1 INTRODUCTION

Zabuye Salt Lake (ZSL) in Xizang is the only saline lake in the world with natural crystalline lithium carbonate, and as such represents an important lithium production base for China. Since the 1990's, the lake has been expanding rapidly; several studies previously reported that lake expansion (LE) was closely related to global climate change (CC) (Zheng et al., 2000, 2004). The Tibetan Plateau (TP) climate is undergoing rapid change, and is mainly characterized by rising air temperatures, increasing precipitation, decreasing potential evaporation, and shifts from dry to wet conditions (Du, 2001; Wu et al., 2005; You et al., 2010). Because of CC, glaciers are shrinking and snow cover is decreasing, a series of environmental hazards is in play (Li et al., 2008; Liu et al., 2009; Yang et al., 2010). Impacted by increased precipitation, increased glacier melting rates, and thawing permafrost (Ye et al., 2008; Wang et al., 2013; Du et al., 2014), the water levels in most Tibetan lakes have significantly increased (Bianduo et al., 2009; Lei et al., 2013; Yan and Zheng, 2015; Ma et al., 2016). Changes in lakes exert a considerable impact on mineral production and daily life (Zheng et al., 2000, 2004). For example, ZSL expansion may complicate mineral extraction processes and damage surrounding grasslands and salt pans, therefore, expansion is a serious issue. Several studies have reported that LE in the TP region may be caused by increased precipitation and melting glaciers (Qi and Zheng, 2006; Yao et al., 2007). However, further research is required to determine how these changes impact surrounding grasslands and salt pans.

Remote sensing (RS) technology provides longterm continuous data accessibility, while multitemporal RS images are ideal for monitoring snow cover, vegetation phenology, and other environmental factors (Harris, 1994; Hall et al., 2002; Zhang, 2005; Wang et al., 2010; Zhou et al., 2010; Karimi et al., 2014; Robson et al., 2015; Ji et al., 2018). RS data, particularly medium-resolution data, have been used to observe saline lakes, e.g., Yan and Zheng (2015) used Landsat images to identify saline lakes with areas greater than 20 km2, and analyzed dynamic changes in surface areas, from 1973 to 2010. Yagmur et al. (2021) analyzed changes in freshwater, saline, and brackish lakes using Landsat satellite images, and observed minimal changes in freshwater lake surfaces over the study period. Moreover, Nhu et al. (2020) used multi-temporal Landsat 7 ETM+ images to monitor and assess water fluctuations in Lake Urmia in Iran and its environmental consequences. Because of multi-time phase, low costs, and high accuracy capabilities, RS is important for monitoring saline lakes and surrounding environments.

Many water, glacier, and vegetation extraction methods have been previously used and described in RS monitoring: Liu et al. (2016) used Huanjing (HJ) satellite data to analyze the cause and effect of sudden floods in Zhuonai Lake, proposing CH4/CH2 < 1 and CH4 < 0 threshold methods to extract lake boundaries; Ma et al. (2007) used normalized difference water index (NDWI) (McFeeters, 1996) and normalized difference vegetation index (NDVI) methods to extract the water body; Zhang et al. (2013a) extracted TP lake data with the modified normalized difference water index (MNDWI) method (Xu, 2005), and analyzed lake, snow cover, and vegetation data in the Namtso Basin; Hall et al. (1995) proposed the normalized difference ice and snow index (NDSI) method using MODIS 4 and 6 bands; and Bajracharya et al. (2014) used the object-oriented method to extract glacier data. Vegetation index (Zhang et al., 1997; McGwire et al., 2000), spectral matching (Price, 1998; Wei and Li, 2011), mixed pixel decomposition (Tao et al., 2004; Li et al., 2010), support vector machine and deep learning (Li et al., 2010; Zhang et al., 2013b; Hasan et al., 2019), etc., are all applied to vegetation extraction. These data extraction methods are unsuitable for all situations, and some new method are too cumbersome to observe large-scale salt lake changes, therefore, a new extraction algorithm for the lake in TP is warranted.

