The Dynamics of Hongjian Nur, the Largest Desert Freshwater Lake in China, during 1990–2017

: China’s largest desert freshwater lake, Hongjian Nur (HN), which is the largest habitat of relict gull ( Larus relictus ), has rapidly changed in recent years. However, it is difﬁcult to quantitatively monitor the dynamics of the lake and determine the causes of its changes due to the lack of in situ observation. In this study, a remote sensing-based approach was utilized to overcome these limitations. The monthly water areas during 1990–2017 were ﬁrst extracted from Landsat multispectral images via an improved method based on the ﬂoating algae index (FAI). Then, lake surface elevations measured by real-time kinematics (RTK) were used to calculate the variations in the water storage of HN. Finally, the driving factors of the rapidly changed HN in different periods were investigated by correlation analysis. The result indicated that the drivers affecting the water storage of HN in different periods were not the same. Climate change was the main driving factor of lake level ﬂuctuation during the HN relatively stable stage (1990–1998). Drought and the intensiﬁcation of human activities were the main factors for the rapid shrinkage of the HN during 1999–2010. Human activities, especially coal-related industries and reservoir impoundment, likely was the primary factors driving the decrease in the water storage of HN from 2010 to 2015. After 2015, the policies that decreased the water consumed by human activities formulated by the government and humid climate were the main factor for the expansion of HN.


Introduction
Lakes in arid and semi-arid regions play an important role in local water circulation and ecological environment protection and therefore have special significance for local human inhabitants, animals, and plants [1][2][3]. As lakes are highly sensitive to changes in the climate and environment [4][5][6][7][8][9], quantifying the changes of the lake area, water level, and water storage is crucial to understanding various ecosystem processes, including the water balance [10,11], carbon cycle [12,13], and surface-groundwater dynamics [14]. In recent years, lake ecological environments have continued to deteriorate under the effects of climate change and human activities; as a result, many lakes have shrunk or even dried up completely in China's arid and semi-arid regions [11,15,16] In order to prevent further deterioration of lake environments in arid and semi-arid regions, monitoring changes of the

Study Area
HN is located at the boundary between the Shaanxi Province (SX) and Inner Mongolia Autonomous Region (IM), China. In the mid-1990s, the lake surface area was 54.5 km 2 (Ma et al. 2014, Shen et al. 2005a [44,45]. The HN basin is a naturally closed basin with an elevation of 1200-1417 m and a drainage area of 1352 km 2 . Seven seasonal rivers, the Qibosu, Zhashake, Mudushili, Songdao, Erlintu, Haolai, and Maogaitu Rivers, flow into HN. There are no hydrological and meteorological stations in the HN Basin (Yin et al. 2008).

Materials
To identify the effects of different driving forces on the water balance of HN, five different datasets were collected and used in this study: (1) Landsat surface reflectance data for 1990-2017 were used to extract the lake areas.
Only cloud-free images from April to November were selected to avoid cloud contamination and to exclude periods when the lake was frozen.

Materials
To identify the effects of different driving forces on the water balance of HN, five different datasets were collected and used in this study: (1) Landsat surface reflectance data for 1990-2017 were used to extract the lake areas.
Only cloud-free images from April to November were selected to avoid cloud contamination and to exclude periods when the lake was frozen. (2) The x, y, and z coordinates in WGS84 of two routes around HN were measured with RTK positioning in November 2017 to retrieve the height of the lake surface in different historic periods ( Figure 3). (3) Pan evaporation and precipitation of two nearest meteorological stations (Yulin and Dongsheng) away from the HN basin for 1990-2017 were downloaded from the Greenhouse Data website (http://data.sheshiyuanyi.com/WeatherData/, 15 August 2020). In this study, the mean values of the evaporation and precipitation data of these two meteorological stations are used as the evaporation and precipitation data of the HN Basin. (4) The data of gross industrial output value and population came from China Statistical Yearbook (volume of towns and villages). The data for agriculture and livestock before 2015 are derived from the literature [18].

