Energy Exchange and Evapotranspiration over the Ejina Oasis Riparian Forest Ecosystem with Different Land-Cover Types

: Investigating the energy and water vapor exchange in oasis riparian forest ecosystems is of signiﬁcant importance to improve scientiﬁc understanding of land surface processes in extreme arid regions. The Heihe Watershed Allied Telemetry Experimental Research (HiWATER) provided many observations of water vapor and heat ﬂuxes from riparian forest ecosystem by using a network of eddy-covariance (EC) systems installed over representative surfaces in the Ejina Oasis, which is located in the downstream areas of the Heihe River Basin, northwestern China. Based on EC ﬂux measurements and meteorological data performed at ﬁve stations and covering representative surface types of Populus euphratica tree with associated Tamarix chinensis shrub, Tamarix chinensis shrubland, cantaloupe cropland, and barren-land, this study explored the spatio-temporal patterns of heat and water vapor ﬂuxes over the Ejina Oasis riparian forest ecosystem with ﬁve different surface types over the course of a growing season in 2014. Energy balance closure of the ﬂux data was evaluated; footprint analysis for each EC site was also performed. Results showed that energy balance closure for the ﬂux data was reasonably good, with average energy balance ratio (EBR) of 1.03. The seasonal variations in net radiation (R n ), latent (LE), and sensible heat ﬂux (H) over the ﬁve contrasting surfaces were similar, and a reverse seasonal change was observed in energy partitioning into LE and H. Remarkable differences in R n , LE, and H between the ﬁve surfaces were explored preliminarily, associated closely with the soil properties and foliage phenology. Over the growing season (May–October) in 2014, the total ET ranged 622–731 mm for mixed forest of P. euphratica trees with associated T. chinensis shrubs with average daily ET of 3.6–4.2 mm; ET from T. chinensis shrubland was about 541 mm, with average daily ET of 3.6 mm. ET for barren-land was 195 mm. The total ET in irrigated cantaloupe cropland with plastic mulch was 431 mm for its four-month growing period with a total average of 3.8 mm d − 1 . Determination of ET over riparian forest ecosystem helps to improve reasonable use of limited water resource in the Ejina Oasis.


Introduction
Many important processes of terrestrial ecosystems, such as photosynthesis and evapotranspiration, are associated with the energy and mass exchange between the land surface and atmosphere [1][2][3][4]. Understanding the exchange processes of water vapor and energy between the terrestrial ecosystem and atmosphere is important in modeling ecosystem production, water balance of terrestrial ecosystems, and atmospheric circulation [4,5]. The terrestrial ecosystem latent and sensible heat fluxes strongly interact with the overlaying atmosphere and consequently affect the characteristics of the planetary boundary layer [6], ultimately influencing the local and even regional climate [1,3,7,8]. In return, the water vapor and energy fluxes can be altered by local environmental variables that affecting a (EWDP) was implemented in the HRB to restore the deteriorated ecosystems in the Ejina Oasis. Since the reallocation of surface water, the dried-up terminal lake (Sogo Nuur) has been recharged with water since 2002 [47]. Consequently, the spatiotemporal distribution of water resources in the Ejina Oasis has changed dramatically, and the riparian forest ecosystem has been restored to some degree [48][49][50][51]. Furthermore, the area of cropland has also been expanded extensively, resulting in the overextraction of groundwater and inappropriate use of surface water for agriculture [46,51]. Quantification of ET over the Ejina Oasis ecosystem with different land-cover types is thus crucial to reasonable allocation of the limited water resources for ecosystem rehabilitation [45].
To date, several pioneering works on surface energy and water exchange of riparian forest ecosystem under extreme arid regions have been conducted [52][53][54][55][56][57][58][59]. Yuan et al. [54] found the seasonal processes of energy and water exchanges over a riparian Tamarix stand with the vegetation cover of 0.65 and leaf area index (LAI) of 1.15 in the lower TRB were highly related with plant phenology, and ET depend on the groundwater, rather than the precipitation. Yuan et al. [55] compared the ET and its controlling mechanism over the Tamarix shrubland and a Populus euphratica forest (with vegetation cover of 0.49 and LAI of 0.57) in the lower TRB, and found temporal patterns of ET over the two species were similar, but the water consumption of Populus euphratica was much higher. In the lower HRB, Si et al. [52] measured the ET over Tamarix stand (with LAI of 3.2) by using the Bowen ratio method, and the mean daily ET was 1.62 mm over a 153-day growing season. Hou et al. [53] estimated the ET over a Populus euphratica forest with the canopy density of 0.8 using the Bowen ratio method; the maximum daily ET reached to 6.7 mm, and the variation pattern of crop coefficient (Kc) was similar to that of ET during the growing season. Yu et al. [56,57] measured the riparian forest ET of Tamarix (with vegetation cover of 0.52 and LAI of 1.97) and the Populus euphratica (with vegetation cover of 0.47 and LAI of 2.15) using the EC method, and found that the measured daily ET (with a mean of 2.8 mm) over Tamarix shrubland was comparable to the results from the TRB; while the Populus euphratica daily ET was 4.0 mm on average, its seasonal pattern was associated with the LAI. Additionally, the water use between the two contrasting riparian forest in the lower HRB was comparable without consideration of soil evaporation [58]. Moreover, the differences in ET between two different aged Populus euphratica forests (with vegetation cover of 0.52 and 0.3, LAI of 2.27 and 1.78) were mainly related to flood irrigation by surface water from Heihe River [59]. The above-mentioned results demonstrated that groundwater influences the processes of ET by controlling the growth and spatial distribution of riparian forests. Nevertheless, these studies just focused on Populus euphratica forests and Tamarix chinensis shrublands. Therefore, new and long-term energy and water exchange studies on riparian forest ecosystem with various land covers, including cropland, are needed to improve the understanding of the eco-hydrological processes and biophysical functioning of these groundwater-dependent ecosystems. Such studies will bridge considerable knowledge gaps in understanding surface energy exchange and evapotranspiration over these contrasting landscapes in extreme arid regions.
The main objective of this study was to further explore and better understand the temporal and spatial patterns of water vapor and energy exchanges over the Ejina Oasis riparian forest ecosystem with a variety of surface types, including riparian forests of Populus euphratica with associated Tamarix chinensis, Tamarix chinensis shrubland, cropland, and barren-land. Based on surface water vapor and energy flux observations performed at five EC sites and covering all representative surface types, continuous measurements of meteorological factors and soil moisture, and groundwater depth observations made at the riparian forest ecosystem in the Populus euphratica forest reserve of Ejina Oasis in 2014, the present study (1) evaluated the surface energy balance closure and spatial representativeness of the multi-site EC flux measurements to ensure a high-quality flux data; (2) characterized and compared the temporal and spatial variation in energy flux partitioning and evapotranspiration in the riparian forest ecosystem with five contrasting surface types over the course of a growing season to explore the exchange patterns of surface energy and water vapor; and (3) identified the main factors controlling evapotranspiration and energy flux over these five contrasting landscapes to further understand the land surface processes of the riparian forest ecosystem in the Ejina Oasis.

