Tracking Evapotranspiration Patterns on the Yinchuan Plain with Multispectral Remote Sensing

Abstract

Evapotranspiration (ET) is a critical component of the hydrological cycle, and it has a decisive impact on the ecosystem balance in arid and semi-arid regions. The Yinchuan Plain, located in the Gobi of Northwest China, has a strong surface ET, which has a significant impact on the regional water resource cycle. However, there is a current lack of high-resolution evapotranspiration datasets and a substantial amount of time is required for long-time series remote sensing evapotranspiration estimation. In order to assess the ET pattern in this region, we obtained the actual ET (ET_a) of the Yinchuan Plain between 1987 and 2020 using the Google Earth Engine (GEE) platform. Specifically, we used Landsat TM+/OLI remote sensing imagery and the GEE Surface Energy Balance Model (geeSEBAL) to analyze the spatial distribution pattern of ET over different seasons. We then reproduced the interannual variation in ET from 1987 to 2020, and statistically analyzed the distribution patterns and contributions of ET with regard to different land use types. The results show that (1) the daily ET_a of the Yinchuan Plain is the highest in the central lake wetland area in spring, with a maximum value of 4.32 mm day⁻¹; in summer, it is concentrated around the croplands and water bodies, with a maximum value of 6.90 mm day⁻¹; in autumn and winter, it is mainly concentrated around the water bodies and impervious areas, with maximum values of 3.93 and 1.56 mm day⁻¹, respectively. (2) From 1987 to 2020, the ET of the Yinchuan Plain showed an obvious upward and downward trend in some areas with significant land use changes, but the overall ET of the region remained relatively stable without dramatic fluctuations. (3) The ET_a values for different land use types in the Yinchuan Plain region are ranked as follows: water body > cultivated land > impervious > grassland > bare land. Our results showed that geeSEBAL is highly applicable in the Yinchuan Plain area. It allows for the accurate and detailed inversion of ET and has great potential for evaluating long-term ET in data-scarce areas due to its low meteorological sensitivity, which facilitates the study of the regional hydrological cycle and water governance.

1. Introduction

Evapotranspiration (ET) is a complex physical process involving factors including vegetation transpiration, soil, and river and lake evaporation, playing an important role in the hydrological cycle and energy exchange in ecosystems such as the biosphere and the atmosphere [1]. Simultaneously, it is crucial for water resource management, improving water security, and achieving the allocation of water for agricultural crops. Studies have shown that 70% of the precipitation on the earth’s surface is returned to the atmosphere through ET [2]. Especially in agricultural ecosystems of arid and semi-arid zones, ET is an important factor in maintaining water and energy balance. In these areas, the evaporative losses account for more than 80% of the total water consumption [3]. The surface heat flux and water balance are critical factors in achieving water budget closure at the plot scale. Simulating their variations using remote sensing models is an essential method used for accurately quantifying the hydrological components of regional ecosystems [4,5], thereby contributing to a more comprehensive and precise assessment of local evapotranspiration. Remote sensing technology has been used to estimate surface evapotranspiration (ET) since the 1970s [6], effectively monitoring and quantifying water exchange processes [7]. In addition to measuring the daily surface ET using spatial and subsurface parameters, as well as the vegetation and soil information obtained from remote sensing [8,9,10,11], existing remote sensing data can also be utilized to predict ET variations; this approach helps mitigate issues related to the scarcity of observational data [12]. Models for estimating ET based on remote sensing can be mainly classified into three categories: statistical empirical methods, energy residual methods, and full remote sensing models.

Statistical empirical methods are easily affected by vegetation cover and weather conditions [13,14], and the full remote sensing model can achieve good inversion results in some arid areas. However, the inversion of ET in croplands or forest needs to be further researched and validated [15]. The residual method is used to invert the net radiative flux, soil heat flux and sensible heat flux by remote sensing technology to calculate the latent heat flux residual according to the surface energy balance equation, so as to derive the surface ET [16]. The residual method can be generally categorized into two types: single-layer models and double-layer models. The single-layer model does not distinguish the fluxes between soil and vegetation but take that as a unit for calculating the heat flux, and is commonly used to estimate the surface ET, which is strongly influenced by the land surface temperature (Ts). In cases of high vegetation cover, single-layer models exhibit greater accuracy in estimating land surface evapotranspiration. Additionally, the aerodynamic resistance parameters in the model can be determined based on surface characteristics and conventional meteorological observations. Furthermore, this single-layer model simplifies land surface processes and requires fewer parameters, meaning that it has broader applicability. Widely used single-layer models include SEBI (Surface Energy Balance Index) [17], SEBAL (Surface Energy Balance Algorithm for Land) [18,19,20,21,22], SEBS (Surface Energy Balance System) [23,24,25], and Simplified Surface Energy Balance (SSEBop) [26]. These models have different sensitivities to Ts and meteorological inputs. The SEBS model, which is an extension of SEBI, is more suitable for estimating the actual evapotranspiration over heterogeneous surfaces. The SSEBop requires high-quality daily meteorological aggregation data inputs [27,28,29], which has a large impact on the applicability of ET estimation, especially in areas lacking ground meteorological data. On the other hand, the SEBAL model is suitable for estimating the actual evapotranspiration over homogeneous vegetation cover, as it has a lower sensitivity to the meteorological data input and a higher sensitivity to Ts [30,31]. Additionally, the SEBAL model does not incorporate air temperature into the calculation of sensible heat, allowing for fewer parameters and internal adjustments, thus avoiding empirical adjustments to residual aerodynamic resistance. This characteristic enhances its stability in long-term crop evapotranspiration estimation compared to other single-layer models, meaning that it has demonstrated good applicability and reliability in estimating crop ET in the Gansu Plain in China [32], Shahrekord Plain, Chaharmahal and Bakhtiari Province, Iran [33], and Habra Plain in western Algeria [34]. The double-layer models proposed by Shuttleworth [35,36,37] in 1988 not only consider the energy exchange at the surface, but also incorporate vegetation characteristics and soil heat conduction. Compared to single-layer models, this kind of model can more accurately describe the surface fluxes in sparse vegetation areas, integrate geometric optics and consider the effects of variations in sensor observation angles, and avoid empirical corrections for residual impedance. However, the double-layer model is not extensively utilized because its impedance calculations employ empirical methods, making them more sophisticated and vulnerable to geo-graphical limitations.

