Evapotranspiration dynamics and their drivers in a temperate mixed forest in northeast China

Site description

EC measurements were taken at the Yichun broad-leaved Korean pine forest experimental station (48.2292°N, 129.2661°E, 342 m.a.s.l) from January 2016 to December 2017. This station was located in the middle of Heilongjiang Province, which was the northeastern-most province of China (Fig. 1A). The experimental ecosystem had a flat topography with an evenly distributed mixed forest (14,141 hm⁻²). The mean fetch of the tower in all directions was lower than 400 m, with the prevailing southwest and northeast wind directions (Fig. 1B). The forest was primarily comprised of Korean Pine (Pinus koraiensis), Mongolian Oak (Quercus mongolica), Amur Linden (Tilia amurensis), and Maple Birch (Betula costata), with a density of 1,152 plants per hm⁻² and an average stand age of 220-years (Fig. 1C). Korean Pine (Pinus koraiensis), Mongolian Oak (Quercus mongolica), Amur Linden (Tilia amurensis), and Maple Birch (Betula costata) accounted for 10–15% of all plants in the study area, respectively. The mean canopy height was 26 m with a maximum and minimum LAI of 6.2 and 0.5 m² m⁻², respectively. The experimental forest had a temperate continental climate with a long-term mean annual air temperature (MAT) of 0.63 °C and a mean annual precipitation (MAP) of 610.7 mm. The highest T_a appeared in July with a long-term mean (1981–2010) value of 20.4 °C, while the lowest T_a occurred in January with a value of −22.5 °C. The greatest amount of precipitation occurred in July (147 mm), while the lowest precipitation occurred in February (6.7 mm). Precipitation was commonly seen as snowfall from November to April, and totaled approximately 90 mm year⁻¹. The soil in the experimental ecosystem was characterized as dark brown soil with high soil organic matter content.

Data measurements

We employed a flux tower with a height of 70 m to measure the ET and conventional meteorological variable dynamics.

The EC system was composed of a datalogger (Model CR5000; Campbell Scientific Inc., Logan, UT, USA), a 3-D sonic anemometer (Model CSAT3; Campbell Scientific Inc., Logan, UT, USA), and an infrared gas analyzer (Model LI-7500; Licor Inc., Lincoln, NE, USA). These were used to measure the net H₂O exchange between the terrestrial ecosystem and the atmosphere, which can be considered as the ET. The EC sensors, including CSAT3 and LI-7500, were mounted at a height of 50 m, which was about twice of the canopy height. The EC measurement sampled the raw data with a frequency of 20 Hz and recorded the flux data at 30-min intervals.

Conventional meteorological variables were also measured, including air temperature (T_a), relative humidity (RH), soil temperature (T_s), precipitation, and soil heat flux (G). T_a and RH were measured at different heights (1.5, 20, 30, 40, 50, 60 m) with shielded and aspirated probes (Model HMP45C; Campbell Scientific Inc., Logan, UT, USA). T_s was measured using thermocouple probes (Model 105T; Campbell Scientific Inc., Logan, UT, USA) at different depths (0, 5, 10, 15, 20, 40, 80, 160, 320 cm). Precipitation was recorded with a rain gauge (Model 52203; R.M. Young Company, Traverse City, MI, USA) at the height of 38 m. G was measured with two flux plates (Model HFP01SC; Campbell Scientific Inc., Logan, UT, USA) at depths of 5 and 15 cm, respectively. Conventional meteorological variables were measured at a frequency of 1 s and recorded at 30-min intervals with dataloggers (Model CR10X & CR23X; Campbell Scientific Inc., Logan, UT, USA).

Radiation data, including global radiation (R_g), reflected radiation (R_r), downward long wave radiation (DLR), and upward long wave radiation (ULR), were measured with a radiation heat balance station (CAMS620-SP, Huatron, China) at 1-h intervals. The measured 1-h data were linearly interpolated into half-hour scale to match the time series of ET and other environmental variables. Net radiation (R_n) was calculated as the sum of net short-wave radiation (R_ns) and net long-wave radiation (R_nl), where R_ns was the difference between R_g and R_r while R_nl was the difference between DLR and ULR.

Data processing

Data analysis

Uncertainty analysis

A simple Monte Carlo experiment was conducted to assess the uncertainty in the annual estimates of ET following Desai et al. (2005). A random number generator was used to remove 10–50% of the existing observed data. The generated data gaps were then randomly selected and filled with modeled ET (described in Flux processing), which was run 100 times for each year (2016 and 2017).

