Hydroclimatic changes of Lake Bosten in Northwest China during the last decades

Bosten Lake, the largest inland freshwater lake in China, has experienced drastic change over the past five decades. Based on the lake water balance model and climate elasticity method, we identify annual changes in the lake’s water components during 1961–2016 and investigate its water balance. We find a complex pattern in the lake’s water: a decrease (1961–1987), a rapid increase (1988–2002), a drastic decrease (2003–2012), and a recent drastic increase (2013–2016). We also estimated the lake’s water balance, finding that the drastic changes are caused by a climate-driven regime shift coupled with human disturbance. The changes in the lake accelerated after 1987, which may have been driven by regional climate wetting. During 2003 to 2012, implementation of the ecological water conveyance project (EWCP) significantly increased the lake’s outflow, while a decreased precipitation led to an increased drought frequency. The glacier retreating trend accelerated by warming, and caused large variations in the observed lake’s changes in recent years. Furthermore, wastewater emissions may give rise to water degradation, human activity is completely changing the natural water cycle system in the Bosten Lake. Indeed, the future of Bosten Lake is largely dependent on mankind.


Results
Bosten Lake climatic variability in the past five decades. The PDSI, the SPI and the SPEI are used extensively to assess drought and wetness variability. These indices are formulated by climatic water supply (precipitation) alone or water supply and demand (evaporation) 30,33 . Although drought and wetness variability is primary determined by precipitation, drought severity can be aggravated by higher atmospheric evaporative demand as a consequence of global warming 46 . Calculating the evaporative demand is very critical to monitor drought and wetness variability. The FAO-endorsed Penman-Monteith equation (P-M), is the most popular and physical-based one, employed for the calculation of the potential evaporation (PET) 33 .
To quantify the recent changes in drought and wetness conditions over the Bosten Lake basin, Fig. 2a shows a time series of the SPEI, SPI and sc_PDSI indices, and Fig. 2b compares the frequency based on the SPEI for the three epochs. As shown in Fig. 2a, the SPEI is a good match with sc_PDSI and SPI. The SPEI had a positive correlation with SPI (R = 0.58, p < 0.01) and sc_PDSI (R = 0.47, p < 0.01) of the Bosten Lake basin. Furthermore, it illustrates that the three major periods are 1961-1987, 1988-2002 and 2003-2016. The epoch from 1988 to 2002 can be qualified as exhibiting high wetness conditions, while the other two remained at a drought level (Fig. 2a). Despite the smaller overall trend (−0.007 ~ 0.025 per year, p > 0.05) in the drought indices from 1961-2016, there was a switch from 1988 onwards to a significantly decreasing trend (−0.023, −0.064 and −0.063 per year at the 0.01 significance level, correspond to SPEI, SPI and sc_PDSI, respectively). SPI-based drought severity has become significantly aggravated relative to that based on the SPEI, which is independent of the effect of the decreased precipitation.
In comparison to the epoch from 1961-1987, intensified wet conditions can be concluded from the wetness category frequency during 1988-2002, where the number of wet months increased from 9.3 to 37.1 months per decade (Fig. 2b). For the period 2003-2016, the drought frequency also increased by a factor of several relative to the period 1988-2002. On average, the drought frequency increased from 6.1 to 14.7 events per decade, and the extreme drought frequency rose rapidly to 1. Bosten Lake level, area and storage changes.  (Fig. 3a). In tandem with these changes, the level changes were divided into four epochs: 1961-1987, 1988-2002, 2003-2012, and 2013-2016. Overall, there was a decreasing trend (−0.024 m per year) (Fig. 3a). The lake level presents a decrease between 1961 and 1987, thereafter it shows a dramatic increase, and then a continuous decrease, but a dramatic increase of lake level is mapped in recent years. The average rate of change in the lake levels is  In addition, the multi-year average of lake storage was 7.22 ± 1.06 km 3 . The rate of change in the water storage in Bosten Lake was −0.088 km 3  Overall, Bosten lake's levels, area and storage changes varied in four stages: a decrease (1961-1987), a rapid increase (1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002), a drastic decrease (2003-2012) and a recent drastic increase (2013-2016). The drastic changes in the lake levels and area after 1987 accelerated, which could be a driven regime shift arising from regional climate changes 47,48 . Observations from the CMA stations in the lake catchment indicate that temperature and precipitation exhibited a dramatic change during 1961-2015 (Fig. 3d). Temperatures experienced a sharp increase in 1997, and since then remained highly volatile. Precipitation exhibited a sharp increase in 1987, and since then the increasing trend has diminished during the past two decades (Fig. 3d). These changes are  [49][50][51] . The drought severity was aggravated by a greater evaporative demand caused by the significant temperature rise 32,46 . Dramatic changes in the climatic conditions may bring some adverse effects, and the drought and wetness variability could have played a leading role in altering the lake.
