Spatiotemporal variations and driving factors of evapotranspiration in the Yunnan-Guizhou Plateau from 2003 to 2020

ET

evapotranspiration

YGP

Yunnan–Guizhou plateau

Ec

vegetation transpiration

Es

soil evaporation

Ei

vaporization of intercepted rainfall

ETa

annual mean ET

ETs

seasonal mean ET

ETm

monthly mean ET

Eca

annual mean Ec

Ecs

seasonal mean Ec

Ecm

monthly mean Ec

Esa

annual mean Es

Ess

seasonal mean Es

Esm

monthly mean Es

Eia

annual mean Ei

Eis

seasonal mean Ei

Eim

monthly mean Ei

PML-V2

Penman–Monteith–Leuning Version 2

Sen–MK

Theil–Sen median trend method with Mann–Kendall test

BEAST

Bayesian estimation of additive seasonal and trend

GED

geodetector mode

YP

Yunnan Province

GP

Guizhou Province

Rn

net radiation

NDVI

normalized difference vegetation index

Tem

monthly average temperature (a product of temperature)

Pre

monthly average precipitation (a product of precipitation)

2Tem

monthly average 2m temperature (another product of temperature)

Tpre

total precipitation (another product of precipitation)

Wind

10m wind speed

RHum

relative humidity

SPre

surface pressure

TCC

total cloud cover

DEM

elevation

Slp

slope

Asp

aspect

Kop

Köppen climate types

CLCD

China land cover dataset

TC

trend component in BEAST

SC

seasonal component in BEAST

QTP

Qinghai–Tibetan plateau

LP

Loess Plateau

Surface evapotranspiration (ET) is a vital component of both the surface energy and water balances, ranking as the second-largest water flow in the terrestrial water cycle, trailing only precipitation. It encompasses key elements such as vegetation transpiration (Ec), soil evaporation (Es), vaporization of intercepted rainfall (Ei), water evaporation, and snow and ice sublimation (Sharma et al. 2016; He et al. 2022). Globally, over 60% of land precipitation is cycled back to the atmosphere through ET, concurrently utilizing approximately three-fifths of the net surface radiation. The associated latent heat, with its cooling effect, positions ET as a crucial link connecting the water and heat cycles (Oki & Kanae 2006).

ET is one of the main parameters of land-surface hydrothermal processes, and studying its distribution pattern and characteristics is crucial for understanding the repercussions of climate change (Nistor et al. 2016). However, due to the complexity of its occurrence mechanism and the significant influence of net radiation, precipitation, temperature, wind speed, and water–air pressure difference (Xie & Zhu 2013), ET remains a challenging variable to estimate. Previous studies have shown that Ec accounts for about 56%–74% of ET, and increased CO₂ concentration is thought to lead to the closure of vegetative stomata, thereby affecting transpiration (Good et al. 2015). As the hydrological cycle changes, the risk of floods and droughts will increase, which will inevitably have ecological and human social impacts. Therefore, monitoring of ET is of great significance.

In recent years, data-driven retrieval methods and process-driven physical retrieval methods have become the two main categories of ET retrieval techniques (Zhang et al. 2019). Data-driven remote-sensing inversion methods for estimating ET establish relationships based on features closely related to ET. These methods rely on various data sources, including satellite remote-sensing data, meteorological data, hydrological data, and flux observation data (Elbeltagi et al. 2022a; Das et al. 2023). Additionally, data-driven machine-learning methods have also often been utilized in ET estimation, such as reference ET (Elbeltagi et al. 2022b; Kushwaha et al. 2022) and actual ET (Bai et al. 2021; Rajput et al. 2024), owing to their exceptional capabilities in classification and regression prediction (Reichstein et al. 2019). Process-driven models have also been employed to produce various actual ET products, such as the MOD16, GLEAM, and PML-V2 products (Zhang et al. 2019). These ET products are commonly used to analyze the impacts of global or regional climate change, drought, or flood events.

