Investigating the Patterns and Controls of Ecosystem Light Use Efficiency with the Data from the Global Farmland Fluxdata Network

Abstract

Ecosystem light use efficiency (ELUE) is generally defined as the ratio of gross primarily productivity (GPP) to photosynthetically active radiation (PAR), which is an important ecological indictor used in dry matter prediction. Herein, investigating the dynamics of ELUE and its controlling factors is of great significance for simulating ecosystem photosynthetic production. Using 35 site-years eddy covariance fluxes and meteorological data collected at 11 cropland sites globally, we investigated the dynamics of ELUE and its controlling factors in four agroecosystems with paddy rice, soybean, summer maize and winter wheat. A “U” diurnal pattern of hourly ELUE was found in all the fields, and daily ELUE varied with crop growth. The ELUE for the growing season of summer maize was highest with 0.92 ± 0.06 g C MJ⁻¹, followed by soybean (0.80 ± 0.16 g C MJ⁻¹), paddy rice (0.77 ± 0.24 g C MJ⁻¹) and winter wheat (0.72 ± 0.06 g C MJ⁻¹). Correlation analysis showed that ELUE positively correlated with air temperature (T_a), normalized difference vegetation index (NDVI), evaporative fraction (EF) and canopy conductance (g_c, except for paddy rice sites), while it negatively correlated with the vapor water deficit (VPD). Besides, ELUE decreased in the days after a precipitation event during the active growing seasons. The path analysis revealed that the controlling variables considered in this study can account for 73.7%, 85.3%, 75.3% and 65.5% of the total ELUE variation in the rice, soybean, maize and winter wheat fields, respectively. NDVI is the most confident estimators for ELUE in the four ecosystems. Water availability plays a secondary role controlling ELUE, and the vegetation productivity is more constrained by water availability than T_a in summer maize, soybean and winter wheat. The results can help us better understand the interactive influences of environmental and biophysical factors on ELUE.

1. Introduction

To reveal the ability of plants to convert radiation energy into carbohydrate at the ecosystem level, ecosystem light use efficiency (ELUE) is routinely defined as the ratio of gross primary production (GPP), or the above ground net primary production, to incident photosynthetically active radiation (PAR) [1,2,3]. Moreover, ELUE is an indicator of how sensitive photosynthetic production is to both environmental and physiological regulations [4], as well as an underlying basis for estimating the carbon cycle in the satellite-based light use efficiency models [5,6]. Herein, investigating the dynamics of ELUE and its controlling factors is of great significance for simulating ecosystem photosynthetic production.

ELUE is the combined effects of environmental controls, e.g., incoming solar radiation, air temperature, soil and atmosphere dry-wet conditions [7,8], physiological factors (canopy conductance and foliar age) [9], vegetation index [10,11], etc. Previous studies suggested that soil moisture, air temperature, vapor pressure deficit and crop management practice have significant influence on ELUE variability through regulating stomatal aperture and the related photosynthetic reaction [12,13], while some others concluded that vegetation indices (i.e., NDVI or LAI) are the dominant control factors on the process of carbon exchange by affecting the fraction of radiation absorbed by photosynthesis [14,15]. An analysis conducted at the 35-eddy covariance (EC) flux sites across various terrestrial ecosystems revealed that rainfall determined the inter-annual ELUE variation [7]. It seemed that the dominating factors of ELUE varied spatially among ecosystems and temporally over time [13,16]. Furthermore, the effects of one factor on ELUE can be masked by other covarying factors [7,17]. For example, Garbulsky et al. (2010) demonstrated that water stress significantly affects ELUE more than air temperature in the Mediterranean environment [7]. Therefore, a persistent challenge for simulating the carbon cycle and energy transformation in general has been the lack of systematically understanding the concerning factors and their interactions controlling ELUE [7,18]. Therefore, it is necessary to investigate the important factors that control the ELUE variation, which is significant for revealing how ELUE changes in face of climate change. Path analysis is a typical multivariate statistical analysis approach to study the relationship between multiple variables [19,20]. For example, Fei et al. (2019) revealed the critical roles of GPP and vapor pressure deficit in controlling ELUE by path analysis [21]; Jiang et al. (2020) quantified the effects of climatic factors and the leaf area index on crop canopy water consumption and carbon sequestration [22]. Therefore, the path coefficients in the analysis as standardized partial regression coefficients can allow us to examine the possible causal link between the independent variables and their relative effects on the dependent variable.

