Increased mass over the Tibetan Plateau: From lakes or glaciers?
Key Laboratory of Tibetan Environmental Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
Jiangxi Province Key Laboratory for Digital Land, East China Institute of Technology, Nanchang, Jiangxi, China
Corresponding author: G. Zhang, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Bldg. 3, Courtyard 16, Lincui Rd., Chaoyang District, Beijing 100101, China. (firstname.lastname@example.org); H. Xie, Laboratory for Remote Sensing and Geoinformatics, University of Texas at San Antonio, San Antonio, TX 78249, USA. (email@example.com)
 The mass balance in the Inner Tibet Plateau (ITP) derived from the Gravity Recovery and Climate Experiment (GRACE) showed a positive rate that was attributed to the glacier mass gain, whereas glaciers in the region, from other field-based studies, showed an overall mass loss. In this study, we examine lake's water level and mass changes in the Tibetan Plateau (TP) and suggest that the increased mass measured by GRACE was predominately due to the increased water mass in lakes. For the 200 lakes in the TP with 4 to 7 years of ICESat data available, the mean lake level and total mass change rates were +0.14 m/yr and +4.95 Gt/yr, respectively. Compared those in the TP, 118 lakes in the ITP showed higher change rates (+0.20 m/yr and +4.28 Gt/yr), accounting for 59% area and 86% mass increase of the 200 lakes. The lake's mass increase rate in the ITP explains the 61% increased mass (~7 Gt/yr) derived from GRACE [Jacob et al., 2012], while it only accounts for 53% of the total lake area in the ITP.
 Using GRACE satellite gravimetry data, Jacob et al.  showed that the mass balance rate in Tibet and Qilian Shan increased by 7 ± 7 Gt/yr between 2003 and 2010. They attributed this gain to an increase of glacier mass balance [Bamber, 2012; Jacob et al., 2012]. Other field-based evidences for the Inner Tibet Plateau (ITP) (an endorheic basin), however, showed that glaciers have been retreating in recent decades [Bolch et al., 2010; Kang et al., 2007; Yao et al., 2007, 2010, 2012]. For example, based on field measurements of glaciers in the ITP started in 2005, the mean annual mass balances in three glaciers of ITP between 2005 and 2009 ranged from −549 to −312 mm [Yao et al., 2012]. The glacier area and length in the ITP also indicated an overall decrease tendency over the study period [Bolch et al., 2010; Kang et al., 2007; Yao et al., 2007, 2012]. In this study, our hypothesis is that a significant part of the mass gain has been coming from increased water mass in the lakes rather than from the mass gain of glaciers.
 Lakes over the Tibetan Plateau (TP) play a crucial role in maintaining the water balance of the great river basins of Asia and are an important indicator of water dynamics associated with climate as well as glaciers and snow cover changes. Most lakes over the TP are located in the ITP [Yao et al., 2008], where the India monsoon, westerlies, and the East Asian monsoon join together [Yao et al., 2012]. The TP has the greatest number and area of lakes in China. Two detailed lake investigations, conducted in 1960s to 1980s [Wang and Dou, 1998] and 2005–2006 [Ma et al., 2011], showed that there were more than 1000 lakes (>1 km2 each) with a total area of 4.07 × 104 km2 in the region. Furthermore, most of the lakes that have newly appeared or whose area has increased in the last half-century in China are found in the TP [Ma et al., 2010].
 Only a few gauging stations in the entire plateau are available for lake level observations, such as those at Qinghai Lake, Nam Co, and Yamzhog Yumco. Satellite radar/laser altimetry such as TOPEX/Poseidon (T/P) (1992–2002), ENVISAT (post-2002), and the Ice, Cloud, and land Elevation Satellite (ICESat) (2003–2009) have provided the best measurements of water level. The absolute lake level and changes can be derived from ICESat observations with an accuracy of better than 10 cm [Zhang et al., 2011a., 2011b]. In addition, the elevation differences between satellite altimetry data and Digital Elevation Models (DEMs) have been efficiently used to monitor glacier mass balance [Gardelle et al., 2012; Kaab et al., 2012; Nuth and Kääb, 2011; Willis et al., 2012].