In this study, using Landsat RS data, band analysis, and calculations, water extraction method EW (extraction of water), glacier extraction method EG (extraction of glaciers), and grassland extraction method EGS (extraction of grassland with salinity) were designed. A shuttle radar topography mission (SRTM) digital elevation model (DEM) was used to simulate expanded lake waters using submergence analysis (SDA). When combined with meteorological data, we analyzed relationships between salt lake area changes, glaciers, and various climate factors, to assess area changes in a series of salt lakes on the TP, represented by ZSL. Additionally, using overlay analysis (OA), LE rates and potential damage of ZSL to surrounding grasslands and salt pans were estimated.

2 MATERIAL AND METHOD 2.1 Study area

Zabuye Salt Lake is located in the hinterland of the TP and the northern foothills of the Western Gangdise Mountains. Its geographical coordinates are 83°57′10″E–84°15′08″E and 31°27′10″N–31°34′ 30″N (Fig. 1). ZSL is recharged by river water, atmospheric precipitation, and groundwater (Tian et al., 2005). The water system around the lake, including Sangmujiuqu, Luojuzangbu, Langmengaqu, Quanshui River, and many springs, is relatively well developed (Fig. 1). The climate in Zabuye area, which is affected by the westerly belt, is characterized by large temperature differences, strong radiation, less precision, and large evolution (Qi and Zheng, 2006). Annual precipitation is only 192 mm, but annual evaporation can reach 2 269 mm. ZSL is rich in mineral resources, especially lithium, boron, and potassium, which have high economic value. Furthermore, a large area of grassland and a salt pan are distributed around ZSL, which will be endangered by expansion of the lake.

2.2 Dataset and pre-processing 2.2.1 Landsat RS data

The terrain of the TP is complex, and there are usually many clouds distribute around high mountains, which poses a challenge to remote sensing observation. Medium-resolution Landsat data, which has provided free and up-to-date satellite images worldwide for 50 years, is one of the most abundant data resources and was chosen for long-term monitoring of ZSL. The Landsat data covers the spectral range from visible-near-infrared to thermal infrared. The radiometric resolutions of Landsat 7 and Landsat 8 are 8 bit and 12 bit, respectively. Multispectral bands of these types data have a spatial resolution of 30 m, and panchromatic band can reach a higher resolution of 15 m, which is of appropriateness for the long-term and large-scale salt lake change research. In this study, dynamic monitoring was accomplished using Landsat 7 and Landsat 8 multispectral images and panchromatic band (band 8). In order to avoid the rainy season and more effectively monitor the impact of glaciers on the salt lake, the imaging period from April to June was deemed more suitable because of less cloud cover. After comparison, the following five images were selected from 2000 to 2020:

1: LE07_L1TP_143038_20000613_20170211_ 01_T1

2: LE07_L1TP_143038_20050627_20170114_ 01_T1

3: LE07_L1TP_143038_20100609_20161214_ 01_T1

4: LC08_L1TP_143038_20150514_20170409_ 01_T1

5: LC08_L1TP_143038_20200628_20200708_ 01_T1

Before the extraction of RS information, a series of pre-processing techniques were performed, including radiometric calibration, atmospheric correction, and Gram-Schmidt (GS) pan sharping fusion, which can correct the radiation distortion during image acquisition and improve the spatial resolution so that RS images more accurately reflect the ground objects. If the image was a Landsat 7 image, restoration was performed in advance to remove the strip. Preprocessed images was shown in Fig. 2.

2.2.2 SRTM DEM data

The data (Fig. 3) used to simulate submerged area was a SRTM DEM provided by the National Aeronautics and Space Administration (NASA). This series of DEM covers a latitude range of 56°S–61°N with a resolution of 30 m, and is less affected by terrain elements, resulting in the best accuracy in the TP (Wan et al., 2015; Gao et al., 2019). Because the SRTM DEM image has been divided into many grids, this data was mosaicked prior to use.

2.2.3 Meteorological data

The meteorological data used in this study were provided by ZSL field observation station of the Chinese Academy of Geological Sciences. Since construction in 1990, the station has continuously observed and recorded lake meteorology, hydrology, and hydrochemistry data. Collection instruments surround the lake and data are regularly operated. The temperature was collected at a fixed time every day, and precipitation and evaporation rates were collected once a week. The measurement of these meteorological factors was consistent with regular meteorological observations. Meteorological data included annual mean temperatures, total precipitation, and total evaporation, from 2000 to 2018.

