Investigation of dynamic lake changes in Zhuonai Lake – Salt Lake Basin, Hoh Xil, using remote sensing images in response to climate change (1989 – 2018)

The area covered by the four lakes in the Zhuonai Lake – Salt Lake Basin in Hoh Xil (Zhuonai Lake, Kusai Lake, Heidinor Lake, and Salt Lake) has changed signi ﬁ cantly over the past 30 years. In this study, remote sensing image data gathered via the Landsat thematic mapper, enhanced thematic mapper plus, and operational land imager from 1989 to 2018 were used to extract the areal parameters of four lakes. The total area of the four lakes had increased by 18% in the past 30 years due to climate change. Interpolated results based on the meteorological data from 28 meteorological stations in the basin were used for trend analysis. A single-layer lake evaporation model was utilized to study the changes in the annual lake evaporation in the basin. The annual lake evaporation slightly increased from 1989 to 1995, followed by a sharp decrease from 1995 to 2018. From 1989 to 2018, the annual evaporation in the basin ranged between 615.37 and 921.66 mm, with a mean of 769.73 mm. A mass balance model was developed to estimate the changes in the lake volumes due to precipitation and evaporation. The increase in precipitation and the decrease in the annual lake evaporation promote the expansion of the four lakes. Lake evaporation is the main factor inducing changes in the lake areas.


GRAPHICAL ABSTRACT INTRODUCTION
Lakes in the Tibetan Plateau (TP) can act as indicators of global climate change (Li et al. a). The average elevation of TP is over 4,000 m and contains more than 1,200 lakes (>1 km 2 ; Zhang et al. ). In the past 30 years, the total number of lakes and their surface areas have increased significantly (Li et al. a). Due to minimal human activities, the water balance of these endorheic basins is mainly controlled by the regional climate (Sheng & Yao ; Haas et al. ). Climate change has been an important factor affecting the evolution of high-altitude lakes (Lei et al. ) because sustained warming has resulted in the retreat of permafrost and glaciers, supplying water to the lakes. In addition, the general increase in precipitation and the decrease in evaporation contribute water to alpine lakes (Yang et al. ). Therefore, long-term monitoring of the lake dynamics can be used to uncover the effects of climate change on water resources and the environment in the TP (Beniston et al. ; Li et al. a). Song & Sheng ) and northwestern TP (Qiao & Zhu ). During the past few decades, the amount of precipitation has increased, especially in the central TP. This increase correlates with observed lake changes and is considered to be the main reason for these changes (Liu et al. ; Lei et al. ; Song et al. a, b). Hydrological modeling results indicate that precipitation is also the main factor driving lake expansion (Biskop et  Siling Co Lake during 1961-2015 (Guo et al. ). The results of the simulations agree well with those of the eddy covariance system. In addition, the primary advantage of the single-layer model is that it explicitly considers the heat storage in the water using the output water temperature (Xu et al. ). Therefore, it may be preferable to simulate the long-term evaporation of the TP lakes using the single-layer model. Based on Hwang et al. (), the collapse was due to the combined action of heavy rainfall and two earthquakes. Liu et al. () suggested that the accelerated melting of glaciers contributed to the event. Although these studies were conducted to determine the cause of the collapse of Zhuonai Lake, they provided insights into the dynamic changes of the four lakes before and after the collapse. Before Zhuonai Lake collapsed, all four lakes experienced different degrees of expansion. In previous studies, the effects of lake evaporation and rainfall on lake expansion were not quantified.
Therefore, the aims of this study are to analyze the dynamic changes of the four lakes from 1989 to 2018 and their response to climate change and to quantify the contribution of lake evaporation and precipitation to lake expansion.
Based on data recorded at the Wudaoliang meteorological station, the multiyear average temperature, annual mean precipitation, average wind speed, and annual average sunshine duration in this area are À4.86 C, 313.8 mm, 4.2 m s À1 , and 2,792.6 h, respectively. The total area of the Zhuonai Lake-Salt Lake Basin is 8,728.78 km 2 , accounting for 23.37% of the Hoh Xil Heritage Area (Hu ). Details about the four lakes in the Zhuonai Lake-Salt Lake Basin, Hoh Xil, are presented in Table 1.
The Zhuonai Lake-Salt Lake Basin is located at the northern margin of the Qinghai-Tibet Plateau climate zone. The basin is characterized by distinct cold and hot seasons, relatively high temperature, ample precipitation, long winters, short summers, small annual but large daily temperature differences, long sunshine hours, and strong solar radiation. The plant growth period in the basin is short and lacks a frost-free stage. The average temperature is low, and the cold season lasts up to 7 months. The area is characterized by semi-arid climate, with evaporation exceeding precipitation. The main source of water recharge is snow and ice meltwater, followed by spring water and atmospheric precipitation. The regional precipitation gradually decreases from the southeast to the northwest. Snow disasters comprise the major meteorological disasters in the region (Hu ).
The Zhuonai Lake-Salt Lake Basin is surrounded by mountains on three sides, i.e., the Kunlun Mountains in the north, Haoriarijiu and Yueba mountains in the south, and Wuxuefeng, Dakanding, and Heishi mountains in the west (Hu ). The Hudong and Pingding mountains are within the basin. Zhuonai Lake, also known as Huotongnuoer, is located in the west of the series of lakes, in the uppermost reaches. Kusai Lake is the sixth largest lake in the Hoh Xil Nature Reserve and forms an important saline wetland area together with the nearby Heidinor Lake and Salt Lake in the northeast of the reserve (Hu ).