Statistical analysis

MATLAB 2014a (Mathworks Inc., Natick, MA, USA) was used to process the data including the flux quality control and gap filling. Linear regression and non-linear regression were used to analyse the effects of various factors on dynamics (including the diurnal and seasonal dynamics) of ET. Stepwise regression was used to analyze the multiple linear regression on ET seasonal variations. The minimum P-value for a variable to be added or removed from the model in the stepwise regression was 0.10. A path-analysis was conducted to evaluate the dependence of ET on various factors. The significant level was set to α = 0.05.

The energy balance closure

The energy balance closure indicated by the regression slope between the available energy (R_n− G) and energy fluxes (LE + H) (Fig. 2), suggested that our ecosystem experienced an energy imbalance. The slope between R_n− G and LE + H during the measuring period was 0.54, with an intercept of 13.87 W m⁻² (Fig. 2A), which was obviously lower than 1. Additionally, different seasons had divergent regression slopes (Figs. 2B and 2C). During the active growing season, the regression slope could be 0.56 with an intercept of 21.68 W m⁻² (Fig. 2B). Though the regression slope during active growing season was less than 1, it was still higher than that of the whole period (Fig. 3A) and the non-growing season, specifically (Fig. 2C). The regression slope during the non-growing season was only 0.39, with an intercept of 8.88 W m⁻² (Fig. 2C).

The energy balance closure, as indicated by the energy balance ratio (EBR), also reflected the energy imbalance of the studied ecosystem. During the measuring period, EBR was 0.62 (Fig. 2A), which was also clearly lower than 1. However, the EBR differed among seasons (Figs. 2B and 2C). During the active growing season, the EBR measured 0.66 (Fig. 2B), which was still lower than 1 but higher than that of the whole period (Fig. 2A) and the non-growing season (Fig. 2C). The EBR during the non-growing season only measured 0.47 (Fig. 2C).

The seasonal dynamics of ET and environmental factors

ET and its related environmental factors all exhibited obvious seasonal dynamics (Fig. 3).

Environmental factors including T_a, VPD, R_g, precipitation, and LAI all exhibited obvious single peak seasonal dynamics (Figs. 3A–3F). T_a ranged from −17 to 29 °C, with the highest values appearing at July, which was consistent across both years measured (Fig. 3A). VPD had higher values during the active growing season and lower values during the non-growing season; the highest VPD appeared in May (Fig. 3C). R_g ranged from 0.2 to 25 MJ m⁻² d⁻¹ and had similar dynamics to those of VPD (Fig. 3D). Precipitation also differed among seasons, with the most precipitation occurring from May to September (Fig. 3E). LAI ranged from 0.2 to 4.9 m² m⁻², with the highest values in July. RH did not show much variation across the months (Fig. 3B).

The seasonal dynamics of ET also exhibited a single-peak pattern, with the higher values appearing in the summer (Fig. 3F). However, the peak values of the seasonal dynamics of ET differed between the years measured (Fig. 3F). In 2016, ET achieved its peak (>6 mm d⁻¹) in the middle of July and then decreased. In 2017, ET showed an increasing trend until the end of August, and achieved its highest value for 2017 (>7 mm d⁻¹) at the end of August.

The diurnal dynamics of ET and environmental factors

ET and environmental factors all exhibited similar, obvious diurnal dynamics during the active growing season and non-growing season, but their values differed between seasons (Fig. 4).

During the day, T_a and R_g both showed single-peak patterns (Figs. 4A, 4B and 4G, 4H). These factors did not vary much from midnight to sunrise, but rapidly increased after sunrise, achieved their peak values around noon, and started to decrease at sunset. The factors were then stable until the end of the day. The time that T_a achieved its peak values (approximately 13:00 local time) was later than that of R_g. However, RH exhibited a concave shape (Figs. 4C and 4D), which indicated that RH began to decrease after sunrise. VPD did not vary much during the non-growing season but showed a single peak diurnal pattern in the active growing season (Figs. 4E and 4F).

There were also similar diurnal dynamics for ET (Figs. 4I and 4J). ET showed single-peak diurnal patterns during both seasons, with the peak values occurring at 12:00 pm. However, the peak values differed between seasons. During the non-growing season, the ET peak value was no more than 0.05 mm h⁻¹, while the ET peak values during the active growing season could be 0.28 mm h⁻¹. These results were consisting across both years measured (Figs. 4I and 4J). Additionally, the highest diurnal peak value for ET differed between measured years, with the peak value during non-growing season in 2016 higher than that of 2017 (Fig. 4I), while the peak value during active growing season in 2016 was lower than that of 2017 (Fig. 4J).

In addition, the diurnal variations of ET and environmental factors showed some fluctuations (Fig. 4), which may source from the unequal number of available data in each half-hour.

