Spatiotemporal patterns of evapotranspiration in the Xiangjiang River Basin from 2000 to 2020

Evapotranspiration (ET) is defined as the actual amount of water that is separated from land surfaces including vegetation and the ground due to the process of evaporation–transpiration (Venturini et al. 2012). While ET plays an important role in the hydrological cycle and the water and energy balance in land surface–atmosphere systems (Batra et al. 2006), it becomes an indicator of global climate change, meteorological factor prediction and agricultural irrigation management (Huo et al. 2013). Therefore, it is of great significance to better understand mechanisms in spatiotemporal patterns of ET for the management of water resources and the restoration of the ecological environment.

Before the rise of remote sensing technology, the conventional methods of estimating ET are based on meteorological daily data, which are limited to small regions or specific land use for short intervals (Brutsaert & Stricker 1979; Granger & Gray 1990). In the last four decades, several models have been developed for measuring ET and analyzing its distribution characteristics in larger spatial and temporal scales. These methods can be categorized as follows: (1) empirical methods that use site relationships among different factors (Priestley & Taylor 1972; Granger & Gray 1989). (2) Indirect methods that are usually physically based and involve several factors (Boone et al. 2000). (3) Residual methods that measure ET as the residual of the surface energy balance, including Surface Energy Balance Algorithm for Land (SEBAL) (Bastiaanssen et al. 1998) and Surface Energy Balance System (SEBS) (Su 1988). Moreover, a method based on the new parameter F, which is a dimensionless coefficient varying from 0 to 1 and can be approximated as the ratio of ET to potential ET, was proposed to compute ET with remotely sensed data and no site-specific relationships (Venturini et al. 2008).

In recent years, though many researchers focus on estimating ET through observations and performing satisfied precision, ET studies that are based on limited observation sites have limitations when the research area is larger than a field or the study period is over several years. Thus, deriving ET from the remote sensing data has gradually become a main direction. Since various remote sensing data have different spatial and temporal resolutions, studies based on remotely sensed analysis have examined the spatiotemporal patterns of ET on different scales, ranging from continent to regional levels. On the continent scale, the products of MODIS (Mu et al. 2007) are widely used in the research of ET, such as East Asia (Hwang & Choi 2013), conterminous United States (Velpuri et al. 2013) and Siberia (Shi et al. 2022). Some experts use other remote sensing data for ET mapping like Landsat (Yang et al. 2017; Tan et al. 2019) or ERA5 (Fan et al. 2022; Li et al. 2022b). However, many studies did not consider the annual and seasonal change of ET, which is more available on the regional scale. MODIS data are often employed in studying the pattern of spatial–temporal change on ET in smaller regions, including the US Midwest (Basso et al. 2021), North China Plain (Mo et al. 2011), Poyang Lake Basin (Wu et al. 2013), Liaohe River Delta wetland (Liu et al. 2020) and agro-pastoral ecotone in Northwest China (Li et al. 2019). Currently, studies in China mainly focus on river basins of arid and semiarid climates, while studies on the area of subtropical monsoon climate are relatively insufficient. The Xiangjiang River Basin within Hunan Province of China (XRB) represents a region with a typical East Asian subtropical humid monsoon climate. While increasing trends in the summer extreme rainfall and streamflow can be observed in the upper and middle XRB (Du et al. 2019), some researchers estimated that water resources and green water storage (mainly composed of ET) in the XRB will decrease by applying a distributed hydrological model (Feng et al. 2021). Therefore, understanding the characteristics of ET changes over a long period is helpful for water assessments in the XRB and other regions with similar climates.