Study Area and Site Description
The study area is located in the Ejina Oasis (Figure 1a), which is in the downstream areas of the HRB (97 • 09 36 -102 • 58 48 E, 37 • 52 12 -43 • 39 00 N). This oasis is one of the most arid regions in China according to local meteorological data from 1990 to 2010, with mean annual precipitation of less than 50 mm and annual potential evaporation of approximately 3755 mm [42]. Mean annual air temperature is around 9.7 • C [46]. Natural vegetation in this oasis is dominated by the species of Populus euphratica (P. euphratica) with associated Tamarix chinensis (T. chinensis) and other desert species. The P. euphratica trees reach 4-20 m in height, and the T. chinensis shrubs have a height in the range of 1-5 m [60]. In this oasis, the P. euphratica tree's leaf area reaches a maximum value from July through August, then its leaves go through senescence and subsequent fall stage by late October [61]. For the T. chinensis shrub, the leaves sprout at the end of April and start to go into senescent stage in mid-September. The agricultural crop is dominated by the cantaloupe (a type of muskmelon). In 2017, the total production area of cantaloupe in the Ejina Oasis was more than 4000 hectares, with an annual yield of more than 110,000 tons. The cantaloupe grows rapidly by early July, reaches its maximum height of approximately 0.4 m within one month, and is harvested at late September. The plastic mulch is usually used by farmers on crop field to conserve soil moisture and reduce soil evaporation. The crop field is irrigated regularly with local groundwater. Throughout the crop growing period in 2014, there are usually four rounds of irrigation (around 11 June, 17 July, 2 August, and 18 August) [62].
During the Heihe Watershed Allied Telemetry Experimental Research (HiWATER [42]), the region from Erdaoqiao (Bridge-2) to Qidaoqiao (Bridge-7) in the oasis (known as 'the core area of the P. euphratica forest reserve') was selected as the HRB downstream investigation area (Figure 1b). A key area in this reserve (centered at 101 • 08 00 E, 42 • 00 00 N) has been selected for intensive and long-term observations since July 2013 (Figure 1b). The intensive observation area located in the Sidaoqiao (Bridge-4), approximately 3 × 3 km 2 , has land cover that is 38% T. chinensis, 13% P. euphratica, and 39% barren-land, as well as 7% cropland and 3% grass ( Figure 1c). In this intensive observation area, two paths of large aperture scintillometer (LAS) systems (named LAS_E and LAS_W; BLS900, Scintec, Germany) were installed parallelly to measure the path-averaged sensible heat fluxes across the intensive observation area. Each path of a LAS system consists of a pair of transmitter and receiver, as shown in Figure 1c. These two LAS systems measure over a sparse riparian forest with same measurement height of 25.5 m and path lengths of 2390 and 2380 m, respectively. Along the two LAS transects, five observation sites were deployed over the dominant surface types within this area. Spatial distribution of each site is shown in Figure 1c, with CL of T. chinensis shrubland, MF of P. euphratica and associated T. chinensis, PF of sparse P. euphratica with associated T. chinensis, CL of cantaloupe field, and BL of barren-land (Table 1).
Each observation site was equipped with an eddy-covariance (EC) system and an automatic weather station (AWS) to observe the sensible and latent heat fluxes, carbon dioxide fluxes, and meteorological elements. Details of the EC system equipment at the five observation sites are given in Table 1. Measurements made at each AWS tower included a four-component net radiation (downward/upward shortwave and longwave radiation), wind speed and direction, air temperature and relative humidity, air pressure, precipitation, soil temperature and moisture depth profile, and soil heat flux. Further details on the meteorological instruments equipped at each AWS are listed in Table 2. The shrubland station (SL) was equipped with wind speed/air temperature/relative humidity measurements at heights of 5, 7, 10, 15, 20, and 28 m, and soil temperature and moisture measurements at 0.02, 0.04, 0.1, 0.2, 0.4, 0.8, 1.2, and 1.6 m depths. During the experimental period from 20 to 31 July 2014, soil samples were also taken at these depths in the vicinity of each tower for the measurements of soil moisture. The LAI of T. chinensis shrubs and cantaloupe was measured using a LAI-2200 canopy analyzer (LI-COR; Lincoln, NE, USA) [63]. The mean LAI values of T. chinensis at SL and cantaloupe at CL were 3.47 and 3.07, respectively. At each observation station, soil heat flux was measured with three heat flux plates installed 0.06 m below the surface. In addition, three groups of thermal dissipation sap flow probes (TDPs) were installed to measure the canopy transpiration of P. euphratica tree (Figure 1c). Nine representative trees with diameters ranging from 33.8 to 51 cm at breast height were selected in the vicinity of the observation sites. To monitor the spatial and temporal variations in shallow groundwater depth, five groundwater wells have been dug with a depth of 6 m and, respectively, equipped with a HOBO Water Level Logger (U20-001-04; Onset, Bourne, MA, USA) since June 2014 [62], as shown by the numbers from 1 to 5 in Figure 1c. shrubland station (SL) was equipped with wind speed/air temperature/relative humidity measurements at heights of 5, 7, 10, 15, 20, and 28 m, and soil temperature and moisture measurements at 0.02, 0.04, 0.1, 0.2, 0.4, 0.8, 1.2, and 1.6 m depths. During the experimental period from 20 to 31 July 2014, soil samples were also taken at these depths in the vicinity of each tower for the measurements of soil moisture. The LAI of T. chinensis shrubs and cantaloupe was measured using a LAI-2200 canopy analyzer (LI-COR; Lincoln, NE, USA) [63]. The mean LAI values of T. chinensis at SL and cantaloupe at CL were 3.47 and 3.07, respectively. At each observation station, soil heat flux was measured with three heat flux plates installed 0.06 m below the surface. In addition, three groups of thermal dissipation sap flow probes (TDPs) were installed to measure the canopy transpiration of P. euphratica tree (Figure 1c). Nine representative trees with diameters ranging from 33.8 to 51 cm at breast height were selected in the vicinity of the observation sites. To monitor the spatial and temporal variations in shallow groundwater depth, five groundwater wells have been dug with a depth of 6 m and, respectively, equipped with a HOBO Water Level Logger (U20-001-04; Onset, Bourne, MA, USA) since June 2014 [62], as shown by the numbers from 1 to 5 in Figure 1c.     The duration of meteorological measurement made at the SL, MF, PF, CL, and BL stations is consistent to that of the EC measurement, as listed in Table 1.

