Implications of climate change for water management of an arid inland lake in Northwest China

Abstract

To evaluate potential effects of climate change on the water budget of Bosten Lake, the largest inland freshwater lake in China, we evaluated trends and step change points between 1980 and 2011 in evaporation, air temperature, and precipitation. Significant increases in air temperature accelerated glacier melt, increasing flow in the Kaidu River, which ultimately discharges into Bosten Lake. The increased inflow resulted in increased lake water depth and 10% greater water surface area. Increased surface area along with higher temperatures led to increased lake evapotranspiration. If glaciers continue to recede and snowpacks continue to decline with projected warmer temperatures under climate change, inflows to Bosten Lake will decrease substantially in the future, with critical implications for long-term water resource management. Our calculations suggest that the water surface area of Bosten Lake should be maintained >963.14 km² to maintain low lake water salinity (total dissolved solids: TDS ≤ 1.5 g/L), 3-year and 5-year means for annual net inflows to Bosten Lake must be approximately 21.80 × 10⁸ m³ and 23.50 × 10⁸ m³, respectively, and Kaidu River mean runoff must be approximately 35.83 × 10⁸ m³ to maintain that surface area while supplying 14.42 × 10⁸ m³/yr of water to meet public water demand in the Kongqi Basin. Water should only be diverted from Bosten Lake to the lower Tarim River for ecological restoration in years when Kaidu River annual runoff is >35.83 × 10⁸ m³. To minimize evaporation loss, water diversions from Bosten Lake should occur between May and September.

Key words:

• Bosten Lake
• climate change
• diversion
• evaporation loss
• glacier snow melt

In arid and semi-arid regions, lakes and reservoirs are vital water resources for economic and social development and environmental conservation. Water loss from evaporation can be exceedingly large in these regions, presenting challenges for water storage (Helfer et al. 2012). For example, in Australia, as much as 40% of total water storage capacity is lost each year to evaporation (McVicar et al. 2010). Estimates of evaporation are essential for water management planning, water quality protection, and land management decisions (Elsawwaf et al. 2010).

Water surface area, heat budget, percent humidity, and wind speed are all important factors that affect evaporation rates, but temperature is the determining factor. Under climate change, changes in air temperature and precipitation will influence lake evaporation rates by changing the heat budget and water surface area, and hence water availability and storage. Long-term changes in evaporation rates can dramatically alter hydrological processes and water availability, and thus influence agricultural production, economic development, and ecosystem function. Therefore, determining the potential effects of climate change on evaporation from lakes is essential for informing long-term water resource management, including water allocation, irrigation management, and revegetation activities, particularly in water-scarce countries (Lei and Yang 2010, Liang et al. 2010, McVicar et al. 2010). In addition, understanding the causes of change in evaporation is critical for making predictions for specific regions and hydroclimatic zones within regions because the interactions and feedbacks between climate change and hydrological processes vary spatially and temporally (Rodriguez-Iturbe and Porporato 2005).

Figure 1 Sketch map of the study area (the photo is a Landsat TM5 image).

Both free surface evaporation (E) and potential evapotranspiration (PE) are critical components of water balance and water loss in the hydrological cycle (Chattopadhyay and Hulme 1997, Lofgren et al. 2011, Spence et al. 2011). Evaporation can be difficult to estimate because it is a function of complex interactions between the components of the land–atmosphere system (Singh and Xu 1997). At present, a multitude of methods are used to estimate PE using different models (Delclaux et al. 2007, Gianniou and Antonopoulos 2007, Rosenberry et al. 2007, Lofgren et al. 2011); however, model estimates of PE are, by definition, uncertain. Many models predict that evaporation will increase with warmer global temperatures, yet in many regions observed evaporation has decreased as air temperature has increased (i.e., the “evaporation paradox”; Chattopadhyay and Hulme 1997, Peterson et al. 2002). It is therefore important to estimate how actual evapotranspiration will respond to climate change, especially actual evaporation loss from large lakes critical to regional water supply. There are relatively few methods for estimating actual evaporation. The Penman-Monteith formula is commonly used for calculating evapotranspiration from farmlands but is inaccurate for study areas without vegetation (Ward 1999). The advection-aridity (AA) model (Wang et al. 2011) and estimates from satellite data (Li and Zhao 2010) also measure evapotranspiration from soil and vegetation in addition to open water, and involve numerous complex parameters. Lake evaporation can be estimated using the isotope mass balance approach (Gibson and Edwards 1996), but this requires advanced and expensive instrumentation. Alternatively, where evaporation data are available from a hydroclimatic monitoring network, open water evaporation can be estimated using actual measurements of evaporation from evaporation pans, multiplied by a rational conversion coefficient.