The Yincha Plain, one of the three major alluvial plains of the Yellow River (including the Yinchuan Plain, Hetao Plain, and North China Plain), is characterized by its flat terrain, fertile soil, and convenient irrigation conditions, which provide significant advantages for local agricultural development. The hydrological cycle is a crucial component of agricultural production. However, the underdeveloped economic conditions and the absence of comprehensive water resource monitoring infrastructure contribute to a lack of high-precision hydrological observation data. Moreover, few studies have applied remote sensing methods to invert ET using high-resolution data series. Meanwhile, due to the complexity of traditional remote sensing methods, surface ET inversions have been conducted in 5-t10-year intervals and without spatial and temporal continuity. To compensate for this issue, we utilized a new tool called geeSEBAL that was developed based on the GEE platform and SEBAL for the estimation of surface ET, using Landsat5(TM+) and Landsat8(OLI) remote sensing images. The state-of-the-art meteorological reanalysis ERA5-Land data were used to obtain the multi-year actual evapotranspiration (ET_a) in the Yinchuan Plain area from 1987 to 2020. The main objective was to analyze the changes in the ET of the Yinchuan Plain area over different seasons, years and land types.

2. Materials and Methods

2.1. Study Area

The Yinchuan Plain is situated in the central part of the Ningxia Hui Autonomous Region, flanked by the Yellow River on both sides. Bounded by the Ordos Plateau to the east, the Helan Mountains to the west, the Loess Plateau to the south and Shizuishan to the north, the study area lies between approximately 37°29′ to 38°53′ N and 105°49′ to 106°53′ E (Figure 1). Although the plain is situated in a temperate arid zone, the annual average precipitation reaches merely 200 mm. The annual water discharge of the Yellow River passing through the region reaches an average of 300 × 10⁸ m³, making it an important industrial and agricultural hub in the middle and upper reaches of the Yellow River, and an important area for the national economic development of the Ningxia Hui Autonomous Region and surrounding areas. The annual evaporation of the Yinchuan Plain is as high as 1600 mm, which has a significant impact on the ecology of the Yinchuan Plain. This area relies heavily on the river runoff supply, which can become an important restricting factor the development of local industries and agriculture [38].

2.2. Data Sources

Since the Yinchuan Plain covers a wide area of 7790 km², it requires high-quality remote sensing images. In this study, some 46 high-quality Landsat5(TM+) and Landsat8(OLI) images from between 1989 and 2020 were selected to cover the whole study area. These images were employed to estimate the ET_a from the surface parameters. The ERA5 Land Reanalysis dataset was used as the meteorological data input for geeSEBAL [39,40]. This dataset contains hourly data on air temperature at 2 m, the dew point temperature, and the eastward and northward wind speed at 10 m. The relative humidity estimation follows Shuttleworth [41], and more detailed information on data input can be found in [42].

The meteorological data were collected from Ningxia Meteorological Bureau (

Table 1. Ningxia Meteorological Bureau Observed Data. Here, 53518, 53519, 53614 and 53615 include a large evaporator (L) and small evaporator (S), both of which are 24 h ET values.

