Estimation of Groundwater Evapotranspiration Using Diurnal GroundwaterLevel FluctuationsunderThreeVegetationCovers at the Hinterland of the Badain Jaran Desert

Accurate estimation of groundwater evapotranspiration (ETG) is the key for regional water budget balance and ecosystem restoration research in hyper-arid regions. Methods that use diurnal groundwater level (GWL) fluctuations have been applied to various ecosystems, especially in arid or semi-arid environments. In this study, groundwater monitoring devices were deployed in ten lake basins at the hinterland of the Badain Jaran Desert, and the White method was used to estimate the ETG of these sites under three main vegetation covers. /e results showed that regular diurnal fluctuations in GWL occurred only at sites with vegetation coverage and that vegetation types and their growth status were the direct causes of this phenomenon. On a seasonal scale, the amplitudes of diurnal GWL fluctuations are related to vegetation phenology, and air temperature is an important factor controlling phenological amplitude differences. /e estimation results using the White method revealed that the ETG rates varied among the observation sites with different vegetation types, and the months with the highest ETG rates were also different among the sites. Overall, ETG was 600∼900mm at observation sites with Phragmites australis during a growing season (roughly early May to late October), 600∼650mm in areas with Achnatherum splendens, and 500∼650mm in areas with Nitraria tangutorum and Achnatherum splendens. Depth to water table and potential evapotranspiration jointly control the ETG rates, while the influence of these two factors varied, depending on the specific vegetation conditions of each site. /is study elucidated the relationship between diurnal GWL fluctuations and vegetation in desert groundwater-recharged lake basins and expanded the application of the White method, providing a new basis for the calculation and simulation of regional water balance.


Introduction
In arid and semi-arid regions where precipitation is scarce, most vegetation depends on groundwater for survival. Estimation of groundwater evapotranspiration (ET G ) is an important component of regional water-balance studies. Vegetation growth is usually closely related to groundwater via complex feedback mechanisms [1], and the spatiotemporal variations of vegetation are largely determined by groundwater availability [2,3]. In the previous research, the depth to water table (DTWT) determined the spatial distribution of riverside vegetation [4], and evapotranspiration was highly correlated with the spatial distribution of vegetation [5] in arid and semi-arid regions. us, in hyper-arid desert ecosystems, it is critical to fully understand the interactions between groundwater and vegetation and accurately estimate the amount of groundwater consumed by vegetation for the management of regional groundwater resources.
Conventional ET G calculating methods (e.g., the eddy covariance method, lysimeter method, Penman model, and remote-sensing inversion model) either cannot directly determine ET G or demand high research cost to cover in large-scale study regions. Moreover, some observational methods may be too complex to carry out under harsh field conditions. In areas with a shallow DTWT, diurnal groundwater level (GWL) fluctuations can usually be observed, which is attributed to regular water consumption by phreatophytes when other factors are negligible [6,7]. ET G can be estimated from diurnal GWL fluctuations, which was first proposed by White [6]. is estimation approach is thus referred to as the White method and has since been frequently used for evapotranspiration calculation. Compared with other methods, ET G calculation methods such as the highly complex and costly eddy covariance method, as well as the assumption of land surface homogeneity [8], the White method has the advantages of being cost-effective, relatively simple, and applicable to long-term continuous observations [9][10][11].
ese characteristics highlight the practicality of the White method for ET G estimation, and thus it has been continually developed and revised since its proposal [12][13][14][15]. Currently, these methods have been applied to various ecosystems, such as wetland environments [16,17] and riverside oases in arid and semi-arid regions [11,18,19]. In the hyper-arid hinterland of the Badain Jaran Desert, vegetation growing in groundwater-recharged lake basins depends on groundwater for survival. us, the White method could be used to estimate ET G in this hyper-arid desert lake ecosystem.