Bosten Lake, located in an arid area in Northwestern China, is the largest inland freshwater lake in China. Inflows to Bosten Lake come from glacier melt, snowmelt, and rain in alpine mountains north and west of the lake. Any changes in the water budget of Bosten Lake will have major consequences for hydrological processes in the downstream Bosten Lake Basin. In this study, we examine the response of estimated annual actual evaporation loss in Bosten Lake to climatic variation between 1980 and 2011 and consider the implications for water management and water diversions. These implications demonstrate how information on water budgets can be used to develop sustainable management practices for inland lakes and reservoirs in arid and semi-arid areas.

Study site

Bosten Lake (41°44′∼42°14′N, 86°19′∼87°28′E) is an inland lake located in a semi-closed, arid basin (Yanqi Basin) at the southern foot of the Tianshan Mountains in Xinjiang Province, China (Fig. 1). The lake functions as a central water control and allocation facility for the local area, regulating droughts and floods and providing drinking water and water for agricultural and aquatic production in the downstream Bosten Lake Basin. It also supports an important fishery, provides a source of reeds for Xinjiang Province, and influences the microclimate of Yanqi Basin (Zhong 1988). Given the importance of Bosten Lake to the regional economy, society, environment, and climate, it has been incorporated into the “Lake Governance Agenda in 21^st century” and “Program of 1311 environment protection action in Xinjiang” as an important water resource of the Xinjiang Water Resource Comprehensive Management Plan (Li et al. 2003). Moreover, it became one of the first lakes of the national program “Pilot projects to protect the ecological environment of lakes” in 2012 (Zhou et al. 2014).

Bosten Lake is located in middle Asia, a region featuring abundant light and heat, and scarce precipitation. The Bosten Lake Basin is approximately 13.34 × 10⁴ km², is composed of Bosten Lake and the Kaidu and Konqi river basins, and has a mean annual temperature of 8.4 C and mean annual total precipitation of 68.2 mm. The lake is approximately 55 km long from east to west, 20 km wide from south to north, and shaped like a deep dish. When the mean surface water elevation is 1048.75 m a.s.l, the total lake water surface area is 1002.4 km², the total water volume is 8.8 × 10⁸ m³, the mean depth is 8.8 m, and the maximum depth is 17 m. Bosten Lake discharges to the Kongqi River through a pump. Its contributing rivers include the perennial Kaidu River and the seasonal Huangshuigou and Qingshui rivers. The Kaidu River originates from the Hargat and Jacsta valleys in the Sarming Mountains and is fed mainly by glacier melt and snowmelt from the alpine zone, as well as by precipitation from the mountain zone.

Materials and methods

Evaporation capacity calculation

The annual evaporation capacity of the Bosten Lake water surface was estimated using a simple equation: (1) where E_water is the annual evaporation capacity per unit area of the lake surface (mm/yr); E_ϕ20 is evaporation measured from a ϕ20 evaporation pan with a 20 cm diameter (mm/yr); and k is the correction coefficient of evaporation of the lake water surface, estimated as follows, based on the methods of Zhong (1988), Li (2001), and Wang (1993): (2) where E₆₀₁ is evaporation (mm/yr) measured from an E601 evaporation pan, a cylindrical pan with a conical bottom, maximum surface area of 3000 cm², and maximum depth of ∼7.5 cm; and E_m20 is evaporation (mm/yr) measured from a 20 m² evaporation tank.

Evaporation measurements were collected by the Bohu County meteorological bureau at the Bosten Lake inflow from 1996 to 2009 using a 20 cm evaporation pan. Additional evaporation measurements were collected by the national meteorological bureau in Yanqi County at a location 24 km from the lake (the nearest available location) from 1980 to 1995 and 2010 to 2011 using a 20 cm evaporation pan. Zhong (1988) and Li (2001) suggested k₁ and k₂ values of 0.57 and 0.83, respectively (k = 0.473), based on data from 2-year evaporation observation experiments on an island in southeastern Bosten Lake in 1984 and 1985. Similarly, Wang (1993) estimated k₁ and k₂ values of 0.52 and 0.90, respectively (k = 0.468), based on data collected from 1982 to 1987 from E601 and 20 cm evaporation pan measurements at the Yanqi national meteorological station and 20 m² evaporation tank measurements at a water balance experiment station at a reservoir on the Aksu River, the nearest 20 m² evaporation tank measurement station, 400 km from Bosten Lake. The estimates of Zhong (1988) and Wang (1993) did not differ significantly, although Zhong's (1988) values were higher. Because the available evaporation data were collected in different locations for different years, we used values for k₁, k₂, and k of 0.57, 0.83, and 0.473, respectively, for 1996–2009, and 0.52_, 0.90, and 0.468, respectively, for 1980–1995 and 2010–2011.