Quantifying the effects of the hydro-climatic components to runoff changes. Lake inflow is dominated by components that supply the lake with water and regulate changes in its level. Approximately 95% of the total lake inflow to the Bosten Lake comes from the Kaidu River 40,41 . The runoff from Kaidu River is mainly supplied by mountainous precipitation and melting glacier water, which accounts for 61.5% and 38.5%, respectively 51 . Therefore, the runoff change experienced by the Kaidu River is dominated by changes in mountainous precipitation, glacier melting water and evaporation.
The climate elasticity method was used to quantify the effects of the above components to river runoff for the Kaidu River. The values of the climate elasticity coefficient of annual runoff to precipitation ∂ are 0.78 ± 0.005, −0.05 ± 0.01 and 0.27 ± 0.01, respectively. This means that a 10% increase in precipitation (glacier melting water) will result in a 7.8% (2.7%) increase in runoff, and that a 10% increase in evaporation will result in a 0.5% decrease in runoff. This illustrates that the change of lake water inflow from the Kaidu River is more sensitive to changes in precipitation and glacier mass loss than to changes of evaporation.
Bosten Lake Water Balance. The Bosten Lake water balance was estimated based on observations of several components, including precipitation (land and lake surface), evaporation (land and lake surface), glacier melting water, and lake input and output. The lake exhibited drastic changes in two phases (1988−2002 and 2003−2015). Variations in the components of the water balance in both phases are listed in Fig. 4 and Table 1. The annual mean and trends of precipitation and evaporation were spatially different in the basin. The average annual precipitation on land surfaces was 287.5 mm and 290.9 mm in both phase (Fig. 4a), while the precipitation on lake surfaces was 91.2 mm and 70.2 mm, respectively (Fig. 4a,b). Precipitation on land surfaces increased significantly in both phases, while the precipitation on lake surfaces decreased. The rate of change in the average annual precipitation on land surfaces was 52.9 mm per year and 13.9 mm per year for the two epochs, Lake water balance components for the Bosten Lake in three phases (1960-1987, 1988-2002 and 2003-2015) were listed in Fig. 5. The mean annual precipitation over the lake was 0.29, 0.29 and 0.31 km 3 in three phases (similarly hereinafter), respectively, surface runoff water input generated by precipitation was 1.93, 2.71 and 2.18 km 3 , glacier melting water into the lake was 0.52, 0.73 and 0.59 km 3 , evaporation from the lake was 1.03, 1.07 and 0.88 km 3 , water outflow from the lake was 1.15, 1.55 and 1.83 km 3 , groundwater inflow and error was −0.67, −0.82 and −0.85 km 3 , and the change in lake storage was −0.12, 0.31 and −0.48 km 3 . This means that lake water increased due to the increased water input (R in and G) higher than water output (E L and R out ) for the 1988-2002. In 2003-2015, the lake water decreased was caused by the decreased runoff (R in and G) and increased water output (R out ).

Discussion
Climate-driven regime shift and their influence to lake changes. Over the past 50 years, studies showed that the climate exhibited a dramatic change in northwestern China, which is becoming warmer and wetter 47,48,[53][54][55] . Dramatic changes in the climatic conditions may bring some adverse effects 32 , where the dramatic fluctuations of Bosten Lake's levels are a case in point. In this study, we compared the time series and frequency between the SPEI, SPI and sc_PDSI indices, and found that SPEI is a good match with sc_PDSI and SPI. Actually, SPEI is an effective tool to monitor recent climates under the global warming, due to it combines the multi-scalar  Table 1. Lake water balance in Bosten Lake from the 1961 to the 2015. The lake water balance components include changes in the lake water storage (ΔL), lake precipitation (P L ), runoff generated by precipitation (R in ), melting glacier water (G), lake evaporation (E L ), lake outflow (R out ), and groundwater exchange and errors (R g ± ε).

