Satellite Data Revealed That the Expansion of China’s Lakes Is Accompanied by Rising Temperatures and Wider Temperature Differences

Abstract

Lake surface water area (LSWA) and lake surface water temperature (LSWT) are critical indicators of climate change, responding rapidly to global warming. However, studies on the synergistic variations of LSWA and LSWT are scarce, and the coupling relationships among lakes with different environmental characteristics remain unclear. In this study, the relative growth rate of LSWA (RK_LSWA); the absolute growth rates of annual maximum, mean, and minimum LSWTs (i.e., K_{LSWT_max}, K_{LSWT_mean}, K_{LSWT_min}); and the absolute growth rates of the difference between maximum and minimum LSWT (LSWT_mmd) (K_{LSWT_mmd}) were investigated across more than 4000 lakes in China using long-term Landsat data, and their coupling relationships among different lake types (i.e., permafrost and non-permafrost recharge, endorheic or exorheic lakes, and natural and artificial lakes) were comprehensively analyzed. Results indicate significant differences in the trends of LSWA and LSWT, as well as their interrelationships across various regions and lake types. In the Qinghai–Tibet Plateau (QTP), 57.8% of lakes showed an increasing trend in LSWA, with 2.4% of the lakes showing moderate expansion (RK_LSWA values of 0.1–0.2), while over 27.5% of lakes in the South China (SC) region displayed shrinkage in LSWA (RK_LSWA values were between −0.1~0%/year). Regarding LSWTs, 49.8% of lakes in the QTP exhibited a K_{LSWT_max} greater than 0, and 47.9% of lakes showed a K_{LSWT_mean} greater than 0. In contrast, 48.1% of lakes in the Middle and Lower Yangtze River Plain (MLYP) had a K_{LSWT_max} less than 0, and 48.5% of lakes had a K_{LSWT_mean} less than 0. Additionally, lakes supplied by permanent permafrost demonstrated more significant growth in both LSWA and LSWT than those supplied by non-permanent permafrost. Further analysis revealed that approximately 20.2% of the lakes experienced a concurrent increase in both mean LSWT and LSWA, whereas around 18.9% of the lakes exhibited a simultaneous rise in both LSWT_mmd and LSWA. This suggests that the expansion of lakes in China is correlated with both rising temperatures and greater temperature differences. This study provides deeper insights into the response of Chinese lakes to climate change and offers important references for lake resource management and ecological conservation.

1. Introduction

Lakes play an important role in the hydrologic cycle, providing essential ecological services [1,2,3]. The lake surface water area (LSWA) is an important indicator of sensitivity to global and regional climate change and a key parameter for basin water resource assessments and water balance analyses [4]. The lake surface water temperature (LSWT) is an important parameter for energy exchange between lakes and the atmosphere and is often used to assess climate change impacts on lakes [5,6,7,8,9]. In the context of global warming, LSWA and LSWT are changing rapidly and responding to natural and anthropogenic factors [10,11,12]. Therefore, understanding the dynamics of LSWA and LSWT is essential for studying lake ecosystems and environmental changes [13,14,15].

Many efforts have been paid to investigate changes in LSWA and LSWT at regional and global scales with the intensification of global warming. Based on Landsat imagery data, the area of 3.4 million lakes globally showed an upward trend with a net increase of 46,278 km² from 1984 to 2019; the Himalayan glacial lakes enlarged by about 14.1% between 1990 and 2015 [16], and the overall increase in lake area in the Qinghai–Tibet Plateau (QTP) has accelerated significantly since 2000 [17], with the number of lakes increasing from 868 to more than 1200, and the total area of lakes expanding from 38,823.3 km² to 48,793.0 km² [18]. However, global warming not only affects LSWA but also has a significant impact on LSWT, with a significant increase in the number of severe heatwave events in lakes globally by nearly 40 events per decade since 1995 [19] and an average warming rate of 0.34 °C per decade in LSWT during the global warm season [20]. From 1985 to 2022, results based on the air2water model combined with Landsat data indicate that the warming rate of LSWT in China’s 2260 lakes is 0.16 °C per decade without removing extreme high temperatures [21]. Over a relatively large time span of 1978–2017, studies also utilizing the air2water model with MODIS data have shown that annual LSWT of lakes on the QTP increased significantly during 1978–2017, by 0.01–0.47 °C per decade [22]. During 2001–2016, the LSWT of 169 large lakes in China generally showed an increasing trend, with an average increase rate of 0.26 °C per decade [23]. The above studies show that both LSWA and LSWT can well reflect the dynamic changes of global warming; however, most of the existing studies only focus on LSWA or LSW and lack a systematic study on the coupling relationship between the two.

In addition, different types of lakes may respond differently to climate change. Compared to non-glacier-recharged lakes, glacier-recharged lakes exhibit significantly greater increases in water volume and delayed LSWA peaks, while endorheic lakes show more pronounced seasonal area fluctuations than exorheic lakes [24]. Moreover, endorheic lakes typically experience more distinct seasonal cycles and faster water level rises than exorheic lakes. This difference may stem from the reliance of endorheic lakes solely on inflowing water, whereas exorheic lakes balance both inflows and outflows [25,26]. Lakes recharged by permafrost can lead to lake expansion through thermal erosion of lake margins and geomorphological changes, especially in areas with high ground ice content, whereas lakes that are not recharged by permafrost may not have such expansion possibilities [27]. Natural lakes and artificial reservoirs exhibit marked differences in LSWA dynamics, with reservoirs generally showing greater surface area variability than natural lakes of comparable size [28]. Additionally, bottom waters in reservoirs are often warmer than those in lakes, which may be attributed to differences in flushing rates and associated stratification mechanisms [29]. Different types of lakes exhibit distinct characteristics, which may influence the coupling relationships between LSWA and LSWT. Investigating these relationships across various lake types is essential for understanding their hydrological and thermal dynamics.

