The study was carried out in a managed poplar (Populus euramericana cv. “74/76”) plantation at the Daxing Forest Farm, which is located in the southern suburbs of Beijing, China (116∘15′07′′ E, 39∘31′50′′ N, 30 m a.s.l.). The trees were planted in 1998 with 3 m × 2 m spacing; dead or low-vigor trees were replaced with new saplings in 2001 and 2003. The stand characteristics over the 4 years of study are provided in Table 1. At the end of 2009, the average height of the trees was 16.2 ± 1.6 m (mean  ± SD), and the diameter at breast height (DBH) was 14.1 ± 1.6 cm. The average leaf area index (LAI) of the stand increased over time. During the growing season, understory shrubs were kept at low density by manual removal. Perennial herbs included Chenopodium glaucum Linn., Medicago sativa L., Melilotus officinalis (L.) Lam., Salsola collina Pall., and Tribulus terrestris L.

The local climate is classified as a sub-humid warm temperate zone, with a mean (1990–2009) annual temperature of 11.6 ∘C; maximum and minimum temperatures are 40.6 and -27.4 ∘C, respectively. The annual precipitation ranges from 262 to 1058 mm (1952–2000), with an average of 556 mm, of which 60–70 % falls from July to September (Daxing Weather Station, 116∘19′56 E, 39∘43′24 N). The annual frost-free period lasts 209 days, and the total hours of sunshine reaches 2772 h per year with 15.5 MJ m-2 d-1 of incoming solar radiation. The average wind speed is 2.6 m s-1and it mostly comes from the southeast (during the growing season) and the northwest (outside of the growing season).

The study area is on the alluvial plain of the Yongding River, and is flat with an average slope of < 5∘. The top 2 m of the soil is mostly composed of well-drained fluvial sand with a bulk density of 1.43–1.47 g cm-3, and a pH of 8.25–8.39. The soil porosity is about 40 % and capillary porosity is 32 %. The mean groundwater depth over the past 9 years (2001–2009) was 16.5 ± 0.2 m, and declined at an average rate of 0.6 m per year. The maximum pan evaporation occurs from May through June, exceeding precipitation for the same period. Severe drought during the beginning of the growing season (from April to June) in northern China is common. The site was irrigated using pumped groundwater, and the amount of water supplied was estimated from the water meter records at the three adjacent wells on a weekly basis from 2006 through 2009. Other management practices have included tilling and weeding since the establishment of the plantations.

The micrometeorological and eddy flux measurements were conducted at a 32 m tower in the center of the study site, which was established in June 2005. The footprint of the eddy flux covariance system, was about 1 km × 1 km in size, with a fetch of at least 300 m in all directions. Fluxes of CO2, sensible heat, and latent heat were calculated based on eddy-covariance (EC) principles. The sensors included a CO2/H2O infrared analyzer (Li-7500; LI-COR, Inc., Lincoln, NE, USA) and a three-dimensional sonic anemometer (CSAT-3; Campbell Scientific, Inc., CSI, UT, USA). The anemometer head was installed towards a predominant wind direction (southeast), and the IRGA was installed at a slight vertical angle tilted northward (< 20∘) between the sonic path and anemometer body. The IRGA was calibrated every year. The EC sensors were mounted initially at a height of 16 m in 2006. This was increased to about 18 m before the start of the growing season in 2007, and again to 20 m in February 2009 to ensure that the sensors remained well above the tree canopy.

Net radiation was measured with net radiometers (Q7.1, REBS, Seattle, WA, USA) and (CNR-1; Kipp and Zonen, Delft, Netherlands) at 26 m above the ground. Photosynthetically active radiation (PAR) was measured with a PAR quantum sensor (LI-190SB; LI-COR, Inc.) mounted at 20 m. The atmospheric pressure was measured by a barometric pressure sensor (CS105, CSI) at a height of 21 m. Air temperatures and humidity were measured with a temperature and relative humidity probe (HMP45C; Vaisala, Helsinki, Finland) at 5, 10, 15, and 20 m above ground. Precipitation was measured with a tipping bucket rain gauge (TE525-L; Texas Electronics, USA) at 22.5 m. Soil heat flux and soil temperatures were each measured with three soil heat transducers (HFT3, CSI) and three thermocouples (TCAV107; CSI) located at depths of 5, 10, and 20 cm below the soil surface. Soil water content was measured with TDR sensors (CS616; CSI) buried at 20 and 50 cm.