Figure 5.
Water balance in three phases over the Bosten Lake. The lake water balance components include lake precipitation (P L ), runoff generated by precipitation (R in ), glacier melting water (G), lake evaporation (E L ), lake Outflow (R out ), groundwater exchange and errors (R g ± ε) and average change in lake storage (ΔL). The map was created using Matlab 2012a, http://cn.mathworks.com/products/matlab/.
character of the SPI and the sensitivity of PDSI to changes in evaporative demand 31 . The lake level change is highly similar to the SPEI time series, lake level rose rapidly from 1988, and the SPEI increased at the same time (Fig. 6). SPEI exhibited a sharp increase in 1988, and since then the increasing trend decreased during the past two decades. Meanwhile, the drought severity was aggravated by a significant rise in temperature coupled with an insignificant increase in precipitation 32 . Undoubtedly, while climate-driven regime shifts contribute a part of the lake level changes, and there are also other factors including ecological water conveyance, agricultural irrigation and water consumption.
Retreating glaciers and dramatic lake changes. The river runoff from the Tianshan Mountains heavily depends on glaciers (snow), hence glacier meltwater occupies an important role in the total amount of river discharge 50 . Glacier melt and their retreat in northwest China has been accelerated by global warming, where 82.2% of glaciers are retreating, and their total area has decreased by 4.5% 56 . The retreating trends have been exacerbated since the 1990s 57 , and the Bosten Lake Basin is no exception. From the climate elasticity method and lake water balance estimate in Bosten Lake, glaciers are a key factor for lake expansion/retreat. Indeed, glaciers show a retreating trend in the Bosten Lake Basin during 1963 to 2000. The decrease in glacier area is 38.5 km 2 , and the decrease rate was 0.31%/a 56 . Table 2 provides an overview of the glacier area and volume changes in the Bosten Lake Basin. The glacier area and volume changes rate are −15.3% and −19.5% during the 1960s to 2000s, respectively. Most of the glaciers in the basin are small-scale ones, which account for 72% of the total number of glaciers 58 . There is an obvious retreat for small-sized glaciers (glacier area <1 km 2 ), where the glacier area and volume changes rate were 23.9% and 31.4% during 1963 to 2004, respectively 58 . Figure 7 shows the average glacier mass change rates in the Bosten Lake Basin during 1961 to 2012 as calculated with the ensemble of models. The average glacier mass change rate was −0.63 ± 0.31 × 10 3 kg m −2 yr −1 during 1961 to 2012. In the period 2003−2009, the average glacier mass change rates was −0.69 ± 0.28 × 10 3 kg m −2 yr −1 and −0.68 ± 0.43 × 10 3 kg m −2 yr −1 as calculated from the models and the ICESat data, respectively 52 . From the analyses of glacier changes, the lake area dramatically expanded from 1988 to 2002, and dramatically shrank from 2003 to 2015, while showing an overall decrease in the total glacier area and mass from 1961 to 2012. Therefore, the dramatic changes of the lake, and the continuous glacier retreat are not coupled in the Bosten Lake Basin. It was also suggested that the changes experienced by Bosten Lake were dramatic and caused primarily by variations in the precipitation in the basin, and secondarily by the accelerated rate of melting glaciers. Similar results were found in the Yamzhog Yumco Basin, Tibetan Plateau 59 .
Glacier recession rates were obviously larger in the 2000s than in previous epochs, which results from the continuously rising temperatures. From 1961 to 2016, the annual mean temperature, maximum temperature and minimum temperature all experienced a significant increasing trend in the Tianshan Mountains, with rates of 0.34 °C/10a, 0.19 °C/10a and 0.56 °C/10a, respectively. Meanwhile, substantial glacier mass loss in the Tianshan Mountains was reported over the past 50 years 52 . A study suggested that the temperature increase had a large effect on the melting of small scale glaciers 60 . Thus, the rising temperatures clearly accelerated the rate of melting glaciers due to the dominance of small-sized ones.