2.3 Method

The technical roadmap is shown in Fig. 4. To monitor dynamic changes in ZSL and Lunggar Glaciers (LG), and assess threatened grasslands and salt pans, four ground features (lake, glaciers, grasslands, and salt pans) and lake area after expansion were extracted. As shown in Fig. 4, lake, glacier, grassland, and salt pan extraction processes were completed by band calculation (BC) using Landsat images, and extraction of lake areas, after expansion, was done by 3-dimensional analysis (3DA) and SDA using SRTM DEM. Additionally, OA was used to identify intersections between grasslands, salt pans, and lake areas after expansion, thereby indicating potentially affected areas, while intersections between glaciers and the Zabuye basin showed how glaciers affected the lake.

2.3.1 Extraction of lake, glaciers, grassland, and salt pan

In our study area, ground features were divided approximately into water, glacier, land, salt pan grassland, and cloud. Each ground feature had different absorption and reflection characteristics for different electromagnetic radiation wavelengths. Therefore, they also had different spectral curves. We extracted spectral curves features from Landsat images, selected a typical feature and enhanced the image. Extracted spectral curves are shown in Fig. 5, and the BC for lakes, glaciers, grasslands, and salt pans extractions were based on the spectral characteristics (SC) of ground objects.

2.3.2 Lake extraction

Lake extraction (EW) involved summing the green band, the near-infrared band (NIR), and the short-wave infrared band (SWIR):

The essence of lake extraction is water extraction. Although more mineral resources are present in a saline lake, its SC is approximately the same as in water, which is highly unique (McFeeters, 1996; Xu et al., 2017). Because water absorbs more electromagnetic radiation, the radiation value of each band of water was typically lower than other ground features. As shown in Fig. 5, water reflectance was significantly lower than in glaciers, salt pans, and clouds, from green to NIR bands, and lower than land (including grasslands) in SWIR bands. Therefore, the sum of several water bands was lower than other ground features. After BC, water appeared black with low values, whereas other features were grey/white, with relatively high values. Additionally, an unsupervised classification (UC) operation was used to transform images into separable parts, which was convenient for generating lake vectors and calculating lake areas.

2.3.3 Glacier extraction

Glacier extraction (EG) involved calculating NIR to SWIR band ratios:

As shown in Fig. 5, glacier reflectance was higher in the NIR band and lower in the SWIR band when compared with other ground features. After the ratio was calculated, the numerical difference was amplified, and the only prominent ground features were glaciers. All other features were black. The entire extraction process for glaciers was basically the same as for water bodies: first, the extraction index EG was used to increase differences between the water body and other ground features, and then a UC method was used to directly extract the results after index enhancement, to finally generate glacier extraction results. As well as these operations, basin analysis (BA) and OA were conducted. LG is some of the most important glaciers affecting the ZSL. However, not all of these glaciers affected the lake, as only some glaciers in the southeast corner of the lake melted and indirectly flowed into the lake. Thus, the Zabuye basin was obtained using BA and glaciers affecting the lake were extracted by OA, both of which were accomplished in ArcGIS.

2.3.4 Grassland and salt pan extraction

As ZSL is high in salt, the surrounding grasslands are covered in a layer of salt. Thus, the SC of grasslands, affected by salinization, were changed, and conventional extraction methods were no longer applicable to these areas.

Extracting high-salinity grasslands (EGS) involved calculating the ratio of the NIR band to the sum of the blue wave band and SWIR band:

To extract saline vegetation, we considered the SC of grasslands and salt pans. Spectrum analyses showed that the spectral values of these features were highest in the NIR band among all bands. Therefore, the NIR band was used to distinguish grasslands and salt pans from other features. However, grassland values in the NIR band were not higher than glaciers and clouds. Therefore, the NIR band alone was insufficient to fully distinguish both ground feature types. As the maximum values of other features were mainly distributed in the blue wave or SWIR band, a combination of the blue wave, NIR, and SWIR bands was used to distinguish these features. This method produced positive eigenvalues for salt pans and grasslands, but negative eigenvalues for all other features.