With the exception of the rain gauge, all microclimatic data were recorded with a data-logger (CR23X; CSI) at 30 min intervals and the fluctuations in wind speed, sonic temperature and CO2 and H2O concentrations were sampled at 10 Hz and recorded by a CR5000 data-logger (CSI).

The 30 min mean fluxes were calculated from raw 10 Hz data with an EC Processor software, version 2.3 (Noormets et al., 2010). The program is designed for reprocessing EC flux data and can calculate half-hour mean eddy-covariance fluxes of carbon, water, and energy. The wind coordinates were rotated using the planar fit method (Paw U et al., 2000; Wilczak et al., 2001). Fluxes were corrected for additional sensor heating (Burba et al., 2008) and fluctuations in air density (Webb et al., 1980). The data quality controls included: screening of the 30 min mean eddy-covariance fluxes based on instrument quality flag, integral turbulence characteristics (Foken and Wichura, 1996), flux stationarity, atmospheric stability, and adequate turbulent mixing (Goulden et al., 1996). The threshold of friction velocity (μ*) below which flux loss occurred was determined from the seasonal binned relationship between nighttime turbulent flux of CO2 and friction velocity (μ*; Schmid et al., 2003). The threshold was consistent across different seasons, but differed slightly between years: 0.18 (2006), 0.12 (2007), 0.14 (2008) and 0.13 m s-1 (2009). Data gaps shorter than 2 h were filled using linear regressions between the flux of interest and net radiation (Rn), gaps between 2 h and 7 days in length were filled using the mean diurnal variation (MDV) method (Falge et al., 2001), and gaps longer than 7 days were not filled.

The 4-year study period was distinctively classified into “wet” and “dry” years. A dry year referred to a year with annual precipitation less than 85 % of the 20-year average according to the National Standard of the People's Republic of China (GB/T 20481-2006; China, 2006) and “wet” when above this average. Years 2007 and 2008 were classified as `wet' while 2006 and 2009 were `dry' years. We focused on the growing season when the environmental forcing (e.g., solar radiation and temperature) for energy and water fluxes and the physiological response of vegetation were usually strong. In this study, the strongest forcing days occurred approximately between day 100 (mid-April) and day 300 (late October). The daytime was defined as the period between sunrise and sunset with PAR > 4 µmol m-2 s-1. The regulation of surface energy and gas exchange is often different during nocturnal periods (Mahrt, 1999), with heat fluxes at night typically weaker and markedly less stationary than during the daytime (Wilson et al., 2002b). Midday was defined as the period from 10:00 to 15:00 LST when the coupling between vegetation and the atmosphere was the strongest.

The availability of relative extractable water (REW) content was calculated to analyze the ecosystem response to drought stress. According to Granier et al. (2007), soil water stress was assumed to occur when the REW dropped below the threshold of 0.4. Daily REW is calculated as REW=VWC-VWCminVWCmax-VWCmin, where VWCmin and VWCmax are the minimum and maximum soil volumetric water content, respectively, across the 4 years.

The Bowen ratio (β) reflects the influence of microclimate and the hydrological cycle on the energy partitioning and water use of the ecosystem (Perez et al., 2008). The midday β is calculated as Eq. (2): β=HLE. Based on the daytime half-hourly and daytime totals of turbulent energy fluxes, the energy balance ratio (EBR) is calculated as Eq. (3): EBR=∑(H+LE)∑(Rn-G-S), where S is the latent and sensible heat storage in the air-column below the EC system and is calculated as in Eq. (4) (Dou et al., 2006): S=∫0hcρcp∂T∂tdz+∫0hcρcpγ∂e∂tdz, where hc is the height of the eddy flux system measurement (32 m), T is air temperature in the air-column below hc, and e is water vapor pressure.

During midday periods (10:00–15:00 LST), the Penman–Monteith approximation was inverted to calculate the surface resistance (Rs) in Eq. (5) (Kumagai et al., 2004): Rs=ρcp(δe/LE)γ+Δγβ-1Ra, where Rs is the surface resistance to water vapor transport (s  m-1), representing four components: bulk stomatal resistance of the canopy, bulk boundary layer resistance of the vegetation, bulk ground resistance, and bulk boundary layer resistance of the ground (Admiral et al., 2006; Cho et al., 2012; Perez et al., 2008; Wilson et al., 2002b).