Lake changes and water balance estimates. According to the above analyses, precipitation and glacier melting water are the main sources of recharge for lake water input, and influence whether a lake expands or retreats. Table 1 lists the average values of the Bosten Lake water balance components and their proportions between 1961 and 2015. Lake precipitation, inputs generated by precipitation, and glacier melting water accounted for  respectively. Therefore, the slight change in evaporation during both phases was not able to reduce the lake water as the increased input of lake water was much more than the increased loss arising from evaporation. The average lake water supply increased by 0.8 km 3 per year during 1988 to 2002 in comparison with that in 1961−1987. The portion derived from precipitation (including lake surfaces and generated runoff) and glacier melting water contributed 73.75% and 26.25% of the total supply. Indeed 5% and 50% of water was lost due to an increase in evaporation and output, while 53.75% of them contributed to the lake enlargement. The average lake water storages decreased by 0.79 km 3  Irrigation and water conveyance exacerbated lake water shortage and degradation. The annual runoff from the Kaidu River contributes about 95% of the total lake water input 40,41 . Figure 8a also shows has a clear positive correlation between the river runoff and lake inflow. However, water consumption in the Yanqi Oasis clearly influences the lake water inputs via agricultural irrigation, industrial and domestic water use. Figure 8(b) displays the variations of annual water consumption and river runoff during 1961 to 2010. The average annual river runoff totaled 3.51 km 3 , however, the average annual water consumption in the basin was 1.31 km 3 , and water consumption accounted for 37.3% of the total river runoff. For the different periods, water consumption accounted for 40.4%, 30.2% and 42.5% of the total river runoff for the three periods mentioned above, respectively. The Bosten Lake Basin is an irrigation agricultural area, and is mostly composed of farmland. The area of the farmland was about 5.47 × 10 4 ha in 1990, and then it grew rapidly in recent years, reaching a value of 11.75 × 10 4 ha in 2010 51 . Hence irrigation accounted for 90% of the total water consumption in the lake basin 61 . All these indicate that water consumption accounts for over one half of the total river runoff, implying that water consumption by irrigation has largely influenced the dramatic changes seen in the lake basin.
Bosten lake basin is a main source of the Tarim River, which is the longest inland river in China, and has undergone major ecological degeneration caused by unbridled utilization of water resources 62,63 . The Chinese government implemented the Ecological Water Conveyance Project (EWCP) in 2000 in order to recover the "Green Corridor" in the lower reaches of the Tarim River 64,65 . The EWCP transferred water from the Bosten Lake to the Daxihaizi Reservoir, and finally to the Taitema Lake 64 . The EWCP implementation significantly increased groundwater levels and effectively restored degraded vegetation 66,67 . However, water used by the EWCP was mainly provided by the Bosten Lake and irrigation saving water, which further exacerbated water scarcity in the Bosten Lake Basin. The total ecological water taken from Bosten Lake was 1.91 km 3 during 2002 to 2010, and stop watering after 2010 due to low-flow years (Fig. 7c). The total dissolved solids (TDS), synonymous with the water salinity, is a key indicator of the water environment 68 . Its changes in Bosten Lake also experienced three stages: a rapid increase (1960-1987), a drastic decrease (1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002), and a recent increase (2003−2015), but it showed a clearly anti-phase with the lake level changes (Fig. 3e). This shows that the TDS decreased as lake levels increased. The smallest value of TDS was 0.6 g/L in 1960 while the largest value was 1.87 g/L in 1987 (Fig. 3e). The lake has experienced salinization due to the large-scale water reclamation in the source areas, excessive exploitation and utilization of water resources, reduction of inflow and increase of salt flux into the lake 44 . The TDS changes are significantly (negatively) correlated with changes in lake inflow, lake outflow and lake level, with correlation coefficients of −0.51, −0.84 and −0.69 (p < 0.01), respectively. The TDS changes were dominated by the lake level, and the wastewater emissions is also an important factor affecting salinization of the lake. A correlation analysis  42 . The industrial wastewater emissions exceeded more than 50% of the total wastewater emission in the lake basin 42 . The wastewater emissions lead to the increase of salt flux into the lake, expediting its salinization, and finally restricting the freshwater environment.
All these demonstrate that irrigation and water conveyance exacerbated the water shortages in the Bosten Lake Basin, and that wastewater emissions gave rise to water salinization. The intensity of human disturbances in the Bosten Lake Basin was 62~67.7% before 2000, and reaching more than 80.8% in the 21 century 42 . The water system is clearly fragile in the Bosten Lake Basin, and the uncertainty of volatile water resources is exacerbated by global warming. However, human activity is largely changing the natural water cycle system in the Bosten Lake basin. The future of Bosten Lake is largely dependent on human activities.