Data Processing
Ten-minute AWS observations were averaged to half-hour periods. AWS data were removed during times of known instrument malfunction and if values exceeded physically reasonable ranges. Soil heat flux measurements at the depth of 0.06 m were corrected to surface values using the methods proposed by Liebethal et al. [64] and Xu et al. [65].
EC raw data sampled at 10 Hz were processed carefully using the free EddyPro software (Version 7.0; Lincoln, NE, USA). Data processing procedures included spike removal, time lag correction, coordinate rotation (2D rotation), sonic virtual temperature correction, frequency response correction, and corrections for density fluctuation [66][67][68]. Data quality assessment for every 30 min turbulent fluxes was performed according to the flagging scheme with three different quality control levels (0, 1, and 2) [69]. The 30 min flux data with flag 0 and 1 were selected for the latter analysis; those with flag 2 representing low-quality data were discarded. The flux data within 1 h of precipitation were also rejected. The nonlinear regression method (establishing a regression between the net radiation and latent or sensible heat flux) was used to gap-fill the missing 30 min latent-and sensible heat fluxes. When calculating the accumulative ET, the half-hourly latent heat fluxes were converted to ET by dividing by the coefficient of latent heat vaporization (λ), which is calculated with air temperature. The daily total ET was obtained by summing the 30 min ET observations of each day.

Data Analysis
Considering the time period for instrument recalibration and crop filed management in this oasis, all observations including the half-hourly EC flux measurements and simultaneous meteorological data taken during the growing season from mid-May to October 2014 were selected for the detailed analysis.

Footprint Analysis of EC Flux Measurements
Footprint models have been regarded as diagnostic tools to quantify the representativeness of flux tower measurements [70,71]. In this study, the Kljun's parametrization Flux Footprint Prediction (FFP) [72], a sophisticated backward Lagrangian footprint model, was applied to calculate the flux footprint of EC measurements. For the calculation of the EC flux footprints with FFP model, the values of the input parameters were directly obtained from the corresponding EC measurements, which include measurement height, wind direction, wind speed, friction velocity, Obukhov length, and standard deviation of lateral velocity fluctuations. Other required parameters, such as the roughness height and zero-plane displacement, were obtained using the method referred to Chehbouni et al. [73]. Here, 70% of the total flux contribution source area was set to limit the extent of footprints [74]. The normalized flux footprints of each EC site were separately superimposed on a 0.61 m spatial resolution land-cover classification map to determine the relative contribution of all land-cover classes in the footprint to the total flux measured by EC.

Energy Balance Closure Assessment
The energy balance equation expresses the partitioning of net energy into energy and mass fluxes between the land surface and the atmosphere. The surface energy balance equation can be written as follows (all units in W m −2 ): where R n is the net radiation, G 0 is the soil surface heat flux, H is the sensible heat flux, LE is the latent heat flux, and ∆S is the heat storage term, including soil and canopy heat storage, the energy storage by photosynthesis, and respiration. Net radiation is the sum of net short radiation and net longwave radiation [75]: where K ↓ and K ↑ are the downward and upward shortwave radiation at the surface, respectively, L ↓ and L ↑ are the downward and upward longwave radiation, respectively. Downward longwave radiation originates from the atmosphere, both from the clear sky and from clouds; upward longwave radiation is mainly emitted by Earth's surface. The energy balance closure for the EC flux measurements (H and LE) were evaluated by the energy balance ratio (EBR). The Rn and G 0 were measured by using the fourcomponent net radiometer and heat flux plates at each site. As reported in the literature, the heat storage terms ∆S represent about 2-3% of the net radiation, particularly in the environments with lower canopy and leaf density [65,76]. In this study, the ∆S in Equation (1) was neglected in the EBR [76,77]. The EBR was obtained according to the ordinary least squares (OLS) method, which is the linear regression between the sum of the 30 min turbulent energy fluxes (H and LE) and the available energy (R n minus G 0 ). The linear Water 2021, 13, 3424 8 of 23 regression coefficients (slope and intercept) for energy balance closure can be derived from the ordinary least squares (OLS) method. Ideal energy budget closure is when the intercept is zero, and the regression slope and coefficient of determination (R 2 ) are equal to one [65,77]. In this study, the linear regression is used to assess the EBR and is performed by forcing the slope through zero with an intercept of 0, following the work of [78]. Figure 2 presents the flowchart of this study. Careful data processing and quality control for the AWS and EC raw data were performed to ensure high-quality meteorological measurements and turbulent fluxes. Based on continuous measurements of meteorological factors, water vapor and energy flux, and groundwater depth, this study focused on investigating the temporal patterns of energy fluxes exchanges and evapotranspiration over the Ejina Oasis riparian forest ecosystem with five contrasting surface types.