Meteorological Stations	Latitude	Longitude	Altitude(m)	Measured Instruments	Measured Time Series
53518	3904	10635	1126.9	L	1991–2018
				S	1991–2014
53519	3921	10676	1092.2	L	1991–2018
				S	1991–2014
53610	3856	10634	1106.0	S	1991–2014
53614	3847	10621	1111.6	L	1991–2018
				S	1991–2014
53615	3881	10670	1101.0	L	1991–2018
				S	1991–2014
53618	3822	10620	1118.4	S	1991–2014
53619	3812	10630	1115.7	S	1991–2014

) and used to estimate the reference ET based on the FAO–Penman–Monteith method [43]. Due to the lack of flux tower observations, the large and small ET data from meteorological stations were used as geeSEBAL ET validation data. The large evaporation pan was made from a 60 cm wide fiberglass plastic drum buried in the ground, and the evapotranspiration was close to the surface ET_a. The small evaporator (20 cm in diameter) was made from galvanized iron or other alloys. In dry areas or dry seasons, the outer wall of the pan had higher temperatures given its small size, which makes the observed values significantly larger than the real water surface evaporation. Therefore, the obtained values cannot accurately represent the real free water surface evaporation. Likewise, estimating and tracking the actual land surface evaporation from these values become challenging, but they have certain reference value for understanding the changes and trends in water surface evaporation over time [44]. More detail about the evaporator used can be found in the Specification for Surface Meteorological Observations [45].

The China Land Cover Dataset (CLCD), used to calculate the ET for different subsurfaces, is provided by Xin Huang from Wuhan University [46], who produced the annual CLCD based on 335,709 Landsat images from Google Earth Engine. This dataset contains annual land cover information for China from 1990 to 2020. The author elicited spatiotemporal features based on all available Landsat data on GEE using the Random Forest Classifier. The author also proposed a post-processing method including spatiotemporal filtering and logical inference to further improve the consistency of the CLCD. Finally, the overall accuracy of CLCD reaches 80% based on 5463 visual decoding samples, meeting the accuracy requirements for this experiment.

2.3. Methods

2.3.1. The geeSEBAL Algorithm

SEBAL is a commonly used large-scale remote sensing ET assessment model that uses the energy of longwave and shortwave radiation from the sun and the atmosphere to estimate soil and surface-layer heating and ET. It divides the net surface radiative fluxes into three components: soil heat flux, sensible heat flux, and latent heat flux. The energy balance equation is as follows:

$$\lambda E=R_n-H-G$$

(1)

where $\lambda E$ is the latent heat flux (W/m²), $R_n$ is the surface net radiation flux (W/m²) (Equation (2)), $H$ is the sensible heat flux (W/m²) and $G$ is the soil heat flux (W/m²) (Equation (3)).

Net surface radiation (netradiation) is the main source of energy for the process of heat flux transfer and exchange in surface water. Soil heat flux is the state of heat exchange between the surface and deeper soil layers.

$$R_n=(1-\alpha )R_S\downarrow +R_L\downarrow -R_L\uparrow -(1-{\epsilon }_0)R_L\downarrow$$

(2)

$${\displaystyle \frac{G}{R_n}}=\alpha (T_s-273.15)(0.0038\alpha +0.0074{\alpha }^2)(1-0.98NDV{I}^4)$$

(3)

where $\alpha$ is the surface albedo, calculated according to Tasumi and Ke [38,47], $T_s$ is the surface temperature, $R_S\downarrow$ is the incoming shortwave radiation, $R_L\downarrow$ is the incoming longwave radiation, $R_L\uparrow$ is the outgoing longwave radiation, and ${\epsilon }_0$ is the surface thermal emissivity.

The daily ET is computed by upscaling instantaneous $\lambda E$ to $E{T}_a$($\wedge$ is the intermediate variable) (Equation (4)).

$$E{T}_{a24h}={\displaystyle \frac{\wedge R_{n24h}}{\lambda }}$$

(4)

where $E{T}_{a24h}$ is the daily ET (mm); $R_{n24h}$ is the daily average net radiation (W/m2); $\wedge$ is the evaporative fraction; and $\lambda$ is the latent heat of water evaporation (J/Kg).

The evaporative fraction $\wedge$ is the intermediate variable.

$$\wedge ={\displaystyle \frac{\lambda E}{R_n-G}}$$

(5)

where $\wedge$ is the evaporative fraction; $\lambda E$ is the latent heat flux (W/m²); $R_n$ is the net radiation (w/m²); and $G$ is the soil heat flux (w/m²).

The SEBAL algorithm used in this paper was implemented in the GEE platform using the JavaScript language [42]. The GEE platform provides well-established Landsat datasets and meteorological data such as ERA5. ERA5 is the fifth generation of the ECMWF (European Centre for Medium-Range Weather Forecasts) atmospheric reanalysis dataset of the global climate from January 1950 until now. This dataset provides hourly estimates of a large number of atmospheric, terrestrial and oceanic climate variables, which allows geeSEBAL to handle ET on global to regional scales, especially in areas where ground meteorological data are limited. geeSEBAL provides long-term and comprehensive reference ET values. The operation process of this program can be found in the references [42].