Accurate ET G estimation using the White method is based on an understanding of groundwater dynamics and its relationship with vegetation. e information extracted from diurnal GWL fluctuations is also used to study the interaction between groundwater and vegetation [10,20]. For example, Engel et al. [21] observed diurnal GWL fluctuations in wooded areas during the growing season, but the neighboring grassland did not exhibit this phenomenon during the period. De Castro Ochoa and Reinoso [22] found that elevated temperatures caused an increase in the vegetation transpiration rate. When rising temperatures reach a critical point, transpiration ceased due to leaf stomatal closure, and these changes were reflected in daily GWL fluctuations. Another study revealed that vegetation types, meteorological conditions, and soil properties jointly determine the magnitude of diurnal GWL fluctuations [7]. Overall, the existing research highlights the close relationship between phreatophytes and their surrounding environment, but the actual field observation studies remain limited [23,24]. erefore, the dynamic characteristics of groundwater and its relationship with external conditions require further study. Moreover, the analysis of major factors affecting the ET G rate would help elucidate the complex relationship between groundwater and vegetation. e formation mechanism of the 110 permanent lakes at the hinterland of the Badain Jaran Desert in northwestern China is controversial, and the uncertainty of water consumption in the lake basin is the primary cause of this dispute. e lake basins in the desert hinterland have a considerable area with shallow groundwater, where vegetation flourishes during approximately half of the year. erefore, ET G cannot be neglected and is a critical component of the water balance calculation in the region. However, the desert hinterland presents an adverse environment for field workers, hindering the long-term research. Most monitoring methods can only be performed at a single site for in situ observations or in a small-scale area, which hinders the effective monitoring of the vegetation-driven spatiotemporal variations in water requirements. Furthermore, water resource management usually requires the prediction of future water demands based on the current ecosystem water demands under the intervention of human activities. erefore, the daily dynamic characteristics of GWL and the relationship between vegetation and groundwater dynamics must be clarified first, after which a simple and practical estimation approach to calculate the water requirements of oasis vegetation in the desert hinterland can be developed. Such an approach would support the rational usage of water resources and provide scientific evidence for the aforementioned unsolved problems. erefore, this study sought to monitor the shallow groundwater in the desert hinterland-recharged lake basins of the hyper-arid Badain Jaran Desert to estimate the ET G .

Study Site.
e Badain Jaran Desert (39°04′15″∼42°12′23″N, 99°23′18″∼104°34′02″E) is located in the Alashan Plateau, western Mongolian Autonomous Region, China. It is roughly to the south of the ancient Juyan and Guaizi Lakes, north of the Heli and Beida Mountains, west of the Yabrai and Zong Nai Mountains, and east of the Gurinai Plain, with an area of approximately 52,200 km 2 [25]. e study area lies within the northwestern marginal region of the East Asian summer monsoon with a continental climate. e summer and winter mean daily temperatures are 25.3°C and − 9.1°C, respectively [26]. e mean annual precipitation is ∼100 mm, which is mainly concentrated in May to September and exhibits large interannual variability [27]. e mega-dunes and lakes in the desert are interdependent, and there are 110 perennial groundwater-recharged lakes, most with an area <1 km 2 [28]. e groundwater recharge process mainly occurs through cretaceous and tertiary sandstone, and the space that allows for shallow groundwater circulation is dominated by the quaternary gravel, fine sand, and fine silty sand [29].
e Badain Jaran Desert has closed freshwater lakes, saltwater lakes, and salt/brine lakes classified by the total dissolved solids content in the water. e vegetation landscape of the lake basin is characterized by ring zone distributions around the water [30]. Waterfront regions are swampy meadows, with a groundwater depth <0.5 m, and short and dense vegetation including species such as Triglochin maritima and Glaux maritima. e second belt around the water is mostly saline meadow, with a groundwater depth of ∼1 m and vegetation featuring Achnatherum splendens, Phragmites australis, and Glycyrrhiza uralensis. e outer belt has a groundwater depth of ∼2 m and vegetation cover comprising Nitraria tangutorum and Artemisia salsoloides. e outermost edge of vegetation is distributed among fixed and semifixed sand dunes, connected to quicksand. e lake ecosystem in the Badain Jaran Desert, with minimal human activities, is ideal for the study of ecohydrological processes in hyper-arid areas. e research team has established ten GWL monitoring sites in the desert lake basins since 2010 ( Figure 1); the present study represents the first analysis of the data from these sites.