 Previously, when ICESat altimetry data were utilized to study lake level changes in the TP, the lake boundary was delineated from the Moderate Resolution Imaging Spectroradiometer (MODIS) with a spatial resolution of 500/250 m. This limited the number of lakes with enough ICESat data (e.g., 74 lakes at the 500 m MODIS pixel scale and 154 lakes at the 250 m MODIS pixel scale) to be examined [Phan et al., 2012; Zhang et al., 2011b]. In this study, the water mask (lake boundary) is derived from the Shuttle Radar Topography Mission (SRTM) DEM data sets (90 m pixel size). This results in not only more accurate lake boundaries, but also more lakes with enough ICESat data (200 lakes with available data of 4–7 years) to examine the lake level changes. Furthermore, the differences between ICESat elevations and SRTM are used in the study to extend the trend of lake level change from 2003–09 to 2000–09.
2 Data and Methods
 Comparing elevation differences derived from SRTM DEM and ICESat, the lake level and mass changes in the TP from 2000 to 2009 are examined.
2.1 SRTM DEM Data Sets
 The Shuttle Radar Topography Mission (SRTM) collected interferometric Synthetic Aperture Radar data during an 11 day mission in February 2000, providing a near-global scale (56°S to 60°N) highly valuable elevation database [Farr et al., 2007]. A global validations of SRTM show absolute geolocation and height errors are 7.2–12.6 m and 6.0–9.0 m, respectively, which exceeds the 16 m (90%) performance goal [Rodriguez et al., 2006]. The SRTM-derived elevation data was referenced to the World Geodetic System of 1984 ellipsoid and the vertical datum Earth Gravitational Model of 1996. The SRTM Water Body Data (SWBD) created by the U.S. National Geospatial-Intelligence Agency (NGA) offers high-resolution (90 m) worldwide ocean, lake and river shorelines (http://dds.cr.usgs.gov/srtm/version2_1/SWBD/). The changes of lake level from 2000 to 2003–2009 are calculated using the elevation differences between SRTM and ICESat. The HydroSHEDS (Hydrological data and maps based on SHuttle Elevation Derivatives at multiple Scales), which offer a suite of drainage and water flow data sets (http://hydrosheds.cr.usgs.gov/), are used to delineate the great river basins.
 A total of 4345 lakes with a total area of 4.07 × 104 km2 were mapped using the SRTM data in the TP (Figure 1). There are 1123 lakes with an area greater than 1 km2 (3.98 × 104 km2, 99.5% total area); of them, 76 have an area larger than 100 km2 (total area 2.75 × 104 km2, 68.75%), and 3 lakes are greater than 1000 km2 (Qinghai Lake, Nam Co, and Selin Co). Lake area for the 200 lakes and for all lakes in each basin are summarized in Table 1. It shows that the 200 lakes have a total area of 2.30 × 104 km2, i.e., 56% of the total lake area (4.07 × 104 km2) in the TP, and that the 118 lakes in the ITP have a total area of 1.36 × 104 km2, i.e., 53% of the total lake area (2.56 × 104 km2) in the ITP.
Table 1. The Number of Lakes (Percentage %), Water Level Change Rate (m/yr, Increasing (I) or Decreasing (D)), Area of Lakes With ICESat Data (104 km2) and Mass Change Rate (Gt/yr) of Basins in the Inner Tibet Plateau (I), Yangtze (II), Indus (III), Brahmaputra (IV), Yellow (V), Qaidam (VI), Amy Darya (VII), Salween (VIII), and Hexi Corridor (IX)
Area of all lakes in each subregions.
Estimated mass change rate of all lakes according to area ratio.
2.2 ICESat/Geoscience Laser Altimeter System Laser Altimetry Data
 The Geoscience Laser Altimeter System onboard the NASA ICESat provides global elevation measurements of flat surface with absolute accuracy of 10–15 cm between 2003–2009 [Zwally et al., 2002, 2008]. ICESat sampled the Earth's surface with footprints of ~70 m diameter spaced at 172 m intervals [Zwally et al., 2002]. ICESat laser altimetry was designed to measure thickness changes of Greenland and Antarctic ice sheets. Its great capacity for monitoring changes in elevation of terrestrial lakes and rivers has also been demonstrated [Urban et al., 2008; Zhang et al., 2011b]. The ICESat/GLA14 Release 633 data are acquired from the National Snow and Ice Data Centre (NSIDC) at http://nsidc.org/data/icesat/. The elevation and related information are extracted through the Interactive Data Language (IDL) code developed by NSIDC, and are converted from the T/P ellipsoid to the World Geodetic System of 1984 ellipsoid. The ICESat Release 633 data referenced to EGM2008 geoid are then converted to Earth Gravitational Model of 1996 geoid through the geoid conversion tool (F477) provided by the National Geospatial-Intelligence Agency, which is used for SRTM (http://earth-info.nga.mil/GandG/wgs84/gravitymod/egm96/egm96.html). For each ICESat track intersecting with a lake, outliers are removed to ensure the footprints are sampled over lake water surface without contamination from clouds and other environmental factors [Zhang et al., 2011b].