It is worth noting that in the process of calculating the actual vapor pressure and relative humidity in the geeSEBAL model, the specific humidity is a crucial variable, but it is not archived directly in the ERA5 dataset in the GEE platform. However, the ERA5 contains the near-surface (2 m from the surface) temperature, dew point temperature, and surface pressure, from which one can calculate the specific humidity at 2 m. The specific humidity at 2 m $q_{sat}$ was calculated according to Equations (6) and (7) (for more information, refer to [48]). In these equations, the dew point temperature $T$ and surface pressure $p$ (which is approximately equal to the pressure at 2 m) are approximated at the height of 2 m.

$$q_{sat}={\displaystyle \frac{\frac{R_{dry}}{R_{vap}}{e}_{sat}(T)}{p-(1-\frac{R_{dry}}{R_{vap}})e_{sat}(T)}}$$

(6)

where $R_{dry}$ and $R_{vap}$ are the gas constants for dry air and water vapor. The constants are defined in [42]. The saturation vapor pressure is expressed with Teten’s formula:

$$e_{sat}(T)=a_1\exp \left\{a_2({\displaystyle \frac{T-T_0}{T-a_3}})\right\}$$

(7)

with the parameters set according to Buck (1981) [49] for saturation over water ($a_1$ = 611.21 Pa, $a_2$ = 17.502 and $a_3$ = 32.19 K) and according to the AERKi formula of Alduchov and Eskridge [50] for saturation over ice ($a_1$ = 611.21 Pa, $a_2$ = 22.587 and $a_3$ = −0.7 K), with $T_0$ = 273.16 K.

2.3.2. Hot and Cold Endmembers: Automated Calibration

There are two important assumptions in the SEBAL model: (i) there is a linear relationship between the geothermal temperature difference dT and the surface temperature Ts (Equation (8)); (ii) there are both cold and hot image elements in the study area, with the H of the cold and hot image elements being 0 and ($R_n-G$), respectively.

$$dT=a+b\cdot T_s$$

(8)

where $a$ and $b$ are the empirical regression coefficients.

The first assumption serves to simplify the problem of solving the linear coefficients of the surface radiant temperature $T_s$; the second assumption, which introduces two cases of ET extremes (0 for the cold image element H, 0 for the hot image element LE), will thus solve for the coefficients of the linear relationship described above, and thus for H over the entire region.

geeSEBAL supports a simplified automated statistical algorithm to select the hot and cold endmembers, which is based on the version of the Calibration using Inverse Modeling at Extreme Conditions (CIMEC) algorithm used in METRIC [51]. The NDVI and T are usually used as the cold and hot endmember candidates, for example, in a semi-arid climate. Here, the cold endmember is selected within a set of candidates with the highest NDVI (5%) and the lowest Ts (20%) percentiles. Conversely, the hot endmember is selected based on the lowest NDVI (10%) and the highest Ts (20%) percentiles [51]. In this study, we performed the selection of hot and cold end members based on the NDVI and Ts of each image according to the thresholds provided in

Table 2. Seasonal images and end member criterion.

Season	Image Data	NDVIcold	Tscold	NDVIhot	Tshot
Spring	LC08_129034_20200306	5%	20%	10%	10%
Summer	LC08_129033_20200813	5%	20%	10%	20%
Autum	LC08_129033_20201016	5%	20%	10%	20%
Winter	LC08_129033_20140117	5%	10%	1%	10%

and

Table 3. Remote sensing images and end member criterion.

Remote Sensing Image	NDVIcold	Tscold	NDVIhot	Tshot
LT05_129033_19870920	5%	20%	10%	20%
LT05_129033_19890824	5%	20%	10%	20%
LT05_129033_19910830	5%	20%	10%	1%
LT05_129033_19920816	5%	20%	10%	20%
LT05_129033_19930616	5%	20%	10%	20%
LT05_129033_19940822	5%	20%	10%	20%
LT05_129033_19960811	5%	20%	10%	1%
LT05_129033_19990820	5%	20%	10%	1%
LT05_129033_20010809	5%	20%	10%	1%
LT05_129033_20020625	5%	20%	10%	1%
LT05_129033_20060908	5%	20%	10%	1%
LT05_129033_20070709	5%	10%	1%	10%
LC08_129034_20140728	5%	20%	10%	20%
LC08_129033_20170906	5%	20%	10%	20%
LC08_129033_20180824	5%	10%	1%	10%
LC08_129034_20190811	5%	20%	10%	20%
LC08_129033_20200813	5%	10%	1%	10%

.

2.3.3. Statistical Analysis

RMSD (Root Mean Square Deviation) and the bias value were used to evaluate the consistency between the ET estimated by geSEBAL and the ET measured by meteorological stations (Equations (9) and (10)). R² is used as the correlation analysis between ET_a and the potential evapotranspiration by PM model to comprehensively evaluate the applicability of geeSEBAL in the Yinchuan Plain.

$$RMSD=\sqrt{{\displaystyle {\sum }_{i=1}^n\frac{{(Y_i-{\stackrel{-}{Y}}_i)}^2}{n}}}$$

(9)

$$Bias={\displaystyle {\sum }_{i=0}^n\frac{(Y_i-{\stackrel{-}{Y}}_i)}{n}}$$

(10)