The number and variety of lakes in China, distributed under different geographical and climatic conditions, provide a unique chance for exploring the correlation between LSWT and LSWA in different regions and types of lakes [30,31]. Although two censuses of lakes in different regions of China were conducted in 1960–1980 and 2005–2006 [32,33], the changes in LSWA and LSWT of different types of lakes with the current changes in lakes and their drivers remain unclear. Zhang et al. [34] improved the limitations of studies on LSWA and LSWT of lakes in China by investigating the changes in water bodies in different regions of China from the 1960s to 2015, but their study did not explore the changes in LSWA and LSWT of different types of lakes and their coupling relationships. China’s lakes cover a huge area, support nearly half of the country’s centralized drinking water sources, and play a crucial role in human ecosystem services [35]. Over the past four decades, persistent and dramatic climate change, rapid urbanization, and economic growth have dramatically altered the thermal conditions of lakes. In addition, lakes in China are diverse in character and geographic location and are therefore subject to multiple climate change factors [30,36]. Therefore, it is urgent to study the long-term change pattern of overall lakes in China.

This study utilized time-series Landsat data and a comprehensive dataset of lakes in China, combined with the random forest algorithm, to extract annual LSWA and LSWT data for over 4000 lakes with areas exceeding 1 km². Regression analyses were employed to statistically analyze the trends in LSWA and LSWT. The analysis focused on examining the variations and coupling relationships of LSWA and LSWT across different lake types and regions. This study can provide a reference for lake ecological assessment, monitoring of extreme hydrological droughts [37], and provides a scientific basis for lake protection and drought risk assessment.

2. Materials and Methods

2.1. Study Area

The study area covers 73°33′E to 135°05′E and 3°51′N to 53°33′N, covering a wide range of climatic zones from temperate to tropical, as shown in Figure 1. It can be divided into nine distinct geographical regions: Northeast China Plain (NCP), Yunnan–Guizhou Plateau (YGP), Northern Arid and Semi-Arid Regions (NAASR), South China (SC), Sichuan Basin and Surrounding Areas (SBASR), Middle and Lower Yangtze River Plain (MLYP), QTP, Loess Plateau (LP), and Huang-Huai-Hai Plain (HHHP). The geographic zoning vector data were obtained from the Chinese Academy of Sciences Resource and Environment Data Cloud Platform (http://www.resdc.cn, accessed on 1 March 2025). In these regions, natural resource endowments—including topography, climate, soil, vegetation, and socio-economic development—are relatively balanced, which minimizes the impact of major natural and societal factors [38]. This results in a clearer reflection of regional differences in lake distribution [39].

The study area encompasses 4093 lakes with surface area exceeding 1 km². Lakes have different hydrological characteristics and recharge mechanisms and can be categorized based on their representation of regional hydrological conditions and the availability of reliable data. In this study, the classification considers factors such as the source of recharge (whether the lake is recharged by permafrost, i.e., permafrost or non-permafrost), the direction of water flow in the hydrological process (endorheic or exorheic), and the type of lake formation (manmade reservoirs or natural lakes). These criteria are used to reflect both the natural attributes and hydrological characteristics of the lakes. Based on these factors, the lakes are classified into eight types: permafrost recharge into endorheic artificial reservoirs (PIA), non-permafrost recharge into endorheic artificial reservoirs (UPIA), permafrost recharge into endorheic natural lakes (PIN), non-permafrost recharge into endorheic natural lakes (UPIN), permafrost recharge into exorheic artificial reservoirs (POA), non-permafrost recharge into exorheic artificial reservoirs (UPOA), permafrost recharge into exorheic natural lakes (PON), and non-permafrost recharge into exorheic natural lakes (UPON). In this classification, P denotes lakes fed by permafrost meltwater, while UP refers to those replenished by non-permafrost meltwater. O signifies exorheic lakes, and I denotes endorheic lakes. Additionally, N designates natural lakes, and A indicates artificial reservoirs. The distribution of these lake types is intrinsically linked to regional climate, topography, and recharge processes, collectively shaping the hydrological dynamics and ecological diversity of the study area. The total number of lakes by region and type is shown in

Table 1. Summary of the number of lakes in different geographical regions and by lake type.

Distinct Geographical Regions	Number of Lakes	Different Types of Lakes		Number of Lakes
NCP	548	permafrost or non-permafrost	recharge from permafrost	1045
YGP	185		non-recharge from permafrost	3048
NAASR	641		total number of lakes of this category	4093
SC	222	Endorheic or exorheic	endorheic	1212
SBASR	67		exorheic	2881
MLYP	1051		total number of lakes of this category	4093
QTP	1014	Manmade reservoirs or natural lakes	manmade reservoirs	852
LP	57		natural lakes	3241
HHHP	308		total number of lakes of this category	4093
Total number of lakes studied				4093

.