Observation and Data
Processing of GWL and Meteorological Parameters. Ten groundwater observation wells were established in different lake basins with shallow groundwater, which were constructed with PVC screens with a diameter of 8 cm. Hourly GWL was measured with a pressure transducer (Solinst 3001; Solinst Canada Ltd., Georgetown, ON, Canada), which had a measurement accuracy of 0.1 cm, clock accuracy of ±1 min/year, and work temperature range of − 10°C to 40°C. e DTWT was 0∼2 m at the observation wells, and the transducers were fixed at ∼30 cm below the water table. e transducer measured both the total pressure of the water column above the probe and the atmospheric pressure, which was revised using a Barologger barometer (Solinst). e daily changes in GWL in the observation wells were calculated during the growing season, which was defined as the difference between the maximum and minimum values within a day (amplitude in cm). Because missing data could affect the calculation, data were only considered if more than 90% of the normal daily total amount of hourly data were collected; otherwise, the amplitude for that day was excluded from the analysis. Before groundwater series were applied to the White method, a median smoothing filter in MATLAB was applied to eliminate noise.
To obtain meteorological parameters, a weather station (MAWS-301; Vaisala, Vantaa, Finland) was established on the flat ground between Sumujilin South and North lakes at the hinterland of the Badain Jaran Desert. A QMH102 sensor (Vaisala) was used to observe temperature (°C) and relative humidity (RH, %) with an observation interval of 10 s and an output interval of 10 min. An NR01 net radiation sensor (Hukseflux ermal Sensors, Delft, the Netherlands) was used to observe radiation (Rg, W/m), with output every 30 min. Precipitation (mm) was monitored with a HOBO RG3-M (Onset Computer Corp., Bourne, MA, USA) tilting rain gauge, recorded in rain events with an accuracy of 0.2 mm/gauge. Based on the observed data, the daily mean temperature, daily mean RH, daily Rg, and daily  Advances in Meteorology precipitation were used for subsequent analysis. Rg and precipitation were the total amounts in a day, and the temperature was the mean of values at 03 : 00, 09 : 00, 15 : 00, and 21 : 00 (China Standard Time). Moreover, daily maximum and minimum temperature, RH, Rg, and wind speed (WS, m/s) were used to calculate potential evapotranspiration (PET) based on the Penman-Monteith (FAO56) method [31]. e data involved in this study were all observation records for the year 2014.

Estimation of ET G from the Diurnal Fluctuations in GWL.
In the revised White method, Loheide improved the calculation accuracy in areas with shallow DTWT and ET G can be estimated on an hourly basis [32]. us, this improved method will be henceforth referred to as the Loheide method, which was applied to estimate the ET G of eight observation sites under various vegetation conditions. Several assumptions were analyzed and discussed to explain the suitability and reduce the uncertainty of the approach. Changes in groundwater storage near the observation wells can be represented by the changes in GWL with time (dWT/dt). e changes in storage are controlled by the net inflow or outflow of nearby groundwater (r(t)[L/T]) and ET G : where S y is the specific yield. When ET G is zero, equation (1) can be simplified as follows: e recharge rate is a function of time [33]. Loheide [32] assumed that the head of the recharge source was constant.