Flowchart of This Study
The energy balance closure for the EC flux measurements (H and LE) were evaluated by the energy balance ratio (EBR). The Rn and G 0 were measured by using the four-component net radiometer and heat flux plates at each site. As reported in the literature, the heat storage terms ΔS represent about 2-3% of the net radiation, particularly in the environments with lower canopy and leaf density [65,76]. In this study, the ΔS in Equation (1) was neglected in the EBR [76,77]. The EBR was obtained according to the ordinary least squares (OLS) method, which is the linear regression between the sum of the 30 min turbulent energy fluxes (H and LE) and the available energy (Rn minus G0). The linear regression coefficients (slope and intercept) for energy balance closure can be derived from the ordinary least squares (OLS) method. Ideal energy budget closure is when the intercept is zero, and the regression slope and coefficient of determination (R 2 ) are equal to one [65,77]. In this study, the linear regression is used to assess the EBR and is performed by forcing the slope through zero with an intercept of 0, following the work of [78]. Figure 2 presents the flowchart of this study. Careful data processing and quality control for the AWS and EC raw data were performed to ensure high-quality meteorological measurements and turbulent fluxes. Based on continuous measurements of meteorological factors, water vapor and energy flux, and groundwater depth, this study focused on investigating the temporal patterns of energy fluxes exchanges and evapotranspiration over the Ejina Oasis riparian forest ecosystem with five contrasting surface types.

Meteorological Conditions
The seasonal variations in the wind speed, air temperature, relative humidity, and pressure as well as cumulative precipitation in 2014 is shown in Figure 3. The maximum wind speed occurred in May, with a monthly average of 4.1 m s −1 . Wind speed decreased gradually with the increase in vegetation coverage in this area and reached a minimum average of 2.4 m s −1 in August (Figure 3a). The annual average wind speed was 3 m s −1 in 2014. Air temperature increased from winter to summer, with the highest values (23-27 • C) during June-August, and the annual average was about 10 • C ( Figure 3b); whereas the air pressure exhibited a distinctly opposite variation, with a lowest value of 899 hpa in July (Figure 3d). In Figure 3c, the relative humidity of air throughout the whole year was almost less than 40%. The total precipitation was 25 mm in this area. Scarce precipitation had less impact on the increase in air humidity during June-August. All in all, the study area experienced a hot and dry summer in 2014. As shown in Figure 4, there are two prevailing wind directions at this area, namely northwesterly and southeasterly winds. The maximum frequencies of northwesterly and southeasterly winds were about 3% and 2.5%, respectively. Both wind speeds and frequencies of northwesterly winds were a little larger those of southeasterly winds. For northwesterly winds, its wind speed can reach a maximum value larger than 8 m s −1 .

Meteorological Conditions
The seasonal variations in the wind speed, air temperature, relative humidity, and pressure as well as cumulative precipitation in 2014 is shown in Figure 3. The maximum wind speed occurred in May, with a monthly average of 4.1 m s −1 . Wind speed decreased gradually with the increase in vegetation coverage in this area and reached a minimum average of 2.4 m s −1 in August (Figure 3a). The annual average wind speed was 3 m s −1 in 2014. Air temperature increased from winter to summer, with the highest values (23-27 °C) during June-August, and the annual average was about 10 °C ( Figure 3b); whereas the air pressure exhibited a distinctly opposite variation, with a lowest value of 899 hpa in July (Figure 3d). In Figure 3c, the relative humidity of air throughout the whole year was almost less than 40%. The total precipitation was 25 mm in this area. Scarce precipitation had less impact on the increase in air humidity during June-August. All in all, the study area experienced a hot and dry summer in 2014. As shown in Figure 4, there are two prevailing wind directions at this area, namely northwesterly and southeasterly winds. The maximum frequencies of northwesterly and southeasterly winds were about 3% and 2.5%, respectively. Both wind speeds and frequencies of northwesterly winds were a little larger those of southeasterly winds. For northwesterly winds, its wind speed can reach a maximum value larger than 8 m s −1 .

Representativeness of EC Flux Matrix Measurements
The spatial representativeness of the five EC sites in the intensive observation area were performed by overlaying the flux footprint with a land-cover classification map. Table 3 lists the relative contribution of all land-cover classes present in the daytime-averaged footprint (70% flux contribution source area). For Cropland station (CL) and Barrenland station (BL), the surface types dominated in their footprint areas were cantaloupe cropland and barren-land, respectively. For Shrubland station (SL), the T. chinensis occupied large proportion of its source area; while for Mixed forest station (MF) and Populus forest station (PF), the dominant vegetation types present in their source areas were P. euphratica and T. chinensis. The barren-land also contributed greatly to the total flux, particularly for PF, indicating that major part of the surface within its footprint area was even free of vegetation.
For the three natural vegetation covered stations (SL, MF, and PF), the relative weight of natural vegetation within their flux footprint areas varied slightly as the prevailing wind direction changed. When winds were southeasterly, at SL, the relative contribution of T. chinensis shrubland to the total flux was about 0.92. For MF and PF, the total weights of P. euphratica and T. chinensis in their footprint areas were 0.74 and 0.53, respectively. However, their total contributions to the measured flux altered to 0.81 and 0.47, respectively, when wind directions became northwesterly; and the flux contribution of T. chinensis decreased to 0.62 for SL. The results indicated that, in general, the EC flux matrix measurements in the intensive observation area of HiWATER were representative of the dominant surface types within the oasis riparian forest ecosystem.