Trend analysis is significant to understand mechanisms of spatiotemporal variation in ET in current research. The Mann–Kendall test, simple linear regression and multivariate regression are widely applied for trend analysis in long time series with observed data (Shan et al. 2015; Zhang et al. 2015), remote-sensed image (Yang et al. 2017; Wang et al. 2022) or assimilation data (Lu et al. 2019; Dang et al. 2020; Zhang et al. 2020). The research on factors that have impacts on ET spatiotemporal change is also abundant. Multiple lines of evidence indicate that ratio of transpiration (T) and total ET has increased under climate change (Wang et al. 2013). The soil moisture fluctuation signals, which derive from soil water storage through the calculation of the integral, can be used to indicate ET due to the interaction between soil and the atmosphere system (Wang et al. 2021). Some experts investigate the divergence of potential and actual ET (ET_r) by exploring the influence of meteorological, hydrological and botanical factors on ET (Liu et al. 2021). Although these studies have disclosed many features of spatial and temporal variation in ET, there are several problems that need to be further investigated: (1) how to map the pattern of spatial–temporal change in ET and (2) how to measure the effects of natural and human factors.

The XRB is characterized by a subtropical monsoon climate of East Asia, and the average annual per capita water resources has resulted in the value of 2,158 m³ which is relatively high in China. Nevertheless, urban expansion has directly resulted in negative impacts on maintaining natural and artificial wetlands (Zhang & Zeng 2018). Thus, the worsening ecological environment is the main threat to the sustainable development of the XRB (Peng et al. 2022). Moreover, against the background of global warming, extreme weather events such as high temperatures, torrential rain, floods, and droughts have shown an increasing trend in the XRB (Luo et al. 2008).

We used MOD16 ET data obtained from NASA (https://earthdata.nasa.gov/) to analyze the spatiotemporal patterns of ET. MOD16 ET datasets are available at 500 m grid cells and mask out land surfaces including water bodies, built-up areas and barren lands. According to the research area, we used 500 m, annual datasets to analyze the whole trend and driving factors of ET over the period of 2000–2020. Then we accessed 500 m, 8-day datasets to discuss the seasonal spatiotemporal pattern of ET. To validate remote sensing ET, the statistical data of the Hunan Provincial Water Resources Bulletin were applied to obtain the actual ET (E) based on the water balance formula (see Section 3.7). The MOD16 ET data were processed to map the annual ET distribution of the XRB in the past 20 years and then compared with E calculated by the water balance equation (Table 1).

The Mean Absolute Error (MAE) was 137 mm and the Mean Relative Error (MRE) was 21%. The relative errors in some years are higher than the Mean Relative Error (MRE). These errors are mainly caused by the occurrence of extreme weather, which increases the error in ET that is estimated by the water balance equation but not due to the remote sensing data itself (Yang et al. 2015). Hence, the accuracy of MOD16 ET data is acceptable for this study on the XRB.

Datasets of meteorological stations were accessed from the China Meteorological Data Sharing Service System (http://cdc.cma.gov.cn/). The data series were dated from 2000 to 2020, and 25 observation stations were extracted to cover the XRB. The observed air temperature and precipitation were used to detect the potential driving forces of the ET spatiotemporal changes in the XRB. This study removed the outliers and replaced them with the values of adjacent years. Through the spatial interpolation method, the temperature and precipitation are converted to raster data with a spatial resolution of 500 m, which is convenient for correlation analysis with MOD16 ET data.

We obtained the remote sensing land cover data of Globeland30 (http://www.globeland30.org/) with 30 m pixels in 2000 and 2020. Details of land cover type information are below (Table 2).

STRMDEM data (http://www.gscloud.cn/) with a pixel scale of 90 m are used for generating the boundary and water system of the XRB in ArcGIS.

This study used the kriging method and inverse distance weighting (IDW) method to interpolate meteorological station data. The kriging interpolation method can better reflect the changes of variables in large areas and has been widely used in the interpolation of temperature and elevation, but it also has certain limitations (Li et al. 2013). In our study area, the terrain characteristics are complicated for a river basin with a relatively large area. However, when the elevation difference between the mountain and the plain is too large, we find that the kriging method will show poor accuracy after spatial interpolation with limited observations (24 national meteorology sites for our study). Instead, the IDW method is more suitable for precipitation interpolation in areas with fewer meteorological stations (He et al. 2005). Thus, the IDW method was applied for spatial interpolation analysis in our study.

For the water balance of the catchment, there is a relationship between the water balance parameters, which can be defined as the water balance equation. The decisive input variable is precipitation, which is transformed in the hydrological system into the output variable evaporation, runoff and soil water. Since variables other than precipitation are often difficult to measure directly, changes in soil water can be ignored in the annual scale (Everson 2001).