2.3.4. Mann–Kendall Trend Analysis

The Mann–Kendall (MK) nonparametric test was used in this study to reveal the trends in the ET changes in the Yinchuan Plain area. The MK method has no requirements regarding the distribution of the sample data, which works well for not normally distributed hydrological data [52]. The MK method defines the statistic S as Equation (11):

$$S={\displaystyle {\sum }_{j=1}^{n-1}{\displaystyle {\sum }_{k=j+1}^{n}s(Y_k-Y_j)}}$$

(11)

where $Y_1$, $Y_2$…, $Y_{\mathrm{n}}$ are the time series variables and $n$ is the length of the time series.

$$s(Y_k-Y_j)=\left\{\begin{array}{l}+1,\hspace{1em}{Y}_{\mathrm{k}}>Y_j\\ 0,\hspace{1em}{Y}_k=Y_j\\ -1,\hspace{1em}{Y}_{\mathrm{k}}<Y_j\end{array}\right.$$

(12)

where $Y_j$ and $Y_k$ are the corresponding measurements for the years j and k, respectively, and k > j.

$$Z=\left\{\begin{array}{l}{\displaystyle \frac{S-1}{\sqrt{V(S)}}},\hspace{1em}S>0\\ 0,\hspace{1em}S=0\\ {\displaystyle \frac{S+1}{\sqrt{V(S)}}},\hspace{1em}S<0\end{array}\right.$$

(13)

where $Z$ is a normally distributed statistic and $V(S)$ is the variance. At a given α confidence level, $Z>0$ indicates an upward trend and $Z<0$ indicates a downward trend. The statistics $\left|Z\right|\ge 1.96$ and $\left|Z\right|\ge 2.58$ mean significance at 95% and 99% confidence levels, respectively.

3. Results

3.1. geeSEBAL ET Validation

Four meteorological sites with longer time series and large ET values were selected for validation (2000–2020) (Figure 2). Meanwhile, to ensure the accuracy of the long ET time series, meteorological stations with relatively complete and small ET values from 1990 to 2010 were selected as geeSABAL validation sites (Figure 3).

The comparison of geeSEBAL ET and the observed values (Figure 2) showed moderate-to-high accuracy in the Yinchuan plain area. The average bias was estimated as −0.11 mm day⁻¹ close to 0, ranging from −0.87 mm day⁻¹ to 0.80 mm day⁻¹. The average RMSD was measured as 1.12 mm day⁻¹ without a clear bias pattern, ranging from 0.83 to 1.47 mm day⁻¹, which demonstrates a wide range of applicability.

The comparison of geeSEBAL values with the large evapotranspiration values at some of the meteorological stations, such as meteorological stations 53615 and 53519 distributed along the Yellow River, showed lower R² values. Station 53615 had the highest RMSD and bias at 1.47 mm day⁻¹ and −0.87 mm day⁻¹, respectively, and the lowest R² at 0.12. To further investigate the reasons for the low R² values, we calculated the potential evapotranspiration using the PM model based on local meteorological data (Figure 3) and the result showed a significant improvement in R². Additionally, the R² values for large evapotranspiration and the potential evapotranspiration from the geeSEBAL results are consistent with stations 53518, 53519, and 53614, and station 53615 exhibited the lowest R² for both large and potential evapotranspiration. This indicates that these results are not due to the geeSEBAL model itself, but rather because the ET calculated is influenced by the meteorological or image conditions affecting the actual ET. On the one hand, the Landsat remote sensing images were scarce in dry seasons. On the other hand, the limited number of years with actual large ET values resulted in large time spans between the data. Although an R² was not as high as expected in four meteorological stations when comparing large ET values, the R² showed a higher correlation for most of the stations with small ET values, reaching up to 0.82 (Figure 4). As for stations 53614 and 53615, there was a good consistency between the large and small ETs. Although a weak correlation was detected between the geeSEBAL and observed ET values, the estimated values were still within a reasonable range, with an RMSD of 1.47 mm day⁻¹.

Due to the large span of the Yinchuan Plain and its complex geomorphology, the surface ET results for the remote sensing inversion are uncertain. This paper analyzes the inversion results from the geeSEBAL and PM based on quantitative aspects. The surface ET obtained by inversion based on the geeSEBAL model was analyzed in correlation with the values obtained from the potential ET(ET_p) calculated by the P-M model (Figure 3). Only in site 53619 was a weak correlation with an R² of 0.43 detected. The rest of the stations showed strong correlations, with the R² ranging from 0.62 to 0.79. In summary, the surface ET inversion accuracy on the Yinchuan Plain using the geeSEBAL model was relatively high, indicating that the model has high applicability in arid and semi-arid areas.

Figure 3. Comparison between ET_p and geeSEBAL ET (a–d). Compared with large-scale evapotranspiration, R² has significantly improved, indicating that the model correlation is influenced by external factors.

Figure 3. Comparison between ET_p and geeSEBAL ET (a–d). Compared with large-scale evapotranspiration, R² has significantly improved, indicating that the model correlation is influenced by external factors.