New measurement techniques, including EC-based method and model inversions based on EC, Geographic Information System (GIS) or Moderate Resolution Imaging Spectroradiometer (MODIS) surface reflectance products, are the most common methods in current research to investigate the dynamics of site-level and regional-level ELUE [23,24]. EC measurement determines carbon fluxes by the covariance between the vertical wind velocity pulsation and the concentration of CO₂ in the air mass [25]. The measurement provides a time-continuous and spatially integrated data set simultaneously with the advantage of minimal disturbance to the surrounding vegetation [26]. The open-access availability of fluxes and tower-based supplementary meteorology data provide important support to reveal the spatiotemporal variations and controlling variables of ELUE.

Cultivated croplands covers about 12% of the ice-free land surface on the earth, provides a great deal of the food and fiber for humans and has great potential for greenhouse gas mitigation by Grow-Harvest-Grow management of the global ecosystem [26,27,28]. Summer maize, winter wheat, paddy rice and soybean are the most widely produced grains for humans and livestock globally [24]. However, the complex interactions among the environmental and biophysical factors on ELUE have seldom been systematically analyzed in the agroecosystems. In this study, we used EC fluxes and meteorological data collected from 11 EC towers with 4 different coverages (maize, winter wheat, paddy rice and soybean) globally to analyze ELUE. The objectives of the paper are (1) to reveal the dynamics of ELUE diurnally and daily, (2) to explore the mechanisms of how environmental and physiological variables affect ELUE dynamics and (3) to quantify the relative effects of the control factors on ELUE through the path analysis. This study is important for predicting the process of carbon storage and grain production of croplands in the context of climate change.

2. Materials and Methods

2.1. EC Fluxes and Meteorological Data

The data from 11 EC flux sites from around the world with a measurement duration of at least two years (

Table 1. Descriptions of the eddy covariance (EC) flux sites of the croplands studied.

Crop Type	Site ID	Country	Latitude (º)	Longitude (º)	MAT (°C)	MAP (mm)	Period	References
Paddy rice	US-Twt	USA	38.11	−121.6	15.6	421	2011–2013	[12]
	JAN-MSE	Japan	36.05	140.02	13.7	1200	2004–2006	[33]
	IT-Cas	Italy	45.06	8.66	13.2	576	2007–2009	[34]
Soybean	US-Ne2	USA	41.16	−96.47	10.1	789	2002/2004/2006	[35]
	US-CRT	USA	41.62	−83.34	10.1	849	2011–2012	[36]
	US-Ne3	USA	41.17	−96.43	10.1	784	2002/2004/2006	[35]
Summer maize	FR-Gri	France	48.84	1.95	12.0	650	2005/2008/2011	[34]
	IT-BCi	Italy	40.52	14.95	18.0	600	2005–2007	[34]
	US-Ne1	USA	41.16	−96.47	10.1	790	2005–2007	[37]
Winter wheat	CH-Oe2	Switzerland	47.28	7.73	9.8	1155	2006–2007/2008–2009/2010–2011	[38]
	BE-Lon	Belgium	50.55	4.64	10.0	800	2004–2005/2006–2007/2008–2009	[39]
	FR-Gri	France	48.84	1.95	12.0	650	2005–2006/2009–2010/2011–2012	[34]

Note: MAT means annual temperature and MAP means annual precipitation in Table 1.