us, the recharge rate of an observation well can be obtained from the observed water table records, as expressed in equation (2): e method assumes that the head of the recovery recharge has a similar change trend as the observed water table record; therefore, the trend included in the GWL can be removed as follows: where WT DT (t) is the detrended GWL, WT(t) is the observed GWL, m T is the trendline slope, and b T is the trendline intercept. Γ(WT DT ) is a function of dWT DT /dt and WT DT (t), and a best-fit estimate of the function can be obtained based on the detrended GWL from 00 : 00 to 06 : 00 of the day of interest and the following day. en, to obtain the recharge rate function, Finally, ET G is calculated as e estimated uncertainty of S y (specific yield) is the major factor that causes ET G estimation errors [23,34]. e simulation experiments conducted by Loheide et al. [35] to estimate S y demonstrated that diurnal GWL fluctuations and antecedent moisture conditions had almost no influence on S y . Furthermore, Meyboom [36] suggested that the readily available S y value should be half of the standard definition for S y , whereas Loheide et al. [35] thought the suggestion should be based on the specific situation. To obtain the S y value of the study area, soil samples at two depth ranges (0-0.8 m and 0.8-1.5 m below the ground) were collected. e soil moisture curve of the soil samples was determined using a pressure plate extractor (Daiki-3404; Daiki Rika Kogyo Co., Ltd, Saitama, Japan) to obtain important parameters used in the van Genuchten model, and the properties of the soil samples were analyzed using a Mastersizer 2000 laser diffractometer (Malvern Panalytical, Malvern, UK). e gravel (>2 mm), sand (0.0625-2 mm), silt (0.004-0.0625 mm), and clay (<0.004 mm) contents of the 0-0.8 m soil sample were 0, 98.63, 1.37, and 0% and those of the 0.8-1.5 m sample were 0, 78.58, 19.10, and 2.32%, respectively. e van Genuchten parameters θ r (unitless), θ s (unitless), α (cm − 1 ), and n (unitless) for the 0-0.8 m sample were 0.0291, 0.3873, 0.0425, and 2.363 and those for the 0.8-1.5 m sample were 0.0153, 0.3732, 0.0259, and 1.753, respectively. e S y estimation method proposed by Crosbie et al. [37] was used: where θ s is the soil saturated moisture content, θ r is the residual moisture content, z i and z f are the initial and final DTWT, and α and n are parameters in the van Genuchten model.

Relationship between Diurnal GWL Fluctuations and
Vegetation. From the observed water table records, regular diurnal GWL fluctuations were detected at eight groundwater observation sites, except wells G1 and G2, which were almost entirely comprised of bare sand (Table 1). e observation sites where the fluctuations were detected were covered with various types of vegetation (see Table 1 for main vegetation types). is phenomenon emerged from May to October, which was consistent with the growing season of the desert lakeside vegetation at the observation sites (see Figure 2(d) for wells G6 and G8). e findings mentioned above indicated that the diurnal GWL fluctuations were related to the vegetation covering lakeside areas.
On a daily scale, the diurnal GWL fluctuations exhibited a characteristics pattern whereby the water table decreased continuously during the daytime and rose gradually at night, reaching a maximum in the morning and a minimum in the afternoon. is diel cycle is generally thought to be induced by regular daily water consumption of phreatophytes, which essentially represents the dynamic balance between the groundwater lateral recharge and the consumption of groundwater by vegetation [6,[38][39][40]. On a seasonal scale, there were variations in diurnal GWL fluctuations at the eight sites, which was attributed to the interaction of vegetation with the surrounding environments [1,7,41]. To further understand the relationship between the diurnal GWL fluctuations and the vegetation at the observation sites on a seasonal scale, the diurnal GWL fluctuations during rainless periods at observation sites G4, G6, and G9 in different months are illustrated in Figure 3. In mid-April, the diurnal GWL fluctuations of the three observation sites were not obvious due to the low temperature (daily average maximum and minimum temperature: 20.0°C and 3.9°C) and no germinative vegetation. Figure 3 illustrates evident diurnal GWL fluctuations from June to September when desert vegetation was in the growth stage and large amounts of groundwater were consumed. e diurnal GWL fluctuations in July were more obvious at observation sites G6 and G9, although the DTWT in July was higher compared with that in June, showing a larger amplitude of diurnal GWL fluctuations in July. e meteorological conditions in July that drove higher vegetation transpiration (daily average maximum and minimum temperature: 33.5°C and 17.3°C) could account for mentioned observations. At the end of November, the vegetation entered the dormant period and the diurnal GWL fluctuations decreased because of the lower temperature in the desert (daily average maximum and minimum temperature: 8.9°C and − 6.7°C). e diurnal GWL fluctuations exhibited variations among the observation sites. As illustrated in Figure 3, the amplitudes of diurnal GWL fluctuations at observation site G4 from June to September were larger than those of observation sites G6 and G9. To have a full understanding of these variations, the amplitudes of diurnal GWL fluctuations of each observation site in the vegetation growing season and nongrowing season were plotted as separate boxplots ( Figure 4). Observation sites G3-G10 (with vegetation cover) had larger amplitudes of diurnal GWL fluctuations in the growing season, but the amplitudes in the nongrowing season were negligible. At sites G1 and G2 (the bare sand sites), no evident diurnal GWL fluctuations were observed in either the vegetation growing season or the nongrowing  season. Furthermore, the amplitudes of diurnal GWL fluctuations in the growing season varied among the sites G3-G10, which was attributed to the vegetation type. For instance, the amplitudes at sites G4 and G5, mainly covered by Phragmites australis, were larger than those of the sites G6-G10, with Achnatherum splendens and Nitraria tangutorum. Interestingly, site G3, also mainly covered with Phragmites australis, had a smaller amplitude of diurnal  GWL fluctuations than sites G4 and G5, indicating that vegetation type was not the only factor determining the variations in diurnal GWL fluctuations.