Water surface area and actual evaporation loss calculations

The annual actual evaporation loss of Boston Lake was estimated using the formula: (3) where E_lake is the actual evaporation loss of the lake (m³/yr), E_water is the annual evaporation capacity of the water surface (m/yr), s is the surface area of the lake (km²), and 10⁶ is the conversion of km² to m².

Data on the water surface area of Bosten Lake from 1980 to 1989 were obtained from the Bazhou Hydrology and Water Resources Survey Bureau. Water surface area data from 1990 to 2011 were interpreted from Landsat TM5 images. Because the lake surface area varies monthly, the mean value from January to December was used as the annual lake surface area. Clear Landsat images taken between January and December in the same location each year were generally unavailable. Every month, we selected only one image in which weather conditions were relatively favorable with the fewest clouds present. In a July 2011 image, we selected 37 random points as the surface control points of the lake (Fig. 1). We used this July 2011 image as a standard to correct for differences in the locations and scales of the images for different years. The July 2011 image was treated with geometric rectification and clipped using the study area border and defined lake surface control points. The root mean square error (RMSE) of each registration was maintained below one pixel. Radiometric calibration and atmospheric correction procedures were conducted to ensure that the change detection analysis truly detected changes at the Earth's surface rather than at the sensor level, and truly detected changes in surface area, not differences in solar illumination or atmospheric conditions (Li and Zhao 2010). Validation of the corrected image with the 37 random sampling points indicated that interpretation accuracy was 85%. Images taken in other months from 1990 to 2011 were adjusted to match the rectified July 2011 TM image based on the locations of the defined lake surface control points and then resampled with 30 m resolution. The lake shape was extracted, and the surface area was calculated using Arc GIS 9.3.

Salinity and water surface area prediction

Although Bosten Lake is naturally freshwater and an important source of freshwater to Bosten Lake Basin, it has changed gradually into a slightly salty lake since the 1980s. Therefore, a key question for water management is how to reduce total dissolved solids (TDS) in Bosten Lake. To examine the implications of our results for Bosten Lake water quality and water management, we constructed a simple model of Bosten Lake salinity and water surface area: (4) (5) (6) in which T is the TDS of the lake water column, S is the water surface area, s is the water surface area in the previous year, I is net inflow, O is outflow, E_lake is evaporation loss, E_water is evaporation capacity of the water surface, R is Kaidu River runoff, and a, b, c, d, and r are constants. We fitted these equations using curve estimation, with measurements of S, s, T, E, I, O, and R from 1980 to 2011 (r > 0.5). Annual water surface area, evaporation capacity, and evaporation loss measurements calculated in the present study were entered into the model. Annual mean TDS data (measured monthly from 11 points around the lake; Fig. 1) in Bosten Lake from 1958 to 2011, as well as annual inflow and outflow data from 1980 to 2011, were obtained from the Bazhou Hydrology and Water Resources Survey Bureau. Annual Kaidu River runoff data for 1980 to 2011 were obtained from the Dashankou hydrological station, a mountain pass station on the Kaidu River. This simple model was designed as a conceptual exercise to explore the implications of changes in water surface area for lake water quality and water management and does not include all the details necessary to predict actual Bosten Lake TDS.

Other data collection and analysis

Annual mean temperature and precipitation measurements from 1980 to 2011 were obtained from 5 meteorological stations that encompassed the main landscape characteristics and variability of Boston Lake Basin, including Bayinbuluk meteorological station in the alpine mountains; meteorological stations in Yanqi, Bohu, and Heshuo agricultural counties on the plains, and a meteorological station in Korla, a mainly industrial city on the plains. Air temperature and precipitation were averaged across the 5 stations to estimate mean annual air temperature and precipitation in Bosten Lake Basin by arithmetical averaging and the Thiessen Polygons method, respectively (Rhynsburger 1973).

The nonparametric Mann-Kendall (MK) trend test was used to test for significant monotonic, temporal trends in evaporation capacity, water surface area, actual evaporation loss, and climatic variables over the study period. The nonparametric Mann-Kendall–Sneyers test was used to identify significant step change points (i.e., abrupt temporal changes) in evaporation capacity, water surface area, actual evaporation loss, and climatic variables over the study period. These statistical models followed the procedures of Yue and Wang (2002) and Chen et al (2012).

Analytical software

The nonparametric Mann-Kendall trend test and step change point analysis were conducted using DPS 7.05 (Data Processing System, Tang QY, China, 2010). Remote sensing image interpretation was conducted using ENVI 4.6 (The Environment for Visualizing Images, Exelis Visual Information Solutions Inc., USA, 2009). Pearson correlation analysis and curve estimation were conducted using SPSS 13.0 (Statistical Product and Service Solutions, SPSS Inc., USA, 2005). Figures were created using ArcGIS 9.3 (ArcGIS Desktop, Esri Inc., USA, 2007) and SigmaPlot 12.5 (Systat SigmaPlot, Systat Software Inc., USA, 2012).