2.2. Data Sources

2.2.1. Landsat Data and Joint Research Centre Global Surface Water (JRC Global Surface Water) Dataset

This study utilized Landsat series satellite surface reflectance and temperature products in conjunction with the JRC Global Surface Water dataset, both accessible via the Google Earth Engine (GEE) platform. The Landsat series provides remote sensing imagery with high spatial, spectral, and temporal resolution, making it well suited for long-term environmental analysis [40]. Specifically, the surface reflectance data from Landsat 5 and 7 were generated using the Landsat Ecosystem Disturbance Adaptive Processing System LEDAPS (LEDAPS) algorithm [41], while data from Landsat 8 were processed using the Land Surface Reflectance Code LaSRC (LaSRC) algorithm [42]. All Landsat series of standardized Land Surface Temperature products are derived using an operational single channel (SC) algorithm based on the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Emissivity Dataset Version 3 (GEDv3), which provides physically retrieved emissivity values that enable more accurate and consistent correction of Landsat thermal data at their native spatial resolution [43]. Compared to the MODIS and Sentinel satellite series, Landsat satellites offer higher spatial resolution in the thermal infrared bands. Additionally, their coverage effectively encompasses the study area, providing superior data support for detailed land surface temperature monitoring and analysis. The data volume provided by each satellite is illustrated in Figure 2.

The JRC Global Surface Water dataset, derived from Landsat 5, 7, and 8 imagery spanning from 1984 to 2019, provides a global surface water coverage map at a 30 m resolution. This dataset enables the tracking of surface water dynamics over time, identifying regions where water levels have increased, decreased, or remained stable globally. It includes data on the probability of water presence at each pixel, inter-annual variability, spatial distribution, and seasonal fluctuations. Specifically, the dataset classifies trends into non-water areas, seasonal water bodies, and permanent water bodies, offering insights into their dynamic changes, spatial distribution, and the duration (in months) of seasonal and permanent water presence. Despite its strengths, the JRC dataset mainly offers statistical surface water information and does not effectively capture variations in LSWA. Nevertheless, it can be utilized as a supplementary dataset for correction purposes.

2.2.2. GLAKES Global Lakes Dataset

The GLAKES global lake dataset is derived from the Global Water Frequency Dataset and a deep learning-based semantic segmentation model. It provides detailed boundary information for lakes worldwide, covering 3.4 million lakes with an area greater than or equal to 0.03 square kilometers. The dataset records the maximum spatial extent of lakes from 1984 to 2019. This comprehensive coverage significantly reduces inaccuracies in lake LSWA extraction caused by lake movement. Compared to existing similar datasets, GLAKES offers a broader coverage, higher spatial resolution, and stronger spatiotemporal consistency, making it capable of reflecting long-term trends [44]. Additionally, GLAKES includes several auxiliary variables that describe the geographic characteristics of lakes, such as whether the lake is fed by glaciers or permafrost, whether it is an endorheic lake, or whether it is an artificial reservoir. These variables further enhance the value of GLAKES in the fields of lake hydrological processes, ecological environment assessment, and water resource management.

2.3. Method

2.3.1. Lake Surface Water Body and Temperature Extraction

Given that the random forest (RF) algorithm integrates Classification and Ensemble learning, Regression Trees, Boosting, and Bagging, it exhibits high classification accuracy, robustness to outliers and noise, computational efficiency, internal error estimation capability, model strength, variable importance assessment, and ease of parallelization [45,46,47,48,49,50,51,52]. RF was employed to classify water bodies using Landsat 5, Landsat 7, and Landsat 8 satellite images covering terrestrial regions of China from 2000 to 2020 in this study. To extract the distribution of water bodies, a combination of several water indices was applied (as shown in

Table 2. Indices used for lake water body extraction.

Index	Formula	Serial Number
MNDWI	${\displaystyle \frac{Green-MIR}{Green+MIR}}$	(1)
NDBI	${\displaystyle \frac{TIR-SWIR1}{TIR+SWIR1}}$	(2)
NDVI	${\displaystyle \frac{SWIR1-NIR}{SWIR1+NIR}}$	(3)
EWI	${\displaystyle \frac{Red-SWIR1-TIR}{Red+SWIR1+TIR}}$	(4)
NDWI	${\displaystyle \frac{Red-SWIR1}{Red+SWIR1}}$	(5)
LSWI	${\displaystyle \frac{SWIR1-TIR}{SWIR1+TIR}}$	(6)
NBR2	${\displaystyle \frac{TIR-SWIR2}{TIR+SWIR2}}$	(7)
AWEI	$4\times \left(Red-TIR\right)-(0.25\times SWIR1+2.75\times SWIR2)$	(8)

The formulas in the table are used for calculating various remote sensing indices, with each element representing specific bands: green refers to the green band; red corresponds to the red band; NIR represents the near-infrared band; SWIR1 and SWIR2 denote shortwave infrared bands 1 and 2; MIR refers to the mid-infrared band; and TIR represents the thermal infrared band.

), including the Modified Normalized Difference Water Index (MNDWI), Normalized Difference Built-up Index (NDBI), Normalized Difference Vegetation Index (NDVI), Enhanced Water Index (EWI), and Land Surface Water Index (LSWI). These indices were integrated into an RF model for classifying water bodies, resulting in annual water body images. Classification accuracy was evaluated using multiple metrics, including overall accuracy, user accuracy, producer accuracy, and the Kappa coefficient. The model achieved an overall accuracy of 0.9899, with both user and producer accuracies exceeding 95% and a Kappa coefficient of 0.9795 (as shown in

Table 3. Accuracy metrics for image random forest classification.

Evaluation Indicators	Precision Value
Validation overall accuracy	0.98989898989899
User accuracy	[[0], [0.9772727272727273], [1]]
Producer accuracy	[[0], [1], [0.9821428571428571]]
Kappa	0.9794988610478361

). To derive annual data on permanent water bodies from 2000 to 2020, the JRC dataset was used as a spatial mask to filter out non-water pixels from the annual water body images. Pixels with a water occurrence frequency greater than 70% were designated as permanent water bodies. Subsequently, the GLAKES lake dataset was used to extract the area of these permanent water bodies within the lakes, which was then calculated as LSWA.