us, climatic conditions, soil properties, and groundwater dynamic processes might also cause differences in water consumption and recharge balance, manifesting as a spatial difference in diurnal GWL fluctuations.

Estimation of ET G under
ree Vegetation Covers. e White method was used to estimate the ET G rate of observation sites with three vegetation covers.
e GWL records of observation sites G3-G10 during the vegetation growing season were selected to analyze the ET G rate, and the statistical results are summarized in Table 2. e ET G rate was associated with vegetation type, especially in the middle of the growing season. From the relationship between vegetation type and the daily average ET G rate, the observation sites could be grouped into three categories: the ET G rate at sites G4 and G5, with well-grown Phragmites australis coverage, was ∼5 mm/d; the daily average ET G of observation sites G6 and G7 dominated by Achnatherum splendens coverage was 3∼4 mm, including site G7 with the best vegetation coverage in this category, where the ET G rate reached 4 mm/d; the remaining three observation sites G8-G10 were mainly covered by Nitraria tangutorum, and the daily average ET G rate was relatively low. Overall, the order of groundwater consumption by the three vegetation covers in this study was Phragmites australis > Achnatherum splendens > Nitraria tangutorum.
ese results were consistent with the ET G rate of similar research areas. For instance, an area covered with Salix psammophila in the Mu Us Desert, China, had a DTWT of 1∼1.5 m and daily ET G rate of 3∼4 mm in the growing season [42], which was similar to the estimates for the observation sites with Nitraria tangutorum and Achnatherum splendens in this study.
is similarity supports the accuracy of the estimated results in representing the ET G rate of desert lakeside vegetation.
In this study, ET G rates were also related to vegetation vitality. For example, site G3 with Phragmites australis, compared with sites G4 and G5 with similar vegetation, had poorer vegetation conditions with sparse patches and weak viability. As mentioned above, the amplitudes of diurnal GWL fluctuations at this site were narrower, which were ultimately reflected in a lower ET G rate than that at sites G4 and G5. Other sites also showed a similar situation, which was reflected in the maximum daily ET G rate. Observation site G3 only had a maximum daily ET G rate of 6.2 mm/d, which was similar to that of observation sites with Nitraria tangutorum. In contrast, the maximum daily ET G rate at site G6, with Achnatherum splendens, reached 11.3 mm/d, which was slightly lower than that of the sites with well-grown Phragmites australis.
On a seasonal scale, ET G is controlled by vegetation phenology [7]. In this study, the estimated ET G was higher from July to August than in other stages of the growing season ( Figure 5). Moreover, the period during which the peak daily mean ET G occurred was not perfectly consistent among the observation sites. e highest daily mean ET G at G3, G6, G9, and G10 occurred in early July, whereas the ET G rates of observation sites G4 and G7 were markedly higher from mid-July to early September than other growth stages. is indicated that the optimal growth periods varied among sites with different vegetation types and vitality. Due to the better vegetation coverage and growth condition of sites G4 and G7, the vegetation maintained relatively good growth and consumed large amounts of groundwater even when the vegetation in other sites entered their final growth stage in September. To explain the discrepancy, a previous study suggested that differences in the types of deciduous and nondeciduous vegetation led to different periods reaching a maximum ET G [43]. e vegetation in the present study comprised herbaceous and small shrub plants, with different vegetation coverages from those of the study mentioned above. us, it is logical that the unsynchronized maximum daily mean ET G was driven mainly by vegetation type and growth vitality.