Results

Temporal change in evaporation capacity and water surface area of Bosten Lake

Water surface evaporation capacity of Bosten Lake did not change significantly between 1980 and 2011. Although Bosten Lake evaporation capacity increased slightly over the study period (Fig. 2A), the Mann-Kendall trend test was not significant (P > 0.05; Table 1). Further, the curves in the Mann-Kendall–Sneyers test fell entirely within the confidence interval, indicating no significant step change point in annual evaporation capacity between 1980 and 2011 (Fig. 2B).

Table 1 Mann-Kendall (MK) trend tests of annual water surface evaporation capacity, water surface area and actual evaporation loss in Bosten Lake, and annual temperature and precipitation in Bosten Lake Basin from 1980 to 2011.

			MK trend test
Items	Mean value	Trend rate	Z	Significance level
Annual evaporation capacity (mm/yr)	896.1	0.001	0.73	P > 0.1
Surface area (km^2/yr)	975.1	4.16	3.07	P < 0.01
Annual actual evaporation loss (m^3/yr)	8.8	0.05	2.45	P < 0.01
Temperature (C)	8.9	0.39	3.67	P < 0.01
Temperature at alpine mountain station (C)	4.2	0.33	3.58	P < 0.01
Precipitation (mm)	152.3	1.94	1.57	P > 0.1
Precipitation at alpine mountain station (mm)	273.0	22.50	2.07	P < 0.05

Figure 2 Annual evaporation capacity of Bosten Lake from 1980 to 2011: (A) temporal trend and (B) Mann-Kendall–Sneyers test (P > 0.05).

In contrast, the water surface area of Bosten Lake increased significantly between 1980 and 2011 (Table 1; Figs. 3A, 3B). Water surface area was greatest in 2002, at 1251.10 km². The Mann-Kendall–Sneyers test indicated a step change point in water surface area in 1995. From 1980 to 1994, mean annual water surface area was 924.11 km², and from 1995 to 2011, mean annual water surface area was 1020.01 km², a 10% increase. Visual comparisons of the spatial distribution of the water surface area during these 2 stages suggest that the water surface area expansion after 1994 occurred mainly in the eastern and northwestern portions of the lake. The lake surface area also varied within years, with the largest changes between May and September and little change from October to April (data not shown).

Temporal change in actual evaporation loss from Bosten Lake

Similarly, annual actual evaporation loss from Bosten Lake increased significantly between 1980 and 2011 (Table 1; Fig. 4A). Annual evaporation was greatest in 2002, at 11.57 × 10⁸ m³, due to the largest water surface area during that year. A Mann-Kendall–Sneyers test indicated a step change point in annual actual evaporation loss in 1995, consistent with the step change point in water surface area (Fig. 4B). Between 1980 and 1994, the annual evaporation loss was small, with a mean evaporation loss of 8.13 × 10⁸ m³, whereas between 1995 and 2011, annual evaporation increased substantially, with a mean evaporation loss of 9.29 × 10⁸ m³, a 14% increase.

Figure 3 Water surface area of Bosten Lake from 1980 to 2011: (A) temporal trend and (B) Mann-Kendall–Sneyers test (P < 0.05).

Figure 4 Annual actual evaporation loss of Bosten Lake from 1980 to 2011: (A) temporal trend and (B) Mann-Kendall–Sneyers test (P < 0.05).

Bosten Lake TDS between 1980 and 2011 was significantly and negatively correlated with both Bosten Lake water surface area and actual evaporation loss (Fig. 5B; P < 0.01), with an R² of 0.71 and 0.73, respectively.

Effect of climate on evaporation from Bosten Lake

Temporal change in the climate of Bosten Lake Basin

Mean annual air temperature in Bosten Lake Basin increased significantly between 1980 and 2011 (Table 1). The Mann-Kendall–Sneyers test detected a step change point in air temperature in 1996, consistent with the step change points of Bosten Lake water surface area and actual annual evaporation loss (Fig. 6A). Between 1980 and 1995, mean annual temperature was 8.59 C, whereas between 1996 and 2011, mean annual temperature was 9.21 C, a 0.62 C increase. Both climate variability and human activities can affect temperatures, especially in farming regions and cities. To isolate effects of climate variability from human activities, we analyzed change in temperature in the alpine mountains (Bayinbuluke meteorological station) separately. Temperatures in the alpine mountains also increased significantly between 1980 and 2011 (Table 1), with a step change point in 1996 (data not shown). These results are consistent with the conclusion that China shifted from a cold phase to a warm phase in the 1990s because of greenhouse gases and the Atlantic Multidecadal Oscillation (Information Office of the State Council of the People's Republic of China 2008, Li and Gary 2007).

