Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

# Annual 30-m big Lake Maps of the Tibetan Plateau in 1991–2018

## Abstract

Lake systems on the Tibetan Plateau (TP) are important for the supply and storage of fresh water to billions of people. However, previous studies on the dynamics of these lakes focused on monitoring on multi-year scales and therefore lack sufficient temporal information. Here we present a new dataset comprising annual maps of big lakes (>10 km2) on the TP for 1991–2018, generated by utilizing all available Landsat images in conjunction with Google Earth Engine. The annual lake maps with high overall accuracy (~96%) highlight distinctive lake distribution and lake changes: (1) about 70% number and area of lakes concentrated in the Inner basin; (2) generally increasing trends in both the area (by 33%) and number (by 30%) of lakes from 1991 to 2018; (3) the total area changes were dominated by larger lakes (>50 km2) while more fluctuations in the lake number changes were found in medium lakes (10−50 km2). Our dataset infills temporal gaps in long-term inter-annual variations of big lakes, contributing towards enhanced knowledge of TP lake systems.

 Measurement(s) lake area and number Technology Type(s) satellite images and coding Sample Characteristic - Organism lake Sample Characteristic - Environment plateau-climate Sample Characteristic - Location Tibetan Plateau

## Background & Summary

The Tibetan Plateau – often referred to as the Third Pole and ‘Water Tower of Asia’ – has the highest number and largest area of lakes (approximately 50,000 km2 or 50% of the total lake area) in China1. These lakes are sensitive to global and regional environmental changes, which makes them ideal indicators of climate change2,3. Monitoring lake area change is therefore essential with respect to regional environmental and climate change issues; it benefits an enhanced understanding of local cryosphere changes and their impact on ecosystems, energy and hydrological cycles, and on livelihoods4,5. However, most studies rely upon lake mapping results obtained for irregular or large time intervals (i.e., five-year interval) to analyze lake responses to climate change, and so a continuous annual history of detailed lake variations across the whole TP has not yet been examined6,7,8.

Most existing lake extent maps were generated using satellite imagery acquired at discrete time points, which limits their inter-year comparability since lakes undergo areal changes every year and even every day1,3,9,10,11,12,13,14,15,16,17,18. Some studies have also only focused on a specific lake type or area of the TP1,6,19,20,21,22,23. There are only a few datasets providing continuous annual lake maps2,24, however, they either do not provide lake maps with high spatial and temporal resolution over the TP since they address global- or national-scale issues, or they contain a relatively large degree of uncertainty due to problematic data or limitations in the suitability of the methods employed to map lakes on the TP2. For instance, the use of top-of-atmosphere reflectance data without atmospheric correction would likely introduce uncertainty into the global surface water body dataset generated by Pekel et al., since atmospheric correction is a necessary step to extract quantitative information from the satellite imagery to remove the effects of aerosols, clouds and cloud shadows2,25,26. Moreover, the use of ancillary maps and data products in generating this global water dataset could also have introduced additional inaccuracies and inconsistencies to the resultant maps, since these were generated from different resources using different standards and mapping approaches and likely have varying degrees of accuracy.

The most recent annual maps of China’s surface water area and storage (1989–2016) have poor data quality for 1989 and 1990 in the TP area due to limited Landsat imagery and, without further removal of glaciers or rivers, the data on lakes smaller than 30 km2 on the TP carries large uncertainties27. Therefore, it is challenging to assess the lake dynamics on the TP using any of the existing datasets. Consequently, the incomplete and non-continuous historic lake dynamics record hampers attempts to fully understand the underlying driving hydrologic mechanisms on the TP. To help overcome this, a more detailed and complete historic record of changes in the number and size of lakes on the TP is required.

This study presents a new continuous dataset of annual lake maps on the TP for 1991–2018, which is generated by using all of the available archived Landsat imagery in conjunction with the Google Earth Engine (GEE) cloud-computing platform. The dataset focuses on lakes >10 km2 in size (hereinafter termed ‘big lakes’) as they account for more than 90% of the total lake area on the TP9,17. Given the dominance of big lakes and the challenges in identifying smaller lakes in high mountain regions in satellite imagery (e.g., insufficient spatial resolution, obscuring effects of clouds, snow and topographic shadow), most studies – including the present one – utilize only lakes larger than 10 km2 to capture a reliable representation of lake change dynamics across the TP6,8,28. To achieve this, we apply an automatic mapping approach to detect water bodies on the TP that involves the use of spectral indices and the concept of water frequency to generate annual maps of the big lake extents. The result is a detailed annual lake record for the three most recent decades, which can provide an enhanced understanding of lake change dynamics across the TP and help establish the underlying driving mechanisms.