Representativeness of EC Flux Matrix Measurements
The spatial representativeness of the five EC sites in the intensive observation area were performed by overlaying the flux footprint with a land-cover classification map. Table 3 lists the relative contribution of all land-cover classes present in the daytimeaveraged footprint (70% flux contribution source area). For Cropland station (CL) and Barren-land station (BL), the surface types dominated in their footprint areas were cantaloupe cropland and barren-land, respectively. For Shrubland station (SL), the T. chinensis occupied large proportion of its source area; while for Mixed forest station (MF) and Populus forest station (PF), the dominant vegetation types present in their source areas were P. euphratica and T. chinensis. The barren-land also contributed greatly to the total flux, particularly for PF, indicating that major part of the surface within its footprint area was even free of vegetation. For the three natural vegetation covered stations (SL, MF, and PF), the relative weight of natural vegetation within their flux footprint areas varied slightly as the prevailing wind direction changed. When winds were southeasterly, at SL, the relative contribution of T. chinensis shrubland to the total flux was about 0.92. For MF and PF, the total weights of P. euphratica and T. chinensis in their footprint areas were 0.74 and 0.53, respectively. However, their total contributions to the measured flux altered to 0.81 and 0.47, respectively, when wind directions became northwesterly; and the flux contribution of T. chinensis decreased to 0.62 for SL. The results indicated that, in general, the EC flux matrix measure-ments in the intensive observation area of HiWATER were representative of the dominant surface types within the oasis riparian forest ecosystem. Figure 5 presents a scatter plot of the half-hourly available energy (R n -G 0 ) and the sum of the turbulent heat fluxes (H + LE) at five EC sits during the period from mid-May to October in 2014. For the five EC sites in the intensive observation area, the average EBR during the five and half months was 1.03. The Barren-land station has the lowest EBR (0.84) (Figure 5e). The EBR for other four vegetation covered sites is a little larger than 1. As shown in Figure 5, both the Shrubland station and the Populus forest station showed an EBR of 1.08; while the EBR values in the Mixed forest station and the Cropland station were 1.06 and 1.12, respectively.

Temporal Variations in Energy Flux Partitioning over A Riparian Forest Ecosystem
To investigate the variation in net radiation (Rn), sensible (H), and latent (LE) heat fluxes during the growing season, half-hourly measurements of Rn, H, and LE taken at five sites (SL, MF, PF, CL, and BL) over the period through May to October in 2014 were selected. Figure 6 gives the variations in daily averages of Rn, H, and LE over five typical surfaces (riparian forests of P. euphratica with associated T. chinensis, T. chinensis shrubland, cropland, and barren-land) in the Ejina Oasis. The ratios of monthly H and LE to Rn over these five surfaces are summarized in Table 4.
In Figure 6, the net radiation increased gradually from May, reached the maximum in July, and then declined gradually, starting at the end of August. The temporal variations in turbulent energy fluxes (H, LE) over the four contrasting vegetation covered surfaces showed very similar patterns for the growing season. The temporal pattern of the sensible heat flux showed maximum values in May and a gradual decrease through June to August, increasing slowly in early September (Figure 6b). In contrast, the latent heat flux exhibited a gradual increase from May to August as plants leafed out and quickly developed, and in mid-September it started to decline (Figure 6c). As shown in Table 4, the ratios of H/Rn for the four vegetation covered sites were relatively greater than that of LE/Rn in May and October. Nevertheless, during the period of June-September, a large proportion of Rn was partitioned into LE. For barren-land, more than 40% of net radiation was partitioned into sensible heat flux for the whole growing season. An energy deficit of 10% was still observed in EBEX-2000 even after considering all the possible influences [79]. The energy imbalance may be attributed to the mesoscale or secondary circulations induced by spatial heterogeneity [80][81][82][83]. The EC systems and radiometers used in the data analysis were inter-compared over an even and open Gobi Desert before the HiWATER intensive observation experiment was conducted [67]. Additionally, careful maintenances were performed for all instruments during the intensive observation period. Agam et al. [84] found spatial variation in the soil heat flux is remarkable under sparse/clumped vegetation covered sites, and inadequate spatial sampling of soil heat flux measurements in the observation sites may lead to the surface energy imbalance. Furthermore, an EBR larger than 1 may be attributed to the impact of wave phase difference in each surface energy balance component [85].
The average EBR in this study was similar to, or a little higher than, the results of previous studies in diverse ecosystems. For instance, the EBR was 0.86 for a barren surface in LITFASS-98 [86]; 0.7-0. 8 [78]; and 0.87-1.00 (annual value of 0.75) for a seasonally dry tropical forest in a semi-arid land of Brazil [76].
The ERB for the five selected EC sites was reasonably good as compared to the former results obtained from other ecosystems in semi-arid and arid regions, indicating the measurement systems provide reliable estimates of surface energy balance components. Thus, the energy imbalance in the EC flux measurements used in this study was not corrected by using any energy-balance correction methods [91].