Controlling Factors of ET G Rate.
Vegetation type, meteorological parameters, and groundwater dynamics jointly led to spatiotemporal variations in the diurnal GWL fluctuations at the observation sites, which ultimately was reflected in the ET G . To further explore the key factors affecting the ET G rate at the observation sites and the mechanisms that control ET G in the Badain Jaran Desert, the relationship between ET G rate with DTWTand meteorological parameters was analyzed. Taking observation site G6 as an example (Figure 6), there was a significant positive correlation between ET G rate and air temperature, which was consistent with the result that temperature was the key factor controlling the seasonal-scale variations in the diurnal GWL fluctuations obtained from Figure 3. Moreover, the ET G rate at site G6 was positively correlated with solar radiation but had no significant relationship with wind speed, relative humidity, and other factors. ese results indicated that air temperature and solar radiation are the key meteorological factors controlling the ET G rate at this observation site.
e relationship between the ET G rate and the external factors at site G6 indicated that the meteorological conditions in the Badain Jaran Desert were among the most critical factors controlling the ET G rate. To verify this conclusion, the relationship between ET G rate and PET at eight observation sites (i.e., as a comprehensive measure of meteorological condition) were investigated. Significant positive correlations (R 2 � 0.38∼0.54) were observed at all sites except G7 (Figure 7), which demonstrated the influence of meteorological factors on ET G rate. For comparison, a previous study in a semi-arid area suggested that ET G was weakly positively correlated with PET (R 2 � 0.02∼0.32) [19], and Lautz [23] also reported a similar correlation between PET and ET G in a riparian region. In contrast, the ET G observed by Yuan et al. [18] showed a significant positive correlation with PET in desert riparian forests. Mazur et al. [17] also recognized a positive correlation and estimated the ET C (i.e., interpolated ET rate) based on the significant relationship between the estimated ET G and PET when the White method was not appropriate for ET G estimation. erefore, the correlation between the estimated ET G and PET differed among the observation sites and the   [17]. In this study, estimated ET G and PET showed relatively high correlations at most observation sites, and there were few precipitation events. erefore, the White and Mazur's estimation methods could be jointly used to quantify the annual groundwater consumption (see Figure 8 for the estimated results at observation site G6). Observation sites G3, G5, G6, G8, G9, and G10 were selected for this application based on the significant relationships between the estimated ET G and PET, where the total ET G in the whole growing season (total 184 days) was obtained. e ET G ranged from 600 to 900 mm at observation sites with Phragmites australis and varied depending on the coverage conditions, and it ranged from approximately 600 to 650 mm at the sites covered with Achnatherum splendens. At sites with Nitraria tangutorum and Achnatherum splendens, the ET G ranged from 500 to 650 mm.

Advances in Meteorology
In general, the amount of groundwater available for vegetation roots decreases as DTWT increases, resulting in a negative relationship between DTWT and ET G rate [44]. Furthermore, Cooper et al. [45] found that a 1.6 m water level decline resulted in a 62% reduction in ET G , highlighting the impact DTWTchanges on ET G . In this study, the DTWT range varied among the observation sites, and their relationships with vegetation type, vegetation vitality, and evapotranspiration rate were consistent with the trends observed by Jackson et al. [44]. For instance, observation sites G4 and G5, which were mainly covered densely with Phragmites australis, had a DTWT of 0.7∼1.5 m. Meanwhile, sites G8 and G9 covered sparsely with Nitraria tangutorum, had a DTWT of 1.2∼1.7 m. Correspondingly, under the same meteorological conditions, the ET G rate of the sites with Phragmites australis was markedly higher than that of sites with Nitraria tangutorum. e estimated total ET G at each observation site in the growing season was greater than the total annual precipitation in that year (103.7 mm), but there was no difference in the DTWT at the beginning and end of the growing season. Overall, the water level first declined and then returned to the original level (Figure 9), indicating that groundwater was the primary recharge source of the lakes in this study area, with only a small amount of precipitation recharge. Furthermore, the total amount of lateral recharge and precipitation recharge of groundwater was greater than the total ET G . erefore, groundwater is the main water source for vegetation growth in this case, and the ET G process is dependent on groundwater recharge. However, a positive correlation was indicated between ET G rate and DTWT at observation site G6 (Figure 6), and the water table decreased by ∼0.2 m from May to August, while the estimated ET G showed an increasing trend (Figure 9). is seemingly contradictory phenomenon can be explained by the fact that the ET G rate was jointly controlled by the DTWT and PET,  and the effect of PET exceeded that of DTWT in that period. During the observed period in this study, changes in DTWT had no marked influence on the ET G rate because DTWTs were shallow (i.e., all within 2 m) and presented relatively small changes, all within 2 m. erefore, observation sites with a wider DTWT range can be set up in future research to further discuss the influence of DTWT on the ET G rate.