## Methods

### Study area

The TP is situated in central Asia (25°59′ N–39°49′ N; 73°29′ E–104°40′ E) and is the most extensive and highest plateau in the world, with an average elevation of more than 4000 m above sea level (a.s.l.) and an area of 2.5 × 106 km2,29. There were ~1400 lakes of >1 km2 in size on the plateau in 2018, and 453 of these were larger than 10 km2,17. The changing hydrology on the TP has exerted huge impacts on ecosystems and human society, since it is the original water source for many essential rivers in Asia, supplying water to ~22% of Earth’s population for agricultural, industrial, and domestic use30. The climate across the TP varies significantly owing to its heterogeneous geography with annual average temperature from −4.1 °C in the Inner basin to 1.7 °C in Brahmaputra basin31. With its high elevation and broad surface, the plateau also acts as a barrier for the westerlies and monsoon atmospheric circulation. The plateau lies in the arid portion of the monsoon region and experiences a gradient of decreasing precipitation from east to west of 335–430 mm/yr32. The monsoonal climate area (Brahmaputra and Salween basins outlined in purple in Fig. 1) experiences summer rains and winter droughts, whereas the westerlies-controlled area (Tarim and Indus basins outlined in blue in Fig. 1) is extremely arid in winter with relatively more precipitation in summer. The other basins receive influences from both the Indian summer monsoon (ISM) and westerlies to some extent33, as indicated in Fig. 1.

### Source data

We obtained Landsat Collection 1 Tier 1 surface reflectance satellite imagery for Landsat 5 Thematic Mapper I, Landsat 7 ETM + , and Landsat 8 Operational Land Imager (OLI)/TIRS from the United States Geological Survey (USGS). All data in the collection have been atmospherically and geometrically corrected, and cross-calibration between different sensors has been applied34. Lakes were mapped using all the available imagery acquired during the period 1991–2018. It was not possible to map lakes at high temporal resolution prior to 1991 because the coverage is incomplete for the study area2,35. A summary of the source imagery used is shown in Table 1. Modern glacier polygons derived from Glacier Area Mapping for Discharge from the Asian Mountains (GAMDAM)36 were also used to avoid the possibility of glaciers being incorrectly mapped as lakes.

The study area is divided into ten basins as shown in Fig. 11,37, due to the vast extent of the TP and its spatial heterogeneity. The boundaries of the TP were taken from the Datasets of the Boundary and Area of the Tibetan Plateau (DBATP), which is a result of long-term fieldwork and was released and revised in 201429.

### Annual lake map generation

An approach based on water frequency – which is the ratio of the number of water observations at each location to the total number of images used in a year38 – has proven effective for obtaining reliable representations of annual water bodies27. This is especially important for the TP, where significant seasonal variations in lake boundaries are observed13,14. The frequency map approach can also further help reduce the impact of cloud shadows and other cloud artifacts to improve the accuracy of the mapping. Also, this method can avoid the drawbacks of manual visual interpretation and mapping of lakes based on their appearance on remotely sensed imagery, which can be time-consuming over large areas and somewhat subjective. The mapping approach employed involves several steps, including image preprocessing, water body map generation, and annual lake map production based on the water frequency, as shown in Fig. 2.