Conclusions
Bosten Lake, the largest inland freshwater lake in China, has experienced drastic change over the past five decades. This study aimed to identify annual changes in the lake's water components and investigate water balance During 2003 to 2012, implementation of the ecological water conveyance project (EWCP) significantly increased the lake's outflow, while a decreased precipitation led to an increased drought frequency. Furthermore, wastewater emissions may give rise to water degradation, human activity is completely changing the natural water cycle system in the Bosten Lake. Methods Lake Water Balance. The Bosten Lake is an open lake with several outlets; therefore any changes in the lake's overall water level (ΔL) includes precipitation over the lake's surface (P L ), surface runoff inflow to the lake (R in ), glacier melting water (G), evaporation from the lake surface (E L ), water outflow from the lake (R out ), and the groundwater inflow and its error (R g ± ε). The lake water balance model can be expressed as: Climate Elasticity Method. For a given catchment, the change of river discharge (R) for different phases can be expressed as: where ΔR C and ΔR H are changes in R arising from climatic variability and human disturbances, respectively. The Kaidu River basin is in an uninhabited natural system because human disturbance is very slight 42,69 . Runoff is mainly supplied by mountainous precipitation and melting glacier water 42 . Therefore, the change in R is dominated by climatic variability (ΔR C ), which can be approximately estimated as: are the contributions of changes in P, ET 0 and G to R, respectively. Therefore, these relationships can be expressed as: Based on the Budyko hypothesis 70 , the climate elasticity coefficients can be calculated as follows: Thus, the contributions of changed G to changed R can be expressed as:  74 . The quality of the Landsat-and MODIS-data was assessed against available observations 74 . The total dissolved solids (TDS) of Bosten Lake from 1960-2012 were collected from the Xinjiang Environmental Protection Academy of Science in China (http:// www.xjaeps.com/). The monthly meteorological data were providing by the China Meteorological Data Service Center (http:// data.cma.cn/) for five ground-based meteorological stations (Korla, Yanqi, Hejing, Heshuo and Bohu station) for the periods of 1961-2015. Mean data from the 5 stations were used for the basin-scale analysis. The original monthly data includes the air temperature, precipitation, maximum temperature, minimum temperature, pressure, and relative humidity. To guarantee consistency, the monthly data were checked to ensure that they met the expected standards. The standard requires strict quality control processes including extreme inspection, time consistency check, and others before releasing these data. The annual observed data of the river's discharge, lake inflow and outflow series  were collected from the Dashankou (DSK), Baolangsumu (BLSM) and Tashidian (TSD) controlling sites from the Xinjiang Tarim River Basin Management Bureau in China (http:// www.tahe.gov.cn/Category_1/Index.aspx). The pan evaporation measurements were taken from the Yanqi meteorological station collected from Xinjiang Meteorological Bureau, whose measurements ended in 2009. These data were used to represent the evaporative demands of the atmosphere instead of actual evaporation from the lake's surface. Traditionally, the actual amount of evaporation was determined using the pan evaporation multiplied by a conversion coefficient, which was set to 0.47 for Bosten Lake 21 .
Scientific RepoRts | (2018) 8:9118 | DOI:10.1038/s41598-018-27466-2 In this paper, we used the PDSI 24,75 , the SPI 25 and the SPEI 26 to assess the drought and wetness variability. The potential evaporation (PET) is a key factor when calculating the PDSI and SPEI values. To calculate the PET based on the Penman-Monteith model, we collected measurements of the air temperature, maximum temperature, minimum temperature, pressure, relative humidity from the meteorological stations in the lake basin. The meteorological data in this study are providing by the China Meteorological Data Service Center (http://data.cma. cn/). The self-calibrating PDSI (sc_PDSI) was reported by the Dai et al. 75 , with the spatial resolution of 0.5 degree (https://www.esrl.noaa.gov/psd/data/gridded/data.pdsi.html). We used the regional average sc_PDSI value in the lake basin region (40. 5°N-42°N, 85°E-88°E). More detailed information on the calculation of the three drought indices are presented by Vicente-Serrano et al. 46 , Dai et al. 75 and McKee et al. 25 .