There were few bands in Landsat and many ground feature types, therefore, it was difficult to only extract salt pans. Their most prominent feature was that they exhibited the highest spectral values in the NIR band, and salt pans could not be accurately extracted using only simple single-band threshold division (TD) because the threshold was difficult to determine. Nearly dry salt pans with less water content can be extracted by EGS; however, salt pans used for mineral production often had different water content. Some salt pans were newly injected into lake water, which was consistent with the SC of lake water. Therefore, salt pans could not be extracted completely by EGS. Visual interpretation (VI) was the ideal extraction method for such features with large changes in spectral features, fixed shapes and locations, and small numbers. In this study, salt pans were visually interpreted, and EGS used to extract grasslands and salt pans together. Based on EGS results, salt pans were removed to generate grasslands. After both extractions were completed, vector data for grasslands were obtained using grid-vector transformation for subsequent statistical analysis (STA).

2.3.5 Extraction of expanded water areas

SDA based on a DEM can achieve rapid simulation and evaluation of submerged areas under the condition of incomplete observation data. SDA is typically divided into passive SDA and active SDA, each of which are suitable for different conditions (Guo and Long, 2002; Zhang et al., 2016). As Zabuye basin is a typical inland basin, and the salt lake is the lowest point (Wang et al., 2019), passive SDA was selected because it only considers the ground elevation. Passive SDA was performed in ArcGIS using Raster Calculator Tool. Grids with values less than or equal to the target water level represented water bodies. Therefore, by setting the recent water-level thresholds extracted in the last step and the water levels after LE, the SDA was able to divide the entire area into submerged and unsubmerged areas, where submerged areas indicate an expansion of the lake area. Finally, OA was conducted to obtain the intersection between grassland, the salt pan and the lake area after expansion. This intersection represents the area that may be affected by LE.

3 RESULT 3.1 Extraction of ZSL

The unique climatic and geological characteristics of the Tibetan Plateau provide surface features with SC distinct to other regions. Lake water is distinguished from wet lakeshores by EW, which was more suitable for extracting shallow lakes such as ZSL. The effects of other water extraction methods were not ideal. As shown in Fig. 6, in water body index extraction results, the whole water body was divided into several parts and represented by different colors so that the water body index cannot play the role of extracting the water body, such as Fig. 6b, c, f, & g. Some water body index extraction results were incomplete, such as Fig. 6d, e, h, i, & j. NDVI, new water index (NWI) (Ding, 2009), NWI8 (Ni and Liu, 2015), desert lake water index (DLWI) (Zhu et al., 2010; Li et al., 2015), MNDWI, and enhanced water index (EWI) (Yan et al., 2007) were suitable for detecting deep lakes, and shallow water areas cannot be well proposed, the surrounding wet shore is also mistaken for water. High resolution water index (HRWI) (Yao et al., 2015) is of better outperformance than NDWI and interactive data language water index (IDLWI) (Li et al., 2018); however, near-water shores were still considered part of the lake. According to VI, the water extraction effects of EW were better.

The accuracy of extraction results (Fig. 6) were evaluated in Table 1. The overall accuracy of EW was >98%, and the Kappa coefficient was >91%. Whether the VI effect or the accuracy evaluation result, we observed a better water extract effect for EW. Also, EW was advantageous in that it was less dependent on the threshold. The sum of the three bands was used to calculate differences between the body of water and other features. The discrepancy increases, and water bodies can be proposed quickly and easily directly by UC, without the need for TD.

Water body changes from 2000 to 2020 were extracted by EW. As shown in Fig. 7, the lake area changed substantially over the previous 20 years, from 88.86 km² to 194.46 km², with an average annual increase of 5.28 km². From 2000 to 2005 and from 2015 to 2020, the lake rapidly expanded, whereas from 2005 to 2015, expansion was slower. Although the growth rate changed, ZSL has continued to increase in size. Based on these change trends, the lake will likely continue to expand in the future.