Temporal Variations in Energy Flux Partitioning over a Riparian Forest Ecosystem
To investigate the variation in net radiation (R n ), sensible (H), and latent (LE) heat fluxes during the growing season, half-hourly measurements of R n , H, and LE taken at five sites (SL, MF, PF, CL, and BL) over the period through May to October in 2014 were selected. Figure 6 gives the variations in daily averages of R n , H, and LE over five typical surfaces (riparian forests of P. euphratica with associated T. chinensis, T. chinensis shrubland, cropland, and barren-land) in the Ejina Oasis. The ratios of monthly H and LE to R n over these five surfaces are summarized in Table 4.
In Figure 6, the net radiation increased gradually from May, reached the maximum in July, and then declined gradually, starting at the end of August. The temporal variations in turbulent energy fluxes (H, LE) over the four contrasting vegetation covered surfaces showed very similar patterns for the growing season. The temporal pattern of the sensible heat flux showed maximum values in May and a gradual decrease through June to August, increasing slowly in early September (Figure 6b). In contrast, the latent heat flux exhibited a gradual increase from May to August as plants leafed out and quickly developed, and in mid-September it started to decline (Figure 6c). As shown in Table 4, the ratios of H/R n for the four vegetation covered sites were relatively greater than that of LE/R n in May and October. Nevertheless, during the period of June-September, a large proportion of R n was partitioned into LE. For barren-land, more than 40% of net radiation was partitioned into sensible heat flux for the whole growing season.    There were distinct differences in R n , H, and LE over the five contrasting surfaces ( Figure 6). By comparison, the cropland had the lowest net radiation, followed by the barren-land, while the mixed forest exhibited the highest net radiation, followed by the shrubland and the P. euphratica forest (Figure 6a). No significant relationship was found between net radiation and elevation, as the terrain of the study area is almost even (Table 1), while cloud cover and types of clouds have considerable effect on the day-to-day variability of surface solar radiation, as well as the net radiation [75]. The downward and upward longwave radiation and albedo of the five surfaces were compared in Figure 7. In Figure 7a, the DLR increased gradually due to an increase in air temperature ( Figure 3b) and humidity (Figure 3c). There was very small difference in DLR between the five contrasting surfaces as a result of their similar air temperature. However, little difference in ULR between the five surfaces was obviously observed, particularly in July and August, indicating that the differences in surface temperature increases; and the vegetation covered surfaces emitted less longwave radiation than the barren-land (Figure 7b) as the surface temperature of the plants was lower, particularly for CL. This in turn is due to the fact the plants use a large part of the available energy for evapotranspiration, rather than to heat up their leaves [75]. Figure 7c presents the temporal variations of the albedo for the five surfaces. In addition to CL, the albedo decreased gradually as the vegetation canopy greened and grew, starting in May, and remained more or less steady in July, increasing sharply late in the season. The temporal pattern of the albedo had a great impact on the temporal trend in R n . The values of the albedo for the five surfaces differed markedly (Figure 7c). Furthermore, the cropland had much larger albedo as compared to the barren-land, mainly associated with the plastic mulch used on the fields, which would enhance the surface reflection and thus result in lower net radiation. Dramatic differences in albedo and surface temperature between the five surfaces provide a partial explanation for the observed differences in R n across sites (Figure 6a), as net effects of the lower albedo and lower surface temperature of the vegetated surface are responsible for the retainment of more energy [4]. The Cropland station (CL) was covered by the artificially planted cantaloupe. The first round of flood irrigation with groundwater over the fields was conducted before planting (about 11 June in 2014, Figure 6d) to provide ample soil moisture for crop growth (Figure 7d). For this irrigation period, the latent heat flux was much larger than the sensible heat flux (Figure 6c). In mid-June the cantaloupe began to plant. As a result of its rapid growth and subsequent second round of groundwater irrigation, LE increased dramatically and even exceeded to R n in August (Figure 6c and Table 4); meanwhile, many negative H values occurred, mostly attributed to significant advection of heat from the surrounding dry land [93]. As wet evaporation surfaces can become cooler than the air above, when this condition occurs, the sensible heat is extracted from the warm air and used in evapotranspiration [3,94]. This special phenomenon, known as the 'oasis effect', was dis- Significant differences in the energy partitioning into sensible and latent heat flux between the five surfaces can be found ( Figure 6 and Table 4). The sensible heat flux over the barren-land was much larger than latent heat flux throughout the growing season as compared to the vegetation covered surfaces, as a large part of the available energy was used for heating up the soil. In addition, the latent heat flux increased a little after rainfall events or flood irrigation by the water from the Heihe River (Figure 6d). Over the T. chinensis shrubland (SL), the sensible heat flux accounted for a large proportion of the net radiation (65-78%) in May and October, while during June to September, more net radiation was partitioned into latent heat flux, and the ratios of LE to R n ranged between 0.48 and 0.80. Values of LE/R n ratio for SL resembled data reported for other T. chinensis shrubland in the Tarim River Basin [92]. The temporal patterns of the sensible and latent heat fluxes at MF and PF were similar to the behavior observed over SL, but with much higher LE/R n values (Table 4). Consistent finding was reported by Yuan et al. [55] The above-mentioned results showed that the surface heat and water vapor exchange of riparian forest ecosystem was mainly controlled by the phenology of vegetation regardless of soil water content, as reported by Hao et al. [1] in the Inner Mongolia steppe.
The Cropland station (CL) was covered by the artificially planted cantaloupe. The first round of flood irrigation with groundwater over the fields was conducted before planting (about 11 June in 2014, Figure 6d) to provide ample soil moisture for crop growth (Figure 7d). For this irrigation period, the latent heat flux was much larger than the sensible heat flux (Figure 6c). In mid-June the cantaloupe began to plant. As a result of its rapid growth and subsequent second round of groundwater irrigation, LE increased dramatically and even exceeded to R n in August (Figure 6c and Table 4); meanwhile, many negative H values occurred, mostly attributed to significant advection of heat from the surrounding dry land [93]. As wet evaporation surfaces can become cooler than the air above, when this condition occurs, the sensible heat is extracted from the warm air and used in evapotranspiration [3,94]. This special phenomenon, known as the 'oasis effect', was distinctly observed in the oasis agroecosystem on clear days in summer [2,68,93,95].

