the continuous lake level increase from 2007 to 2009, even with decreased SCA from 2007 to 2009. The Selin Co drainage basin is large (Figure 1), precipitation (and precipitation-caused runoff) could have played an important role for the lake level increase.

Figure 7.

(bottom) Mean seasonal SCA (%) in the Selin Co drainage basin, and precipitation and pan evaporation observed at station Xainza during HY2001–2010, (top) lake level changes derived from ICESat data during HY2003–2009.

4.2.3. In Nam Co Basin

[31] Snow cover variations in the Nam Co basin is shown in Figure 8 and Table 3. The seasonal mean snow cover of 28.78% in spring of HY2001 is the second largest cover in all years for the basin. The HY2003 presents a continuous snow cycle of long SCD from around October 2002 to June 2003. The 31.50% in autumn was the largest seasonal snow cover in all years for the basin, while the summer snow cover of the year was only 7.48%, less than the previous 2 years and the following year. This means the majority of annual snow cover melted away in summer to supply more water resource to the lake, contributing to the lake level increase. As shown in Figure 9 (top), the lake level shows a 1.33 m increase from February to September 2003. This suggests that snow melt indeed played an important role for this lake's water level change. In addition, snow cover could have melted away rapidly in later summer of HY2004, to greatly contribute the lake's water level (∼1 m) again from February 2004 to October 2004 (Figure 9, top).

Figure 8.

Time series of snow cover area over the basin area (%) from flexible multiday combined MODIS images, with seasonal means (unit %) (left to right: autumn, winter, spring, and summer) on top of each hydrological year, in the Nam Co basin during HY2001–2010.

Figure 9.

(bottom) Mean seasonal SCA in the Nam Co basin during HY2001–2010, precipitation and pan evaporation measured at station Baingoin during HY2001–2010, and (top) lake level changes derived from ICESat data during HY2003–2009.

[32] Similar to the HY2003, HY2005 shows a continuous snow cycle with long snow covered days from October 2004 to June 2005. The summer snow cover of only 6.13% suggests most of the snow cover melted away to supply water to the Lake, to greatly contribute in increasing the lake level (∼0.9 m) from 4724.38 m on 27 February 2005 to 4725.27 m on 30 October 2005.

[33] The HY2006 shows a smaller snow extent, and could not provide enough melted water to increase the lake level. This is consistent with the ICESat observations that lake level presents a slight drop (Figure 9, top). The snow cover in HY2007 melted away rapidly in summer (only 4.78%), which contributes to the lake level of a 0.39 m increase from 4724.75 m on 20 March 2007 to 4725.14 m on 24 October 2007. The HY2010 was the least snow cover year of all years examined for the basin. Overall, the maximum SCA peak of a year is usually either in October–November or April for the basin.

[34] As shown in Figure 9 there is an overall much higher mean SCA (as compared with Figures 5 and 7) and a notable interannual variability of annual SCA. HY2003 has the greatest snow extent (not HY2007 as in Figures 5, 7, and 11). As mentioned before, the melt water from the greatest SCA in HY2003 should have contributed a lot to the lake level increase of ∼1.3 m from February to September 2003. The larger precipitation in HY2004 (454.9 mm) against HY2003 (419.8 mm), increased SCA and decreased ETpan in 2005 against 2004 contributed a lake level increase of ∼1 m in 2004 and 2005, respectively. HY2006 was smaller in terms of SCA and precipitation, and larger ETpan, which is consistent with a small decrease of lake level. The comparison of SCA, precipitation and ETpan in HY2007 and 2008 show that increased precipitation and decreased ETpan played an important role for the 2007 and 2008 lake level increase. The decreased precipitation, increased SCA and ETpan resulted in a slight decrease of lake level in HY2009. A high correlation (r = 0.87, p = 0.336) between SCA, precipitation and ETpan changes together and lake level changes is found, although insignificant at the 95% level. This may have been due to precipitation and ETpan measured outside of this basin.