Ri is the climatological resistance (s m-1) indicating the atmospheric demand (Wilson et al., 2002b) and is calculated as Ri=ρcpδeγA, where A is the available energy (Rn-G); ρ is air density (kg  m-3), cp is the specific heat of the air (1005 J kg-1 K-1); δe is the atmospheric vapor pressure deficit (Pa); LE is the latent heat flux; Δ is the change of saturation vapor pressure with temperature (Pa K-1); γ is the psychrometric constant (≈ 67 Pa K-1); and β is the Bowen ratio.

Ra is the aerodynamic resistance of the air layer between the canopy and the flux measurement height (s m-1), that reflects the aerodynamic properties of turbulent transport in the near-surface boundary layer (Holwerda et al., 2012; Zhang et al., 2007). Ra is calculated following Hossen et al. (2011) and Migliavacca et al. (2009): Ra=ra, m+rb=μμ∗2+6.2μ∗-2/3, where ra, m is the aerodynamic resistance for momentum transfer, rb is the quasi-laminar boundary-layer resistance, μ is the wind speed, and μ* is the friction velocity.

The decoupling coefficient (Ω) explains the degree of coupling between the atmosphere and the vegetation, and describes the relative control of evapotranspiration by surface resistance and net radiation (Pereira, 2004). The Ω value ranges from 0 to 1, with values approaching zero indicating that LE is highly sensitive to surface resistance and ambient humidity deficit. The Ω value approaching 1 indicates that LE or evapotranspiration is mostly controlled by net radiation (Jarvis and McNaughton, 1986): Ω=Δ+γΔ+γ(1+RsRa). The equilibrium evaporation (LEeq) is the climatologically determined evaporation (atmospheric demand) over an extensive wet surface and is dependent only on Rn and temperature. It is calculated as LEeq=Δ(Rn-G)Δ+γ. The ratio LE / LEeq, which is also known as the Priestley–Taylor α, reflects the control of evaporation by atmospheric and physiological factors. LE / LEeq characterizes the surface dryness of an ecosystem, indicating whether soil water supply for evapotranspiration of an ecosystem was limited. An LE / LEeq of < 1 indicates water stress and suppressed evapotranspiration. Conversely, LE / LEeq > 1.26 indicates unrestricted water supply, and only available energy limits evapotranspiration (Arain et al., 2003). The LE / LEeq is dependent on the leaf area index (LAI), soil water content, meteorological conditions (e.g., wind speed, solar radiation, VPD (vapor pressure deficit, air stratification stability, convection, and advection surface resistance), vegetation types, and altitude (Guo et al., 2008).

Repeated ANOVA (SPSS) measurements were used for quantifying the changes of all the biophysical variables, energy fluxes, and bulk parameters across years. The t test was used to compare the differences of biophysical variables among different studies. The partial correlation analysis was used to distinguish the impacts of each of the three resistance parameters (Rs,Ri, and Ra) on the Bowen ratios. All analyses were accessed at α= 0.05.

The annual precipitation rates in the 4 years of study differed from the long-term (i.e., 1990–2009) average (556 mm yr-1). Thus, years 2006 and 2009 were drier and 2007 and 2008 were wetter than the mean (Table 1). The interannual contrast was exaggerated by the seasonality of rainfall. Generally, over 90 % precipitation of each year occurred in April–October, but with different timing and magnitude among the years. The study site was irrigated during the dry years of 2006 and 2009 to mitigate drought conditions (Fig. 1). Seasonal drought stress (REW < 0.4) occurred during periods in the late growing season of 2006 and 2009, the spring of 2007 and 2009, but not at all in 2008 (Fig. 2a–d). In 2006, precipitation during the growing season reached 467 mm, of which 51 % had occurred by July. The amount of irrigation was 35 mm in April, 21 mm in May, and 30 mm in September. The two separate seasonal drought periods were #1_06 (from DOY – day of year – 164 to 192) and #2_06 (from DOY 231 to 300). The total rainfall in 2007 and 2008 was similar, but more evenly distributed throughout the year in 2008. In 2007, drought stress occurred during DOY 110–143 (#1_07) and 151–200 (#2_07). A single rain event in late May (57 mm) and a few large precipitation events (> 25 mm d-1) in July were recorded. The amount of rainfall in 2009 was the smallest among the 4 years, during which 195mm of irrigation was applied from March to September. There were several short and scattered droughts across the growing season in 2009 (Fig. 2d). Despite the higher than normal rainfall in the 2 wet years, there was no flooding or overland runoff.