The overall flow direction of the XRB is from south to north. The south of the XRB is higher in elevation, and the woodland and shrubs account for a large proportion. By contrast, the north of the XRB has a lower elevation and is greatly affected by the CZT urban agglomeration, which is the engine of economic development in Hunan Province. As a result, more land construction leads to lower annual ET in the north of the basin. Similarly, the central part of the basin centered on Hengyang, the second largest city in Hunan, also showed low values of ET. Due to the higher vegetation coverage, the eastern part of the XRB performs a high ET value.

In addition, the outliers of ET in the forest were the most, indicating that the degree of dispersion of the data was large. The fluctuation of artificial surfaces was the smallest from 530 to 1,028 mm, and the outliers were the least, indicating that the degree of dispersion is relatively small. From the maximum value on each pixel, the ET of different land covers in 2000 was forest land > agricultural land > grassland > artificial surface in descending order. From the mean value, the ET under different land covers was grassland > artificial surface > forest land > agricultural land in descending order.

In 2020, the fluctuation range of forest ET is still the largest, which is between 495 and 1,509 mm (Figure 6(b)). The range of artificial surface ET is still the smallest, which is between 477 and 1,438 mm. These are both increased comparably to that of 21 years ago. From the maximum value and the mean value in 2020, the ET of each land cover is in the order of forest land > grassland > agricultural land > artificial surface. The ET of forest, grassland, and agricultural land increased by 144, 90, and 51 mm, respectively, while the ET of artificial surfaces decreased by 82 mm (Table 3).

In addition, since the MOD16 data define ET in some built-ups and waterbodies as null value, these areas are not considered (instead of setting them as 0 mm) in spatial analysis for the study, and thus, ET of the XRB is generally high in statistics.

From 2000 to 2011, the RCR of ET in 2003 and 2009 is significantly high, which was mainly related to the occurrence of drought. In fact, 2003 was the year when Hunan Province was hit hard by natural disasters. Compared with insect disasters and floods, the drought was the most serious and the degree of drought was rare (http://tjj.hunan.gov.cn/tjfx/jmxx/2004tjxx/201507/). As a consequence, with more solar radiation and less cloud and precipitation, the ET in the XRB increased significantly when compared with previous years.

From 2012 to 2020, the ET in other years was above the annual mean ET, except for ET in 2014, which was lower than the annual mean ET. Accordingly, it is worth considering that climate change is also a potential factor affecting ET in the basin. The Hunan Climate Change Monitoring Bulletin announced ‘The average temperature in Hunan Province increased by 0.16 °C every ten years from 1961 to 2014. Especially in 2014, more than 100 extreme heavy precipitation events occurred in Hunan.’ Evidence is mounting that climate change will affect small- to medium-scale regions. Thus, the relatively low ET value in 2014 may be related to the extreme weather under climate change.

The overall characteristics are as follows. In 2000, the migration magnitude of the ET center was the smallest in the north–south direction (26.66°N–26.72°N). In 2010, the migration path of the seasonal ET center shifted to the NW, most notably in summer (26.73°N–26.78°N). In 2020, the migration path has the largest north–south span (26.64°N–26.74°N), while the east-west span is the smallest (112.63°E–112.66°E). Compared with the same period in 2000 and 2010, the center of ET in the XRB obviously migrates north in spring and summer and migrates south in autumn and winter in 2020. Moreover, the magnitudes of ET center migration were significantly reduced along the latitude.

In the study period, T showed a decreasing trend from SE to NW, which corresponded with the decreasing trend of solar radiation from low latitude to high latitude. The high-value area of T is distributed in the south, and the low-value area is located in the NW. Since the summer monsoon is blocked by the Nanling Mountains, P decreases in the north of the basin, and the area with a low P value extended from NW to the south. The center and south of the basin are higher in P (greater than 1,500 mm). On the contrary, the NW is a low-value area in P (between 1,344 and 1,395 mm). Thus, T and P in the northwest are lower, and the ET is also low in these areas.