Precipitation, in contrast, did not change significantly during the study period (Table 1). Although there was an increase over time, the Mann-Kendall trend test was not significant (P > 0.05; Table 1), and there was no significant step change point in precipitation between 1980 and 2011 (Fig. 6B). There were, however, large differences between the plains and mountain regions. On the plains, there was no significant change in precipitation from 1980 to 2011, whereas in the mountains near the headwaters of the Kaidu River (Bayinbuluke meteorological station), precipitation increased significantly from 1980 to 2011 (Table 1), and the Mann-Kendall–Sneyers test detected a step change point in 1993 (data not shown).

Figure 5 (A) Temporal trends in the TDS of Bosten Lake water from 1958-2011 and (B) regression analyses of TDS and both water surface area and actual evaporation loss of Bosten Lake from 1980 to 2011 (P < 0.05).

Figure 6 Mann-Kendall–Sneyers tests of annual temperature (A; P < 0.05) and precipitation (B; P > 0.05) in Bosten Lake Basin from 1980 to 2011.

Response of evaporation loss of Bosten Lake to climatic variation

Pearson correlation analyses indicated no significant relationship between Bosten Lake Basin temperatures and estimated Bosten Lake evaporation capacity (Table 2). In contrast, correlations between air temperature and both water surface area and actual evaporation loss were significant, suggesting that increased air temperature may have directly or indirectly influenced water surface area and actual evaporation loss. There was no significant correlation between mean precipitation or precipitation on the plains and Bosten Lake evaporation capacity, water surface area, or actual evaporation loss; however, correlations between precipitation in the mountains (Bayinbuluke meteorological station) and both water surface area and actual evaporation loss were significant (Table 2). Moreover, correlations between precipitation in the mountains and both evaporation loss and Kaidu River runoff were significant (R² = 0.71 and 0.57, respectively; P < 0.01), suggesting that precipitation in the plains did not strongly affect evaporation loss from the lake, but that increased precipitation in the mountains increased evaporation losses by increasing Kaidu River runoff, and thus increasing Bosten Lake inflows and water surface area.

Discussion

How did climatic variation affect the water surface area and actual evaporation loss of Bosten Lake?

The water balance of Bosten Lake is composed mainly of inflows from the Kaidu, Huangshuigou and Qingshui rivers, outflow to the Konqi River, and evaporation loss. In comparison, factors such as precipitation, confined aquifer replenishment, and seepage flow are small and relatively stable over time (Sun and Wang 2006). The mean annual Kaidu River runoff was 34.80 × 10⁸ m³ between 1956 and 2004 (Sun and Wang 2006, Ma et al. 2010), and 36.14 × 10⁸ m³ between 1980 and 2011 (Fig. 7). This constitutes the main inflow to Bosten Lake, accounting for >84% of the total annual inflow in 1996 (Table 3). In contrast, the seasonal Huangshuigou and Qingshui rivers contributed only 4% and 1% of total inflow, respectively, in 1996 (Table 3), and their lower reaches have dried up gradually since then, supplying no water to Bosten Lake since 2002 (Sun and Wang 2006).

Table 2 Pearson correlation coefficients for correlations between Bosten Lake hydrologic variables and Bosten Lake Basin climatic variables.

Items	Temperature	Precipitation	Precipitation in mountains	Precipitation on plains
Evaporation capacity	0.101	−0.228	0.037	0.023
Surface water area	0.649**	0.002	0.519*	0.142
Evaporation loss	0.641**	-0.105	0.503*	0.123

**P < 0.01, *P < 0.05.

Figure 7 (A) Linear regression analysis of Kaidu River annual runoff and Bosten Lake water surface area and (B) temporal trends in net inflow and outflow of Bosten Lake and Kaidu River annual runoff from 1980 to 2011.

An annual mean of 15.19 × 10⁸ m³ of water has been pumped from Bosten Lake to the Konqi River each year between 1980 and 2011 (Fig. 7), accounting for ∼50% of total outflow. For example, 11.34 × 10⁸ m³ of water was pumped from Bosten Lake to the Konqi River for industrial and agricultural development in the Konqi Basin in 1996, accounting for 55.34% of total outflow (Table 3). In addition, 13.00 × 10⁸ m³ of water (between 0.26 and 3.47 × 10⁸ m³/yr) was pumped from Bosten Lake to the lower reaches of the Tarim River between 2000 and 2006 for ecological restoration (Huang and Pang 2010). Actual evaporation loss from Bosten Lake, however, is another major outflow, with an annual mean of 8.75 × 10⁸ m³ between 1980 and 2011. In 1996, evaporation loss from Bosten Lake was similar in magnitude to outflows to the Konqi River and was almost 6 times greater than precipitation inputs to the lake (Table 3). Moreover, the significant increase in actual evaporation loss between 1980 and 2011 widened the gap between precipitation inputs and evaporation losses.