Plateau regions, particularly the Qinghai–Tibetan Plateau (QTP) and Loess Plateau (LP) in China, are sensitive indicators of global climate change (Yang et al. 2021). Academics have extensively investigated climate-change impacts on various phenomena in these regions, including phenology, vegetation, biological community succession, lake changes, snowpack changes, water resources, land surface temperature, and human activities (Zhang et al. 2020; He & Tang 2023). While prior research has mainly focused on the QTP and LP, there is a relative scarcity of studies on ET in the YGP (Cao et al. 2020; Guo et al. 2022). Despite the rich forest vegetation types in the YGP and its susceptibility to climate-induced droughts and floods, studies on ET in this region have been limited (Cheng et al. 2020). Considering the significance of ET as a surface hydrological parameter, investigating its spatial and temporal distribution in the YGP is crucial for understanding and addressing climate-change impacts in the region.

Currently, numerous studies are dedicated to time-series analysis of ET, for instance, utilizing various trend analysis methods to investigate the changing trends in regional ET characteristics (Huang et al. 2023; Su et al. 2023). In equatorial Africa, research has identified key drivers of ET, highlighting the substantial impact of factors such as air temperature, soil moisture, and relative humidity (Nooni et al. 2023). In the Amazon, ET is primarily influenced by precipitation and energy availability (Zhang et al. 2022). In southwest China, the primary factor influencing ET is sunshine duration (Zhang et al. 2023). Thus, it is evident that the dominant factors influencing ET vary significantly across different regions. Exploring the dominant factors influencing ET on the YGP is crucial for understanding the causes of drought and wildfires. In the aforementioned studies, researchers have primarily focused on the annual or seasonal scales of ET, with relatively limited attention given to the monthly scale. Given that the influence of different driving factors varies across different months (Wang et al. 2023), it is important and necessary to analyze the trends in ET and their driving factors on a monthly scale in regions similar to YGP, where dry and rainy seasons are distinctly defined.

In summary, ET products derived from various methods are often used to analyze the causes of changes in ET in typical regions around the globe, thereby assessing the impacts of climate change. Although numerous studies have revealed the dominant factors influencing ET in specific regions, these factors vary across different regions, especially in plateau regions, and there is still insufficient research analyzing the spatiotemporal variations and dominant factors affecting actual ET at monthly scale in the YGP. These are essential for the prevention and management of droughts, landslides, and other hazards in the YGP.

Therefore, this study aims to: (1) investigate the spatiotemporal patterns and trends of ET in the YGP at various different temporal scales; (2) analyze the major driving factors influencing ET in the YGP at monthly and annual scales; (3) detect the breakpoints of ET in the YGP across different temporal scales.

The YGP exhibits significant spatial variation, requiring high-resolution data for analysis. This study utilizes an ET dataset derived from the PML-V2 (China) dataset, which employs the Penman–Monteith–Leuning Version 2 model. Compared with global ET datasets, this product demonstrates superior accuracy in simulating ET within China (He et al. 2022). Further assessments show that this new product surpasses other commonly used ET estimation products such as MOD16A2, GLEAM, MOD17A2H, SEBAL, EC-LUE, and VPM (He et al. 2022).

This study utilized 15 driving factors, including daily net radiation (Rn), normalized difference vegetation index (NDVI), monthly average temperature (Tem), monthly average precipitation (Pre), monthly average 2 m temperature (2Tem), and total precipitation (Tpre), 10 m wind speed (Wind), relative humidity (RHum), surface pressure (SPre), total cloud cover (TCC), elevation (DEM), slope (Slp), aspect (Asp), Köppen climate types (Kop), and China land cover dataset (CLCD). These driving datasets primarily encompass factors such as radiation, meteorology, terrain, land cover types, etc., which exert significant influence on ET. The datasets are introduced as shown in the Table 1. Due to the relatively significant impact of temperature and precipitation on ET, two temperature and rainfall-related datasets from different sources were chosen in order to achieve a more comprehensive analysis.