 A total of 436 lakes have available ICESat data: 200 lakes with ICESat tracks for 4–7 years and 236 lakes with tracks for 1–3 years. The simple linear regression is used to estimate lake-averaged elevation change rate.
2.3 Lake Mass Estimation
 The lake mass change rate is estimated using formula (1)
where V is mass change rate (Gt/yr); L, the rate of lake level change (m/yr); and A, is the area of lake (km2); and ρ, is the density of water (1.0×103 kg/m3 is used in this study).
 The lake area used in this study is derived from SWBD data sets (in 2000). An ideal method is to derive lake area and lake level concurrently, as reported in Zhang et al. . However, for the 200 lakes in this study, a fixed area derived from the same SWBD data sets provides a reasonable estimation of lake mass variations [Phan et al., 2012].
3.1 Lake Level Changes
 The TP is divided into different basins according to great rivers and watersheds provided by HydroSHEDS data sets [Immerzeel and Bierkens, 2012]. Lakes with ICESat data of 4–7 years are mainly in nine basins (Inner Plateau, Yangtze, Indus, Brahmaputra, Yellow, Qaidam, Amy Darya, Salween, and Hexi corridor) (Figure 2).
 The mean rate of lake level change of the 200 lakes was +0.14 m/yr, with 0.21 m/yr for the 152 (76%) lakes with increasing levels and −0.08 m/yr for the 48 (24%) lakes with decreasing levels (Table 1).The ITP with the largest number of lakes (118), has a mean water level increase of 0.20 m/yr. Of these 100 (85%) are increasing with a mean rate of 0.25 m/yr and the remaining 18 (15%) have a mean rate of −0.07 m/yr.
 The ITP is further divided into six sub-basins (zones) (Table S1 of the auxiliary material). Zone A has 30 lakes with a mean rate of +0.16 m/yr, including Nam Co and Selin Co, two of the largest lakes after Qinghai Lake. The snow cover in the basins of Nam Co and Selin Co is an important contribution fraction of lake level changes besides glaciers melting and precipitation [Bolch et al., 2010; Yao et al., 2007; Zhang et al., 2012]. All lakes in zones B (8 lakes) and D (3 lakes) show lake level increase. Zone C has 13 lakes, 10 with lake level increase and 3 with decrease, showing the smallest mean increase rate (0.12 m/yr) and the largest mean decrease rate (−0.12 m/yr) among all subzones of ITP. There are 32 lakes in each of Zones E and F, with 88% and 97% showing lake level increase, of 0.23 and 0.29 m/yr, respectively.
 Around 70% of lakes in the River basins of Yangtze, Indus, Yellow, Qaidam show lake level increase. The Brahmaputra basin is the only basin that has more lakes (62%) with lake level decrease, and has the largest mean decrease (−0.14 m/yr) of all basins in the TP. Yamzhog Yumco in this basin has a rate of −0.38 m/yr, the largest lake level decrease in all lakes in the TP. Only 8 lakes in the basins of Amy Darya, Salween, and Hexi corridor have enough ICESat observations to assure the calculations of lake level change, with 6 showing lake level increase and 2 showing a slight decrease.
 Overall, the lakes in the ITP predominantly show a water level increase. The majority of lakes with water level decrease are found in the Himalayas, particularly in the Brahmaputra basin.
 The 20 lakes with the greatest level increase are examined in detail (Figure 3). 18 lakes (90%) are located in the ITP and 2 in the Yangtze River basin. These lakes have a mean rate of rise 0.51 m/yr (ranging from 0.39 to 0.86 m/yr).