3.2 Extraction of LG

The normalized difference ice and snow index (NDSI) is the most commonly used method for ice extraction (Dankers and De Jong, 2004). However, as shown in Fig. 8, when the NDSI was used to extract glaciers in the plateau, both glaciers and lakes were jointly extracted, while the extraction range was larger than the actual range. Therefore, we used the EG ratio method to extract glaciers. Using EG calculation ratios, the glacier area was determined quickly and effectively; the only prominent ground features were glaciers, while all other features were black. Clouds in images were white as the glaciers in the true color composite image, it seriously interferes with the interpretation of glaciers. After ratio processing, clouds were weakened. The extraction results for glacier areas over 20 years are shown in Fig. 9.

LG is the most important glacier affecting ZSL, however, the whole glacier did not completely affect the lake. BA results (Fig. 10) showed that the salt lake is the lowest point, only a part of the southeast corner melts and flows into ZSL indirectly. The intersection of the whole glacier area and the basin area is the glaciers which will affect ZSL. Result of OA showed that areas in this part of the glacier in 2000, 2005, 2010, 2015, and 2020 were 57.30, 49.48, 50.32, 47.49, and 55.91 km2, respectively (Fig. 11). From 2000 to 2005, glaciers melted rapidly, whereas from 2005 to 2015, glaciers melted more slowly. In contrast to the first four periods, the glacier area increased in 2020. As a whole, the glacier remains in retreat.

3.3 Extraction of grasslands and salt pans

The NDVI (Eq.4) is the most commonly used method for vegetation extraction:

Salt levels in ZSL were relatively high and covered surrounding grasslands with a layer of salt. Using conventional NDVI to extract grasslands, grass with a salty surface could not be identified. As shown in Fig. 12, EGS and NDVI extraction areas were significantly different around the South Lake and the northeastern part of the North Lake. From field investigations around the South Lake, grasslands were identified (Fig. 13), therefore, NDVI extraction was not complete. A false color composite (FCC) image (Fig. 12c) was generated by combining 6, 5, and 4 bands of the Landsat 8 image. These bands were colored red, green, and blue, respectively, highlighted vegetation, and allowed vegetation analysis. By comparing images (Fig. 12a, b, & c), EGS comprehensively extracted grasslands, the most distinctive being northeast of North Lake. As shown in Fig. 12d, this extraction indicated that grasslands around ZSL were predominantly distributed to western and northeastern sides of the North Lake, and at the southeast corner of the South Lake. Salt pans were distributed to the north, east, and west of the South Lake.

3.4 Extraction of expanded water areas

Simulated lake areas, after a water level increases of 1, 2, 3, and 4 m from base water levels, are shown in Fig. 14. The North Lake would undergo distinct changes to the northeast and southwest sides, while almost all solid salt deposits around the South Lake would be submerged, with only the southeast corner being less affected.

The intersections between water, grassland and salt pans were obtained through OA, and potential flooded areas were extracted (red areas in Fig. 15). Grassland to the northeast and southwest of the North Lake and to the southeast of the South Lake will likely be submerged upon lake-level rise, and a large area of the salt pan in the South Lake is forecast to be submerged. Water-level rises of 2, 3, and 4 m are forecast to submerge grassland area of 103.769, 111.308, and 119.448 km² and salt pan area accounts for 8.748, 12.471, and 14.153 km² (the water level is about 4 426 m in DEM at present, respectively).

4 DISCUSSION

The ZSL area has continuously expanded from 2000 to 2020 (Fig. 16). Based on these change trends, the likelihood is that LE will be maintained in the future. From 2000 to 2005, the lake rapidly expanded while the glacier area decreased substantially, suggesting that LE was probably due to glacier melting caused by rising temperatures. From 2005 to 2015, LE slowed, as did glacier melting. Therefore, ZSL expansion was closely related to glacier ablation. The rainy season in the Zabuye area is generally concentrated from midJuly to late September; the 2020 imaging date was June 28, which was close to the rainy season, thus short-term precipitation may have been the reason for the observed increase in glacier area in 2020, while LE in this year may have resulted from the combined effects of atmospheric precipitation and glacial melt-water. A comprehensive observational analysis of the previous 20 years suggested ZSL expansion was primarily caused by melting glaciers.