Owing to the absence of an ET dataset from 2000 to 2002 in some parts of the YGP, we excluded the data from this period to ensure the accuracy and completeness of the analysis. The ET data utilized in this study were the summation of Ec, Es, and Ei in the PML-V2 dataset. From the temporal scale, it can be divided into annual mean ET (ETa), seasonal mean ET (ETs), and monthly mean ET (ETm), which were obtained by summing daily ET over the respective periods. The term ‘mean ET’ refers to the temporal scale rather than the spatial scale.

There are several noteworthy points in the dataset processing: (1) All driving factors were resampled to 500 m and then cropped to match the size of the study area. (2) Annual mean data for Pre and Tpre were obtained through cumulative summation, similar to the process for ET data, while other driving factors were obtained by averaging. (3) To ensure temporal-scale consistency, monthly-scale analyses utilize monthly average data for driving factors, while annual analyses utilize annual average data, except for topography and climate-type data.

In this study, identifying the evolving trend in the long-term series of ET changes is achieved by combining the Theil–Sen median (Sen) trend method (Sen 1968) with the Mann–Kendall (MK) test (Mann 1945). This combined approach is referred to as the Sen–MK method for simplicity in subsequent analyses. The Sen trend method is utilized to assess change trends for each pixel at annual, seasonal, and monthly scales. The calculated values of β from the Sen trend and the Z statistic from the MK test for trend assessment are presented in Table 2.

This study undertakes a comprehensive analysis of ET in the YGP across three temporal scales: annual, monthly, and daily. The analysis at the annual scale focuses solely on trend detection, while the analysis at the monthly and daily scales incorporates both trend and season detection.

Interaction detection examines the interplay between two distinct driving factors, evaluating whether the combined influence of X1 and X2 enhances or diminishes the explanatory capacity for Y. It also assesses whether the influences of these factors on Y are mutually unrelated or independent. The comparison involves evaluating q(X1), q(X2), and q(X1 ∩ X2) (Table 3).

In this paper, the annual mean Ec is abbreviated as Eca, the annual mean Es is abbreviated as Esa, and the annual mean Ei is abbreviated as Eia. At the seasonal scale, these three components are abbreviated as Ecs, Ess, and Eis, and at the monthly scale, as Ecm, Esm, and Eim.

The percentages shown in the figure represent the average contribution of Eca, Esa, and Eia. Specifically, Eca accounts for 56% of ETa, Esa accounts for 37% of ETa, and Eia accounts for 7% of ETa. In the YGP, 92% of the region's Eca falls between 50 and 650 mm, 91% of the Esa falls between 50 and 350 mm, and 91% of the Eia falls between 0 and 80 mm.

In the Supplementary Material, Figure S1(X-1) comprises 16 images illustrating the spatial distribution of ETs and their components across four seasons on the YGP. Several notable features are evident: (1) ETs are highest in summer, followed by spring, autumn, and winter on the YGP. (2) Spatially, during spring, Ecs in the southern YP exceeds that in the northeastern GP, while in summer, the opposite is observed. This could be attributed to the differences in the phenological stages of vegetation between the two areas. Additionally, the southern region of YGP experiences more frequent cloud cover and rainfall during the summer season. It is also possible that abundant moisture suppresses transpiration in forest vegetation, leading to relatively lower rates of Ecs.

Figure 2(e) shows the spatial distribution of ETm over 18 years, with ETm predominantly falling within the range of 10–110 mm for each of the 12 months. In July and August, it can be observed that the ETm in the GP is slightly higher than that in the YP, while in other months, the ETm in the YP is generally higher than that in the GP. In the Supplementary Material, Figure S2(X-1) depicts the spatial distribution of Ecm, with the highest classification predominantly in the range 80–101.9 mm in May, 80–119.8 mm in July, and 80–109.1 mm in August. Analysis combining this data with the CLCD dataset indicates that areas with higher ET values are predominantly forested. Supplementary Material, Figure S3(X-1) and Figure S4(X-1), respectively, illustrate the spatial distribution of Esm and Eim. Especially, in the same month, Esm in the city of Kunming is slightly higher than in the surrounding areas. This could be attributed to the urban heat island effect, where the relatively lower vegetation cover in urban areas compared with forested areas, coupled with higher temperatures, leads to the occurrence of this phenomenon.