The growing season Ta in 2008 was significantly lower than in 2007 and 2009 (dT = 1.3 ∘C, p < 0.05, Fig. 2e–h). The years differed in the spring warm-up and the timing of peak temperature (by up to 35.9 ∘C). The maximum air temperature occurred in June 2006 and 2009, and in July 2007 and 2008. The warmest month was June 2006 (27.1 ± 2.4 ∘C).

The daytime average VPD of the four growing seasons (Fig. 2e–h) was 1.3 ± 0.7 kPa. The mean VPD in wet years (i.e., 2007 and 2008) was 1.2 ± 0.7 kPa, which was significantly lower (F= 6.093, p < 0.01) than in dry years (i.e., 2006 and 2009, 1.3 ± 0.8 kPa). The VPD of the growing season in 2008 (i.e., 1.1 ± 0.5 kPa) was lower than in the other years (p < 0.05). Higher Ta and lower precipitation in May 2007 led to higher VPD compared to the same period in 2006 and 2008 (p < 0.001). Furthermore, the VPD was the highest in June 2009 (i.e., 2.3 ± 1.1 kPa, p < 0.05) and the lowest in 2008 (i.e., 1.0 ± 0.5 kPa, p < 0.01).

The energy partitioning trends of daytime total net radiation (Rn) into latent, sensible heat fluxes (LE and H), soil heat fluxes (G), and heat storage of canopy (S) for the years 2006–2009 were presented in Fig. 3. Among these years, Rn varied with solar radiation (R > 0.95, α = 0.01 level), reached the maximum in July, and gradually decreased until late October (in dry years) or November (in wet years). During the growing season, there were no significant differences in average daytime total Rn between wet and dry years. The average daytime total G during the growing season displayed great seasonal and annual differences among the years (p < 0.05), with a lower value in wet years (2.1 % in 2007) than in the dry years (4.9 % in 2006; p < 0.001). Moreover, the average values of daytime total S among the four growing seasons were 0.46, 0.49, 0.51, and 0.54 MJ m-2, respectively. S/Rn varied from 6.0 % in 2007 to 6.8 % in 2009, showing no differences between the wet and dry years.

Partitioning of Rn into LE and H differed significantly between the wet and dry years (Table 3; F= 17.599, p < 0.001). The dominant turbulent energy flux during the early growing season was sensible heat flux (H) with or without drought stress, except in 2006 when the irrigation was applied (Table 3). Then LE was the dominant driver of energy partitioning during the middle and late growing seasons under drought stress. The average daytime total LE was about 20 % greater in wet years (6.77 MJ m-2) than in dry years (5.72 MJ m-2, p < 0.01). The timing of peak LE was weakly related to drought, peaking in July in 2006, 2008, and 2009, and in August in 2007. The peak values of daytime total LE were 16.61, 17.01, 19.72, and 16.27 MJ m-2, in 2006–2009 respectively. The daily evaporative fraction (LE/(Rn-G)) was significantly higher in wet years (60.3 and 64.8 % in 2007 and 2008, respectively) than in dry years (57.1 and 50.4 % in 2006 and 2009, respectively; p < 0.05).

The seasonal variation of the midday Bowen ratio (β) displayed a rapid and significant trend across the growing season, especially at the beginning (April–June) and the end (September–October) of the growing season (Fig. 4). The Bowen ratios during the middle of the growing seasons were all smaller than 1, and approximately lasted from DOY 180–250 in the dry year and from DOY 180–290 in the wet years. The average midday β in the dry years was greater (1.57) than in the wet years (0.83; F= 19.176, p < 0.001). The Bowen ratio showed differences in response to drought stress across the four growing seasons (Table 3); with much higher values (> 1) during the drought periods in 2007 and 2009, but not in 2006. The Bowen ratio was smaller than 1 during drought-stressed periods in 2008.