Water Area Extraction
The FAI (Hu 2009), which was used to extract the water area, is defined as: where R is the reflectance and λ denotes the wavelength for the red, near-infrared (NIR),

Water Area Extraction
The FAI (Hu 2009), which was used to extract the water area, is defined as: where R is the reflectance and λ denotes the wavelength for the red, near-infrared (NIR), and shortwave infrared (SWIR) bands of Landsat 5, 7, and 8 images. The FAI was calculated for each image for the HN region, and the widely used threshold segmentation method was adopted to delineate the water-land boundary of HN [46][47][48].
To address any potential misclassification caused by the presence of mixed pixels along land-water boundaries [49] and any possible geometric registration errors, a logical checking procedure was introduced to refine the water area extraction results according to the following criterion: a pixel should be classified as water at a higher water level if it was classified as water at a lower water level; for example, if a pixel is submerged at a water level of 50 m, it will also be submerged at 70 m, and vice versa [50]. To eliminate the noise caused by water waves, boats, and other high-reflectivity objects, a morphological reconstruction method was further applied to smooth the results within the inundation area [51]. The interannual variation of water area was estimated using the average lake area every year.

Water Level and Storage
The water boundary of HN at different times was superimposed on the RTK data. The mean of two measurements of the height difference between two points of one water boundary ( Figure 3) in the two routes of RTK measurements for which the difference was small (<10 cm) was used to approximate the water level of HN. A shape-preserving piecewise cubic interpolation function which was employed to interpolate the lake water levels was constructed. In addition, linear extrapolation was applied to estimate the water areas above the minimum inundation level but were submerged in November 2017 when the ground survey was carried out: where H is the lake water level (m) relative to the reference ellipsoid ( Figure 4) and A is the water area (km 2 ).
where H is the lake water level (m) relative to the reference ellipsoid ( Figure 4) and A is the water area (km ). Using the water area time series extracted with Landsat data and the corresponding water levels derived from either interpolation or extrapolation, the hypsometric relationship curve between the water level and water area was built ( Figure 4). The isobath heights of inundation areas less than the area in November 2017 were derived by the curve. Then, all the isobath heights were used to obtain the lake bathmetric map, and the Using the water area time series extracted with Landsat data and the corresponding water levels derived from either interpolation or extrapolation, the hypsometric relationship curve between the water level and water area was built ( Figure 4). The isobath heights of inundation areas less than the area in November 2017 were derived by the curve. Then, all the isobath heights were used to obtain the lake bathmetric map, and the concurrent water storage results were calculated.

Water Balance
The HN basin is an endorheic basin, in which the water budget of major components can be represented by the equation below: where dV is the change in the water storage of HN. The water supply of HN is mainly through surface runoff R, precipitation on lake surface P w , and groundwater exchange G.
The water of HN is consumed mainly by lake surface evaporation E w and human activities H a . The annual surface runoff R throughout HN can be estimated using the following equation (Liu et al. 2009): where P is the annual precipitation in the HN basin (mm), A HN denotes the area of the HN basin, K is the runoff coefficient, and the long-time average annual value of K is 0.11 in an area 20 km away from the study area [52]. The precipitation on the lake surface (P w ) and the evaporation of the lake surface (E w ) were calculated based on meteorological data: where A avg is the annual average area of HN (km 3 ), E is the evaporation of a 20 cm pan (mm), and I is a conversion coefficient between the small evaporation pan and the evaporation of the lake surface, which is considered to be 0.43 in this study area (Yuan 2006). The water consumed by human activities (H a ) mainly includes residential water consumption, livestock water demand, agricultural irrigation water, hydraulic engineering construction, and water consumption for industry.
The domestic water and livestock water consumptions in the three towns within the HN basin were calculated by the population, the sheep population, the domestic water consumption quota, and the livestock water consumption quota (Wang et al. 2011). The water consumption quotas for residents (with the development of the economy, the quota of water used by residents has also increased. Before 2010, it was 39 L/p·d −1 . Since 2010, it has averaged 45 L/p·d −1 ) and livestock were taken from Chen et al. [53]. We could not estimate the amount of water used for irrigation and industry because the detailed data on irrigated land and industry were unavailable.
The effect (dV r ) of the Zhashake and Maogaitu Reservoirs, which were built on two main tributaries of HN, can be estimated by the following formula: where S r is the annual average water storage of reservoirs and E w is the evaporation of the reservoir surface. The amount of water needed to keep the lake area stable can be estimated according to Equation (3) when dV = 0. The water consumed by human activities (H a ) can be approximately estimated using Equation (8):

Statistical Methods
In this study, linear regression analysis and Pearson correlation analysis were used to research the trends in the time-series data and the relationships between lake storage and the driving forces for the changes in lake storage. Linear regression can analyze trends and variation rate of hydrological and meteorological variables in different stages [54].