#### Generation of water body maps

Water body maps were subsequently generated from the preprocessed Landsat images using the Modified Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI), as reported in previous studies25,38:

$$MNDWI=\frac{{{\rm{\rho }}}_{{\rm{Green}}}-{{\rm{\rho }}}_{{\rm{SWIR}}1}}{{{\rm{\rho }}}_{{\rm{Green}}}+{{\rm{\rho }}}_{{\rm{SWIR}}1}}$$
(1)
$$NDVI=\frac{{{\rm{\rho }}}_{{\rm{NIR}}}-{{\rm{\rho }}}_{{\rm{Red}}}}{{{\rm{\rho }}}_{{\rm{NIR}}}+{{\rm{\rho }}}_{{\rm{Red}}}}$$
(2)
$$EVI=\frac{{{\rm{\rho }}}_{{\rm{NIR}}}-{{\rm{\rho }}}_{{\rm{Red}}}}{1.0+{{\rm{\rho }}}_{{\rm{NIR}}}+6.0{{\rm{\rho }}}_{{\rm{Red}}}+7.5{{\rm{\rho }}}_{{\rm{Blue}}}}$$
(3)

where ρBlue, ρGreen, ρRed, ρNIR and ρSWIR1 are the surface reflectance values for the blue, green, red, near-infrared (NIR) and shortwave infrared-1 (SWIR1) bands of the Landsat sensors40,41,42.

The MNDWI enhances water features in the imagery and has been widely applied to identify the presence of water bodies in various regions40,41,42. It also has the added benefit of eliminating any residual effects associated with mountain shadows and light cloud cover40,41,42. Since low-lying vegetation in proximity to wet surfaces is one of the major causes of commission error in open surface water body mapping, in this study we combined MNDWI and vegetation indices (NDVI and EVI) using logical operators to improve the discrimination capabilities of the water body mapping algorithm25,38:

$$Water\,body=\left(MNDWI > EVIorMNDWI > NDVI\right)and\left(EVI < 0.1\right)$$
(4)

where ‘MNDWI > EVI or MNDWI > NDVI’ identifies pixels that have a stronger water signal than vegetation signal25,38, while ‘EVI < 0.1’ ensures that all vegetation pixels or mixed water-vegetation pixels can be removed39. Only pixels meeting the criteria were classified as corresponding to water bodies and all other pixels were classified as non-water pixels25,38. The criteria are effective in distinguishing water body from non-water pixels on Landsat images27, and so it was used to generate water body maps for each of the 101,768 individual Landsat images.

#### Generation of water frequency and annual lake maps

The individual water body maps for each year were collated and then used to compute yearly water frequency maps (28 in total). The water frequency for each pixel in a water frequency map was calculated according to Zou et al.25 as:

$$F\left(y\right)=\frac{1}{{N}_{y}}\mathop{\sum }\limits_{i=1}^{{N}_{y}}{W}_{y,i}\ast 100{\rm{ \% }}$$
(5)

where F is the water frequency,y is the year, Ny is the total number of Landsat observations for that pixel in that year, while Wy,i denotes whether a pixel in a water body map is classed as water (represented by a value of 1) or non-water (a value of 0).

A permanent water surface is underwater throughout the entire year2, which should correspond to an annual water frequency (F) of 100% in a frequency map. However, the water frequency of pixels corresponding to water bodies that do persist all year long can be less than 100% due to the obscuring effects of cloud cover and shadow in the satellite imagery. For instance, some clouds (i.e., optically thin clouds that were not masked) that obscure the presence of an actual permanent water body beneath have a chance of being classified as non-water in the water body maps, which in turn would reduce the water frequency of the corresponding pixels in the annual water frequency map. To overcome this problem, we set a threshold of F ≥ 75% to classify pixels in the frequency maps as permanent water pixels. The choice of the threshold value of 75% is in line with previous studies25,27,38 and has been confirmed that it was the most appropriate threshold value for indentifying permanent water bodies in the TP area. After applying the threshold to the water frequency maps, areas of permanent water were extracted using the Reclassify Tool in Arcpy as pixels comprising water for the vast majority of the year38. Only these permanent water bodies were selected for inclusion in the inter-annual dataset, since the temporary bodies of water could be associated with seasonal inundations and irrigation activities2,41,43.

Next, the permanent water body extents were converted from raster to vector format to produce annual lake maps. These water body polygons were then intersected with the modern glacier polygons to remove any glaciers that were incorrectly mapped as lakes. Furthermore, the presence of other water bodies, such as rivers, was manually detected and removed to further improve the quality of the lake mapping dataset. The remaining permanent water bodies were then considered to comprise only lakes. Finally, after filtering the dataset to retain the lakes with areas larger than 10 km2, the area and perimeter of each big lake were computed to facilitate the determination of inter-annual lake changes.