) were used for this study. The towers were located in four different crop (paddy rice, winter wheat, soybean and maize) fields. Hourly measured and derived EC fluxes as well as meteorological data over the growing seasons (Figure 1) were obtained from the Fluxnet 2015 Dataset (http://fluxnet.fluxdata.org/data/fluxnet2015-dataset/ accessed on 10 October 2019), including downward shortwave radiation (R_g, W m⁻²), net solar radiation (R_n, W m⁻²), air temperature (T_a, °C), vapor pressure deficit (VPD, hPa), precipitation (mm), latent heat flux (LE, W m⁻²), sensible heat flux (H, W m⁻²) and gross primary production (GPP, g C m⁻² d⁻¹ [29,30]. Data associated with equipment failures and stable boundary layer conditions (fraction velocity u_* < 0.25 m/s in this study) were removed [31,32]; then, the data gaps (≤2 h) were filled with linear interpolation while larger gaps (>2 h) were filled using mean diurnal variation (MDV). In the MDV method, a missing observation is replaced by the mean for that time period (half-hour) based on previous and subsequent days. Data windows of days 7 and 14 were usually chosen for averaging within the application [25]. Ultimately, hourly data was aggregated to daily data. Besides, data from rainy days (the amount of daily precipitation > 0.0 mm) were removed when performing path analysis and correlation analysis (see Section 2.4).

2.2. Normalized Difference Vegetation Index (NDVI)

The MODIS 16-day composite NDVI at a resolution of 250 × 250 m at 11 sites was downloaded (MODIS 13Q1, https://modis.ornl.gov/data.html accessed on 1 November 2019). Data with poor quality caused by the interference of clouds and precipitation were also removed. The spikes presented in the raw NDVI data were smoothed by removing unrealistic abrupt short-term changes in the NDVI. The gaps were filled by linearly interpolating the closest available data. Then, the 16-day NDVI was interpolated to a daily scale using a spline function for daily scale analysis.

2.3. Derived Variables

Canopy conductance (g_c), a vital physiological indicator that reveals the weighted integration of the individual leaf’s conductance, is calculated using an inversion of the Penman–Monteith equation (Monteith 1964):

$${\mathrm{g}}_{\mathrm{c}}=\frac{{\mathsf{\gamma }\mathrm{LEg}}_{\mathrm{a}}}{∆\left({\mathrm{R}}_{\mathrm{n}} -\mathrm{G}\right)-\left(∆+\mathsf{\gamma }\right){\mathrm{LE}+\mathsf{\rho }}_{\mathrm{a}}{\mathrm{C}}_{\mathrm{p}}{\mathrm{VPDg}}_{\mathrm{a}}}$$

(1)

$${\mathrm{g}}_{\mathrm{a}}={\left(\frac{\mathrm{u}}{{\mathrm{u}}_{*}^2 }+6{.2\mathrm{u}}_{*}^{-2/3}\right)}^{-1}$$

(2)

where Δ is the curve slope of the saturation vapor pressure (kPa °C⁻¹); γ is the psychrometric constant (kPa °C⁻¹); G is the surface soil heat flux (MJ m⁻² d⁻¹); g_a is the aerodynamic conductance (m s⁻¹); ρ_a is the air density (kg m⁻³); ${\mathrm{C}}_p$ is the air specific heat capacity (J kg⁻¹ K⁻¹); and $u_{*}$ is the friction velocity estimated by the EC system (m s⁻¹).

The evaporative fraction (EF) was calculated:

$$\mathrm{EF}=\frac{\mathrm{LE}}{\mathrm{LE}+\mathrm{H}}$$

(3)

For EF, we only used the values between 0 and 1 from the dataset.

ELUE was estimated by [40]:

$$\mathrm{ELUE}=\frac{\mathrm{GPP}}{\mathrm{PAR}}$$

(4)

where PAR is photosynthetically active radiation, and is set to half of R_g for some sites where PAR is not measured [41,42].