Remote sensing data
In this study, Landsat images (Table 2)

Image preprocessing
In this study, 60 Landsat images were used to extract the areas of the four lakes in the basin from 1989 to 2018.
Owing to the abundant annual precipitation in the basin from May to early October and the long freezing period of the lakes (December to early May), images from mid-October to end of November without cloud cover over the lakes were selected to extract the lake areas to reduce the effects of precipitation and lake freezing. The Landsat images obtained for the four lakes during the study period are shown in Figure 2. Because the Landsat 7 ETMþ images contained stripes, they were visually interpreted after removing the stripes.
Other images were visually inspected after the automatic extraction of waterbodies. Because it is difficult to collect images of the entire basin in the same time period, the seasonal changes of the lake areas were not considered in this study. Geometric correction and calibration were applied to the images to accurately reflect the Earth's surface. Image pixel values were converted to spectral radiance, and an atmospheric correction was applied to radiation image data to obtain spectral reflectance images.  1989/10/10; 1990/11/14; 1991/11/01; 1992/11/ 03; 1993/11/22; 1994/ 11/09; 1995/11/22; 1996/11/14; 1997/11/ 01; 1998

Lake extent mapping
The lake areas were obtained from Landsat satellite images.
The spectral water index was derived from arithmetic operations applied to two or more spectral bands (e.g., the ratio, difference, and normalized difference). Although the images used in this study had the same spatial resolution, different spectral and radiation resolutions were generated during the segmentation process. The spectral water index is based on the fact that water absorbs radiation energy in the near-infrared (NIR) and short-wave infrared wavelength ranges (Phiri & Morgenroth ). In this study, the lake area was extracted using the normalized differential water index, which is defined as follows (McFeeters ): where GREEN denotes the green band and NIR denotes the NIR band. In this method, after band operations, the spectral curves are analyzed to set reasonable thresholds, the number of pixels within the thresholds is counted, and the lake area is calculated according to the resolution of the satellite image.
Climate change is an important factor affecting the lake surface area. To examine the effect of climate change on the changes of lake areas, the inverse distance weighting method was used in this study to interpolate monthly average and annual average meteorological data (temperature, wind speed, sunshine hours, and precipitation) recorded at 28 stations and to obtain meteorological basin data in raster format.

Single-layer lake evaporation model
The single-layer lake evaporation model ( meteorological station. The lake surface temperature and measured evaporation were used to validate the model. The coefficient of determination (R 2 ), Nash-Sutcliffe coefficient (NSE), root-mean-square deviation (RMSD), and mean bias error (MBE) were used to evaluate the model. and where n is the number of data, and X i , X 0 i , X, and X 0 are the measured value, model output, mean of the measured value, and mean model output, respectively.