## Data Records

A total of 28 annual lake maps are provided for the entire TP for the period 1991–2018. The dataset is available at the figshare repository in the Esri shapefile format and a statistical summary file (https://doi.org/10.6084/m9.figshare.13633880)44. The dataset is provided in the ESPG: 4326 (WGS_1984) spatial reference system. The 28 annual big lake maps comprise polygons demarcating the location and extent of each lake, and are attributed with information on the shape (Shape*), perimeter (Shape_Leng) and area (Shape_Area). These annual maps cover the geographic area from 73°29’ E to 104°40’ E longitude, and from 25°59’ N to 39°49’ N latitude. The maps can be visualized and analyzed in ArcGIS, QGIS or any similar software packages.

## Technical Validation

In order to assess the thematic accuracy of the water body mapping algorithm, we chose to validate the accuracy of the 2018 lake map, given that this year had the highest temporal resolution of very-high resolution (VHR) images in Google Earth. In particular, the high number of VHR images acquired between June and October – when the presence of cloud cover is relatively high and lakes are close to their maximum area – increases the likelihood of obtaining cloud-free scenes that capture a representative perspective of the status of the lakes to permit a reliable validation of the mapping results45. A total of 1000 sample points inside the lake polygons and another 1000 samples outside the lake polygons were randomly generated across the TP using the ArcMap software. A validation dataset was then compiled by assigning each sample point as either permanent water or non-permanent-water through visual interpretation based on VHR images in Google Earth. Since the classification of 91 of the validation sample points was not possible due to persistent cloud or snow cover and other image limitations, the remaining 1909 sample points were used to calculate the accuracy through a confusion matrix. This was performed by comparing the designation of the validation points to those in the lake map to determine the proportion of points that were correctly classified. The results revealed an overall accuracy of 95.8% and a Kappa coefficient of 0.92 (Table 2). In addition, the user’s and producer’s accuracies of the lake maps are 92.1% and 98.9%, respectively. We also compared the dataset with another recent dataset of lakes larger than 1 km2 in Tibetan Plateau (V2.0) (mapped several specific years from 1970s to 2018)1, generated by using visual interpretation and NDWI. The area of lakes >1 km2 is only about 3.2–5.6% larger than the area of lakes >10 km2 mapped here, and that dataset also exhibits a similar change trend from 1991 to 2018 that further validates our method. The Joint Research Centre (JRC) dataset2, which represents significant progress in the remote sensing application of surface water mapping, was also used for comparision. Since the JRC dataset includes small lakes and streams, it has a moderately higher total lake area for TP region than that in our dataset. Nevertheless, the two datasets show very similar change trends on the whole (Fig. 3). The largest deviation between the two datasets occurs for 1997, when there is a substantial decrease of lake area in the JRC dataset and only a minor decrease was seen in our data (which has an overall accuracy of 93.4% for this year). The main differences between these two datasets could be due to subtle differences in the range of water bodies mapped (i.e., we excluded all rivers and lakes smaller than 10 km2, whilst the JRC dataset includes these), differences between using Lansdat top-of-atmosphere reflectance and surface reflectance data, as well as inherent differences between the mapping methods. Overall, the high mapping accuracy and comparison with the existing datasets attest the reliability and enhanced potential for ultising the 1991‒2018 annual lake map dataset for additional applications.

Overall, accurate mapping of lakes depends on the quality and spatio-temporal resolution of the remotely sensed data and the choice of water body mapping algorithm8,46. In general, the uncertainty of the lake extent delineation using satellite data is inversely proportional to the spatial resolution of the images used, since the lakes or their associated boundary changes may be smaller in size than that of a pixel8. This is also the case for the identification of emerging and disappearing lakes. In this study, only the mapping of sizeable lakes (>10 km2) was performed, which is larger than the mapping unit (pixel size) of 30 m. While this excludes the analysis of smaller lakes, it is still possible to gain an accurate and reliable representation of lake dynamics across the TP given that 90% of the lakes are >10 km2. Furthermore, the ability to generate a consistent long-term dataset that captures the dynamics of the lakes is only possible due to the extensive archive of satellite imagery provided by the long-running Landsat earth observation program. Whilst satellites with improved spatial and spectral capabilities may facilitate more detailed mapping for the future and recent past, a long-term historical analysis of lakes on the TP is inherently constrained by the imaging capabilities of the Landsat satellites.