For temperature extraction, thermal infrared bands were employed to invert the temperature distribution. The permanent water body image was used as a mask to clip the LSWT of the lake surface. Based on the time series LSWT data, the annual temperature statistics, including maximum, average, and minimum LSWT for each lake from 2000 to 2020 were calculated.

2.3.2. Trend Analysis

Given that linear regression analysis effectively captures trends in time-series data [53], this study employed this method to analyze variations in LSWT and LSWA over specific periods after obtaining the complete LSWA and LSWT dataset for all lakes. The significance of the slope in a linear regression equation is determined using the t-Student test. The t-statistic follows a t-distribution with n − 2 degrees of freedom, where n is the sample size. By comparing the calculated t-statistic with the critical values in the t-distribution table, the corresponding p-value is calculated, and a p-value of <0.05 is used as a criterion for assessing the presence of a significant trend in the time series [54]. The relevant equations are shown in (9)–(12) below:

$$K={\displaystyle \frac{{\sum }_{i=0}^n\left(x_i-\overline{x}\right)\left(y_i-\overline{y}\right)}{{\sum }_{i=1}^n{\left(x_i-\overline{x}\right)}^2}}$$

(9)

$$\mathrm{S}\mathrm{E}\left(\mathrm{K}\right)=\sqrt{{\displaystyle \frac{\frac{1}{\mathrm{n}-2}{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{y}}_{\mathrm{i}}-{\widehat{\mathrm{y}}}_{\mathrm{i}}\right)}^2}{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{x}}_{\mathrm{i}}-{\overline{\mathrm{x}}}_{\mathrm{i}}\right)}^2}}}$$

(10)

$$\mathrm{t}={\displaystyle \frac{\mathrm{K}}{\mathrm{S}\mathrm{E}\left(\mathrm{K}\right)}}$$

(11)

$$\mathrm{P}=2\times P\left(\mathrm{T}>\left|\mathrm{t}\right|\right)$$

(12)

where x_i represents the year (from 2000 to 2020, with indices from 0 to 20), y_i represents the corresponding temperature/area values (i.e., annual maximum, mean, minimum, and difference between maximum and minimum LSWT, and LSWA), ${\hat{\mathrm{y}}}_{\mathrm{i}}$ is the ith prediction, K is the slope of the linear relationship between the year and the temperature including K_{LSWT_max}, K_{LSWT_mean}, K_{LSWT_min}, K_{LSWT_mmd} (with units of °C per year), and K_LSWA (with units of km² per year), and SE(K) is the standard error of the slope.

The relative growth rate, calculated from the ratio of K_LSWA to average LSWA, is used to quantify changes in LSWA due to the significant differences in lake areas. Lakes with larger surface areas may exhibit substantial absolute changes in water body size, but these changes may not accurately reflect their relative importance or variability. The relative growth rate allows for comparison of growth across lakes of differing sizes by normalizing the growth relative to the initial area. This method ensures that the variation in growth rates is comparable across lakes with varying surface areas, providing a more robust metric for understanding temporal changes in lake water bodies. The specific formula is as follows:

$${RK}_{LSWA}={\displaystyle \frac{K_{LSWA}}{mean}}\times 100\mathrm{\%}$$

(13)

where mean is the average LSWA from 2000 to 2020, and RK_LSWA is the relative growth rate of LSWA (%/year).

The differences in the changes of LSWA and LSWT among lakes of different types and regions, as well as their coupling relationships, were analyzed. The flowchart is shown in Figure 3.

3. Results and Analysis

3.1. Changes of Lake Surface Water Area in China

As shown in Figure 4, from 2000 to 2020, 36.5% of lakes in China exhibited a significant increasing trend in LSWA, with 89.0% of these lakes having RK_LSWA values between 0 and 0.1 year⁻¹. In contrast, 18.1% of lakes showed a significant decreasing trend in LSWA, with 91.8% of these lakes having RK_LSWA values between −0.1 and 0 year⁻¹. RK_LSWA box plots and statistical analyses for different lake types and regions are shown in Figure 5 and Figure 6 and

Table 4. Variance and mean of RK_LSWA of lakes in different regions.

Region	The Mean Value of RKLSWA (%/year)	Variance
HHHP	−0.003636	0.003119
LP	0.029050	0.001098
MLYP	−0.000839	0.002282
NAASR	0.021927	0.002273
NCP	0.007927	0.001454
QTP	0.020254	0.001417
SBASR	0.045260	0.002793

and

Table 5. Variance and mean of RK_LSWA in different types of lakes.

Region	The Mean Value of RKLSWA (%/year)	Variance
non-permafrost	0.678165	23.838035
permafrost	2.559002	12.817852
exorheic	0.78819	23.254914
endorheic	2.176142	15.868552
natural lakes	0.929816	20.185217
manmade lakes	3.153732	20.994347

.