2.4. Path Analysis Theory

Multiple linear regression was adopted to investigate the relationship between ELUE and explanatory variables:

$${\mathrm{ELUE}=\mathrm{b}}_0{+\mathrm{b}}_{{\mathrm{x}}_1}{\mathrm{x}}_1{+\text{…}+\mathrm{b}}_{{\mathrm{x}}_{\mathrm{i}}}{\mathrm{x}}_{\mathrm{i}}{+\text{…}+\mathrm{b}}_{{\mathrm{x}}_{\mathrm{n}}}{\mathrm{x}}_{\mathrm{n}}$$

(5)

where x_i indicates the ith factor (one of NDVI, g_c, T_a, PAR, R_n, VPD and EF in this study); n (=7) is the number of the factors. R_n and PAR were chosen as the energy terms, although they are related: PAR relates more to photosynthesis, while R_n relates more to the energy balance and Penman–Monteith model. NDVI and g_c were chosen as the deputations of the crop structural and physiological factors, respectively. T_a and VPD were chosen since both of them were vital climate factors, while EF was chosen to analyze the effect of the energy allocation on ELUE.

A matrix based on Equation (5) can be expressed as:

$$\left[\begin{array}{c}{\mathrm{r}}_{{\mathrm{x}}_1}\\ {\mathrm{r}}_{{\mathrm{x}}_2}\\ \begin{array}{c}⋮\\ {\mathrm{r}}_{{\mathrm{x}}_{\mathrm{n}}}\end{array}\end{array}\right]=\left[\begin{array}{c}1\\ {\mathrm{r}}_{{\mathrm{x}}_2{\mathrm{x}}_1}\\ \begin{array}{c}⋮\\ {\mathrm{r}}_{{\mathrm{x}}_{\mathrm{n}}{\mathrm{x}}_1}\end{array}\end{array}\begin{array}{c}{\mathrm{r}}_{{\mathrm{x}}_1{\mathrm{x}}_2}\\ 1\\ \begin{array}{c}⋮\\ {\mathrm{r}}_{{\mathrm{x}}_{\mathrm{n}}{\mathrm{x}}_2}\end{array}\end{array}\begin{array}{c}\cdots \\ \cdots \\ \begin{array}{c}\ddots \\ \text{…}\end{array}\end{array}\begin{array}{c}{\mathrm{r}}_{{\mathrm{x}}_1{\mathrm{x}}_{\mathrm{n}}}\\ {\mathrm{r}}_{{\mathrm{x}}_2{\mathrm{x}}_{\mathrm{n}}}\\ \begin{array}{c}⋮\\ 1\end{array}\end{array}\right]·\left[\begin{array}{c}{\mathrm{b}}_{{\mathrm{x}}_{1\mathrm{y}}}\\ {\mathrm{b}}_{{\mathrm{x}}_{2\mathrm{y}}}\\ \begin{array}{c}⋮\\ {\mathrm{b}}_{{\mathrm{x}}_{\mathrm{ny}}}\end{array}\end{array}\right]$$

(6)

where $r_{x_i{x}_j}$ is the correlation coefficient between x_i and x_j (r_xjxi = r_xixj); $b_{x_{iy}}$ is the quantified direct effects of x_i on ELUE:

$${\mathrm{b}}_{{\mathrm{x}}_{\mathrm{iy}}}=\frac{{\mathsf{\sigma }}_{{\mathrm{x}}_{\mathrm{i}}}}{{\mathsf{\sigma }}_{\mathrm{y}}},(\mathrm{i}=1,2,\dots ,8)$$

(7)

where ${\sigma }_{x_i}$ and ${\sigma }_y$ represent the standard deviations of x_i and ELUE; $r_{x_i{x}_j}\cdot b_{x_{iy}}$ represent the quantified indirect effects of x_i through x_j on ELUE. The relative effects of unconsidered variables on ELUE were estimated:

$${\mathrm{b}}_{\mathrm{ey}}=\sqrt{1-{\displaystyle \sum }_{\mathrm{i}=1}^7{\mathrm{r}}_{{\mathrm{x}}_{\mathrm{i}}}·{\mathrm{b}}_{{\mathrm{x}}_{\mathrm{iy}}}}$$