In this study, the use of an automatic method was critical for large-scale lake mapping over a long time period, as the manual interpretation of over 100,000 individual Landsat images would have been an extremely laborious task. Also, compared to other studies that use only single images to extract lakes for each year, calculating the water frequency helps to improve the mapping accuracy by mitigating the effect of clouds and their shadows, as well accounting for seasonal changes of the lakes25. This was achieved by applying a water body frequency threshold of 75% to capture only permanent lakes, therefore increasing the reliability of the observed inter-annual variation trends2,38,46. The water frequency approach also mitigates potential differences in the reliability when mapping using imagery acquired from Landsat 5, 7 and 846, particularly those surrounding the number of observations available for each year based on the operational periods of the satellites. Although there were an average of 1790 good observations each year for 1991‒1998 from Landsat 5, 3981 images each year for 1999‒2012 acquired from Landsat 5 and 7, and 5286 satellite images each year for 2013‒2018 from Landsat 7 and 8, all map products performed comparably well (with accuracy >95%2,27) in capturing surface water regardless of the platform or number of observations. Therefore, the influence of using three Landsat sensors on the mapping results is considered negligible. Furthermore, a consistent method of lake mapping was applied to 28 years’ worth of data, which also helps to further reduce any uncertainties in the observed trends38. Finally, the intersection with glaciers and manual removal of rivers further improves the quality of the lake mapping results.

In the future, with improved access to higher spatial resolution data, the mapping of smaller lakes and the analysis of the drivers could be undertaken to analyze contemporary lake change dynamics. In addition, the method outlined here could be applied to imagery as it is acquired, in order to rapidly update the lake maps for the TP in near-real time. The ability to map and analyse seasonal (intra-year changes) is currently restricted by the lack of sufficient observations needed to reliably generate monthly or bi-monthly water frequency maps, due to the prevalent cloud cover over parts of the TP, especially during the summer. However, with access to satellites or constellations of satellites with a shorter revisit time in the future this will likely become possible as higher frequency acquisitions would maximise the opportunity to obtain more cloud-free images.

## Usage Notes

The lakes on the TP are highly dynamic and mapping their inter-annual variations can provide enhanced insights into the effects of climate oscillation or extreme climate events. Tracking the inter-annual lake dynamics can also provide a valuable indication of the future of the TP hydrological system. In this study, we have produced maps with relatively high temporal resolution (1 year) and spatial resolution (30 m) for a long time-series (28 years) for all TP lakes larger than 10 km2. This continuous long-term record therefore allow us to examine the lake changes in detail and ensures that any crucial and relatively rapid changes in the trends are not missed. The annual lake maps are valuable for investigating the spatial heterogeneity of lake change patterns across the plateau and for investigating the changes of specific lakes. Particularly, the high accuracy of the maps ensures the quality and reliability of the obtained change records for those important lakes (Fig. 4). If combined with climatic data of high spatial and temporal resolution or change records of environmental elements like glaciers, permafrost and vegetation, the annual lake maps would allow for an in-depth analysis of the mechanism behind the changes or the linkage between them. The lake boundaries are also important for the validation of global and regional hydrological modelling and can be used to improve the predication capability of models to aid a better understanding of the water resources, which would further inform policy-making and support sustainable development in the region. The data may also contribute to regional and local flood risk mapping across the TP.