Temporal Patterns of Evapotranspiration in the Oasis Riparian Forest Ecosystems
Time series of daily ET under five contrasting surface conditions for the 2014 growing season are shown in Figure 8. Between May and October, day-to-day variation in ET resulted from the temporal variation in atmospheric conditions (Figures 3 and 6a) and vegetation growth. The ET increased significantly on days when rainfall event occurred, particularly in June. The barren-land had the smallest ET values throughout the growing season, but showed a much stronger response to the rainfall events, with a rapid increase in ET on rainy days. The barren-land ET increased a little in mid-October due to flood irrigation by the surface water from the Heihe River (Figure 8). Over the vegetated surfaces, the daily sums of ET showed very similar patterns for the growing season, where ET values increased rapidly in May and reached the maximum in July, declining gradually at the end of August. The groundwater depth at the vegetated area increased gradually during the plant growing period until the groundwater was replenished greatly by large amount of surface water from the Heihe River in mid-September (Figure 9), according to the streamflow observed at Langxinshan gauging station ( Figure 1). It indicated that a gradual increase in the groundwater depth occurred in response to the uptake of groundwater by riparian forest ecosystem via evaporation and transpiration. This finding corresponds well with other reports on riparian forests given by Wang et al. [39]. As shown in Figure 9, the groundwater depth increased to some extent in mid-July, particularly observed at No.5 well nearby the river, as the surface water from Heihe River entered the key area of P. euphratica forest reserve. According to the streamflow observed at Langxinshan gauging station, this round of water allocation lasted about 10 days. Since 9 September, the amount of surface water entering the Ejina Oasis increased dramatically until late October. This revealed that the surface water from Heihe River and groundwater are two main water sources to sustain the natural growth of oasis riparian forests [53,96,97].

Evapotranspiration over A Riparian Forest Ecosystem with Different Land-Cover Types
The monthly cumulative ET for the five surfaces from May to October in 2014 is listed in Table 5. The monthly ET presented a distinctly temporal variation throughout the growing season, with higher values in July and August. The main reason is that the highest rate of plant growth occurs over these two months and vegetation water use increases dramatically [103]. In May, monthly ET values were much greater for the mixed forest (54-58 mm). By September, the monthly sums of ET declined considerably, partly due to the occurrence of leaf senescence resulting in general lower water use by the vegetation. In October, monthly ET from riparian forests was close to that of barren-land, attributed to the dormant and leafless P. euphratica trees and T. chinensis shrubs with no transpiration.
For five and a half months (11 May to 30 October) in 2014, the monthly ET over the barren-land increased from 14.6 mm in May to 52.2 mm in July and decreased to 20.65 mm in September for a total of 194.5 mm. For the irrigated cantaloupe cropland, the total ET from June to September was 431.5 mm, with mean ET of 3.8 mm d −1 , which is comparable to the results estimated from remote sensing based ET model by Zhou et al. [104]. Nevertheless, for the whole growing season, total ET was 545 mm, and mean daily ET value 3.4 mm d −1 over an irrigated maize field in the HRB [2]. In comparison, total ET was

Evapotranspiration over A Riparian Forest Ecosystem with Different Land-Cover Types
The monthly cumulative ET for the five surfaces from May to October in 2014 is listed in Table 5. The monthly ET presented a distinctly temporal variation throughout the growing season, with higher values in July and August. The main reason is that the highest rate of plant growth occurs over these two months and vegetation water use increases dramatically [103]. In May, monthly ET values were much greater for the mixed forest (54-58 mm). By September, the monthly sums of ET declined considerably, partly due to the occurrence of leaf senescence resulting in general lower water use by the vegetation. In October, monthly ET from riparian forests was close to that of barren-land, attributed to the dormant and leafless P. euphratica trees and T. chinensis shrubs with no transpiration.
For five and a half months (11 May to 30 October) in 2014, the monthly ET over the barren-land increased from 14.6 mm in May to 52.2 mm in July and decreased to 20.65 mm in September for a total of 194.5 mm. For the irrigated cantaloupe cropland, the total ET from June to September was 431.5 mm, with mean ET of 3.8 mm d −1 , which is comparable to the results estimated from remote sensing based ET model by Zhou et al. [104]. Nevertheless, for the whole growing season, total ET was 545 mm, and mean daily ET value 3.4 mm d −1 over an irrigated maize field in the HRB [2]. In comparison, total ET was The daily ET values at the five study sites were remarkably different ( Figure 8). The differences in ET of the five ecosystems during the growing period could be attributed primarily to the differences in vegetation coverage (Table 3) and soil moisture conditions (Figure 7d). Consistent findings have been reported in the grasslands and agricultural ecosystems [5] as well as the Leymus chinensis steppe [1]. For the cropland, the greatest values of ET occurred in August (7-8 mm d −1 ), associated closely with the highest rate of crop growth and abundant available water via flood irrigation with groundwater, which was applied four times throughout the whole crop growth period (Figure 8). The four rounds of flood irrigation through pumping the groundwater resulted in the decline of the groundwater depth until the flood irrigation with the water from the Heihe River was conducted over the cropland in mid-October (Figure 9).
The T. chinensis shrubland exhibited higher ET rates in July (6-7 mm d −1 ), where the atmospheric conditions were warm (reaching 25 • C) with high air humidity (averaged 30%) and ample soil moisture (~40%), see Figures 3 and 7d. The mixed forest of P. euphratica and T. chinensis (MF and PF) showed larger values of ET as compared to SL, particularly early in the season when the tree was leafed out and well developed. This difference is also observed by Yu et al. [58]. In the Ejina Oasis, a previous study shows that more than 90% of water transpired from P. euphratica trees comes from the groundwater, when its depth ranges between 2 and 2.5 m [98]. Here, the groundwater depth at the intensive observation area ranged between 1.5 and 3.5 m (Figure 9), which is within the optimal depth of 3-5 m for the survival of P. euphratica tree [96,99,100]. It showed that the groundwater table at this area was high enough to provide abundant available water for natural growth of P. euphratica in the reserve. In addition, greater soil moisture occurred at 0.8 m depth (36%) under the areas covered by the P. euphratica tree, presumably due to its highest absorbent root density at 0.8-1.0 m depth [101,102] and its ability to uptake deeper water sources [4].