The step change point in air temperature in Bosten Lake Basin in 1996 was consistent with step change points in Kaidu River runoff in 1993 and 1995 (Chen et al. 2012), suggesting that warmer temperatures may have been responsible for recent increases in Kaidu River runoff, probably by increasing glacier melt and snowmelt. Mean annual glacier melt and snowmelt in the Kaidu River watershed is ∼5.0 × 10⁸ m³, forming a relatively stable water source and accounting for 15% of mean annual runoff (Chen et al. 2005). Where glaciers and snowpack are abundant, warmer temperatures can increase runoff by increasing glacier melt and snowmelt. For example, the mass balance volume of Urumqi Glacier No. 1 was significantly and negatively correlated with Kaidu River annual runoff between 1959 and 2005; as the glacier receded with climate warming, Kaidu River annual runoff increased (Ma et al. 2010). Kaidu River runoff reached a record high in 2002, during which melting water from glaciers and snow directly supplied 9.6 × 10⁸ m³ of water to the Kaidu River, accounting for >20% of annual runoff (Sun and Wang 2006). Groundwater recharge from glacier melt and snowmelt reached 19.1 × 10⁸ m³, 47% more than the annual mean of 13.0 × 10⁸ m³ (Sun and Wang 2006). Zhang et al. (2004) found that summer temperatures, runoff, glacier melt, and snowmelt have increased in recent decades in the Kaidu River watershed but have been relatively stable in autumn, winter, and spring.

Meanwhile, increasing precipitation in the mountains (Bayinbuluke meteorological station), particularly in summer (Zhang et al. 2004), also has increased Kaidu River runoff. Under global climate change, Northwest China has shifted from a warm-dry to a warm-wet climate regime since the 1990s (Shi et al. 2003); however, increases in temperature, glacier melt, and snowmelt have contributed more to the increases in Kaidu River runoff than have increases in precipitation in recent decades (Zhang et al. 2004).

Table 3 Water balance of Bosten Lake in 1996.

Water balance	Item	Quantity (× 10^8 m^3)	Percentage (%)
Inflow sources	Kaidu River	22.7	84.4
	Huangshuigou River	1.8	4.0
	Qingshui River	0.3	1.1
	Farmland drainage	0.8	3.1
	Confined aquifer replenishment	0.5	2.0
	Precipitation	1.5	5.4
Outflow sources	Konqi River	11.3	55.3
	Evaporation loss	8.6	41.8
	Seepage flow	0.6	2.8
Budget deficit		1.9

Note: data were obtained from Sun and Wang (2006), except for evaporation loss, which was estimated using the formulas in this study.

These results, together with the lack of significant change in precipitation on the plains, suggest that the increases in Bosten Lake water surface area and evaporation loss between 1980 and 2011 were caused by increasing inflow from the Kaidu River due to warmer temperatures, higher glacier melt and snowmelt, and higher precipitation in the mountains. Significant climate changes in the Kaidu River Basin accounted for 90.5% of the increase in Kaidu River runoff since 1993, whereas human activities only accounted for 9.5% (Chen et al. 2012). Bosten Lake evaporation losses were closely linked to Kaidu River runoff, inflows to Bosten Lake, and Bosten Lake water surface area between 1980 and 2011 (Fig. 7). The step change points in Kaidu River runoff in 1993 and 1995 (Chen et al. 2012) were consistent with the step change points in Bosten Lake water surface area and actual evaporation loss in 1995. Further, annual Kaidu River runoff was significantly correlated with Bosten Lake water surface area and evaporation loss between 1980 and 2011 (R² = 0.55 and 0.57, respectively; P < 0.01; Fig. 7a).