### Data statistical properties

We performed a statistical analysis on the dataset to further demonstrate the value of the new annual lake maps. With the presented continuous lake maps, we are able to track the changes associated with big lakes across the TP on an annual basis during 1991–2018. Overall, the TP lakes experienced a significant expansion from 1991 to 2018 (Fig. 5), with a considerable increase of 12976 km2 in area (40%) and 130 in number (47%)44. This corresponds to overall change rates of lake area and number of 463 km2/year and 4.6 lakes/year, respectively. In 2018, there were 409 lakes >10 km2 on the TP, comprising a total area of 45621 km2. In terms of the dynamics, the annual lake change record shows an increase in area from 1991 to 1992, followed by a sharp decrease in 1995 and then a significant and continuous increase since 1995, except in 2009 and 2015 when decreases occurred (Fig. 5a,b). These fluctuations in lake area observed in 1992, 2009 and 2012 would not be detectable in five-year interval datasets, which were the highest temporal resolution lake datasets covering the TP prior to this study (indicated in Fig. 5a). The change in the number of lakes follows a very similar pattern to the lake area (Fig. 5a,b). Similarly, a five-year interval lake number dataset with record of only 1995, 2000, 2005, 2010 and 2015 would lack information for lake dynamics within those five-year intervals so that fewer fluctuations will be seen (Fig. 5a). Overall, the inter-annual fluctuation in lake area and number is highest during 1991–1996 and around 2010 (Fig. 5b). Larger lakes (>50 km2) are seen to have contributed to the overall change in lake area across the TP (Fig. 5c). From 1991 to 2018, the total area of the larger lakes increased considerably by 11521 km2 (39.9%), while the medium-sized lakes (10−50 km2) experienced a smaller areal increase of 1455 km2 (38.7%). Although both the numbers of larger (from 116 to 173) and medium (from 163 to 236) lakes significantly increased during 1991–2018 (Fig. 5d), the fluctuation in the number of medium lakes is greater than that of larger lakes, most notably in 1995 and 2010.

We further calculated both the number and area of lakes for each basin and found that the lakes are unevenly distributed across the TP (Fig. 6). The Inner basin was found to have the highest average proportion of lakes with 73% of the total, followed by the Qaidam basin and Brahmaputra basin (Fig. 6a). On average, there are only 0 to 15 lakes located in Tarim basin, Yangtze River basin and Salween basin, while the Mekong basin does not contain any lakes larger than 10 km2. The Inner basin has the largest lake area (67%), followed by Yellow (15%) and Qaidam (6%). Salween, Yangtze, Mekong and Tarim basins have the smallest total lake area on the TP.

## Code availability

The annual maps of lakes larger than 10 km2 from 1991 to 2018 were produced using GEE platform. Key JavaScript code developed for this work are openly shared with the scientific community at figshare repository44. GEE should be used to access and edit the code.

## References

1. Zhang, G., Luo, W., Chen, W. & Zheng, G. A. Robust but variable lake expansion on the Tibetan Plateau. Sci. Bull. 64(18), 1306–1309 (2019).

2. Pekel, J. F., Cottam, A., Gorelick, N. & Belward, A. S. High-resolution mapping of global surface water and its long-term changes. Nature 540, 418–422 (2016).

3. Zhang, G. et al. Regional differences of lake evolution across China during 1960s–2015 and its natural and anthropogenic causes. Remote Sens. Environ. 221, 386–404 (2019).

4. Yao, T., Pu, J., Lu, A., Wang, Y. & Yu, W. Recent glacial retreat and its impact on hydrological processes on the Tibetan Plateau, China, and surrounding regions. Antarct. Alp. Res. 39, 642–650 (2007).

5. Zhang, G. et al. Response of Tibetan Plateau’s lakes to climate changes: Trend, pattern, and mechanisms. Earth-Sci. Rev. 208, 103269 (2020).

6. Li, Z. et al. Response of Glacier and Lake Dynamics in Four Inland Basins to Climate Change at the Transition Zone between the Karakorum And Himalayas. PloS. One. 10(12), e0144696 (2015).

7. Song, C. & Sheng, Y. Contrasting evolution patterns between glacier-fed and non-glacier-fed lakes in the Tanggula Mountains and climate cause analysis. Climatic Change 135, 493–507 (2016).

8. Zheng, G. et al. Sustained growth of high mountain lakes in the headwaters of the Syr Darya River, Central Asia. Global. Planet. Change 176, 84–99 (2019).

9. Lei, Y. et al. Response of inland lake dynamics over the Tibetan Plateau to climate change. Climatic Change 125, 281–290 (2014).

10. Sun, J. et al. Linkages of the dynamics of glaciers and lakes with the climate elements over the Tibetan Plateau. Earth-Sci. Rev. 185, 308–324 (2018).

11. Zhang, G. et al. Lake water and glacier mass gains in the northwestern Tibetan Plateau observed from multi-sensor remote sensing data: Implication of an enhanced hydrological cycle. Remote Sens. Environ. 237, 111554 (2020).