Limitations and Observations on Applying the Proposed Method.
e application of the White method and its revised methods are based on four basic assumptions: (1) reduction in GWL is induced only by vegetation; (2) the transpiration at night is weak and negligible; (3) the average raise rates of GWL at night can represent recharge rates of groundwater in a given day; (4) the specific yields are representative and reliable. Although the groundwater signal methods have been widely verified and applied in studies on vegetation evapotranspiration of arid regions, the assumptions still have certain limitations. ese shortages are specifically reflected in the following scenarios: lateral GWL recharge rates change on a daily basis and there are errors in recharge rates using the constant rise rates of GWL at night; accurate specific yields are not easy to attain; with the advancement of high-frequency digital data collection devices, recent studies have found that transpiration of some vegetation did not stop completely at night [5,46]. To overcome these deficiencies, many studies proposed several revised methods. In regard to the calculation of groundwater recharge rates, Troxell [33] questioned whether recharge rates were constant on a daily basis. Later, scholars represented by Gribovszki et al. [14] and Loheide [32] proposed new methods to obtain the dynamic daily recharge rates, which improved the ET G estimation accuracy.
Considering the basic assumptions and their limitations, there are some observations to be explained in the application of the groundwater diurnal signal method in Badain Jaran Desert. Due to their diverse mechanisms, GWL diurnal fluctuations can be categorized into various types [40].
e GWL series are easily distinguished from each other except the cycles induced by atmospheric pressure and evapotranspiration, as the two have similar fluctuant shapes. Atmospheric pressure effects can also occur in unconfined aquifers typically mirroring each other [47][48][49]. e diurnal cycles induced by atmospheric pressure and evapotranspiration are both continuous time series. However, there are significant differences in their daily variations. e pressuredominated GWL cycles are closely related to the fluctuations of atmospheric pressure in each season, and the valley and peak values of GWL correspond to the maximum and minimum air pressure throughout the day, respectively. Similar to the GWL diurnal fluctuations observed by Jiang et al. [50], air pressure reached a maximum at approximately 10 : 00 am each day when the water level was the lowest. Conversely, the pressure was lowest at 3 : 00 pm when GWL reached its maximum. In the case of the Badain Jaran Desert, diurnal GWL fluctuations are completely different from the above type and are consistent with the ET-diurnal cycles [40]. e second assumption in the White method (i.e., the transpiration of vegetation at night is negligible) can be confirmed by comparing the results in this study with other research results in nearby desert areas. For example, Yuan et al. [18] thought that groundwater evapotranspiration at night was weak based on observations using the eddy covariance method in areas covered by desert vegetation. e site chosen for observation was dominated by Tamarix, where Phragmites australis, Glycyrrhiza inflata, and other types also grow, which was rather similar to the observation sites in this study. Additionally, other studies based on the White method estimated groundwater evapotranspiration in regions of Mu Us and Gobi Deserts that were covered with vegetation [11,42,43]. Although vegetation transpiration at night could be negligible in most cases, other studies have determined that transpiration at night is considerable in certain environments. erefore, the ET G estimation in these cases is based on the reliability of the assumption. In future research, it is necessary to monitor the vegetation water consumption at night to reduce the ET G estimation uncertainty.   Methods to obtain aquifer specific yields are also important in research involving the White method. Specific yields are not only related to the aquifer soil texture but also affected by the depth to water table, dynamic rates of temporal change of GWL, and drainage time.