Change in the lake mass balance
The variation in the lake water quantity depends on the inflow of land runoff (including glaciers), lake evaporation, and precipitation. In the TP, glacier mass loss may contribute to the lake water balance. Therefore, the lake water balance equation can be written as (Yang et al. ): where A 1 , A b , V 1 , G, and S are the lake area (m 2 ), basin area (m 2 ), lake water volume (m 3 ), glacier volume (m 3 ), and groundwater storage (m 3 ), respectively; R (m s À1 ) is the precipitation-generated land runoff; E 1 (m s À1 ) is the lake evaporation; and P (m s À1 ) is the precipitation.
Based on Equation (6), Let: and where V P (m 3 ) and V E (m 3 ) are the volumes of water change caused by precipitation and evaporation, respectively, and C R is the runoff coefficient (Yang et al. ).
During a period with stable lake volume (i.e., dV 1 =dt ¼ 0), d(G À S)=dt is assumed to be zero. Equation (6) may be simplified as follows: The solution of V P and V E can be obtained using the following procedure.
(2) Calculate the annual V P and V E values of the four lakes.
(3) Calculate the cumulative changes based on steps 1 and 2 according to the stable year of the lake as follows: and

Standardized precipitation evapotranspiration index
The standardized precipitation evapotranspiration index where T, I, and m are the monthly mean temperature ( C), heat index, and coefficient depending on I, respectively; D i,l is the P À PET difference in the first month of year I; α, β, and γ are the scale, shape, and origin parameters, respectively; W ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi À2 ln (P) p and P are the probability of exceeding a determination D value. The constants are C 0 ¼ 2:515517, C 1 ¼ 0:802853,

Temperature
As shown in Figure 6, the annual mean temperature of the basin increased from 1989 to 2018. The annual mean temperature in the basin ranged between À2.08 and 0.22 C, with an average of À0.77 C. The annual mean temperature in the basin was lower than that at the Golmud and Xiaozaohuo stations, but higher than that at the Wudaoliang station.
The changes in the annual mean temperature do not correlate with the changes in the lake areas (Liu et al. ). Note that the effect of the air temperature on the water balance in the lakes is mainly related to glacier retreat and permafrost thawing (Yang et al. ). The temperature increase promoted the lake expansion to a certain extent, but it was not the dominant factor.

Wind speed
From 1989 to 2018, the annual mean wind speed in the basin first declined and then increased, as shown in Figure 7.
The annual mean wind speed in the basin ranged between 3.18 and 4.01 m s À1 , with an average of 3.49 m s À1 . The annual mean wind speed of the basin is similar to that recorded at the Xiaozaohuo Station, lower than that obtained at the Wudaoliang Station, and higher than that A decrease in the wind speed can lead to a decrease in the lake evaporation, which indirectly promotes lake expansion.

Sunshine duration
As shown in Figure 8, from 1989 to 2018, the annual mean sunshine duration in the basin decreased at a rate of   The annual mean sunshine duration in the basin was lower than that at the Golmud and Xiaozaohuo stations, but higher than that at the Wudaoliang Station. Similar to the wind speed, the duration of sunshine influences the lake area by affecting the lake evaporation. The longer the sunshine duration, the greater the lake evaporation (Guo et al. ). Therefore, a continuous decline in the sunshine hours can promote the lake expansion.

Standardized precipitation evapotranspiration index
The SPEI was used to analyze the effects of precipitation and air temperature on lake expansion. Figure 9 shows the 3-, 12-, and 24-month SPEIs for the basin between 1989 and 2018. Based on the SPEI, the basin was mainly dry, but the drought started to decline in 2007. Before 1996, drought may have led to lake shrinkage. However, in 2007, a humid period occurred, which may have accelerated the lake expansion. According to the trend analysis, the change in SPEI can be divided into two periods: from 1989 to 1995, the SPEI increased (0.097 yr À1 ); from 1996 to 2018, the SPEI decreased (À0.012 yr À1 ). As the area of the four lakes changed dramatically after 2010, we compared the Pearson correlation coefficients between the changes in the four lakes before 2010 and the SPEI, as shown in Table 4. There was a negative correlation between the SPEI and lake area change. From 1989 to 1995, owing to the decrease in precipitation and the increase in temperature, the SPEI showed a significant upward trend. From 1996 to 2010, both precipitation and temperature showed an upward trend, while the SPEI showed a downward trend, which was relatively slow. Therefore, the correlation Figure 9 | The 3-, 12-, and 24-month SPEIs of the basin .