In 2020, the scatter density is smaller and the scatter is closer to the 1:1 line than that of other years. Since the range of T is small, a scale factor of 100 is used to enlarge it. The correlation scatter between T and ET is mainly concentrated around 18 °C. T and ET perform greater correlation in the range of 17.5–18.5 °C. Moreover, the change in the scatter density distribution indirectly shows an increase in both P and T from 2000 to 2020.

In terms of temporal changes, the wind direction is not obvious in March, and the wind field is in a relatively stable state. In June, the summer monsoon significantly turns southerly, so that the rainfall regions are pushed northward. In September, the wind direction was shifted again to form a northerly wind, and the wind field became stable. In December, the northerly wind speed increased, and the rainfall region showed an obvious southward shift compared to that of summer. These spatiotemporal changes of precipitation are consistent with the ET change in 2020.

Comparing the migration paths of ET center at 10-year intervals from 2000 to 2020, the migration mainly occurred in the middle of the XRB. On the whole trend, starting from spring, the three centers of ET move to the NE–SE–SW direction with the change of seasons. ET center with the highest latitude appeared in the summer of 2010, and the one with the lowest latitude appeared in the winter of 2020. On annual changes from 2000 to 2020, the spring ET center moves along the SW–NE direction, and the moving distance is larger in 2010–2020 than in 2000–2010. Summer ET center may shift along the NW–SW direction. Autumn ET center moves in the NW–SE direction. Then all migrated to the SE in winter. Apparently, all the ET center of 2020 migrated to the south of XRB in each season compared with 2000, except for spring. This finding verifies the previous conjecture.

This study analyzed the spatiotemporal patterns of ET in the Xiangjiang River Basin from 2000 to 2020 based on a long sequence of MOD16 ET data. The effects of monsoon and urban expansion in ET are also examined. The main conclusions are summarized below.

1. From 2000 to 2020, the overall spatial–temporal pattern of ET shows a spatial trend of increase in the forest, grassland and agricultural land, while the ET of artificial surfaces decreased. ET in the XRB also demonstrates an overall trend of obvious increase based on the change rate, especially after 2012 on the annual analysis.

2. According to the migration route of the ET center in 21 years, the center of seasonal ET all migrated in a clockwise direction each year, forming a circular trajectory with a NE–SW direction on its long axis. The magnitudes of ET center migration were significantly reduced along the latitude in 2020.

3. For correlation analysis on meteorological factors, T and P in the northwest show a low value and the ET is also low in these areas. The correlation between P and ET from 2000 to 2020 is more obvious than that between T and ET. P and ET show higher correlation when P is around 1,500 mm. T and ET perform a greater correlation with T in the range of 17.5–18.5 °C.

4. Through the monthly wind field of East Asia in 2020, the spring and summer monsoons become prevailing in February and August, respectively. With the direct effects of monsoon on precipitation, the spatiotemporal changes of precipitation are consistent with those of ET in 2020. Moreover, while vegetation cover declined under urbanization, the seasonal ET center shows a southward migration in 21 years, except for spring.

Since the zonal vegetation in the XRB is mainly subtropical evergreen broad-leaved forest, the changes are not obvious throughout the year, so the impact of vegetation on ET is not further analyzed here. With the wind field changing in East Asia under global warming and the promotion of urbanization in China, spatial and temporal distribution on ET will change correspondingly in the XRB. That is why the seasonal ET center of spring in 2020 should be investigated. The study provided insights for mapping spatiotemporal patterns of ET in the XRB, which are useful for monitoring drought, estimating water use, and indicating climate change.

The authors give special thanks to editors and teachers whose comments helped improve the paper.

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by S.L., C.F. and L.Y. The first draft of the manuscript was written by S.L. and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

This study was financially supported by the National Natural Science Foundation of China (Grant Nos. 42001024 and 41901026), the Natural Science Foundation of Hunan Province, China (Grant Nos. 2021JJ40011 and 2022JJ40015) and the Scientific Research Project of Hunan Provincial Department of Education, China (Grant Nos. 21B0625 and 21B0646).

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.