3.2 Lake Mass Change
 Like lake level changes, most of lakes in the TP present a positive mass change rate (see Figure S1 in the auxiliary material). Lake Qinghai in the Yellow River basin has a high mass increase rate of 0.6 Gt/yr and Yamzhog Yumco, in the Brahmaputra basin, has the highest mass decrease rate of −0.2 Gt/yr. Most of lakes in the ITP increased in mass. Specifically, lakes Selin Co and Nam Co show a strong mass gain of ~1.2 and 0.5 Gt/yr, respectively. The mass changes of lakes in sub-basins A, B, D, E, and F of ITP show a faster mass increase rate.
 All basins show a positive mass change rate, except Brahmaputra, the only basin with a negative rate of −0.23 Gt/yr. The mean mass change rate is +4.95 Gt/yr for the 200 lakes in the TP and +4.28 Gt/yr for the 118 lakes in the ITP. Therefore the 118 lakes in the ITP account for 59% area and 85% mass increase of the 200 lakes in the TP. This indicates that the lakes in the ITP had an overall faster water mass increase than lakes in other areas of the TP. Moreover, the +4.28 Gt/yr (in the ITP) and +4.95 Gt/yr (in the TP) explain 61% and 71% respectively of the increased mass (~7 Gt/yr) derived from GRACE [Jacob et al., 2012], although they only account for 33% and 56% of total lake area in the TP, respectively. Assuming other lakes without or with insufficient ICESat data have similar mean mass change rates as the 118 (in the ITP) or the 200 (in the TP) lakes examined, we can then approximately estimate the mass change/increase rates for all lakes in the ITP and entire TP are respectively 8.06 and 8.76 Gt/yr according to area ratio shown in Table 1. This is similar to the ~7 Gt/yr for Tibet and Qilian Shan derived from GRACE satellite gravimetry [Jacob et al., 2012].
4 Summary and Discussions
 The +4.28 Gt/yr in the ITP explains the 61% increased mass (7 Gt/yr) derived from GRACE [Jacob et al., 2012], while it only accounts for 53% of the total lake area in the ITP. Therefore, we conclude that the lake level/mass increases alone in the ITP's measured lakes can explain the majority (61%) of the increased mass seen from GRACE. If data on unmeasured lakes (47% of the lake area with no or insufficient ICESat hits) in the ITP were available, they might well explain the remaining portion (39%) of the GRACE mass change. We approximately estimate the mass change rate of all lakes in the ITP as 8.06 Gt/yr according to area ratio.
 Glacier melting can be a major reason for lake level and mass increases [Zhang et al., 2011b]; however, glacier melting into lakes, itself, should not increase the overall mass and may decrease the mass because a portion of the melted water would be lost through evaporation or discharged to rivers that leave the TP. Precipitation in the TP over the past several decades shows a significant increase, especially in the central TP [Xu et al., 2008]. The snow depth during the spring months over the Plateau since the mid-1970s exhibits a sharp increase [Zhang et al., 2004]. Therefore, we conclude that the majority of mass gain for the region should come from the increased precipitation and decreased evaporation that also contributed to the lake level/mass increase [Lei et al., 2013; Yang et al., 2011].
 However, whether glacier mass is unchanged or decreasing is not discernible from the combination of GRACE data, and lake level/mass changes. Other evidence from satellite observations and field measurements for the ITP supports the premise that glaciers have been retreating in recent decades [Bolch et al., 2010; Kang et al., 2007; Yao et al., 2007, 2010, 2012]. Therefore, our analysis is consistent with field measurements that glaciers have been retreating. An increase in glacier mass, however, is inconsistent with the combined data of lake level/mass changes and GRACE.
 In addition, the uplift rate in the ITP from several stations with continuous GPS measurements by China Earthquake Administration is about 0.5 mm/yr. The uplift rate at Lhasa is less than 1 mm/yr [Altamimi et al., 2007; Matsuo and Heki, 2010]. The mean lake level change rate in the ITP is +0.20 m/yr, which is approximately 200–400 times of crustal uplift. So the lake level increase from crustal uplift is ignored in this study.
 This work was supported by National Natural Science Foundation of China (41190081 and 31228021) and China Postdoctoral Science Foundation (2011M500405). Provision of ICESat data through NSIDC and SRTM DEM by USGS are sincerely acknowledged. Constructional comments from Jon Harbor (Purdue University) and one anonymous reviewer to improve the quality of this manuscript are greatly appreciated.
 The Editor thanks Jon Harbor and an anonymous reviewer for their assistance in evaluating this paper.