Long-term variation in the annual lake evaporation in the basin
The surface temperature data measured during June and July 2018 were used to validate the simulated lake surface temperature. The surface temperature is strongly affected by the release of energy stored in the water and by surface water energy exchange processes, such as cooling/warming involving sensible and latent heat, which are affected by meteorological variables (Zhang & Liu ). As shown in    Lake at lower altitude is higher than that of the basin, while the evaporation of Ngoring Lake is lower than that of the basin. This might be due to the different climate and atmospheric conditions.

Response of the lake area to climate change
Assessing the impact of climate change on lake area variations is essential for water resource management and ecological protection. The variations in the areas of the four lakes and meteorological data over the past 30 years obtained using the MK method are shown in Table 5. The temperature affects the water balance of lakes, mainly by affecting evaporation and glacier melting.
The sunshine duration and wind speed do not directly affect the change in the water quantity of the lake, but they affect the lake evaporation. The longer the sunshine duration and the higher the wind speed are, the greater is the lake evaporation (Guo et al. ). For closed lakes, the evaporation is the only output component because there is no runoff.
Response of the lake water volume to evaporation and precipitation changes According to the changes in the areas of the four lakes, Zhuonai, Kusai, and Heidinor lakes were relatively stable in 1990 and their areas insignificantly changed, whereas Salt Lake was relatively stable in 1991. Therefore, 1990 was selected as the base year for Zhuonai, Kusai, and Heidinor lakes and 1991 was the base year for Salt Lake to calculate the changes in lake water volumes due to precipitation and evaporation until 2010. The results are shown in Figure 12 and Table 6 (the time range for lakes Zhuonai, Kusai, and Heidinor is 1990-2010 and that of Salt Lake is 1991-2010). Table 6 shows that the water volumes of lakes Zhuonai, Kusai, and Salt increased due to rainfall and that of Heidinor Lake decreased. The lake water volumes of all four lakes decreased due to evaporation. Both the increase in precipitation and decrease in evaporation promote the expansions of the lakes, but the decrease in evaporation is the main control factor.  1989-1995 1995-2010 1995-2011 2010-2018 2011-2018 1989-2018  Although this study provides monitoring data with high temporal resolution for four lakes in the basin and indicates that the recent lake dynamics may be mainly driven by lake evaporation, it has a few limitations. First, extracting the surface area of lakes from satellite remote sensing data might lead to misestimates (Sun et al. ). Second, we only quantified the contributions of evaporation and precipitation to lake expansion. However, to accurately estimate the effect of different factors on lakes, additional parameters must be considered (Zhu et al. ), including precipitation, evaporation, runoff, other sub-basins, groundwater, and the ice mass loss of glaciers. Thus, detailed research on the interactions between these factors and lakes will be conducted in the future.

CONCLUSIONS
The areas of the four lakes have significantly changed due to climate change. In 2011, Zhuonai Lake burst, and its area continues to shrink. The areas of Kusai and Heidinor lakes increased gradually, whereas the area of Salt Lake sharply increased. Climate change has caused dramatic changes in the areas of the four lakes and affected the hydrological conditions of the lakes, resulting in their integration.
Because the water level of Salt Lake continuously rises, the Figure 12 | Changes in the lake water volume due to precipitation and lake evaporation. lake water may overflow, posing great threats to the Qinghai-Tibet Highway and Qinghai-Tibet Railway. Both the increase in precipitation and decrease in lake evaporation promote the expansion of lakes, but the decrease in lake evaporation is the main control factor. Changes in the precipitation and lake evaporation directly affect the lake water volume. The effect of the air temperature on the water balances of the lakes is mainly related to evaporation and glacier processes. The wind speed and sunshine duration mainly affect lake evaporation, but have no direct impact on lake expansion. In this study, the effects of glacier melt and permafrost thawing were not considered; however, they might have significant effects on lake expansion.