In the SBASR region, 44.8% of lakes displayed an increasing trend in LSWA, with 9.0% of lakes showing RK_LSWA values in the 0.1–0.2 year⁻¹ range. This may be attributed to the reduction in cultivated land area and decreased water consumption due to the conversion of cropland to forest, which has led to an increase in LSWA [55]. In the QTP region, nearly 57.8% of lakes have increased in area, primarily due to water supply from permanent glaciers and permafrost. Increased precipitation has been the main driver of increased lake water storage, while glacier mass loss and permafrost thaw have supplemented this increase [56]. In the LP region, about 42.1% of lakes showed an increasing area trend, which is largely due to artificial rainfall, ecological water replenishment, and other measures that reversed the trend of lake shrinkage [57]. In the NAASR region, over 45.0% of lakes had RK_LSWA values significantly greater than 0. The expansion of lake areas in this region is mainly attributed to the warm and humid climate, while the shrinkage of lakes is driven by increased surface water consumption [58,59]. In SC, more than 27.5% of the lakes showed a decreasing trend in LSWA. The shrinkage of lakes may be related to lake reclamation, lakeside development, and lake enclosure [33]. In other regions such as NCP, HHHP, and YGP, there is no great variability in the LSWA changes of lakes.

Among lakes recharged by permafrost, 63.5% exhibited an increase in LSWA, whereas only 25.8% of lakes recharged by non-permafrost sources showed such an increase. The increase in lake area for permafrost-recharged lakes may be attributed to increased runoff caused by global warming. Among endorheic lakes, 57.2% displayed an increasing trend in LSWA, compared to only 26.2% for exorheic lakes. In terms of artificial reservoirs and natural lakes, over 30.0% of artificial reservoirs and nearly 36.8% of natural lakes showed an increase in LSWA, with only a small difference between the two categories.

3.2. Changes of Lake Surface Water Temperature in China

Spatially, as shown in Figure 7, Figure 8 and Figure 9, in China, 23.7% of lakes exhibit an increasing trend in maximum LSWT, while 28.3% show a decreasing trend. Regarding the changes in mean LSWT, 21.9% of lakes show an increasing trend, while 29.3% exhibit a decreasing trend. As for the minimum LSWT changes, 19.3% of lakes show an increase, while 28.3% demonstrate a decrease. Figure 10 and Figure 11 display box plots of K_{LSWT_max}, K_{LSWT_mean}, and K_{LSWT_min} for different regions and lake types, and

Table 6. Variance and mean of K_{LSWT_max}, K_{LSWT_mean}, and K_{LSWT_min} of lakes in different regions.

Region	The Mean Value of KLSWT_max (°C/Year)	The Variance of KLSWT_max	The Mean Value of KLSWT_mean (°C/Year)	The Variance of KLSWT_mean	The Mean Value of KLSWT_min (°C/Year)	The Variance of Kmin
HHHP	−0.0573	0.013	−0.054	0.0048	−0.0296	0.0021
LP	0.0239	0.0199	0.0083	0.0075	0.0066	0.0037
MLYP	−0.0333	0.005	−0.0309	0.0023	−0.0216	0.0019
NAASR	−0.0043	0.0143	−0.0136	0.0048	−0.0161	0.0071
NCP	−0.0199	0.0087	−0.0184	0.0028	−0.0157	0.0035
QTP	0.0138	0.0079	0.0040	0.0026	−0.0007	0.0039
SBASR	0.0080	0.0023	0.0066	0.0011	0.0053	0.0015

and

Table 7. Variance and mean of K_{LSWT_max}, K_{LSWT_mean}, and K_{LSWT_min} in different types of lakes.

Type	The Mean Value of KLSWT_max (°C/Year)	The Variance of KLSWT_max	The Mean Value of KLSWT_mean (°C/Year)	The Variance of KLSWT_mean	The Mean Value of KLSWT_min (°C/Year)	The Variance of KLSWT_min
non-permafrost	0.0007	0.0014	0.0008	0.0014	0.0023	0.0014
permafrost	0.0227	0.0009	0.0209	0.0009	0.0222	0.0009
exorheic	0.002	0.0014	0.0007	0.0014	0.0028	0.0014
endorheic	0.0189	0.0011	0.0164	0.0011	0.0189	0.0012
natural lakes	0.0057	0.0012	0.0035	0.0012	0.0056	0.0013
manmade lakes	0.0192	0.0018	0.0201	0.0019	0.0211	0.0018

summarize the mean and variance of K_{LSWT_max}, K_{LSWT_mean}, and K_{LSWT_min} for different regions and lake types.

In the QTP region, 49.8% of lakes display K_{LSWT_max} greater than 0, 47.9% of lakes show K_{LSWT_mean} greater than 0, and 36.3% of lakes exhibit K_{LSWT_min} greater than 0. This trend is likely driven by meteorological factors, including increased downward longwave radiation, temperature, humidity, and decreased wind speed [60]. In the NAASR region, 32.8% of lakes show an increasing trend in max LSWT, and 28.4% display an increase in mean LSWT. The warming trend in this region is mainly driven by anthropogenic factors, with human activities contributing the most to climate warming [61]. In contrast, in the MLYP region, 48.1%, 48.5%, and 39.9% of lakes show K_{LSWT_max}, K_{LSWT_mean}, and K_{LSWT_min} less than 0, respectively. This declining trend is likely related to the significant cooling observed between 2007 and 2012 [62]. In the HHHP region, 31.2%, 35.4%, and 33.1% of lakes showed K_{LSWT_max}, K_{LSWT_mean}, and K_{LSWT_min} less than 0, while only 8.1%, 4.9%, and 7.8% of lakes showed K_{LSWT_max}, K_{LSWT_mean}, and K_{LSWT_min} greater than 0, respectively, and this trend suggests that that there may be a decrease in lake temperatures at some point in time in the context of global warming, which is consistent with the decrease in temperature in the region from 2009 to 2013 found by Duo et al. [63]. In other regions such as NCP and YGP, there is no great variability in the LSWA changes of lakes.