12. Zhang, H. Y., Wu, Y. H., Lei, L. P. & Guo, L. A. Remote Estimation of Water Storage Variation of Lakes in Tibetan Plateau over the Past 20 Years. Int. Geosci. Remote Se., 8412-8415 (2018).

13. Zhang, Y., Zhang, G. & Zhu, T. Seasonal cycles of lakes on the Tibetan Plateau detected by Sentinel-1 SAR data. Sci. Total Environ. 703, 135563 (2020).

14. Zhang, G., Li, J. & Zheng, G. Lake-area mapping in the Tibetan Plateau: an evaluation of data and methods. Int. J. Remote Sens. 38, 742–772 (2017).

15. Zhang, G., Yao, T., Xie, H., Kang, S. & Lei, Y. Increased mass over the Tibetan Plateau: from lakes or glaciers? Geophys. Res. Lett. 40, 2125–2130 (2013).

16. Zhang, G. et al. Response of Tibetan Plateau lakes to climate change: Trends, patterns, and mechanisms. Earth-Sci. Rev. 208, 103269 (2020).

17. Zhang, G., Yao, T., Xie, H., Zhang, K. & Zhu, F. Lakes’ state and abundance across the Tibetan Plateau. Chinese Sci. Bull. 59, 3010–3021 (2014).

18. Zhang, Z. X. et al. The response of lake area and vegetation cover variations to climate change over the Qinghai-Tibetan Plateau during the past 30 years. Sci. Total Environ. 635, 443–451 (2018).

19. Zhang, G. et al. Extensive and drastically different alpine lake changes on Asia’s high plateaus during the past four decades. Geophys. Res. Lett. 44, 252–260 (2017).

20. Wang, X. et al. Glacial lake inventory of High Mountain Asia (1990–2018) derived from Landsat images. Earth Syst. Sci. Data 12, 2169–2182 (2020).

21. Zhang, Z. et al. Glacier variations at Aru Co in western Tibet from 1971 to 2016 derived from remote-sensing data. J. Glaciol. 64, 397–406 (2018).

22. Zhang, G., Yao, T., Xie, H., Wang, W. & Yang, W. An inventory of glacial lakes in the Third Pole region and their changes in response to global warming. Global. Planet. Change 131, 148–157 (2015).

23. Zhang, G. et al. Lake volume and groundwater storage variations in Tibetan Plateau’s endorheic basin. Geophys. Res. Lett. 44, 5550–5560 (2017).

24. Chen, F. et al. Annual 30-meter Dataset for Glacial Lakes in High Mountain Asia from 2008 to 2017. Earth Syst. Sci. Data 13, 2 (2020).

25. Zou, Z. et al. Divergent trends of open-surface water body area in the contiguous United States from 1984 to 2016. P. Natl. A. Sci. India. B. 115, 3810–3815 (2018).

26. Liang, S., Fang, H. & Chen, M. Atmospheric correction of Landsat ETM+ land surface imagery. I. Methods. IEEE. T. Geosci. Remote. 39(11), 2490–2498 (2001).

27. Wang, X. et al. Gainers and losers of surface and terrestrial water resources in China during 1989–2016. Nat. Commun. 11, 1–12 (2020).

28. Debnath, M. et al. Glacial lake dynamics and lake surface temperature assessment along the Kangchengayo-Pauhunri Massif, Sikkim Himalaya, 1988–2014. Remote Sens. Appl. Soc. Environ. 9, 26–41 (2018).

29. Zhang, Y., Bingyu, L. I. & Zheng, D. Datasets of the boundary and area of the Tibetan Plateau. Glob. Change Res. Data Publ. Repository http://www.geodoi.ac.cn/WebEn/HTML_INFO.aspx?Id=deb04027-8ab1-4d59-8efd-ba71cec7b7f7 (2014).

30. Xu, J. et al. The melting Himalayas: cascading effects of climate change on water, biodiversity, and livelihoods. Conserv. Biol. 23, 520–530 (2009).

31. Pörtner, H.-O. et al. Special Report on the Ocean and Cryosphere in a Changing Climate. (IPCC, 2019).

32. Hudson, A. M. & Quade, J. Long-term east-west asymmetry in monsoon rainfall on the Tibetan Plateau. Geology 41, 351–354 (2013).