The HN Water Balance
The HN water balance was estimated based on surface runoff ( ), precipitation on the lake surface ( ), groundwater exchange ( ), evaporation of the lake surface ( ), the water consumed by human activities ( ), and the change in the water storage of HN (dV).
The water supply of HN is mainly through surface runoff, precipitation on lake sur-

The HN Water Balance
The HN water balance was estimated based on surface runoff (R), precipitation on the lake surface (P w ), groundwater exchange (G), evaporation of the lake surface (E w ), the water consumed by human activities (H a ), and the change in the water storage of HN (dV).
The water supply of HN is mainly through surface runoff, precipitation on lake surface, and groundwater exchange. As shown in Table 1  * "−" means a decrease in water storage; "+" indicates an increase in water storage.
The annual average of precipitation on the lake surface was 21.2 × 10 6 m 3 y −1 from 1990 to 1998. Due to the shrinking of HN and the change of precipitation, the annual average lake surface precipitation in the five phases was 5.8 × 10 6 m 3 y −1 , 6.9 × 10 6 m 3 y −1 , 7.1 × 10 6 m 3 y −1 , and 2.9 × 10 6 m 3 y −1 less than that during 1990-1998, respectively ( Figure 9). The groundwater exchange was relatively stable, about 6.67 × 10 6 m 3 y −1 [55], due to the HN is located in the lowest part of the HN basin.  * "−" means a decrease in water storage; "+" indicates an increase in water storage.
The annual average of precipitation on the lake surface was 21.2 × 10 m y from 1990 to 1998. Due to the shrinking of HN and the change of precipitation, the annual average lake surface precipitation in the five phases was 5.8 × 10 m y , 6.9 × 10 m y , 7.1 × 10 m y , and 2.9 × 10 m y less than that during 1990-1998, respectively ( Figure 9). The groundwater exchange was relatively stable, about 6.67 × 10 m y [55], due to the HN is located in the lowest part of the HN basin. Since HN is an inner-flowing lake, its water is mainly consumed by human activities and lake evaporation. The annual average of water consumed by human activities (H a ) was 37.1 × 10 6 m 3 y −1 from 1990 to 1998. The H a continued to increase after 1998 and reached a peak of 64.3 × 10 6 m 3 y −1 during 2010-2015. In 2016-2017, the H a dropped to 59.8 × 10 6 m 3 y −1 . The annual average of evaporation of the lake surface (E w ) was 48.5 × 10 6 m 3 y −1 , 41.8 × 10 6 m 3 y −1 , 37.2 × 10 6 m 3 y −1 , 29.2 × 10 6 m 3 y −1 , and 26.5 × 10 6 m 3 y −1 in five phases, respectively, which has the same trend with the lake area ( Figure 9).
The change in the water storage of HN (dV) was caused by the difference of water supply and loss. During 1990-1998, the HN was in a relatively stable state, the annual average change in the water storage only was −1.53 × 10 6 m 3 y −1 . The dV was −17.13 × 10 6 m 3 y −1 , −11.63 × 10 6 m 3 y −1 , −10.03 × 10 6 m 3 y −1 , and 18.47 × 10 6 m 3 y −1 in the four phases that followed, respectively.

Phase I
During this period, there was no significant change in water area, level, and storage (p > 0.05). The annual average precipitation was 389.8 mm (Figure 10a). The annual average 20 cm pan evaporation 2070.8 mm (Figure 10b). The annual average of the lake water supply was about 83.9 × 10 6 m 3 y −1 . The annual average of water needed to keep the HN stable was only 46.8 × 10 6 m 3 y −1 . The difference between them was 37.1 × 10 6 m 3 y −1 , which was probably consumed by human activities.
average change in the water storage only was −1.53 × 10 m y . The dV was −17.13 × 10 m y , −11.63 × 10 m y , −10.03 × 10 m y , and 18.47 × 10 m y in the four phases that followed, respectively.