Figure 2(f) illustrates the average ETm and its components across the entire YGP during an 18-year period, showing that the highest ETm occurs in August at 73.17 mm, and the lowest ETm occurs in December at 21.96 mm. Among the other three components, Ecm and Eim peak in August, with 44.14 and 6.87 mm, respectively, while Esm is highest in June at 31.23 mm. Supplementary Material, Figure S2(X-1) shows that in June, the Ecm in the southern part of the YGP is lower than in May. Conversely, in the northeastern part of the YGP, the Ecm in July is higher than in June. The results show that from May to August, Ecm is lowest in June, higher in May, and highest in August. This could be due to the more pronounced influence of cloud cover and rainfall on vegetation in the southern part of the YGP compared with the northeastern part, which is similar to the reasoning during the analysis of Ecs.

The ETs calculations yielded results shown in the Supplementary Material, Figure S1. Notably, Supplementary Material, Figures S1(a-2) and (m-2) indicate a noticeable increase in ET during spring and winter, mainly in the YP. Conversely, Supplementary Material, Figure S1(e-2) shows a significant decrease in ET across most regions of the YGP in summer. Supplementary Material, Figures S1(b-2) and (n-2) highlight substantial changes in Ec, driving overall ET variations. Particularly, Supplementary Material, Figure S1(c-2) demonstrates a significant decrease in Es in most areas during spring, except for near Dian-Chi Lake and Fu-Xian Lake. The entire YGP experiences increasing ET trends during spring, autumn, and winter by 0.75, 0.19, and 0.80 mm/season, respectively, while experiencing a decreasing trend of 1.53 mm/season in ET during summer. Ec notably increases by 1.47 mm/season during spring, while Es decreases by 1.01 mm/season (Supplementary Material, Figure S5).

Figure 3(b) illustrates ETm calculations, highlighting distinct trends in ET changes. The YP shows significant increases in January, March, November, and December, while the GP experiences a notable increase in October. Conversely, the YP exhibits decreases in July and August, and the GP in September. Mean ET in the YGP varies throughout the months, with increasing trends from January to June and decreasing trends from July to September, followed by increasing trends from October to December. In January, ETm is increasing at a rate of 0.42 mm/year, while in July, it is decreasing at a rate of 1.07 mm/year (Figure 3(d)).

The changes in Ecm, Esm, and Eim are shown in Supplementary Material, Figure S2(X-2), Figure S3(X-2), and Figure S4(X-2), respectively. Ecm shows an increasing trend mainly from January to May and from October to December. However, from June to September, there is a decreasing trend in some areas, with an increase observed in the border region between the YP and GP in June. Esm exhibits an increasing trend in January and a generally decreasing trend from February to December, except for a few noticeable areas with special characteristics such as Kunming in February to April.

In the pixel-by-pixel analysis of ET across the YGP, two time-scales, annual and monthly, were utilized. At the annual scale, TC mutations in the YGP are predominantly clustered between 2007 and 2017, with 2007 being the year with the most mutations in the region. The mutation areas are primarily concentrated in the majority of the YP and the northeastern part of the GP (Supplementary Material, Figure S6(a)). From Supplementary Material, Figure S6(b), it is evident that 72.7% of the regions experienced only one year of TC change per pixel, while 27.2% exhibited two years of change. The estimated slope of the trend is mainly concentrated between −1 and 1 (Supplementary Material, Figure S6c).