The Rs varied widely at the beginning and the end of growing season, but changed steadily within a low range during the middle of growing season by comparison. Moreover, these lower Rs values in the dry years lasted for a shorter period (DOY 190–250) than in the wet years (Fig. 5a). A significantly negative relationship was found between the Rs and LAI during the wet years (Fig. 6). Overall, the seasonal average of Rs normalized by leaf area index (LAI; i.e., Rs:LAI) was lowest during the wettest year (2008, 54.1 s m-1 LAI-1; p < 0.05). The Rs:LAI in the dry years (106.8 s m-1 LAI-1) was 50 % higher than in the wet years (71.2 s m-1 LAI-1; p < 0.001). The Rs:LAI in the seasonal drought-stressed periods in 2006, 2007, and 2009 were much higher than those during unstressed periods (p < 0.001, Table 3).

The average midday Ri peaked in June and decreased in July/August before reaching a second peak in October (Fig. 5b). The seasonal average Ri during the growing season ranged from 68.3 to 77.9 s m-1, with a mean value of 74.4 s m-1, and showed no difference among the four growing seasons (p > 0.05). Figure 5c presents the seasonal and annual variations of midday Ra. The average Ra for the four growing seasons was 23.2 ± 8.5 s m-1, ranging from 10.6 to 43.5, 9.7 to 52.5, 6.5 to 43.1, and 9.7 to 74.5 s m-1, from 2006 to 2009, respectively. Ra in 2007 was significantly higher than in the dry years (p < 0.01), while Ra in 2008 was smaller than in the dry years (p < 0.001). However, there were no significant differences between dry and wet years Ra (p > 0.05).

The seasonal changes of LE / LEeq values varied between 0.4 and 1.0 during most of the growing seasons (Fig. 5d). The average LE / LEeq of the 4 years were 0.76, 0.73, 0.89, and 0.63, respectively. The mean LE / LEeq of the dry years (0.68) was lower than that of wet years (0.81; p < 0.001). Specifically, the value of LE / LEeq in drought periods of 2007 and 2009 were much smaller. A significantly exponential relationship existed between the LE / LEeq and Rs during the growing season (Fig. 7).

The decoupling coefficient (Ω) across the growing seasons peaked in mid-July in 2008 and in early August in the other years (Fig. 5e). The mean Ω for the 4 years was 0.41, 0.46, 0.43, and 0.39 (Table 3), respectively, and was significantly higher in wet years (0.45) than in dry years (0.40; F= 9.460, p < 0.01). Compared to the value during unstressed periods, the decoupling coefficient during the seasonal drought periods (#1_06, #2_06; #1_07, #2_07 and #1_09, #2_ 09, #3_09) was much lower in value.

The energy balance ratio (EBR) in the current study was 0.88 based on daytime 30-min fluxes and > 0.96 based on daytime totals (Table 2). The annual mean EBR at our site was similar to the values of eight ChinaFlux sites, which averaged 0.83 and ranged from 0.58 to 1.00 (Li et al., 2005). The energy budget is also consistent with the 50 site-years of flux data from 22 FLUXNET sites, which had energy closure of 0.34–1.69 (Mean = 0.84, Wilson et al., 2002a). A recent analysis of 173 FLUXNET sites also found an average closure of 0.84 (Stoy et al., 2013), although the authors also detected consistent differences among the biomes based on metrics of landscape heterogeneity. In addition to the known reasons for decreasing energy balance closure (Hernandez-Ramirez et al., 2010; Li et al., 2005; Nakai et al., 2006; Stoy et al., 2013), management operations at our site (e.g., irrigation, tilling, and partial felling) may also affect the energy balance. Although the causes of surface energy balance closure continue to be debated (Stoy et al., 2013) and will not be conclusively answered in the current study, the results reported here are similar to other FLUXNET sites.

The surface energy partitioning to sensible and latent heat depends on water potential gradient and surface resistance (Arain et al., 2003; Baldocchi et al., 2000; Chen et al., 2009). Canopy development (Guo et al., 2010), rainfall dynamics and irrigation (Ozdogan et al., 2010) affect these properties to some extent and could directly lead to a change in soil moisture and the evaporation component of LE, thereby impacting energy partitioning and β (Chen et al., 2009; Ozdogan et al., 2010). However, the impact of precipitation on the Bowen ratio may vary, even at the same site (Tang et al., 2014). In our study, detectable responses of LE/(Rn-G) and the Bowen ratio to drought stress and non-stress periods were observed in response to soil water supply (Table 3) with a 50 mm threshold on average (Fig. 8). The variability of energy partitioning during the growing season was highly sensitive to water availability from precipitation and irrigation. On an annual scale, the Bowen ratio appeared linearly related to the total growing season precipitation (R2= 0.89, p < 0.05). Thus, the Bowen ratio is very responsive to the site water supply. A similar finding was reported by Grünwald and Bernhofer (2007) in a temperate spruce forest.