ere are various methods based on factual research conditions aiming to obtain a representative and reliable specific yield. Gribovszki [51] recently proposed a diurnal method to obtain dynamic specific yield and compared it with traditional specific yield estimation techniques. Moreover, pumping test estimations are considered to be a relatively low-cost and suitable method, as it is the most similar to the dynamic specific yield. While it would be difficult to implement a slug test at the hinterland studied herein, relatively reliable specific yield could be feasibly obtained via a pF curve (a texture-based field capacity value). However, the specific yields obtained by this method are nondynamic and may also become a source of ET G estimation errors. erefore, observations on the specific yield estimation in this study are further explained below.
Specific yields strongly depend on the DTWT in shallow groundwater environments and are dynamic in reality. Existing research [5] combined drainage experiments and simulation modeling methods to obtain the changes in specific yields with DTWT within 0∼2 m. When DTWT was within 0.8 m, specific yields varied as DTWTs became deeper. en, there was a difference when depth is greater than 0.8 m, and the effect of DTWT change on specific yields began decrease and approached a constant value that depended on soil texture. In another case, observations demonstrated that there was no significant correlation between specific yields and DTWT [51], for which the range of GWL changes was thought to be small and depth compensation in groundwater was not a key issue to consider. DTWTs varied from 0.8 to 1.1 m in the second case; therefore, to some extent, the results of the two studies mentioned above were consistent. In the present study of the Badain Jaran Desert, the DTWTs of each observation site was greater than 0.8 m (Figure 7) except the observation sites G3 and G5, which had DTWTs of less than 0.8 m during a short period of the beginning and end in the growing season. From this perspective, the changes in DTWTs at most observation sites had little effect on the specific yield estimations. Nevertheless, the specific yield estimation in this study was nondynamic as it did not account for the dynamic GWL temporal change rates relative to a pumping test estimation or the dynamic diurnal method. erefore, the specific yield estimation may become a source of ET G estimation errors. According to Gribovzxki's comparison, the estimated specific yield in this study will be slightly higher than that calculated by the dynamic method. In future research, soil moisture monitoring instruments are expected to be added at groundwater observation sites to observe dynamic specific yields and thus improve ET G estimation accuracy.

Conclusions
Based on observed GWL and meteorological parameters in the lake basins of the Badain Jaran Desert, the characteristics of the regular diurnal GWL fluctuations in spatiotemporal scales and their relationship with vegetation were analyzed. Afterwards, the ET G was estimated via the White method, and the main controlling factors of ET G were explored. e observed series indicated that diurnal GWL fluctuations were consistent with the vegetation growing season, which was driven by groundwater consumption by phreatophytes. e amplitudes of diurnal GWL fluctuations showed marked differences among the sites, which was attributed to vegetation type and vitality. On a seasonal scale, the air temperature was an important external factor affecting the diurnal GWL fluctuations by controlling the vegetation phenology and groundwater consumption. e estimated ET G was associated with vegetation type and exhibited a degree of association with the vegetation growth stage. e observation sites could be grouped into three categories based on the estimated results and vegetation type, and the ET G followed the general order Phragmites australis > Achnatherum splendens > Nitraria tangutorum. In the lake basins of the Badain Jaran Desert, the vegetation grew in the area where the DTWT was within ∼2 m, and meteorological factors had a greater influence on vegetation transpiration than in other areas with higher DTWTs. Moreover, DTWT and PET are both important factors controlling the ET G rate, while their influence varied among observation sites.
is study further expands the geographical applicability of the White method and offers a new research paradigm for the study of diurnal GWL fluctuations in groundwaterdependent lakeside ecosystems. Additionally, the study results could become a scientific basis to solve the controversy on the formation mechanism of the lakes in the Badain Jaran Desert and support the sustainable usage of regional groundwater resources.
Data Availability e groundwater level data and meteorological data used to support the findings of this study were supplied by the Center for Glacier and Desert Research, Lanzhou University, under license and so cannot be made freely available. Requests for access to these data should be made to the corresponding author.

Conflicts of Interest
e authors declare that they have no conflicts of interest.