In this study, we evaluated the impacts of 15 driving factors. Generally, Rn serves as the primary energy source for ET and holds a key role in its process. NDVI and CLCD determine the type of subsurfaces, and ET varies from different subsurfaces. In addition, various meteorological factors also have an important influence on ET, and the influence of air temperature and precipitation on ET is more obvious.

In the YGP, the spatial variation of ET is collectively influenced by a combination of multiple factors rather than being solely determined by any single factor. The pairwise interactions among these factors provide a more comprehensive explanation for the spatial heterogeneity of ET compared with considering each factor in isolation. It is important to highlight that the interactions among the 15 factors primarily exhibit a bivariate enhancement pattern, except for Tpre and TCC. Of course, there are also a considerable number of factors that show a nonlinear enhanced interaction type. In particular, the three topographic factors basically showed nonlinear enhancement with other factors. Data from Figure 6(e) show that the q value for the interaction of each factor is generally higher than that of the individual factors. Among these, the interaction between Rn and NDVI stands out with the highest q-value of 0.37, followed by the interaction between Rn and Wind with a q-value of 0.35. When comparing the q values of individual factors with those of pairs of factors, the interactions involving Rn, NDVI, and 2Tem exhibit significantly higher q values. This suggests that the combined influence of these three factors plays a substantial role in shaping the spatial distribution of ETa in YGP.

Supplementary Material, Figure S7 illustrates a bivariate enhancement of the two factors across various months, with some exhibiting a nonlinear relationship. Particularly noteworthy is the nonlinear enhancement observed between the three terrain factors and most other factors during different months. Additionally, in January, February, March, October, November, and December, the q values between the two factors demonstrate notably enhanced interactions compared with single-factor q values. This enhancement is particularly pronounced in the interactions between Rn and NDVI and other factors.

The previous section described what are the dominant factors affecting ET in the whole YGP, and this section briefly describes the correlation analyses of ET with each factor. From Figure 8 it can be seen that the correlation between ETa and related factors in the whole YGP is spatially heterogeneous. The correlation between ET and Rn, NDVI, and Tem is basically positive, which is more significant in the central-eastern GP and the southern YP, while the correlation with Pre, RHum, and TCC is basically negative, which is more significant in most parts of the GP and the southern YP.

Supplementary Material, Figure S8 reveals varying correlations between ET and different factors across different months. For instance, while ET and Rn generally exhibit a positive correlation, in March, April, and May, certain regions in the YP show a negative correlation with Rn. ET and NDVI demonstrate a notably positive correlation from January to April, while correlations in other months are less significant. Conversely, correlations between the two temperature products and ET remain consistent, with negative correlations in parts of the YP from February to May, and predominantly positive correlations in other months, except for July and December in the YP where correlations are less pronounced compared with May, and November and December in the GP. Correlations between the two precipitation products and ET are generally negative from April to November, with a few regions showing insignificant or positive correlations. Wind and ET tend to have a negative correlation in most YGP regions from January to May, with insignificant correlations in other months. The correlation pattern between RHum and ET aligns closely with that of precipitation. While the correlation between SPre and ET lacks significance in every month, the correlation between TCC and ET is consistently negative across all months, particularly in the GP.