(8)

2.5. Regression Analysis

The one-factor linear regression model was adopted to explore the responses of ELUE to NDVI, g_c, EF and P; meanwhile, an exponential curve was presented to describe the response of the ELUE to the days after a certain precipitation event.

$$\mathrm{ELUE}=\mathrm{ax}+\mathrm{b}$$

(9)

$${\mathrm{ELUE}=\mathrm{ax}}^{\mathrm{b}}$$

(10)

where a and b are fitted parameters; x represents the variables mentioned above.

3. Results and Discussion

3.1. Variation Patterns of GPP, PAR and ELUE on Different Time Scales

3.1.1. Diurnal Characteristics

Hourly averaged daytime GPP, PAR and ELUE during the growing seasons were shown in Figure 2a–d. A trough of ELUE occurred at around noon and peaks appeared at sunrise/sunset. ELUE varied from 0.07 to 0.13, 0.04~0.10, 0.07~0.14 and 0.05~0.10 mol C MJ⁻¹ in rice, soybean, maize and wheat fields, respectively. A synergistic relationship was found between GPP and PAR, of which the trends were opposite to ELUE. Evrendilek and Ben-Asher (2008) found that leaf LUE exhibited a bimodal behavior, which peaked in the early morning and late afternoon in wheat fields [43]. The midday depressions in net CO₂ assimilation have been reported because stomatal control of photosynthesis results from high T_a or VPD [44]. Besides, the assimilation of 1 mol of CO₂ requires 8 moles of photons at the leaf scale under ideal conditions [45]. However, the measured ELUE values are much lower than the theoretical ELUE values because they are constrained by many photosynthetic reactions [21,46]. Photosynthetic productivity usually saturates far below the maximum solar light intensity; in those conditions, many absorbed photons and the resulting electronic excitations of the pigment molecules can no longer be utilized for photosynthesis. To avoid photodamage, various protection mechanisms are induced that dissipate excess excitations, which otherwise could lead to the formation of harmful molecular species such as singlet oxygen. For example, non-photochemical quenching (NPQ) of chlorophyll fluorescence is thought to be an indicator of an essential regulation and photoprotection mechanism against high-light stress in photosynthetic organisms [47]. Ventre-Lespiaucq et al. (2018) revealed that the depression of light interception is a strategy for avoiding the interception of excessive irradiance on whole-crown and leaf scales [48]. These processes may be the other reasons causing a decline in ELUE at midday.

3.1.2. Seasonal Variations

The dynamics of daily GPP, PAR and ELUE in different fields during the growing seasons were illustrated in Figure 3. PAR in the rice, soybean, and maize fields did not show apparent seasonal changes, while the variations of GPP showed apparent seasonal cycles and had a peak in active growth stages. Daily ELUE of paddy rice, soybean and summer maize increased firstly, peaked during the active growing period and decreased as the crops matured (Figure 3a–c). Winter wheat seed germination occurred typically in the autumn; GPP had a small peak and then fell to near zero after the wheat entered the wintertime (due to lack of PAR and heat during the wintertime) [49]. Thus, ELUE presented quite different seasonal variations that increased firstly after germination and then fell as photosynthetic productivity (namely GPP) was limited due to an unfavorable growing environment (Figure 3d). ELUE exhibited the highest value in summer maize with 0.92 ± 0.06 g C MJ⁻¹, followed by soybean, paddy rice and winter wheat, with values of 0.80 ± 0.16, 0.77 ± 0.24 and 0.72 ± 0.06 g C MJ⁻¹, respectively (

Table 2. Growing season mean + standard deviation of gross primarily productivity (GPP) and ecosystem light use efficiency (ELUE) across the typical croplands.