Phase I
During this period, there was no significant change in water area, level, and storage (p > 0.05). The annual average precipitation was 389.8 mm (Figure 10a). The annual average 20 cm pan evaporation 2070.8 mm (Figure 10b). The annual average of the lake water supply was about 83.9 × 10 m y . The annual average of water needed to keep the HN stable was only 46.8 × 10 m y . The difference between them was 37.1 × 10 m y , which was probably consumed by human activities.
In this stage, the correlation coefficient (Table 2) between the lake water volume and the three human factors were very small, indicating that human activities had little influence on the lake water storage change.  The gross industrial output value of three towns in the HN basin increased about fivefold (Figure 11), but the total was small. The annual average of residential water and livestock water consumption was about 1.6 × 10 m y . According to other studies in the area [55,56], agricultural irrigation consumed about 30 × 10 m y water in this phase. The remained relatively constant. In this stage, the correlation coefficient (Table 2) between the lake water volume and the three human factors were very small, indicating that human activities had little influence on the lake water storage change. The gross industrial output value of three towns in the HN basin increased about fivefold (Figure 11), but the total was small. The annual average of residential water and livestock water consumption was about 1.6 × 10 6 m 3 y −1 . According to other studies in the area [55,56], agricultural irrigation consumed about 30 × 10 6 m 3 y −1 water in this phase. The H a remained relatively constant.

Phase II
The annual average precipitation decreased by 47.2 mm, and the annual average 20 cm pan evaporation increased by 55.7 mm compared with those in phase I. At the same time, human activity had intensified. The increased about 10 × 10 m y compared to those from 1990 to 1998. Among all human activity, the gross industrial output value in the HN basin increased from 1018.3 × 10 to 2507.9 × 10 during this period ( Figure 11). The annual average of residential water and livestock water consumption increased 0.3 × 10 m y ( Figure 12). In addition, although we do not have data on farmland, a reasonable assumption is that the drought led to an increase in water consumption of irrigation. According to the above analysis, the drought climate and the rapid development of industry led to the HN shrinking rapidly in this phase.

Phase III
The climate was persistently dry. The annual average precipitation decreased by 24 mm, and the annual average 20 cm pan evaporation increased by 125 mm compared with

Phase II
The annual average precipitation decreased by 47.2 mm, and the annual average 20 cm pan evaporation increased by 55.7 mm compared with those in phase I. At the same time, human activity had intensified. The H a increased about 10 × 10 6 m 3 y −1 compared to those from 1990 to 1998. Among all human activity, the gross industrial output value in the HN basin increased from 1018.3 × 10 6 CNY to 2507.9 × 10 6 CNY during this period ( Figure 11). The annual average of residential water and livestock water consumption increased 0.3 × 10 6 m 3 y −1 (Figure 12). In addition, although we do not have data on farmland, a reasonable assumption is that the drought led to an increase in water consumption of irrigation. According to the above analysis, the drought climate and the rapid development of industry led to the HN shrinking rapidly in this phase.

Phase II
The annual average precipitation decreased by 47.2 mm, and the annual average 20 cm pan evaporation increased by 55.7 mm compared with those in phase I. At the same time, human activity had intensified. The increased about 10 × 10 m y compared to those from 1990 to 1998. Among all human activity, the gross industrial output value in the HN basin increased from 1018.3 × 10 to 2507.9 × 10 during this period ( Figure 11). The annual average of residential water and livestock water consumption increased 0.3 × 10 m y ( Figure 12). In addition, although we do not have data on farmland, a reasonable assumption is that the drought led to an increase in water consumption of irrigation. According to the above analysis, the drought climate and the rapid development of industry led to the HN shrinking rapidly in this phase.

Phase III
The climate was persistently dry. The annual average precipitation decreased by 24 mm, and the annual average 20 cm pan evaporation increased by 125 mm compared with those from 1990 to 1998. The gross industrial output value increased from 2651.4 × 10 6 CNY to 5868.7 × 10 6 CNY, with an increase of 121.3% (Figure 11). Meanwhile, the annual average of residential water and livestock water consumption increased to 2.7 × 10 6 m 3 y −1 (Figure 12). In addition, the Zhashake reservoirs were built on the Zhashake River in 2005. The annual average water storage of the Zhashake Reservoir was 4.5 × 10 6 m 3 , and the evaporation of the lake surface was 3.0 × 10 6 m 2 y −1 . However, the H a was about 48.8 × 10 6 m 3 y −1 which was almost the same as that during 1999-2005. The reason that H a had not increased significantly was that the area of cultivated land has been reduced by 55 km 2 [18] due to the implementation of policy aimed at converting farmland back into forests and grasslands in the HN basin [56]. The irrigation water use should decrease due to this policy, too. The HN continued to shrink in this phase. In addition to the continuous drought, human activities are the main reason for the shrinkage of the HN. The way of human activities had changed: the water consumption of industrial and reservoir impoundment had increased significantly, while the water consumption of agricultural irrigation had decreased.