We researched the spatiotemporal variation patterns of ET in the YGP from 2003 to 2020 and explored the factors causing spatial differentiation in ET. A precise and clear understanding of the patterns and causes of ET changes is crucial for developing strategies to tackle global climate change (Liu et al. 2021; Zhao et al. 2022). In this research, we employed PML-V2 (China) land surface ET datasets to examine the spatiotemporal variations and factors influencing ET at annual, seasonal, and monthly scales. Notably, there are significant differences in the spatiotemporal patterns of ET between different time-scales. According to our findings, both Ec and Ei in the YGP exhibited a significant increasing trend across different time-scales, while Es showed a significant decreasing trend. This is consistent with the global trend of increased ET due to global greening (Li et al. 2020; Shi et al. 2022). This observation could be linked to the extensive afforestation policies that have been implemented in China in recent years. Vegetation restoration tends to promote both Ec and Ei (Tong et al. 2018; Qiao et al. 2021). In the YGP, the ratio of Ec to Es is 8:5. If we include Ei, the proportion of ET related to vegetation in the YGP reaches around 63%, which means vegetation-changes significantly impact ET (Zhang et al. 2016). Ec, especially during the spring and winter seasons, and particularly in the months of January to May and November to December, shows a significant increasing trend. This may be attributed to the early onset of spring phenology due to global warming (Jeong et al. 2009). Alternatively, it could be due to climate change leading to higher temperatures in the YGP during the dry season. This results in a distinct increase in ET during the dry season in the YGP, making the region more arid (Xu et al. 2015; Qiu et al. 2016). Meanwhile, global-warming-induced changes in circulation patterns result in a significant increase in summer precipitation in the southwestern region of China (Jiang et al. 2021). The heightened precipitation leads to an increase in atmospheric humidity over the YGP, consequently suppressing regional ET. This is reflected in a reduction in ET during July and August, aligning with the negative correlation observed in this study between precipitation, relative humidity, and ET during the rainy season, particularly in July and August (Hong et al. 2023). However, the mechanisms influencing the spatiotemporal variation patterns of ET in the YGP are highly complex. While climate change stands out as a primary driver, the specific mechanisms require further investigation.

There is a current research gap in the monthly spatial analysis of ET distribution in the YGP. The GED model has been introduced to enhance the understanding of ETm spatial distribution patterns (Ersi et al. 2022). The results from the GED model indicate that, at the annual scale, Rn has the greatest impact on the spatial distribution patterns of ET, followed by NDVI. Due to the strong correlation between Rn and sunshine duration, our research findings are in good agreement with the conclusions of another study (Zhang et al. 2023). Similar to previous studies, the variation in ET is a result of the combined influence of various factors. The interaction q values between Rn, NDVI, temperature (Tem/2Tem), and precipitation (Pre/Tpre) show a significant enhancement (Chen et al. 2018; Li et al. 2024). It is evident that in regions with favorable water and thermal conditions, vegetation tends to grow vigorously, leading to relatively higher rates of Ec and Ei. The three topographical factors and their interactions with other factors mainly exhibit nonlinear enhancement, indicating that terrain still has a certain influence on the spatial distribution patterns of ET (Ma et al. 2019).

At the monthly scale, it is evident that the influence of various factors on ETm in the YGP varies across different months. Over an extended period, there is a consistent pattern in the impact of different factors on ET: each factor tends to have a higher q value during the dry season and a lower q value during the rainy season, indicating that seasonality significantly influences ET changes and its driving factors (Young et al. 2022). In the dry season on the YGP, Rn, NDVI, and temperature exhibit particularly significant influences on ET, while in the rainy season, no single factor stands out with a high q value for ET. Initially, we hypothesized that there might be a driving factor whose q value for ET would alternate with the q value of Rn for ET over the 12 months. Although this phenomenon has been observed in other study areas (Wang et al. 2023), it did not appear in our research. This phenomenon warrants further investigation. Another noteworthy observation is that, when comparing q values, the response of NDVI to ET lags behind Rn by one month, and the hysteresis effect is consistent with other research results (Li et al. 2023). Furthermore, the q values obtained from the GED model can only indicate the relative importance of certain factors in influencing ET. They do not provide information on whether these factors promote or inhibit ET. Therefore, it is essential to assess the impact of each factor on ET through correlation analysis. As revealed by the results, both temporally and spatially, ET shows significant heterogeneity in its correlation with various factors (Mangan et al. 2023). These characteristics are inherent in the analysis of large-scale ET and are not unique to the specific region under study (Sun et al. 2020; Wang et al. 2023).