Crop Type	GPP (g Cm−2 d−1)	ELUE (g C MJ−1)
Paddy rice	7.72 ± 2.10	0.77 ± 0.24
Soybean	7.80 ± 1.80	0.80 ± 0.16
Summer maize	9.76 ± 0.80	0.92 ± 0.06
Winter wheat	4.74 ± 0.48	0.72 ± 0.06

). Summer maize possessed the highest GPP of 9.76 g C m⁻² d⁻¹, followed by soybean and paddy rice, with the values of 7.80 ± 1.80 and 7.72 ± 2.10 g C m⁻² d⁻¹, respectively. Winter wheat had the lowest GPP, 4.74 ± 0.48 g C m⁻² d⁻¹, due to undergoing the wintertime during which physiological activity was weak under the lower temperature and PAR conditions [49] (approximately half the PAR of other ecosystems, Table 2). These findings suggested that the C₄ crop (summer maize) generally showed higher photosynthetic efficiency than C₃ crops (paddy rice, soybean and winter wheat), which was in line with previous reports [50,51]. Under similar climatic conditions, C₄ plants possessed a stronger ability of environmental adaptation, absorbing more carbon using radiation energy that resulted in their robust photosynthetic capacity and leaf-cell anatomy. Wang et al. (2018) found that the increase in both biomass production and CO₂ fixation with light intensity and CO₂ concentration in C₄ was faster than that in C₃ [24]. Therefore, the characteristics of C₄ plants determine a more efficient use of light and CO₂ than that of C₃ plants [50,52,53].

3.2. Responses of ELUE and GPP to Bio-Physical and Environmental Factors

3.2.1. Responses of ELUE and GPP to Vegetation and Physiological Factors

NDVI affected crop ELUE significantly with a linear relationship in all the crop fields (p < 0.01, Figure 4), with the highest Pearson correlation coefficient (r) of 0.87 (R² = 0.76) in soybean, followed by summer maize (0.81, R² = 0.66), paddy rice (0.79, R² = 0.62) and winter wheat (0.73, R² = 0.53). In addition, it seemed that NDVI was better fitted to ELUE with an exponential function than linear function, with the R² of 0.78, 0.67, 0.63 and 0.58 in soybean, summer maize, paddy rice and winter wheat, respectively. The consistent relationship showed the potential of utilizing NDVI to estimate ELUE as well as further GPP in the agroecosystems. Similar variation patterns of ELUE with NDVI were reported in other ecosystems [4]. Gitelson et al. (2014) examined the fraction of absorbed PAR (fAPAP) by the photosynthetic tissues in cropland sites and found that there was an exponential function relationship between NDVI and fAPAR [14]. The fAPAR increases when NDVI rises, and PAR is more effectively intercepted for carbon uptake, leading to an increase in ELUE. An ecosystem-level ELUE study showed that NDVI was quite consistent with photosynthesis dynamics and could be used as a proxy for both fAPAR and ELUE to predict GPP on a seasonal time scale [54,55]. Moreover, positive correlation relationships were found between NDVI and GPP (p < 0.01,

Table 3. Pearson’s correlation analysis between daily gross primary production (GPP) and the selected variables in different crop fields.

Crop Types	NDVI	gc	Ta	Rn	PAR	VPD	EF
Paddy rice	0.85 **	0.24	0.47 **	0.39 **	0.28 *	−0.06	−0.07
Soybean	0.91 **	0.68 **	0.27	0.37 *	0.17	−0.20	0.82 **
Maize	0.81 **	0.54 **	0.56 **	0.62 **	0.40 **	−0.14	0.67 **
Winter wheat	0.76 **	0.40 **	0.44 **	0.59 **	0.62 **	−0.28 *	0.42 **

Note: T_a represents daily mean air temperature; VPD is vapor pressure deficit; R_n refers to net solar radiation; GPP means gross primarily productivity; NDVI is normalized difference vegetation index; PAR represents photosynthetically active radiation; EF means evaporative fraction; G_c represents total plant canopy resistance. ** represents significant level p < 0.01,* represents significant level p < 0.05.