Evapotranspiration over a Riparian Forest Ecosystem with Different Land-Cover Types
The monthly cumulative ET for the five surfaces from May to October in 2014 is listed in Table 5. The monthly ET presented a distinctly temporal variation throughout the growing season, with higher values in July and August. The main reason is that the highest rate of plant growth occurs over these two months and vegetation water use increases dramatically [103]. In May, monthly ET values were much greater for the mixed forest (54-58 mm). By September, the monthly sums of ET declined considerably, partly due to the occurrence of leaf senescence resulting in general lower water use by the vegetation. In October, monthly ET from riparian forests was close to that of barren-land, attributed to the dormant and leafless P. euphratica trees and T. chinensis shrubs with no transpiration. For five and a half months (11 May to 30 October) in 2014, the monthly ET over the barren-land increased from 14.6 mm in May to 52.2 mm in July and decreased to 20.65 mm in September for a total of 194.5 mm. For the irrigated cantaloupe cropland, the total ET from June to September was 431.5 mm, with mean ET of 3.8 mm d −1 , which is comparable to the results estimated from remote sensing based ET model by Zhou et al. [104]. Nevertheless, for the whole growing season, total ET was 545 mm, and mean daily ET value 3.4 mm d −1 over an irrigated maize field in the HRB [2]. In comparison, total ET was from 503.1 to 562.4 mm, and mean ET values from 3.47 to 3.54 mm d −1 in an irrigated maize field with plastic mulch (with mean LAI of 3.1) in the Hexi Corridor of northwestern China [34]. Differences in ET over irrigated crop fields are mainly affected by the atmospheric conditions (such as solar radiation and air temperature), crop phenology, and field management.
For the T. chinensis shrubland, monthly ET increased from 23.7 mm in May to 161.3 mm in July and declined rapidly to 86 [54]), which is smaller than our results. This difference might be attributed to the canopy structure and groundwater depth.
Over the mixed forest of P. euphratica trees with associated T. chinensis shrubs, total ET between May and October was from 622.4 to 731.4 mm and mean daily values from 3.6 to 4.2 mm. Our result was a little lower than the mean ET of the P. euphratica forest (4. 5 mm d −1 ) in the TRB (location: 40 • 25 59 N, 88 • 01 34 E; 844 m; [55]). Yu et al. [57,59] observed a total ET of 766.4 mm and a mean ET of 4.0 mm d −1 over a middle-aged P. euphratica forest within the same study area (location: 41 • 59 N, 101 • 10 E, 920 m), which is greater than the results in our study sites. One possible reason is that the transpiration rate is greater for the young tree as compared to the older tree [22,105].
All in all, for the five and a half months growing period in 2014, the total ET was highest for the mixed forest of P. euphratica trees with associated T. chinensis shrubs (622-731 mm), followed by the T. chinensis shrubland (541 mm); the total ET over the irrigated cantaloupe cropland from June to September (431.5 mm) was close to that of T. chinensis shrubland (492.1 mm). Thus, more concern should be raised regarding the large amount of water consumption of cropland in the Ejina Oasis. Since the allocation of water resources to the Ejina Oasis for restoring its natural ecosystem, the expansion of cropland can be considered an inappropriate use of water resources in the HRB [46].

Conclusions
Based on the HiWATER project, the exchanges of heat and water vapor over an oasis riparian forest ecosystem were measured by a network of EC systems installed over five representative surfaces. The energy balance closure for the EC flux measurements at the five study sites was generally good, with average EBR of 1.03 over the observation period between May and October in 2014. Throughout the growing season, the temporal variations in net radiation (R n ), sensible heat flux (H), and latent heat flux (LE) showed very similar patterns for the five study sites, with a gradual increase in R n and LE for the early season, followed by a gradual decline in the late season, while the daily H exhibited a gradual decline trend. Remarkable differences in R n and energy flux partitioning into H and LE over the five surfaces were related mostly to the soil properties (e.g., soil temperature and moisture), the phenology of the P. euphratica and T. chinensis, and the local cropland field management. About 70-90% of R n at riparian forest was partitioned into LE over the course of the middle growing season (June-August).
The daily ET showed significantly temporal variations for diverse surface types within the riparian forest ecosystem, with higher values in July and August. Moreover, the total ET showed considerable differences at a variety of surface types. For five and a half months in 2014, ET ranged 622-731 mm for the mixed forest of P. euphratica tree with associated T. chinensis shrub, with the average daily ET of 3.6-4.2 mm; for the T. chinensis shrubland, the total ET was close to 541 mm and the average daily ET was 3.6 mm, which were comparable to the results of other study sites in extreme arid regions of northwestern China. ET for the barren-land was approximately 195 mm. The total ET in irrigated cantaloupe cropland with plastic mulch was 431.5 mm for its four-month growing period with a total average of 3.8 mm d −1 . Clearly, more attention should be paid to the large amount of water consumption of cropland in the Ejina Oasis.
Investigating the energy and water vapor exchange in an oasis riparian forest is quite important to improve scientific understanding of land surface processes of the groundwaterdependent ecosystems in extreme arid regions. Additionally, this work will play a crucial role in developing the algorithms and improve the parameterization schemes of relevant remote-sensing based ET models and land surface process models. Quantification of ET over Ejina Oasis riparian forest ecosystem can provide fundamental indications for allocating the limited water resources reasonably for its ecosystem rehabilitation. Furthermore, this work will be extended to the water balance study of the oasis riparian forest ecosystems, which is significantly practical for hydrology modeling and basin water resource management.
Author Contributions: Conceptualization, W.W. and J.W.; methodology, F.X.; software, F.X.; validation, W.W., J.W. and F.X.; formal analysis, F.X.; investigation, F.X.; resources, F.X.; writing-original draft preparation, F.X.; writing-review and editing, W.W.; supervision, W.W. and J.W.; project administration, W.W.; funding acquisition, W.W. All authors have read and agreed to the published version of the manuscript. Data Availability Statement: All the meteorological data and the flux measurements from eddycovariance systems used in this study are available at http://data.tpdc.ac.cn on request (accessed on 12 October 2021).