This study utilized PML-V2 data, which offers high spatiotemporal resolution, enabling a detailed analysis of ET changes in the spatially heterogeneous YGP. Consequently, the PML-V2 (China) datasets are well suited for analyzing ET in this region. Several scholars have affirmed the accuracy and applicability of PML-V2 ET data in China (Bai et al. 2022). However, comparing different ET products, such as MOD16A2 and GLEAM, can potentially provide a clearer understanding of the spatiotemporal distribution patterns of ET across the entire YGP (Parajuli et al. 2022).

ERA5, GLASS, and other products produced by various scholars were employed in this study. Trend analysis, changepoint analysis, attribution analysis, and correlation analysis were utilized to attribute the changes in ET. However, these methods may entail limitations and uncertainties as follows. Firstly, resampling data to a uniform resolution may cause information loss and uncertainty in the results (Lyons et al. 2018). Future research could address this issue by using meteorological data with higher spatial resolution. Additionally, different ET analysis methods may lead to varying conclusions. For example, while the BEAST algorithm identifies changes in mean ET across different scales, alternative methods like the BFAST algorithm may yield different results, further contributing to uncertainty in the analysis (Li et al. 2022). Furthermore, using GED and Pearson correlation analysis for ET attribution detection may not fully apply to regions with high spatial heterogeneity. Comparative analyses using alternative methods such as multivariate linear regression, the Budyko–Choudhury–Porporato model, or partial correlation analysis could reduce the uncertainty introduced by specific models (Zhang et al. 2020; Guan et al. 2021; Su et al. 2022). ET is a complex process influenced by various factors. The intricate terrain and topography of the YGP, with strong spatial heterogeneity, further complicate understanding the driving mechanisms behind ET changes. This study focused on common climatic factors, topography, vegetation, and land use. In the future, integrating additional data such as meteorology, soil moisture, vegetation structure, and human activity (Liu et al. 2023) could offer a more comprehensive exploration of ET mechanisms. This holistic approach could better clarify the relationship between ET and climate change, essential for water resource management in the face of global climate change.

This article provides a comprehensive analysis of the variation patterns of ET in the YGP from 2003 to 2020 at annual and monthly scales, as well as the mechanisms influenced by climate change across different spatial and temporal scales. The principal conclusions of this study are as follows:

• (1) The ratio of Ec to Es and Ei in ETa on the YGP was 56:37:7. The ETa showed an increasing trend with a magnitude of 0.18 mm/year. ETm exhibited a decreasing trend mainly in July, August, and September, while in other months, it generally showed an increasing trend, which was mainly influenced by Ecm.

• (2) From 2003 to 2020, one breakpoint was detected in the TC of ETa, while four breakpoints and one breakpoint were detected in the TC and SC of ETm, respectively. At the pixel level, the most breakpoints in ETm's TC occurred in January 2007, while the most breakpoints in ETm's SC happened in June 2018.

• (3) In the YGP, the factors with the highest explanatory power for ETa, in descending order, were Rn, NDVI, temperature, wind speed, and precipitation. Additionally, for ETm, the explanatory power of each factor followed the pattern: higher in the dry season (May to October) and lower in the rainy season (November to April of the following year). Rn and NDVI remained the dominant driving factors with strong explanatory power for ETm. Moreover, the interaction among these driving factors typically amplified their influence on the spatial distribution of ET.

• (4) The correlation between ET and various driving factors in the YGP displayed notable spatial heterogeneity. The correlation patterns with each driving factor exhibited significant variations each month, primarily distinguishing between the dry and rainy seasons.

We would like to acknowledge the datasets are provided by the National Tibetan Plateau Data Center, the National Earth System Science Data Center, National Science & Technology Infrastructure of China, the Climate Data Store, and the NASA EARTHDATA website.

This work was supported in part by the National Natural Science Foundation of China under Grant 42230109, in part by the National Natural Science Foundation of China 42301454, in part by the Yunling Scholar Project of the ‘Xingdian Talent Support Program’ of Yunnan Province under Grant KKRC202221002, and in part by the Platform Construction Project of High Level Talent in the Kunming University of Science and Technology (KUST).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.