Water resource management in Bosten Lake under future climate change

Although warmer temperatures seem to have increased Kaidu River runoff and Bosten Lake water surface area since 1995, future warming may decrease Kaidu River runoff and hence Boston Lake inflows. Water from glacier melt and snowmelt will decrease as (1) large glaciers become thinner, (2) small and mid-sized glaciers disappear, and (3) snow lines recede with rising temperatures in the future. Glacier melt water to the Kaidu River is supplied by 722 glaciers covering approximately 445 km² (Ma et al. 2010). Glaciers in the middle Tianshan Mountains are already retreating with warmer temperatures. Between 1963 and 1986, 2 of 8 typical glaciers in the middle Tianshan Mountains retreated at a mean rate of <5 m/yr, whereas between 1986 and 2000, 7 of those glaciers retreated at a mean rate of 10–15 m/yr (Sun and Wang 2006). In addition, snow surface area in the Kaidu River Basin declined significantly between 2000 and 2010 in summer and winter (Li et al. 2012). Correlation analyses between climate variables and snow area suggested that precipitation did not significantly affect snow area, but air temperature was negatively correlated with snow area (Li et al. 2012). A statistical downscaling model (SDSM) predicted that mean daily temperatures of Bosten Lake Basin will rise in coming decades while annual precipitation will decline (Qiu et al. 2010). As glaciers and snowpacks shrink with warmer temperatures in the future, the water surface area and water depth of Bosten Lake likely also will decline. In fact, although not yet part of a clear temporal trend, Kaidu River runoff was 20% lower in 2012 than in 2011 at the Yanqi Bridge hydrological station. By the end of November of 2012, the lower reaches of the Kaidu River had gone dry for a total of 32 days, including 12 consecutive days from 28 September to 9 October. The water level of Bosten Lake declined to 1045.25 m a.s.l, close to the threshold level (1045 m a.s.l) below which the pumping systems in Bosten Lake cannot function. Further pumping from Bosten Lake into the Kongqi River below this threshold would dry out wetland ecosystems around the lake, substantially affecting fish and wildlife that depend on those wetlands. Without that pumping, however, the Konqi River could run dry because it depends entirely on artificial pumping from Bosten Lake. Thus, current pumping measures and quantities of pumped water may not be suitable for future water management of Bosten Lake.

In addition, the salinity of Bosten Lake has increased dramatically since the 1960s (Fig. 5A). Before 1970, Bosten Lake was a typical freshwater lake with TDS <1 g/L. TDS increased from 0.39 g/L in 1958 to 1.87 g/L in 1986, the historically highest record for the lake. In the 1990s, TDS decreased steadily and reached 1.17 g/L in 2003, the lowest since 1972. Since then, TDS has increased again. The mean TDS of Bosten Lake between 1980 and 2011 was 1.52 g/L. These higher TDS levels are considered slightly saline (1 ≤ TDS < 3 g/L) (Li et al. 2003) and can be detrimental to agricultural production, drinking water quality, and lake ecology. Aquatic biodiversity in Bosten Lake has declined significantly, and some fishes, such as bighead carp (Aristichthys nobilis), big-head schizothoracin (Aspiorhynchus laticeps), and a species of ray-finned fish (Schizothorax biddulphi) have disappeared due to the high TDS of the water (Li et al. 2003). Crop yields and soil quality have also declined in the Bosten Lake Basin because of the saline irrigation water pumped from the lake (Wang 2007). Higher inflows to Bosten Lake, which increase Bosten Lake water surface area, likely dilute Bosten Lake salts, leading to lower TDS.

Given the importance of Bosten Lake as a municipal, agricultural, industrial, and ecological water supply, several questions regarding water management should be considered. In the context of projected rising temperatures, decreasing precipitation, and reduced Kaidu River runoff due to glacier recession, how will the water resource management strategy for Bosten Lake need to be changed? How much water can be diverted from Bosten Lake annually without causing the water quality and quantity of Bosten Lake to deteriorate? Finally, when during the year should water diversions occur?

According to equations 4–6, the water surface area of Bosten Lake must be >963.14 km² to maintain TDS <1.5 g/L. Evaporation from Bosten Lake with this water surface area is approximately 8.61 × 10⁸ m³. This water level can be maintained while diverting 14.42 × 10⁸ m³/yr to the Konqi River (the minimum required to meet current agricultural and industrial water demand; Li et al. 2003) if 3-year and 5-year means for annual inflows to Bosten Lake are approximately 21.80 × 10⁸ m³and 23.50 × 10⁸ m³, respectively. When Kaidu River annual runoff is 35.83 × 10⁸ m³ (slightly greater than the multi-year mean annual runoff), it is barely sufficient to support annual Bosten Lake inflows of 21–23 × 10⁸ m³ after upstream water losses. Therefore, if Kaidu River mean annual runoff declines, water resource managers will have to choose between reducing the water supply to the Konqi Basin or maintaining diversions and reducing Bosten Lake water quality. Moreover, if Kaidu River runoff declines consistently and diversions continue apace, then the Bosten Lake water level will eventually reach 1045 m (a.s.l), and diversions to the Konqi River will become impossible. Although public water demand in the Konqi Basin may increase in the future, Bosten Lake will not be able to support greater diversions to the Konqi Basin unless Kaidu River runoff increases instead of decreasing.