Among lakes with permafrost recharge, the percentages of lakes showing an increase in max LSWT, mean LSWT, and min LSWT are 57.0%, 53.3%, and 39.2%, respectively. In contrast, these proportions are significantly lower in lakes without permafrost recharge, where only 12.3%, 11.1%, and 12.4% show increases in the respective LSWT categories. Among endorheic lakes, 47.7% and 44.7% of lakes show K_{LSWT_max} and K_{LSWT_mean} greater than 0, while in exorheic lakes, these percentages are 13.6% and 12.3%, respectively. For natural lakes and artificial reservoirs, the numbers of lakes showing an increase and a decrease in LSWT are approximately equal.

3.3. Synthetic Changes of Lake Surface Water Area and Temperature in China

3.3.1. Coupling Relationship Between LSWA and Mean LSWT in China

In order to further investigate the spatial variation of LSWA and LSWT for different lake types, we first identified lakes with statistically significant RK_LSWA trends (p < 0.05), then selected those exhibiting significant K_{LSWT_mean} (p < 0.05) trends within this subset. These selected lakes were categorized into four classes based on their variation directions: 00, 01, 10, and 11, where the first digit is 0 for K_{LSWT_mean} less than 0 and 1 for K_{LSWT_mean} greater than 0, and the second digit is 0 for RK_LSWA less than 0 and 1 for RK_LSWA greater than 0. The spatial distribution map of the relationship between K_{LSWT_mean} and RK_LSWA, as shown in Figure 12, and the number of lakes in each LSWA–mean LSWT change type (00, 01, 10, 11) in different lake categories is shown in

Table 8. Number of lakes in each LSWA–mean LSWT variation type (00, 01, 10, 11) across different lake categories.

Type	Number of Lakes with Type 00	Number of Lakes with Type 01	Number of Lakes with Type 10	Number of Lakes with Type 11
PIA	0	1	0	2
PIN	32	14	0	429
POA	0	2	0	3
PON	19	4	0	88
UPIA	0	0	0	3
UPIN	61	14	0	81
UPOA	90	40	2	93
UPON	455	59	2	127

and

Table 9. Number of lakes in each LSWA–LSWT variation type (00, 01, 10, 11) across different regions.

Region	Number of Lakes with Type 00	Number of Lakes with Type 01	Number of Lakes with Type 10	Number of Lakes with Type 11
HHHP	57	13	0	13
LP	3	2	0	6
MLYP	275	50	1	54
NAASR	76	20	1	172
NCP	93	10	1	79
QTP	62	25	0	451
SBASR	10	1	0	17
SC	52	6	1	8
YGP	28	7	0	26

. Lakes of types 00 and 11 indicate lakes where the mean LSWT and LSWA are decreasing and increasing together, while lakes of types 01 and 10 indicate lakes where the mean LSWT is decreasing (increasing) and the LSWA is increasing (decreasing). Type 11 lakes constituted 20.2% of all lakes, with 54.6% distributed in the QTP region and 20.8% located in NAASR. Comparatively, Type 00 lakes accounted for 16.0% of the total lake population, of which 41.9% were concentrated in the MLYP region.

In the MLYP region, type 00 emerged as the most prevalent category, comprising 26.2% of the total distribution. The remaining lake types were predominantly classified as 01 and 11, accounting for 4.8% and 5.1% of the total proportion, respectively. Precipitation in the MLYP area is a key meteorological factor influencing LSWA [64] which may lead to independent changes in LSWA and LSWT. Lakes closer to estuaries exhibit more type 10 characteristics, due to moderated temperature fluctuations and increased water vapor from the ocean. Conversely, type 00 lakes are found further inland, where temperature fluctuations are more pronounced. In the NAASR, approximately 26.8% of lakes are type 11, primarily consisting of subtropical outwash lakes, which exhibit substantial seasonal fluctuations in temperature and water area. Temperature variations have a profound impact on snowmelt recharge, with warmer temperatures leading to increased snowmelt, thereby increasing the LSWA [65]. Conversely, as temperatures decrease, snowmelt recharge diminishes, which causes a reduction in RK_LSWA. In the QTP, most lakes are type 11, accounting for 44.5%, and characterized by permafrost recharge. This phenomenon may be attributed to rising temperatures, which accelerate glacier melting, snowmelt, and permafrost degradation, thereby increasing runoff and leading to an increase in LSWA [66]. However, LSWA decline in certain areas may be associated with localized arid climates and reduced precipitation, particularly between the Himalayas and the Tanggula Mountains, as well as north of the Hengduan Mountains, where enhanced evapotranspiration and permafrost degradation may contribute to LSWA reduction [67,68].

Among lakes replenished by permafrost, over 50.0% are classified as type 11 lakes, while 4.9% are classified as type 00 lakes. In contrast, among lakes not replenished by permafrost, the proportions are 10.0% for type 11 lakes and 20.0% for type 00 lakes. This trend is likely driven by rising temperatures, which further melt permafrost and increase lake runoff, leading to an increase in LSWA for these lakes. Within the categories of endorheic and exorheic lakes, 42.5% of endorheic lakes are identified as type 11, whereas only 10.8% of exorheic lakes exhibit the same classification. Regarding natural lakes and artificial reservoirs, 22.4% of natural lakes and 11.9% of artificial reservoirs are categorized as type 11 lakes, and 17.5% of natural lakes and 10.6% of artificial reservoirs are categorized as type 00 lakes. Furthermore, in the categories of PIN and PON lakes, over 55.6% and 34.2% of lakes, respectively, exhibit characteristics of type 11 lakes.