33. Yao, T. et al. A review of climatic controls on δ18O in precipitation over the Tibetan Plateau: Observations and simulations. Rev. Geophys. 51, 525–548 (2013).

34. Google Earth Engine. USGS Landsat 5 Surface Reflectance Tier 1, https://explorer.earthengine.google.com/#detail/LANDSAT%2FLT05%2FC01%2FT1_SR (2020).

35. Allen, S. K., Zhang, G., Wang, W., Yao, T. & Bolch, T. Potentially dangerous glacial lakes across the Tibetan Plateau revealed using a large-scale automated assessment approach. Sci. Bull. 64(7), 435–445 (2019).

36. Nuimura, T. et al. The gamdam glacier inventory: a quality-controlled inventory of Asian glaciers. Cryosphere 9(3), 849–864 (2015).

37. Liu, X. China level two watershed dataset. National Earth System Science Data Center, National Science & Technology Infrastructure of China http://www.geodata.cn/data/datadetails.html?dataguid=243293730193084&docId=10644 (2002).

38. Zou, Z. et al. Continued decrease of open surface water body area in Oklahoma during 1984–2015. Sci. Total Environ. 595, 451–460 (2017).

39. Zhu, Z. & Woodcock, C. E. Automated cloud, cloud shadow, and snow detection in multitemporal Landsat data: An algorithm designed specifically for monitoring land cover change. Remote Sens. Environ. 152, 217–234 (2014).

40. Xu, H. Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. Int. J. Remote Sens. 27, 3025–3033 (2006).

41. Sun, F., Sun, W., Chen, J. & Gong, P. Comparison and improvement of methods for identifying waterbodies in remotely sensed imagery. Int. J. Remote Sens. 33, 6854–6875 (2012).

42. Li, K., Wang, J. & Yao, J. Effectiveness of machine learning methods for water segmentation with ROI as the label: A case study of the Tuul River in Mongolia. INT. J. APPL. EARTH OBS. 103, 102497 (2021).

43. Yang, Z. et al. Mapping Panax Notoginseng Plantations by Using an Integrated Pixel-and Object-Based (IPOB) Approach and ZY-3 Imagery. Remote. Sens. 13(11), 2184 (2021).

44. Zhao, R. et al. Source code for: 28 years annual maps of lakes with area larger than 10 km2 on the TP (1991-2018). Figshare https://doi.org/10.6084/m9.figshare.13633880.v3 (2021).

45. Wang, X. et al. Changes of glaciers and glacial lakes implying corridor-barrier effects and climate change in the Hengduan Shan, southeastern Tibetan Plateau. J. Glaciol. 63, 535–542 (2017).

46. Zhou, Y. et al. Open surface water mapping algorithms: A comparison of water-related spectral indices and sensors. Water 9(4), 256 (2017).

47. Benn, D. I. & Owen, L. A. The role of the Indian summer monsoon and the mid-latitude westerlies in Himalayan glaciation: review and speculative discussion. J. Geol. Soc. London 155, 353–363 (1998).

48. Yao, T. et al. Different glacier status with atmospheric circulations in Tibetan Plateau and surroundings. Nat. Clim. Change 2, 663 (2012).

## Acknowledgements

This study is funded by the Key Research Program of Frontier Sciences (QYZDB-SSW-DQC005) and Strategic Priority Research Program (XDA19040301) of Chinese Academy of Sciences (CAS), and the National Science Foundation China General Program (41971078).

## Author information

Authors

### Contributions

J.D., P.F. and R.Z. designed the study and the methodology. Y.Z. and X.X. wrote the code. R.Z. generated the data, evaluated the resulting maps, analyzed the data, wrote the manuscript. S.G. and G.Z. edited the manuscript.

### Corresponding authors

Correspondence to Ping Fu or Jinwei Dong.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

Zhao, R., Fu, P., Zhou, Y. et al. Annual 30-m big Lake Maps of the Tibetan Plateau in 1991–2018. Sci Data 9, 164 (2022). https://doi.org/10.1038/s41597-022-01275-9

• Accepted:

• Published:

• DOI: https://doi.org/10.1038/s41597-022-01275-9