4.2.4. In Yamzhog Yumco Basin

[35] Snow cover variations in the Yamzhog Yumco basin are shown in Figure 10 and Table 3. An overall small SCA is shown in this basin, particularly, in HY2001, 2008 and 2010. And HY2002–06 show 1–3 peaks above 40% similar to Selin Co basin. The HY2007 shows greater extent snow coverage, which is consistent with other three basins. The HY2009 had big snow coverage in autumn, the largest seasonal mean of 24.78% in all years for the basin. Maximum peak of SCA of each year is either in October to early November or March to April.

Figure 10.

Time series of snow cover area over the basin area (%) from flexible multiday combined MODIS images, with seasonal means (unit %) (left to right: autumn, winter, spring, and summer) on top of each hydrological year, in the Yamzhog Yumco basin during HY2001–2010.

[36] Figure 11 shows the changes of seasonal SCA from HY2001 to 2010 in the Yamzhog Yumco basin. The precipitation presents interannual fluctuations, while ETpan a decreasing tendency from 2001–2004, then an increasing tendency from 2004–2010. The high precipitation as well as low ETpan in HY2004, greater snow extent in HY2007 and increased precipitation in 2008, contributed to small lake level increases, under the overall decreasing tendency in lake level. The high SCA in HY2009 did not result in increasing lake level, since precipitation is relatively small in the year. A high but insignificant correlation (r = 0.86, p = 0.365) is found between lake level change and SCA, precipitation and ETpan changes in this drainage basin. The insignificance could be due to the complex hydrological processes in this drainage basin where the glacier melt water discharges into Lake Puma Yumco, then flows into Yamzhog Yumco (Figure 1), as well as the operation of the Yamzhog Yumco Pump-Storage Power Station, which was active through the study period [Bian et al., 2009].

Figure 11.

(bottom) Mean seasonal SCA (%) in the Yamzhog Yumco basin, and precipitation and pan evaporation observed at station Nagarze during HY2001–2010, (top) lake level changes derived from ICESat data during HY2003–2009.

4.2.5. Remarks on the SCA, Precipitation, ET, and Lake Level

[37] Further analysis shows that precipitation played a major role in lake level increase of both Selin Co and Nam Co, followed by SCA and evaporation, while evaporation played a dominant role in lake level decrease of Yamzhog Yumco, followed by SCA and precipitation. For the Cedo Cake basin, however, the SCA can explain 88% of the lake level variations.

[38] The precipitation and ETpan are point measurements, whereas both lake level derived from ICESat data and MODIS snow cover area are spatial data integrated over larger areas. The topographic factors such as elevation, slope, and exposure in the mountain regimes could affect spatial pattern of precipitation [Basist et al., 1994; Guan et al., 2009; Yin et al., 2008]. Yin et al. [2008] found that satellite rainfall estimates over the TP could be improved significantly by interpolating the geographic location and topographic variables extracted from DEM. ETpan measurement depends on the moisture availability in the region around the pan. Thus, the spatial representativeness of the precipitation and evaporation measurements may impact the regression analysis.

[39] In these relatively dry basins of TP, the actual evapotranspiration (ETa) can be much less than ETpan owing to limited moisture availability. Hobbins et al. [2004] indicated that trends in ETpan do not necessarily represent trends in ETa derived from water budget, i.e., the pan evaporation paradox, and both of the radiative energy and regional advective budgets must be considered together to examine the relationship between ETpan and ETa. The ETpan is a good indicator of potential evapotranspiration (ETp) over the TP [Zhang et al., 2007]. The evaporation used in the paper is ETpan instead of the ETa and the ETpan, more representative of lake (water surface) evaporation, is likely much higher than ETa of the basin's land surface. This difference might have contributed to the unexplained variance in the regressions. Another factor for the unexplained variance is the glacier melt, which should have played an important role in lake level changes and needs to be further examined [Bolch et al., 2010; Yao et al., 2007; Ye et al., 2007; Zhang et al., 2011b]. However, owing to the limited hydrological network and meteorological stations in the study area, our current analysis could not fully involve the unexplained variance of lake level changes in the regressions.