Figure 17, overlaying the extracted lake traces on the simulated water areas, shows that the lake areas in 2000 and 2005 were roughly close to the water surface at the initial water level, and the lake areas in 2010 and 2015 were roughly close to the water level after the 1 meter rise. According to this rate of increase, it can be inferred that the water level in 2000, 2005, 2010, 2015, and 2020 was roughly -0.33, 0, 0.11, 0.22, and 0.72 m respectively. According to the lake volume change Eq.5 provided by Cao et al. (2020), it can be known that the water volume increased by 37.789×10³, 15.853×10³, 16.322×10³ and 86.079×10³ m³ in 2000–2005, 2005– 2010, 2010–2015, and 2015–2020, respectively. And the volume change of the glacier can be calculated according to Eqs.6 & 7 provided by Liu et al. (2015), after calculation, the volumes changed by -23.255×10³, 2.435×10³, -8.145×10³, and 24.737× 10³ m³ in 2000–2005, 2005–2010, 2010–2015, and 2015–2020, separately.

The ZSL is mainly affected by the LG, but not only affected by it, also by other distant glaciers, the amount of water and glacier volume is a rough estimate, however, it can still be seen changes in lake water volume and glacier volume show a synchronous trend (Fig. 18). From 2000 to 2005, the lake volume rapidly increased while the glacier volume decreased substantially, and from 2005 to 2015, lake increase slowed, as did glacier decreasing.

Glacier changes are directly impacted by climate conditions, including temperature, precipitation, and evaporation (Yan et al., 2015). From 2000 to 2020 meteorological data, the annual average temperature showed a distinct upward trend, whereas precipitation and evaporation showed minimal changes (Figs. 19–20). Therefore, atmospheric precipitation and evaporation were not the main factors causing LE, significant temperature increases in the Zabuye area in recent years were factors affecting LE. Moreover, three natural systems can recharge ZSL: atmospheric precipitation, river water, and spring water. River and spring water mainly come from glacial melt-water, while atmospheric precipitation has a certain impact on river and spring water levels. Thus, water in ZSL is partially derived from atmospheric precipitation and glacial melt-water, with no third source. In the absence of precipitation and evaporation changes, the replenishment mechanisms contributing to ZSL expansion were related to glacier melt-water replenishment caused by rising temperatures. In the future, if temperatures in Zabuye continue to rise, the glacier will remain in its current retreat state and ZSL will continue to expand.

Since 2000, the salt lakes in TP have shown overall expansion trends. We conducted similar area monitoring analyses at Jiezechaka, Mami Co, Dangqiong Co, Selin Co, Chaerhan, and Xitaijiner Lakes. Without exception, they had expanded, with expansion rates of approximately 0.49, 1.04, 0.59, 65.66, 12.10, and 4.37 km²/a, respectively. A comparative analysis of changes in surrounding glaciers found that not all lakes and glaciers were highly consistent. There are many reasons for LE, including rising temperatures, increased precipitation, and reduced evaporation. Several studies reported temperatures in the TP showed distinctive upward trends in the previous 20 years; however, changes in precipitation and evaporation were inconsistent. There are various situations. In some areas, precipitation increased and evaporation decreased, while in other places, precipitation and evaporation remained almost unchanged. Unequivocally, rising temperatures will inevitably lead to melting glaciers, which are the main reasons for LE of salt lakes in TP. Whereas increased precipitation will undoubtedly replenish glaciers, rendering lake and glacier area changes less obvious.

The rapid expansion trend identified at ZSL is bound to impact surrounding grasslands and salt pans. After simulations and calculations, for every meter rise in water levels, an average of 7.84 km² of grasslands and 2.7 km² of salt pans would be submerged (Fig. 21). Therefore, appropriate measures should be undertaken based on salt lake changes, e.g., improving extraction technology and abandoning existing grasslands and salt pans.

In 2020, ZSL was in a state between Fig. 14a and Fig. 14b; if water levels increased by approximately 0.5 m to reach the state in Fig. 14b, the lake area would be expanded by approximately 55.28 km². From LE data in the previous 20 years, it would take approximately 10.4 years to reach the state in Fig. 14b. Also, according to the rate in the past 5 years, this expansion would take approximately 6.3 years. Therefore, we hypothesize it will take 6–10 years for water levels to rise by 0.5 m.