), with the r of 0.85, 0.91, 0.81 and 0.76 in paddy rice, soybean, summer maize and winter wheat, respectively. Strong seasonal and positive coupling of leaf area and photosynthesis occurred in annual crops, such as wheat [56,57] and peatland [58], where the findings demonstrated that the higher the LAI is, the higher the GPP is. These findings revealed the fact that the vegetation factor (NDVI) can be recognized as an essential parameter in ecosystem photosynthesis [24,56,59].

There is a significant relationship between ELUE and g_c in the soybean, summer maize and wheat fields (with r of 0.68, 0.56 and 0.41 (p < 0.01), respectively) but not in the rice fields (with r of 0.07) (Figure 5). Besides, GPP was also significantly positively related to g_c in the soybean (r = 0.68), maize (r = 0.54) and wheat (r = 0.40) fields (p < 0.01, Table 3). These results revealed that the physiological factor (g_c) can be well recognized as the control for crop photosynthesis in the typical agroecosystems, except for paddy rice. Vegetation physiological control of water and carbon fluxes between the atmosphere and vegetation canopy is exerted by stomata and the degree of control is quantified in terms of g_c [60]. g_c was conducted using an inversion of the Penman–Monteith equation as described in Section 2.2. The Penman–Monteith equation, which was termed as a “big-leaf” model to estimate ET, can generally obtain satisfying simulation results only in a closed canopy [61,62], when vegetation transpiration was dominant in the composition of evapotranspiration. However, only transpiration was tightly linked with the carbon exchange process [60]; therefore, g_c calculated in this study cannot be recognized as a reasonable control of the photosynthetic process due to the higher water evaporation in long-term flooded paddy rice croplands, where transpiration occupied a relatively smaller proportion of the time compared to other crops (wheat and corn [63]), though it is a common way to calculate g_c based on the inversion of the Penman–Monteith equation. The two-source approach, similar to the Shuttleworth–Wallace model [64], would be a better approach for paddy rice than the single source Penman–Monteith model to inverse g_c using the transpiration component; however, the test cannot be achieved due to the lack of transpiration data.

3.2.2. Responses of ELUE and GPP to Climatic Factors

Based on the linear regression analysis, ELUE was positively correlated to T_a across the four crop fields, but only significantly in paddy rice and summer maize, with the r of 0.48 and 0.50 (p < 0.01), respectively. ELUE was only significantly correlated to R_n in summer maize croplands (r = 0.36, p < 0.01), while no consistent relationship was found between PAR and ELUE in the four agroecosystems. However, GPP was significantly positively correlated to R_n in the fields (Table 3), while significantly related to T_a and PAR in the fields, except for the soybean fields. The results were consistent with previous reports that GPP generally increased linearly with rising T_a and light intensity (reflected by R_n, PAR) [65,66]. Lower light intensity and temperature could inhibit photosynthesis by influencing stomatal behavior and intrinsic biochemical reaction (e.g., thylakoid rection, Rubisco catalytic activity, electron transport capacity) [23]. Hence, the photosynthesis rate was lower during cold weather as was the ELUE [7,67]. However, the low affinity of the enzyme for CO₂ and its dual nature as an oxygenase limit the possible increase in net photosynthesis as temperature rises; although the catalytic activity of Rubisco increases, there can be a rapid fall-off of the photosynthetic rate at high temperatures [68,69]. For example, a quadratic regression result between ELUE and PAR was found in a semiarid savanna ecosystem [4]. However, for farmlands distributed at middle and high latitudes, the temperature usually did not reach the level that incurred a strong negative effect on photosynthesis. Research also demonstrated that T_a tended to exert a more significant positive influence on photosynthesis on colder sites [7]; for instance, T_a was not significantly correlated to the actual ELUE in hot humid ecosystems [18], while T_a determined the intra-annual variability of the ELUE in an energy-limited forest [7]. Therefore, the impact of climate control on carbon fixation will differ depending on the ecosystem types and vegetation growth conditions. More attention should be paid to studying differences in the photosynthetic limiting process according to different species and growth conditions.