3.3.2. Coupling Relationship Between LSWA and LSWT_mmd in China

As in Figure 13, we also identified lakes with statistically significant RK_LSWA trends (p < 0.05), then selected those exhibiting significant K_{LSWT_mmd} trends (p < 0.05) within this subset. These selected lakes were categorized into four classes based on their variation directions: 00, 01, 10, and 11, where the first digit of 0 means K_{LSWT_mmd} is less than 0 and 1 means K_{LSWT_mmd} is greater than 0, and the second digit of 0 means RK_LSWA is less than 0 and 1 means RK_LSWA is greater than 0. Lakes of types 00 and 11 indicate that lakes where the LSWT_mmd and LSWA are decreasing and increasing together, while lakes of types 01 and 10 indicate that lakes where the LSWT_mmd is decreasing (increasing) and the LSWA is increasing (decreasing). The number of lakes in each LSWA–LSWT_mmd change type (00, 01, 10, 11) in different lake categories is shown in

Table 10. Number of lakes in each LSWA–LSWT_mmd variation type (00, 01, 10, 11) across different lake categories.

Type	Number of Lakes with Type 00	Number of Lakes with Type 01	Number of Lakes with Type 10	Number of Lakes with Type 11
PIA	0	1	0	3
PIN	23	49	3	379
POA	0	4	0	2
PON	15	7	1	77
UPIA	1	0	0	2
UPIN	53	28	3	69
UPOA	74	54	13	86
UPON	32	67	50	155

and

Table 11. Number of lakes in each LSWA–LSWT_mmd variation type (00, 01, 10, 11) across different regions.

Region	Number of Lakes with Type 00	Number of Lakes with Type 01	Number of Lakes with Type 10	Number of Lakes with Type 11
HHHP	46	11	6	24
LP	3	6	0	8
MLYP	195	59	30	57
NAASR	61	37	4	167
NCP	76	15	6	93
QTP	54	63	3	387
SBASR	6	3	2	15
SC	38	11	11	4
YGP	19	5	8	18

. Type 11 lakes account for 18.9% of all lakes, among which 50.1% are located in the QTP area and 21.6% in the NAASR area. In contrast, type 00 lakes make up 12.2% of all lakes, with 39.2% concentrated in the MLYP region.

In the QTP region, lakes of types 11 account for 38.2% of the total lake population. The significant rise in summer temperatures has accelerated glacier retreat and permafrost thawing [69], which in turn impacts both the surface water area and temperature of lakes. Additionally, the decrease in minimum temperatures during winter has reduced lake recharge and evapotranspiration, with minimal effects on ice formation. For lakes recharged by permafrost, both maximum and minimum temperatures may have increased, with the rate of increase in maximum temperatures surpassing that of minimum temperatures, resulting in a net rise in the temperature differential. The increase in winter temperatures, combined with reduced snowfall [70,71], has led to increased snowmelt, replenishing the lakes and driving an increase in RK_LSWA. In the NAASR region, lakes of type 11 account for 26.1%, respectively, likely due to the significant impact of seasonal climate variations on lake dynamics, with annual temperature differences affecting evaporation and snowmelt recharge [72]. Precipitation remains a critical factor in determining lake size in these regions [36]. The increase in annual temperature differences exacerbates seasonal variations and precipitation fluctuations [73], which may further expand lake area. In the MLTP region, the proportion of type 00 is 18.6%. The lakes here are close to the river mouth and are less affected by the large temperature fluctuations caused by the ocean, resulting in more stable seasonal temperatures. These lakes show a tendency to shrink.

Among lakes replenished by permafrost, the proportion of type 11 lakes reaches 44.1%, whereas in lakes not replenished by permafrost, this proportion is only 10.2%. Within the categories of endorheic and exorheic lakes, the proportion of type 11 lakes in endorheic lakes is 37.4%, significantly higher than the 11.1% observed in exorheic lakes. Regarding natural and artificial lakes, type 11 lakes account for 10.9% in artificial lakes, compared to 21.0% in natural lakes, and type 00 lakes account for 8.8% in artificial lakes, compared to 13.1% in natural lakes. Furthermore, in the categories of PIN and PON lakes, the proportions of type 11 lakes are 49.2%, and 30.0%, respectively, which are significantly higher than other lake types within the same categories.

4. Discussion

4.1. Comparison with Existing Studies

As analyzed in Section 3.1 and Section 3.2, in the QTP region, 49.7% of lakes exhibited a significant increase in mean LSWT, with the majority of these lakes located in the northern and central parts, whereas lakes with a decrease in mean LSWT were predominantly distributed in the southwestern region. This spatial distribution and about 70% of the lake variation in LSWT of lakes in the QTP region is consistent with the findings of Wan et al. [74]. Additionally, 57.8% of the lakes in the QTP region showed an increasing trend in LSWA, and 7.4% of the lakes showed a decreasing trend in LSWA, which is consistent with Su et al.’s [75] finding that the gain ratios of permanent lake surface area in the Tibetan Inner Basin and the Three-River Headwaters region are 37.78% and 13.46%, respectively. In contrast, the loss ratios of permanent surface area in these two regions are much lower, at only 1.12% and 3.5%, respectively. Furthermore, Zhang et al. [34] showed that the total LSWA of lakes in the QTP increased by about 15% between the 1960s and 2015, with an average annual rate of increase of 103.00 ± 42.49 km²/yr, which is highly consistent with the results of the overall increasing trend of lake area in the QTP revealed in this study. Although Zhang et al.’s study [56] also found a general increase in LSWA in QTP lakes, with more than 80% of the lakes increasing in area between the 1970s and 2010 [76], the relative lake area change in their study was more than 100%, while the RK_LSWA range reflecting the relative change of LSWA in this study was only −30–30%. This is because this study uses linear regression equation to obtain K_LSWA and then divides K_LSWA by the average LSWA of the lake from 2000 to 2020 to obtain the RK_LSWA reflecting the changes of the lake, rather than simply dividing the LSWA of a certain year by the LSWA area of the starting time. The RK_LSWA obtained in this way can better reflect the change trend of lake LSWA according to the observed values in the study time scale and avoid the disastrous impact on the determination of lake change that may be caused by the wrong observed values in a certain year.