Drivers of ET dynamics

Environmental factors including T_a, R_g, and VPD were found to significantly drive the single peak diurnal patterns of ET, but their effects differed among factors and seasons (Fig. 5). During the non-growing season, the increasing T_a, R_g, and VPD were found to significantly affect ET, while their effects differed among factors. T_a exerted the strongest effects on ET diurnal variations during the non-growing season. The equations containing T_a explained 32% and 26% of the variations for ET in 2016 and 2017, respectively (Figs. 5A and 5D). The effects of R_g and VPD on the diurnal variations of ET during the non-growing season were similar but weaker than those of T_a. Therefore, T_a exerted a stronger effect on the diurnal variation of ET during the non-growing season. During the active growing season, the increasing ET was accompanied by an increase in T_a, R_g, and VPD, while their effects differed among factors. In 2016, T_a exerted a stronger effect on ET diurnal variations during the active growing season of 2016 (Fig. 5G), while R_g and VPD exerted weaker effects (Figs. 5H and 5I). The equation containing T_a explained 24% of the diurnal ET variations. In 2017, R_g had the strongest effect the diurnal variations of ET (Fig. 5J). The equation containing R_g explained 42% of the diurnal ET variations. T_a exerted a similar effect with R_g, which were both stronger than VPD (Figs. 5J–5L). Therefore, the diurnal ET variations during the active growing season were governed by T_a and R_g, respectively.

The seasonal variation of ET was primarily shaped by T_a and LAI, while R_g and VPD did not contribute much (Fig. 6). ET increased exponentially as T_a increased with an R² of 0.65 and 0.83 for 2016 and 2017, respectively (Fig. 6A). R_g only significantly increased ET in 2017 with an R² of 0.15 (Fig. 6B).Though the increasing VPD significantly promoted ET, the equations containing VPD explained little of the ET seasonal variations (Fig. 6C). LAI made great contributions to the seasonal variations of ET, and as it increased, ET increased linearly with an R² of 0.69 for 2016 and 0.82 for 2017, respectively (Fig. 6D).

The significant correlations between ET and the environmental variables (Fig. 6) indicated that the seasonal variations of ET were shaped by multiple factors. Stepwise analysis showed that variables entering the equation describing ET seasonal variations differed between years. In 2016, only T_aand LAI entered the equation describing ET seasonal variations, with an R² of 0.75 and an RMSE of 0.68 mm d⁻¹ (Eq. (2)). In 2017, all variables including T_a, R_g, and LAI entered the equation describing ET seasonal variations, with an R² of 0.86 and an RMSE of 0.66 mm d⁻¹ (Eq. (3)).

(2) $$\mathrm{E}\mathrm{T}=0.038T_a+0.604\mathrm{L}\mathrm{A}\mathrm{I}+0.467,{\mathrm{R}}^2=0.75,\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=0.68$$

(3) $\mathrm{E}\mathrm{T}=0.030T_a+0.031R_g+0.754\mathrm{L}\mathrm{A}\mathrm{I}-0.092,{\mathrm{R}}^2=0.86,\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=0.66$

Though many variables entered the equations describing the seasonal variations of ET (Eqs. (2) and (3)), each variable had a divergent role (Fig. 7). LAI exerted the strongest direct effect on the seasonal variation of ET. T_a and R_g also directly affected the seasonal variation of ET but showed a weaker effect than LAI. In addition, the seasonal variation of LAI was dominated by T_a, and supported by R_g and precipitation. The roles of environmental factors in the seasonal variation of ET were comparable in both years (Fig. 7). LAI was shown to exert a direct effect on the seasonal dynamics of ET in our ecosystem.

The annual amount of ET and its related variables

The mean annual ET during the measuring period was 528.26 mm year⁻¹, which were measured as 501.91 ± 5.30 mm year⁻¹ and 554.60 ± 11.24 mm year⁻¹ for 2016 and 2017, respectively (Table 1). The uncertainty of annual ET sourced from gap-filling was less than 2%. The active growing season values from May to October were the primary contributors to ET and accounted for over 90% of its annual value (Table 1). July had the highest ET, exceeding 120 mm month⁻¹ in both years, which was 1 month prior to the measurement of the highest precipitation (Table 1). However, June contributed the second highest ET in 2016 while August had the second highest ET in 2017.

The annual mean air temperature (MAT) were 0.57 and 1.17 °C for 2016 and 2017, respectively, which resulted from a hot summer from May to October and a cold winter from November to April. The annual R_g were not reported as many long data gaps during the measuring period, while VPD for 2016 and 2017 were only 0.33 and 0.37 kPa, respectively. The annual precipitation were 565 and 518.50 mm for 2016 and 2017, respectively. In addition, the annual mean LAI of 2016 and 2017 were 1.53 and 1.47 m² m⁻², respectively.

Therefore, this forest was shown to experience a high evaporate rate, which is crucial to the global water cycle.