Figure 4. Comparison between small ET and geeSEBAL ET ((a–f) presents different meteorological stations in the Yinchuan Plain).

Figure 4. Comparison between small ET and geeSEBAL ET ((a–f) presents different meteorological stations in the Yinchuan Plain).

3.2. Changing Patterns of Evapotranspiration in Yinchuan Plain

3.2.1. Changing Trend of ET

The geeSEBAL was driven by ERA5 Land reanalysis using the optimized percentiles and the standard percentiles recommended by Allen et al. [48]. Table 2 shows the dates of the images in four seasons in the Yinchuan Plain. From the spatial distribution of the spring ET in Figure 5a, it can be seen that the northern and southern regions have lower ET values in the range of 0–3 mm day⁻¹. The ET is greater at water bodies and around Yinchuan City, with a maximum value of 4.32 mm day⁻¹. This is mainly due to the existence of broad lakes and wetlands around Yinchuan City, which is known as the “International Wetland City”. There are five national wetland parks around this city, with the Yinchuan Plain wetland covering more than 80% of the total area and acting as the main source of spring ET in this region.

Figure 5b shows the spatial distribution of ET on the Yinchuan Plain in summer. The ET values from bare land and impervious areas are relatively lower, ranging between 0 and 4 mm day⁻¹. These lower ET values are mainly due to the presence of impervious areas. From the ET spatial distribution, it can be seen that water bodies and arable lands are the main sources of ET in summer, with the highest value being 6.90 mm day⁻¹.

Figure 5c shows that the ET in the fall is more uniformly distributed. In this case, water bodies have the highest ET, at 3.93 mm day⁻¹. Leaving crop lands fallow has resulted in lower ET values in the range of 1–3 mm day⁻¹. Figure 5d shows that the highest ET of 1.56 mm day⁻¹ occurred in winter. Although this result is affected by the quality of the image, it can be seen that the ET in water bodies and impervious areas is generally higher than that in cultivated land and bare land.

In summary, the seasonal ET variation on the Yinchuan Plain is concentrated in agricultural fields, wetlands, rivers, and lakes during spring and summer, and primarily in water bodies like rivers and lakes during autumn and winter. This ET distribution pattern indicates that local agricultural and water resources are relatively abundant. The local government should focus on protecting and developing the ecological environment with regard to these two aspects to promote the sustainable development of key industries.

To obtain long-term ET trends in the Yinchuan Plain area, a total of 22 representative images were selected. These images, acquired from June to August between 1989 and 2020, offer a complete coverage of the study area (Table 3). The trend in the surface ET was obtained using MK trend analysis (Figure 6).

The overall ET changes on the Yinchuan Plain are mostly concentrated in the western section, impervious areas and along the Yellow River, with confidence intervals exceeding 99%. ET changes in the vegetated areas (arable land and grassland) are not obvious, and the majority of the confidence intervals are lower than 95%. The ET in the western region and some areas along the Yellow River showed an increasing trend. However, in several major residential areas (Shizuishan City, Yinchuan City, Wuzhong City), it showed a decreasing trend. Our results showed that the areas with significant ET alterations have undergone considerable land use changes. Land use changes from bare lands to croplands, wetlands and water bodies resulted in a decreasing ET trend. At the same time, land use conversion from cropland to impervious areas resulted in an increasing ET trend. The trend in ET reflects the land use changes to a certain extent. In order to further explore the distribution of ET in the Yinchuan Plain, the contribution of different land uses to ET values is discussed in detail in Section 3.2.2.

3.2.2. Contribution of Soil Evaporation and Transpiration to ET Change

The variation in evapotranspiration (ET) is influenced by meteorology, the vegetation cover type, and other factors. Additionally, there are significant disparities in the physical and chemical properties of different land use types on the subsurface. These disparities result in diverse responses on the surface to solar radiation. Furthermore, there can be substantial differences in the surface radiation and energy distribution within similar subsurface areas, leading to the non-uniform distribution of ET across the land surface. In order to investigate the ET_a changes on different subsurfaces of the Yinchuan Plain, the land surface ET from 1990 to 2020 was used along with the CLCD land use dataset. The land use types on the Yinchuan Plain could be grouped into five main categories: cropland, grassland, water bodies, bare land and impervious land (Figure 7).

As shown in Figure 8, the daily ET of different land use types is in the following order: water bodies > cropland > impervious > grassland > bare land. The ET of different types showed obvious stratified characteristics and highly consistent trends. The fluctuation amplitude of each line reflects the response of different subsurface types to climatic factors. Among them, the ET of water bodies fluctuates between 5.03 and 7.62 mm day⁻¹, being the most sensitive to the influence of meteorological factors such as temperature, wind speed, and surface radiation, etc. Cultivated land and grassland, as the main land use types on the Yinchuan Plain, show opposite trends, with the ET of cultivated land ranging from 4.16 to 6.36 mm day⁻¹, and the ET of grassland ranging from 2.07 to 3.78 mm day⁻¹. This is mainly due to low transpiration in grasslands. On the Yinchuan Plain, rice paddy is the main agricultural land use type where water is more abundant during the growing season in summer. Increased sunlight hours lead to increased evapotranspiration (ET) values, which are twice as large as those in grasslands. Bare lands and impervious areas are the two land use types with high change rates on the Yinchuan Plain. The conversion of bare lands into built-up and impervious areas has altered ET in this region. In this manner, the amount of ET in impervious areas ranges from 3.28 to 5.27 mm day⁻¹, while it ranges from 0.81 to 2.67 mm day⁻¹ in bare lands.