Zhang et al. [33] reported a large number of lakes with decreasing LSWA in MLYP, where the number of disappearing lakes accounts for approximately 40% of the total lake loss in China, and Xu et al.’s study [77] found that the primary reason for the shrinkage of mid-stream lakes in the MLYP region is land reclamation and agricultural irrigation, while the fluctuations and expansion of downstream lakes are mainly influenced by precipitation and runoff. This finding is consistent with the lake area changes observed in the MLYP region in this study. Meanwhile, LSWTs in the MLYP region in this study generally show a decreasing trend, which aligns with the results of Tao et al. [78], who found a cooling effect of the water-temperature interaction in all seasons following the construction of the Three Gorges Reservoir. However, according to Huang et al.’s study [79], LSWTs in the lakes around the MLYP region are generally increasing, and this discrepancy may be due to the smaller number of lakes selected in their study.

4.2. Implications to Water Resource Management

This study reveals the changes in LSWA and LSWTs in different regions and different types of lakes in China from 2000 to 2020, which provides important decision support for water resource management. In the context of global warming, monitoring of water body changes and management of water resources are becoming more and more important, because it directly affects water security and ecological balance [80,81]. By analyzing the changes of LSWA and LSWTs of lakes, the dynamic changes of water resources can be monitored so as to avoid overdevelopment and irrational utilization of water resources and to enable them to give full play to the regulating functions of water storage, water retention, and water release [82] and effectively mitigate the impacts of hydrological events aggravated by climate change such as severe floods, snowstorms, and glacier replenishment [83]. It also helps to identify potential threats to water resources and prevent extreme events, including flooding [84,85].

In addition, lakes play an important role in supporting agricultural irrigation, industrial water supply, and urban drinking water [86], and monitoring lakes can help optimize the allocation and use of water resources, thereby contributing to the achievement of sustainable development goals, especially in the areas of agriculture, urban planning, and nature reserve management [87]. Therefore, this research not only constitutes a major contribution to the scientific community but also holds significant practical implications for policymakers and environmental management agencies.

4.3. Limitations and Future Work

There are some uncertainties and inaccuracies during the data extraction process. One of the major challenges is the absence of satellite images aroused by cloud cover, which has impeded the acquisition of complete and reliable data. This limitation has resulted in gaps and inconsistencies in temporal coverage, reducing the ability to capture continuous lake dynamics. Furthermore, inherent errors during temperature retrieval, such as limitations in inversion algorithms and satellite sensor resolution, may cause discrepancies between the extracted temperature values and the actual surface water temperature. Additionally, atmospheric interference, calibration issues, and the quality of retrieval products have exacerbated these errors.

Moreover, this study has only considered lake changes in China between 2000 and 2020, which presents limitations in terms of both temporal and spatial scale. The short time span of the data has prevented a comprehensive capture of long-term lake dynamics and their responses to climate change, especially in terms of extreme climatic events. Also, the relatively small spatial scale used in this study may not fully reflect the trends of lake changes across larger regions or multiple ecological zones.

To address these limitations, future research should incorporate monitoring data spanning a longer time period and extend to broader geographical areas, providing more representative and generalizable analysis results. Furthermore, the split-window driven temperature-and-emissivity separation (SWDTES) algorithm developed by Wang et al. [88] will be used in future studies to obtain high-precision and high-resolution LST data. Combined with the future scenario data of CMIP6, the impacts of extreme climate events such as heat wave, drought, and rainstorm on LSWA and LSWT in different regions and lake types were simulated to explore the coupling mechanism and driving factors between LSWA and LSWT_mmd. This will contribute to a better understanding of the climate adaptation capacity of lakes.

5. Conclusions

In this study, the random forest algorithm combined with JRC data was used to accurately extract LSWA data of more than 4000 lakes in China from 2000 to 2020. Then, according to the LSWA data of each year, the corresponding LSWTs data were extracted, and the RK_LSWA, K_{LSWT_max}, K_{LSWT_mean}, K_{LSWT_min}, and K_{LSWT_mmd} of each lake were calculated by using regression linear equation. The variation difference of LSWA and LSWT in different types of lakes was analyzed, and the coupling relationship between LSWA and mean LSWT, LSWA, and LSWT_mmd was explored. We found that the variation of LSWA and LSWT in different regions and different kinds of lakes is very different. In QTP and NAASR regions, 57.8% and 47.0% of LSWA showed an upward trend, respectively. The LSWT of lakes in the QTP region is generally on the rise, while the LSWT of lakes in the MLYP region is generally on the decline. Approximately 20.2% of the lakes showed a simultaneous increase in mean LSWT and LSWA, while about 18.9% of the lakes showed a simultaneous increase in LSWT_mmd and LSWA as well. By analyzing the dynamics of lakes across different regions, this study highlights the significant regional differences in both lake area and water temperature changes, providing a better understanding of how climate change impacts lakes in various regions. In regions facing the dual challenges of climate change and human activities, this study offers valuable insights for decision-makers, helping to predict how lakes will respond to climate change.