By contrast, β varied from 0.18 to 0.71, with a mean of 0.35 ± 0.15 during most of the growing season in 2008 and in the non-stressed periods of the other 3 years. This variation was close to 0.42 for deciduous forests (Wilson et al., 2002b) and 0.55 in a temperate Douglas fir (Humphreys et al., 2003), which is also similar to the variations in a ponderosa pine forest in the western United States (Goldstein et al., 2000) and a deciduous broadleaved forest in the southern United States (Wilson and Baldocchi, 2000). Seasonal drought stress had a discernible impact on the Bowen ratio of this poplar plantation. However, compared to the reported β values such as 0.74 in a temperate mixed forest (Wu et al., 2007), 0.81 in a boreal Scots pine forest (Launiainen, 2010), and 0.89 in a loblolly pine plantation (Sun et al., 2010), the average β values in wet years were close to the above values. β was higher in seasonal drought periods and dry years than most temperate coniferous forests (Mean = 1.07, Wilson et al., 2002b), which typically had higher β values. The high β value in this study reflects the semiarid conditions, and suggests a low tree water supply which might be resulted from the combination of low rainfall, the sandy soil's low water-holding capacity and the high plant and atmospheric water demand. It has been suggested that the large-scale establishment of poplar plantations in sandy semiarid regions of northern China could have an adverse impact on the region's groundwater reserves (Li et al., 2014; Petzold et al., 2011). Our findings corroborate the hypothesis that drought would trigger significant changes in the energy partitioning of water-demanding poplar species in a water-stressed region.

The Bowen ratio is dependent on the interactions of climatic and biological factors (Perez et al., 2008; Wilson and Baldocchi, 2000). Ri quantifies the climatic control on energy partitioning and tends to decrease the Bowen ratio. A higher Ri implies a warm and dry climate in continental regions (Raupach, 2000; Wilson et al., 2002b). Rs reflects the physiological control on surface energy exchange of an ecosystem (Costa et al., 2010; Launiainen, 2010; Zhou et al., 2010) and generally increases the Bowen ratio.

In this study, Rs similarly varied seasonally with plant phenology and showed similar seasonal characteristics to other deciduous forests during the course of the growing season (Cabral et al., 2010; Kutsch et al., 2008; Li et al., 2012). As reported by Tchebakova et al. (2002), Rs in seasonal drought-stressed periods was much higher than in non-stressed periods. It has been shown that drought stress during the canopy development affects leaf area and may have lasting effects on canopy gas exchange through the entire growing season, even after the moisture limitation is removed (Noormets et al., 2008), which may explain significant differences in Rs between the wet years of 2007 and 2008 (Fig. 9). Compared with the Rs in other studies, the Rs:LAI in dry years in the current study was close to that of the Euphrates Poplar (Populus euphratica Oliv.; 130.2 s m-1 LAI-1) and smaller than that of the Gansu Poplar (Populus gansuensis Wang et Yang; 189.4 s m-1 LAI-1) in semiarid regions (Chen et al., 2004). In wet years it was similar to that of poplar (58.6 s m-1 LAI-1; Wilson et al., 2002b) and boreal aspen during the full-leaf period (51.8 s m-1 LAI-1; Blanken et al., 1997) in mesic temperate regions. Rs is primarily driven by solar radiation, moisture availability, and VPD (Fernández et al., 2009; Li et al., 2012), and modulated by leaf area and stomatal resistance, which in turn changes as a function of the above factors (Wilson and Baldocchi, 2000). Compared to the strong correlation between Rs and LAI in wet years, the increased scatter in the Rs–LAI relationship during dry years (Fig. 6) suggests that Rs in dry years was also influenced by other physiological and non-physiological (e.g., soil evaporation, canopy structure, and turbulence) factors (Wilson et al., 2002b). The mean Ri in the current study was higher than the mean Ri reported for temperate forests in Wilson et al. (2002b; t= 5.91, df= 741, p < 0.001), but was ∼ 50 % lower than the value reported by Li et al. (2009) in a vineyard in Gansu province in China (t= -29.87, df= 741, p < 0.001), as might be expected given the predominant climatic conditions.