It should be noted that an anomalous phenomenon was observed, where the ET for impervious surfaces exceeded that of grasslands; to clarify the specific reason for this phenomenon in the study, we compared remote sensing imagery, land use classification data, and ET imagery for the region (Figure 9). Locally, instances exist where the ET of grasslands surpasses that of impervious surfaces, particularly in the eastern irrigation areas where grassland types are more prevalent. However, on a broader scale, the primary distribution of grasslands on the Yinchuan Plain differs from that in the eastern irrigation areas. Most grasslands are situated in the western arid region and are quite sparse. As depicted in Figure 3c, the ET from these grasslands is generally lower compared to the ET from impervious surfaces in the central irrigation area. Additionally, the land use types within impervious areas are heterogeneous, encompassing various vegetation types, wetlands, and water bodies. As illustrated in Figure 10, the CLCD dataset lacks finer classifications, which is another significant reason why the overall ET from impervious surfaces exceeds that from local grasslands.

Human activities often have significant impacts on natural systems and may even determine the evolution of regional microclimates and ecosystems in arid and semi-arid regions. For example, human activities have dominated the evolution of hydro-ecological processes in the lower Tarim River for centuries [53]. The most apparent manifestation of human activities in the Yinchuan Plain region could be regarded as land use conversion. In order to more intuitively respond to the distribution pattern of ET on the Yinchuan Plain and under different land use types, the contribution of ET ($\sigma$) of different types (Figure 9) was calculated using Equation (14).

$${\sigma }_C={\displaystyle \frac{E{T}_C}{E{T}_C+E{T}_G+E{T}_W+E{T}_B+E{T}_I}}$$

(14)

where ${\sigma }_C$ is the contribution of cropland to ET, and $E{T}_C-E{T}_I$ is the evaportranpiration of different land use types.

As Figure 11 showed, before 1996, $\mathsf{\sigma }$ was in the order of cropland > grassland > bare land > water bodies > impervious areas. After 1996, the order changed to cropland > grassland > impervious areas > water body > bare land. Among the different vegetated land use types, croplands contributed between 68% and 78% to the total ET, followed by grassland contributing between 13% and 21%. Areas with sparse vegetation including water bodies, bare lands and impervious areas accounted for less than 10% of the total ET. In summary, croplands are the main source of evapotranspiration on the Yinchuan Plain. This implies that the impact of human activities has become one of the most important factors dominating evapotranspiration in the region. The primary focal points for hydrological cycle conservation and water governance in the future appear to be cropland and vegetation management.

4. Discussion

4.1. The Effects on ET Trends

In this paper, the spatiotemporal variation pattern of evapotranspiration on the Yinchuan Plain from 1987 to 2020 was studied from two aspects: one is the distribution of ET in different seasons, and the other is the trend in ET changes in long-term series. Comparing the summer ET distribution in Figure 5b with the summer variation trend in Figure 8, the maximum difference in ET variation among different land types can reach 5.46 mm day⁻¹, while the maximum difference in ET variation among the same land type is only 2.59 mm day⁻¹. This indicates that the variation in surface ET is greatly influenced by different underlying surfaces, and that human activities are one of the important factors affecting regional ET variation. Studies have shown that meteorological factors such as temperature and rainfall also have an important impact on ET. Xiaomei Jin et al. [54] analyzed the impact of the monthly average temperature and rainfall on ET on the Yinchuan Plain using actual data, and found an R² of 0.82 between ET and temperature, while it was only 0.38 between ET and rainfall. At the same time, meteorological factors such as the sunshine hours, wind speed, and relative humidity also affect surface evapotranspiration to varying degrees [55,56]. For example, Wang et al. [57] found that a decrease in the local sunshine hours in the Yangtze River basin would reduce the energy source of surface evapotranspiration, thereby weakening the process of surface evapotranspiration. In summary, it can be seen that the factors affecting surface ET changes are complex and diverse. However, land use types cannot fully explain the changes in ET, and the influence of climate factors is also an important factor.