4.3. Spatiotemporal Changes of Snow Covered Days

[40] Figure 12 shows the snow covered days (SCD) variation in the Cedo Caka basin from HY2001–2010. The small area (∼28.5 km2) of the central-west of the basin always shows the highest SCD in all years and is the only glacier and perennial snow cover area of the basin. Figure 13 shows the SCD in the Selin Co basin. The two small areas: northeast and southwest, with a total area of ∼360 km2, appearing in all years with the highest SCD, are the glaciers and perennial snow covered areas of the basin. The SCD in the Nam Co basin is plotted in Figure 14. The highest SCD areas in all years are located in the southern edge of the basin along the Nyainqentanglha Mountains and they are glaciers and perennial snow covered areas with an area of 198 km2 in 2001 [Bolch et al., 2010]. Figure 15 illustrates that SCD variation in the Yamzhog Yumco basin. The highest SCD areas, glacier and perennial snow cover, are distributed along the northwest edge, southwest tip, and a few small areas along the south edge with a total area of 215 km2 in 2000 [Ye et al., 2007]. The spatial distribution of glacier/perennial snow derived from SCD is consistent with the glacier inventory from GLIMS (the Global Land Ice Measurements from Space initiative) shown in Figure 1.

Figure 12.

Spatial distribution of SCD in the Cedo Caka basin during HY2001–2010. Pixels with white color show no SCD due to cloud cover (the same as in Figures 1315).

Figure 13.

Spatial distribution of SCD in the Selin Co basin during HY2001–2010. This basin includes Selin Co (the largest) and several other lakes also shown as black polygons.

Figure 14.

Spatial distribution of SCD in the Nam Co basin during HY2001–2010.

Figure 15.

Spatial distribution of SCD in the Yamzhog Yumco basin during HY2001–2010. This basin includes Yamzhog Yumco (the largest) and Puma Yumco also shown as black polygon.

4.4. Variation of Normalized Snow Cover Index

[41] Figure 16 shows the normalized snow cover index (NSCI) variations for the four basins. The NSCI is generally comparable with annual SCA (all seasons in 1 hydrological year summed together) for each basin (Figure 5, 7, 9, and 11) and also as shown in Table 2 (SMC). Nam Co basin shows the highest NSCI (i.e., the greatest SCA and days) among the four basins in all years, which is consistent with SCA time series plots (Figures 4, 6, 8, and 10), annual SCA plots (Figures 5, 7, 9, and 11), and SCD maps (Figures 12, 13, 14, and 15). The HY2003 shows the highest NSCI for the Nam Co basin and among the four basins. The smallest NSCI for the Nam Co basin occurs in HY2010. For Cedo Caka, Selin Co, and Yamzhog Yumco, the largest NSCI was in 2007. The HY2001 and 2010 show similar small NSCI for Cedo Caka, Selin Co, and Yamzhog Yumco basins, while the smallest NSCI for the Cedo Caka basin was in HY2004.

Figure 16.

Normalized snow cover index (NSCI) for the basins Cedo Caka, Selin Co, Nam Co and Yamzhog Yumco during HY2001–2010.

5. Summary and Conclusions

[42] The flexible multiday combination with two thresholds of maximum cloud percentage (10%) and maximum composite days (8) removes cloud coverage significantly, which allows efficient and accurate mapping of snow cover. The resulting mean cloud percentage of 4.7% in the four basins is much less than the daily products of ∼40%. In addition, the average of ∼3 days combination for an efficient snow cover image preserves the high temporal resolution of original MODIS data. The SCA time series derived from the flexible multiday combination of MODIS data over the 2001–2010 period for the selected four basins Cedo Caka, Selin Co, Nam Co and Yamzhog Yumco show detailed snow cover dynamics with onset/disappearance date of snow cycle and peaks.