Decisions on diversion of water from Bosten Lake to the lower Tarim River in the future also must be based on Kaidu River runoff. Between 1999 and 2002, mean annual runoff in the Kaidu River was 49.48 × 10⁸ m³, with the highest runoff in 2002 of 57.13 × 10⁸ m³. Correspondingly, Bosten Lake water levels increased continuously between 1999 and 2002. In that context, it was a rational decision for the Tarim River Basin Water Resource Authority to propose pumping water from Bosten Lake to the lower Tarim River to restore desert riparian forest. When Kaidu River annual runoff is relatively low, however, such as in 2012, diversions to the Tarim River may jeopardize Bosten Lake water quality and water supply to the Konqi Basin. According to our calculations, when Kaidu River runoff is <35.83 × 10⁸ m³, inflows to Bosten Lake are insufficient to support diversions to the Tarim River while continuing to meet human water demand in the Konqi River Basin and maintain adequate water quality in Bosten Lake. When Kaidu River annual runoff is >35.83 × 10⁸ m³, water may be diverted from Bosten Lake to the Lower Tarim River, but the diversion must be less than or equal to the difference between Kaidu River annual runoff and 35.83 × 10⁸ m³. Further, even in those instances, it sometimes may be preferable to retain excess water in Bosten Lake to ensure adequate water quality and supply for future, drier years. Predictive models of Kaidu River runoff under climate change are needed to guide longer-term water resource management decisions like these for Bosten Lake.

The appropriate period for pumping water from Bosten Lake is between May and September, when evaporation and public water demand are both greatest. In the Kaidu River Basin, snow begins to accumulate in autumn each year, reaching a maximum snow cover in January or February, and begins to melt in March, reaching a minimum snow cover in July or August (Li et al. 2012). Most Kaidu River runoff occurs during June, July, and August, which sometimes results in summer floods (Chen et al. 2005). Diverting water to keep Bosten Lake water levels low between May and September would not only ensure adequate irrigation water for crop growth in the Konqi River Basin, but also would decrease evaporation loss by reducing water surface area, and could reduce flood risk. Simultaneously, storing water in Bosten Lake from October to April each year after crop harvest would maintain a higher water level in winter and spring when temperatures and evaporation rates are low, thus minimizing total annual evaporation loss from Bosten Lake. Finally, concentrating water diversions during a brief time period might increase water circulation, discharge more saltwater from the lake, and accelerate mixing of lake water with freshwater from Kaidu River runoff, potentially reducing overall lake water salinity.

Recommendations for further analysis

Evaporation measurements at Bosten Lake need to be continued in future. Evaporation loss is an important factor for the water budget of Bosten Lake, accounting for ∼50% of total outflow. Although increases in Bosten Lake evaporation capacity between 1980 and 2011 were not significant, warmer temperatures in the future may lead to significant and important increases in evaporation capacity. Accounting for effects of such changes on actual evaporation losses will be critical for understanding the Bosten Lake water budget and making appropriate water management decisions.

In addition, further research is needed to develop a bathymetric map of Bosten Lake that would allow estimates of lake water volume from lake water depth or surface area. Volume estimates would provide information to water managers on the total quantity of water available and temporal variation in water availability. Further, annual volume estimates could be used to assess how much of the Bosten Lake water budget cannot be accounted for by known inflows, outflows, precipitation, and evaporation. This unaccounted for water may include local runoff, seepage, and confined aquifer replenishment. Volume estimates could be used to test whether these additional water sources also have changed over time.

Conclusions

Climatic variability has significantly affected water inflows, water surface area, and evaporation loss of Bosten Lake. Warmer temperatures beginning in 1995 seem to have increased glacier melt and snowmelt, which in turn increased Kaidu River runoff to Bosten Lake, thus increasing Bosten Lake water surface area and total evaporation. As temperatures have continued to increase, however, shrinking glaciers and snow packs have reduced more recent Kaidu River runoff and inflows to Bosten Lake.

Our calculations using a simple conceptual model suggest that 3-year and 5-year mean annual inflows to Bosten Lake must be approximately 21.80 × 10⁸ m³ and 23.50 × 10⁸ m³, respectively, and Kaidu River mean runoff must be approximately 35.83 × 10⁸ m³ (slightly greater than the mean annual runoff) to maintain adequate water quality in Bosten Lake while supplying 14.42 × 10⁸ m³/yr of water to meet public water demand in the Kongqi Basin. When Kaidu River annual runoff is greater than average, additional water may be diverted from Bosten Lake to the lower Tarim River for ecological restoration, or to the Kongqi Basin to meet increasing demand. If, however, Kaidu River annual runoff declines with climate change, then meeting current public water demand in the Kongqi Basin while maintaining adequate water quality in Bosten Lake will be challenging.

The strong influence of water surface area on actual evaporation suggests that it may be possible to reduce evaporation loss from Bosten Lake by concentrating water diversions during peak inflows between May and September to minimize water surface area. Diversions concentrated between May and September would also meet peak agricultural demand, reduce flood risk, and potentially increase water circulation and dilute lake water salinity.

Funding

This research was supported by the Xinjiang Innovative Talent Project (No. 2014721035), the National Natural Science Foundation of China (Grant No. 41271006), and the Projects of West Light Foundation of the Chinese Academy of Sciences (No. XBBS201008).