Generally speaking, climate factors are a prerequisite for surface energy supply, determining the maximum actual surface evapotranspiration. However, in arid and semi-arid environments, vegetation canopy serves as an interceptor of energy flux and hence determines the range of surface evapotranspiration [58]. As the main vegetation and land use type on the Yinchuan Plain, cropland accounts for about 60% of the total area. In Figure 1, the vegetation cover of the Yinchuan Plain during the growing season (June to August) is shown. In Figure 6, it can be seen that in the time series studied, the evapotranspiration changed only significantly in a small number of areas, such as bare land conversions into farmlands, wetlands, or water bodies. Most areas did not show significant changes in evapotranspiration within the study period, such as farmland. This is consistent with the findings of Xiaomei Jin et al. in their study of the Yinchuan Plain from 2001 to 2014. This indicates that the trend in ET changes in the Yinchuan Plain area is mainly dominated by changes in cropland and land cover types, with meteorological factors being secondary influences. Through studying the changes in ET_a, it is possible to better understand the water consumption of vegetation and land, adjust the allocation of agricultural water resources in a timely manner (such as irrigation) and improve the efficiency of freshwater resource utilization (avoiding inefficient water use or irrigation water shortages). Meanwhile, the attribution analysis of surface evapotranspiration changes on the Yinchuan Plain will be the focus of subsequent research, clarifying the direction of water resource management and sustainable development in the region.

4.2. The Uncertainty Analysis of geeSEBAL

Because of the impact of observational conditions, the validation data gathered from weather stations were not continuous, and there were missing measurements for some study dates and years. Therefore, the large evapotranspiration values from four stations with a relatively complete dataset and the small evapotranspiration values from six stations were selected for validation. Based on our findings, the estimation results of the geeSEBAL were within a reasonable margin of error. Even if there were some stations with weak correlations, these would have been due to the closure mode of the surface energy [42]. According to Laipelt [42], the main challenge faced by the geeSEBAL program is the selection of end members and the definition of a spatial scope. The geeSEBAL model is highly sensitive to the spatial scope, and the calibration function of cold and hot end members can significantly improve the accuracy of ET estimation [59]. This study upgraded the estimation method of ET from single-image to batch estimation in order to achieve high consistency in ET inversion over a spatial range and time span (Figure 12). Therefore, the same calibration end members were set for multiple images on the same date. This affected the estimation accuracy in some areas to some extent, but the overall ET estimation showed satisfactory results (Figure 2). It can also be seen that geeSEBAL has a great potential for estimating ET over small areas, and adjusting the calibration end members based on the actual situation to achieve a higher estimation accuracy.

Due to the scarcity of high-quality remote sensing images in Google Earth Engine and the lack of observation data from meteorological stations, it was difficult to extrapolate ET to both monthly and annual time scales. Therefore, the ET trend analysis for long time series only considered the estimated ET representing dates as an example. By the same token, we conducted no further analysis of the sensitivity and contribution of meteorological factors to ET. This can be used in future research to fully comprehend the pattern of changes in evapotranspiration on the Yinchuan Plain.

Unlike previous studies on ET on the Yinchuan Plain by Wang Zhuoyue et al. [60], the GEE online dataset in this study can be used to quickly obtain continuous high-spatiotemporal-resolution ET values around the world. While the accuracy of meteorological datasets can impact the geeSEBAL model, this problem can be mitigated by using the automatic calibration option of endpoints or by restricting the study area. This ensures that the ET distribution maps generated fulfill the necessary criteria. The ET values for different land use types can also be quickly calculated based on the CLCD, avoiding the cumbersome image processing required in the traditional remote sensing techniques to estimate evapotranspiration. This greatly reduces the inversion time for regional evapotranspiration, and fully leverages the advantages of remote sensing technology in regional water resource management.

5. Conclusions

This study was conducted on the Yinchuan Plain, the irrigation area of the Yellow River Basin. The open-source SEBAL framework was used to verify the applicability of the model to the Yinchuan Plain through the Application Programming Interface (API) to estimate ET and to evaluate its distribution pattern across different underlying surfaces.

In spring, the daily ET of the Yinchuan Plain was found to be mainly concentrated in the central lake wetland area, with a maximum value of 4.32 mm day⁻¹. In summer, the daily ET was found to be concentrated in the cultivated land and water areas, with a maximum value of 6.90 mm day⁻¹. The ET values were highest in autumn and winter in water bodies and impervious areas, with maximum values of 3.93 mm day⁻¹ and 1.56 mm day⁻¹, respectively.

From 1987 to 2020, the ET of the Yinchuan Plain showed an obvious upward and downward trend in some areas with significant land use changes, but the overall ET of the region remained relatively stable without dramatic fluctuations.

The evapotranspiration of different underlying land use types on the Yinchuan Plain region is as follows: water body > cultivated land > impervious areas > grassland > bare land. The evapotranspiration of water bodies ranges from 5.03 to 7.62 mm day⁻¹, that of cropland ranges from 4.16 to 6.36 mm day⁻¹, that of impervious areas ranges from 3.28 to 5.27 mm day⁻¹, that of grasslands ranges from 2.07 to 3.78 mm day⁻¹, and that of bare land ranges from 0.81 to 2.67 mm day⁻¹. Thid study shows that the distribution pattern of regional evapotranspiration is not only influenced by climate conditions, but it is also closely related to human activities.