Phase IV
Rainfall was abundant during this period. The annual average precipitation increased by approximately 38 mm (Figure 10a) and the annual average 20 cm pan evaporation decreased by 9 mm (Figure 10b) compared with those in the first phase. The average annual water supply of the lake is almost equal to those from 1990 to 1998 (Table 1). Nevertheless, the HN continued to shrink. The H a was about 64.3 × 10 6 m 3 y −1 , which was almost 1.7 times as much as those during 1990-1998.
The domestic water consumption increased by 0.237 × 10 6 m 3 y −1 with the increase in population (Figure 12a). The livestock water consumption increased by 1.927 × 10 6 m 3 y −1 from 1990 to 1998. (Figure 12b). The Maogaitu Reservoir was impounded in 2010, and thus, no data indicating its water storage are available. Here, its water storage was estimated as 0.8 × 10 6 m 3 because its water area was one-sixth of that of the Zhashake Reservoir, and the evaporation on the water surface was 0.5 × 10 6 m 2 y −1 . The annual average water storages and evaporation of these two reservoirs were about 8.8 × 10 6 m 3 .
The annual average gross industrial output value in the HN basin increased to 8488 × 10 6 CNY, which was 13 times that during 1990-1998 ( Figure 11) with an annual increase rate of 13.33% (p < 0.001) after 2000.
Although there are no large-scale coal mining activities in the HN basin, the impact of coal mining cannot be ignored. The nearest coal mine is only 10 km away from the HN basin. The large-scale coal mining around the HN basin is likely to have had significant impacts on the groundwater flow within the basin [57][58][59][60].

Phase V
The area of HN began to expand. The largest annual precipitation since 1990 occurred in 2016, which reached 600 mm. The annual average precipitation increased by 160 mm and the annual average 20 cm pan evaporation decreased by 241 mm compared with those from 1990 to 1998.
The H a was about 59.8 × 10 6 m 3 y −1 , which declined 4.5 × 10 6 m 3 y −1 compared with those during 2010-2015. The decrease in H a may be for two reasons as follows: 1.
According to "The 13th Five-Year Plan" drawn up by the Chinese government, coal consumption will drop from 5.7 billion tons to 3.4 billion tons from 2015 to 2020. Many coal mining-related businesses have had to close or reduce capacity.

2.
There was 1.0 × 10 6 m 3 of water discharged each year from the Zhashake reservoir and Maogaitu reservoir to the HN after 2015.

Uncertainties and Limitations
Several uncertainties and limitations exist in this study. First, the lake area was extracted from Landsat data with a resolution of 30 m, which can lead to deviation in the estimated lake area [61]. At the same time, the water level is determined by the lake boundary and RTK data, the deviation of the lake boundary will lead to the bias of the estimated water level. In future studies, higher spatial resolution remote sensing images will help improve the accuracy of water area and water levels. Second, the mean precipitation and evaporation of two nearby meteorological stations was used to estimate the meteorological data, which may not reflect the actual weather conditions in the HN basin. In addition, runoff coefficients in the vicinity of the HN basin may not reflect the true hydrological conditions of the HN basin. Meteorological and hydrologic stations should be established in the HN basin to obtain more accurate weather and hydrologic data for the HN in the future. Third, due to the lack of some social and economic data, it is impossible to accurately estimate the impact of policies drawn by the government on HN.

Conclusions
In this study, the variations in the lake area, level, and storage of HN during 1990-2017 were investigated using Landsat imagery and in situ measurements, and the driving factors of these changes were analyzed. The main conclusions are as follows. (1) The water storage of HN decreased by approximately 2.7 × 10 8 m 3 in total during 1999-2015, and the area of HN began to expand after 2015. (2) In different periods, the human activities affecting the HN were also undergoing great changes due to economic development and policies drawn up by government. (3) The major factors caused by the HN change were different in several phases. Climate drought and the intensification of human activities were the main driving factors for the rapid shrinkage of the HN during 1999-2010. Human activities likely was the primary factor driving the decrease in the water storage of HN from 2010 to 2015. After 2015, the policies that decreased the water consumed by human activities made by the government and humid climate were the main factor for the expansion of HN. The results of this study will provide important scientific guidance for future water management and conservation in HN. In addition, the method used in our study is potentially applicable to similar ungauged basins elsewhere in the world to study their responses to climate change and human activities.

Conflicts of Interest:
The authors declare no conflict of interest.