[43] The relationship between lake level changes (October/November of present year through next year) and SCA, precipitation and ETpan changes of the corresponding lake basin based on hydrological year (September–August) are examined. Results show that lake level changes of the Cedo Caka are highly related with the SCA changes of the basin (r = 0.94, p = 0.06), and lake level changes of the Selin Co are significantly correlated with SCA, precipitation and ETpan changes of the basin (r = 0.99, p < 0.05). The p value of the correlation might be improved if precipitation and ETpan data were available in the Cedo Caka basin. The Nam Co and Yamzhog Yumco lakes show high (r = 0.87 and r = 0.86, respectively) but insignificant correlation at the 95% confidence level. This could be due to the distant meteorological station (Baingoin) from the Nam Co basin and the complex hydrological processes in the Yamzhog Yumco basin, respectively.

[44] The snow covered days (SCD) map illustrates the spatiotemporal changes of snow cover over each basin. Although there are generally big differences in snow cover onset and melt from year to year and from season to season for each basin (Figures 4, 6, 8, and 10), the spatial distribution and pattern of SCD from year to year for the basin seems relatively stable (Figures 12, 13, 14, and 15). This result is similar as found from Gao et al. [2011] for the pacific northwest USA region. The dominant snow covered days per year for all four basins was actually less than 50 days, followed by the 50–100 days. The maximum SCD pixels for each basin are usually the areas of perennial snow/glaciers cover for the basin and they show not much change from 2001 to 2010.

[45] The SCD map shows the spatial distribution of snow cover for each year, but it is still not easy to compare based on visualizing the SCD maps. By using the NSCI in Figure 16, we immediately tell that 2003 (2010) was the combined greatest (smallest) SCA and SCD for the Nam Co basin. Furthermore, it is easy to compare the snow cover conditions (SCA and SCD) for different basins in one plot. This is the advantage of SCI (or NSCI), while the advantage of SCD is the spatial distribution of the snow cover and days, which is lost in the SCI (or NSCI). Together SCA, SCD, and SCI provide useful information for snow cover dynamics and conditions for any area of study.

[46] Based on the information gained from the time series of multiday SCA, seasonal and annual SCA, SCD, and SCI for the selected four basins, there is no trend of snow cover for each basin or for all basins together during the period of 2001 to 2010. Limited precipitation data from three of the four basins do not show any trend of change as well. Statistical analysis shows a high relationship between SCA, precipitation and ETpan together and lake level changes, although it is still currently impossible to calculate the contribution fractions from each of them. Future work is to continue the monitoring of those parameters using MODIS-type remote sensing data and to further examine the glacier and perennial snow cover variations using higher spatial resolution images such as Landsat Thematic Mapper (TM/ETM +) and Advanced Land Observing Satellite (ALOS) for the four basins and beyond.

Acknowledgments

[47] This work was in part supported by National Natural Science Foundation of China (41190081 and 31228021), the Third Pole Environment Program (GJHZ0906), the U.S. NASA grant (#NNX08AQ87G), State Key Laboratory Breeding Base of Nuclear Resources and Environment at East China Institute of Technology (NRE1104), the Cultivation Fund of the Key Scientific and Technical Innovation Project, Ministry of Education of China (708089), and China Postdoctoral Science Foundation (2011M500405). The author G. Zhang wants to thank China Scholarship Council for funding his study for 2 years (2009–2011) at the University of Texas at San Antonio. Provisions of MODIS by NASA's Earth Observation System (EOS) and SRTM DEM by USGS are sincerely acknowledged. Critical reviews and constructional comments from Jeff Dozier, Thomas H. Painter, one anonymous reviewer, and associate editor Michael Lehning to improve the quality of this manuscript are greatly appreciated.