On the seasonal scale, the Bowen ratio and Rs of this poplar plantation were correlated and consistent with Wilson et al. (2002b) and Li et al. (2009), but differed in wet and dry years (Fig. 10). The water limitation during the dry years manifested in disproportional increase in Rs than the Bowen ratio; this response may serve as an indicator as to when water reserves are being depleted. At the extremes, the relationship converges, but as water becomes limiting, stomatal closure and increased Rs do not appear to be able to affect the seasonal dynamics of the Bowen ratio. The partial correlation analysis indicated that Rsand Ri had strong positive and negative effects, respectively, on β in both wet and dry years (Table 4), which could not be detected through correlation analysis (e.g., the impact of Ri and Ra on β). Furthermore, the regulation of the Bowen ratio by Rs and Ri seemed stronger in dry than in wet years. Ra had a significant negative impact on the Bowen ratio in wet years, but not in dry years.

The average LE / LEeq in the growing season was 0.74 at our site, which is similar to deciduous forests (0.72; Wilson et al., 2002b), but smaller than at a temperate broad-leaved forest (0.82; Komatsu, 2005). The average Ω value of 0.42 ± 0.22 (0.39–0.46) was close to the other forests (0.26–0.4, Wilson and Baldocchi (2000); 0.25–0.43, Motzer et al. (2005)). As essentially implied by the Penman–Monteith equation, LE / LEeq exponentially related to Rs during the growing season (Fig. 7), which is equivalent to the logarithmic relationship between LE / LEeq and Gs (surface conductance) reported by other studies (Chen et al., 2009; Hossen et al., 2011; Zhu et al., 2014). The asymptotic value of LE / LEeq in dry years (0.89) and wet years (0.96) were both lower than the 1.1–1.4 range typical in temperate deciduous forest reported by Monteith (1995), indicating that our study site was drier than these reference sites during both dry and wet years. The low LE / LEeq values under dry surface conditions of the ecosystem in this study may also be related to the high porosity of the sandy soil and the low ground water table (Zhao et al., 2013). Overall, as indicated by the lower Ω values and the significant correlation coefficients between LE / LEeq and Rs, Rs was the major factor controlling the LE during the growing season, which was consistent with the relations between Rs and the Bowen ratio. In addition, LE was more coupled to the atmosphere during the dry years and seasonal drought periods across the growing seasons, as reported in other studies (Bagayoko et al., 2007; Bracho et al., 2008; Zha et al., 2013).

As forestry is a long-term endeavor, with the economic payback decades from stand establishment, the availability of resources for the stand to prosper should come naturally to natural resource managers. Supplementing limited resources directly (fertilization, irrigation) or indirectly (competition control, site preparation, thinning) is commonplace in commercial forestry, but it has to be sustainable in the broader context of the region's ecosystems and livelihoods. As we show in a forthcoming study, the water needs of poplar plantation exceed the annual precipitation in the region, and plant survival during dry years depends on irrigation from groundwater. In the current study, energy partitioning to latent and sensible heat and surface resistance was sensitive to climatological drought – even under irrigation – as indicated by low LE / LEeq (< 1) and low values of the decoupling coefficient (Ω; Zhu et al., 2014); the dry surface conditions dominated the poplar plantation in both wet and dry years. In wet years, the plantation itself is in hydrologic balance with the water that arrives as precipitation, with evapotranspiration consuming nearly all of the precipitation. The same is true in dry years, but irrigation increases ET even further by depleting groundwater. Even if the plantations were in hydrologic balance with water delivered as precipitation, their existence and operation could be a threat to adjacent ecosystems and livelihoods if those rely on runoff or groundwater recharge from the areas where the plantation has been sited. In the absence of the plantations it is likely that groundwater recharge would increase, especially given the sandy textured soil that tends to allow rapid infiltration and percolation as well as limit moisture delivery to the atmosphere directly from the soil surface itself. While poplar plantation growth in this water-limited location might be sustained by the modest precipitation in the region, it could still be unsustainable for the broader context of the region's ecosystems and livelihoods. However, further study to truly access these effects is needed by comparing the surface water balance and/or spatial and temporal variations of groundwater levels at an adjacent, similar site without a plantation.