We used the same method for SDA and potential hazard predictions for several salt lakes in Xizang, including Selin Co, Dangqiong Co, and Mami Co. Our analyses indicated the following:

(1) Selin Co has gradually expanded in the previous 20 years, surpassing Nam Co to become the largest lake in Xizang. According to current expansion rates of Selin Co, 13.012 km²/a, if the expansion continues, it will take 3 and 6 years for water levels to rise by 1 m and 2 m, respectively. After rising by 2 m, water will inundate four areas of the road and 23.65 km² of surrounding grasslands.

(2) Dangqiong Co has gradually expanded in the previous 20 years, with a growth rate of 0.596 km²/a. If it continues to expand according to existing trends, it will take 7 years for water levels to rise by 1 m. At this point, water may inundate two areas of the road and 0.28 km² of surrounding grasslands. Two potential points on the S205 road near the lake will be submerged in 7 years.

(3) Mami Co has gradually expanded in the previous 20 years, with a growth rate of 1.04 km²/a. According to current expansion rates, it will take 6.5 years for water levels to rise by 1 m, submerging 2.09 km² of surrounding grasslands.

Undoubtedly, LE will generate many adverse effects toward grasslands, salt pans, and roads, which has attracted local government attention. Animal husbandry and mining are the main future development industries in Xizang. Timely disaster predictions and preventative measures are key to ensuring the long-term economic development of the region, including infrastructural safety and maintenance. After our investigations, two points on the X625 road near Selin Co and two potential points on the S205 road near Dangqiong Co are at risk of being submerged in the future (Figs. 22–23). RS and geographical information system (GIS) technologies are advantageous for salt lake monitoring. In the future, this technology will provide a scientific basis for the development and use of salt lakes, and also monitoring environmental changes.

Our simulations and calculations on submerged grasslands and salt pans are approximate, therefore our results may be influenced by the observation time and resolution of RS data. Additionally, lake water area typically exhibited substantial seasonal variations as did vegetation growth in grassland areas. Therefore, water surface expansion, vegetation changes, and vegetation affected by LE should be carefully monitored during each period. Moreover, grassland and salt pan distribution is controlled by humans, so they are continually changing. This is particularly true for salt pans, where its drying status may have impacted on our observations. For example, if more water was contained in salt pans, it would be more difficult to identify and study. Thus, the extraction of such features, undergoing great change, warrants future research.

5 CONCLUSION

In this study, dynamic monitoring of ZSL was conducted over a 20-year period and four main research objects were extracted using different BC methods according to their specific SC. Moreover, an SRTM DEM was used to estimate the possible extent of grassland and salt pan damage caused by salt-lake expansion. Then, the results were combined with meteorological data to analyse the main reasons for the observed changes. After a series of image processing, statistical and analysis methods, the following conclusions were drawn:

(1) The area of ZSL has changed substantially over the last 20 years, from 88.86 km² to 194.46 km², with an average annual increase of 5.28 km². Although the growth rate has changed over time, the area has increased continually, implying that ZSL will continue to expand in the future.

(2) The continuous expansion of ZSL will have different effects on the surrounding grassland and salt pan. The grassland to the northeast and southwest of the North Lake and to the southeast of the South Lake will be the first to be affected, whereas the salt pan will be affected over a large area. Every metre of water-level rise in the lake will submerge an additional 7.84 km² of grassland and 2.7 km² of salt pan.

(3) CC is responsible for the observed changes of lakes and glaciers in TP. RS data revealed the trend of ZSL changes, as well as the relationship between the lakes, glaciers and CC. For ZSL, rising temperatures and accelerated melting glacier have led to its expansion. If the global climate continues to warm, we infer that ZSL will continue to expand and affect the surrounding grassland and salt pan.

This study demonstrates the successful application of RS and GIS technology to saline lake monitoring in the TP region. RS images can provide long-term continuous information over a large area, whereas GIS technology can analyse and mine deeper information hidden in the original data. We expect that these two technologies will play a more important role in future research on other saline lakes.

6 DATA AVAILABILITY STATEMENT

Data is provided by the Zabuye Salt Lake field observation station of the Chinese Academy of Geological Sciences.