3.2.3. Responses of ELUE and GPP to Water Availability

ELUE was significantly correlated to EF in the soybean (r = 0.83, p < 0.01), maize (r = 0.74, p < 0.01) and winter wheat (r = 0.37, p < 0.05) fields (Figure 6a–c). It was noted that ELUE had a negative relationship with EF in the rice fields (r = −0.07, Figure 6d). A similar relationship between GPP and EF was shown in Table 3. Such results indicated that water availability might not be a limiting factor in the photosynthetic process in long-term flooded paddy rice croplands.

ELUE and GPP were negatively related to VPD, while the relationship between VPD and ELUE in soybean (r = −0.36, p < 0.01), as well as VPD and GPP in winter wheat (r = −0.28, p < 0.05) was significant, respectively. Yuan et al. (2015) found there was a negative correlation between VPD and GPP in the maize and soybean fields [53], since increased VPD may trigger stomatal closure to avoid excess water loss due to the high evaporative demand of the air, leading to a negative carbon balance [70]. Besides, we also found that VPD did not fully characterize the impacts of water availability on vegetation production except for winter wheat. Conversely, EF can better explain the water availability effects on the variability of ELUE and GPP for all but paddy rice in our analysis. EF is estimated by the evaporative fraction (EF = LE/(LE + H)), since increasing the amounts of energy partitioned to evaporated water indicates a greater potential of soil in the water supply [7,71]. Therefore, EF has a more significant correlation with vegetation production than VPD in the agroecosystems, except for paddy rice. VPD tends to exert the effect of atmospheric water restraint but ignore water supply from the soil [72]. Yuan et al. (2015) also indicated that VPD easily decouples with GPP in crop fields [53]. Therefore, VPD might not be a suitable indicator to represent the effect of water availability on photosynthesis [71], though there might be a strong covariance between VPD and soil moisture in semi-arid areas [73].

The relationship between ELUE, GPP, PAR and total precipitation during the crop-growing season was investigated (Figure 7). An upward trend was found between ELUE and precipitation in summer maize and winter wheat, while a downward trend was found in paddy rice and soybean sites (Figure 7a). GPP significantly decreased with increasing total precipitation in paddy rice (r = 0.77, p < 0.05), while GPP increased with increasing total precipitation in the other three agroecosystems (Figure 7b). Increasing precipitation was thought to reduce VPD, which increases GPP by enhancing g_c [74]. Therefore, ELUE rises when precipitation increases and VPD decreases. However, the negative relationship between GPP and precipitation was found in paddy rice in the current study, which was contrary to the other three agroecosystems as well as previously published reports on arid and semiarid regions [75]. Presumably because the effect of precipitation on soil water content (SWC) was weak in long-term flooded paddy rice croplands, more precipitation may indicate additional clouds and the downward trend of PAR; therefore, vegetation photosynthesis may be restricted due to the lack of PAR [7,76,77].

The effect of precipitation on the temporal variations of ELUE was analyzed, and only the data during the active growing season was considered to exclude the effects of growth phenology changes on ELUE. Figure 8 shows that ELUE increased sharply, which may be caused by the decline in PAR when there is rainfall during cloudy conditions, since then ELUE gradually decreased. These results are the first evidence showing that the ELUE variation trend is approximately proportional to the square root of inverse time after a precipitation event; a low and relatively steady ELUE was found during a long period without precipitation or irrigation, reflecting the fact that a soil water deficit will suppress ELUE due to declining GPP (Figure 9). The effect of precipitation on ELUE is mainly in the way of replenishing soil moisture content. The relationship between SWC and ELUE was simple and coherent, exhibiting a trend in which ELUE increased with increasing SWC during the active growing season. Previous studies have concluded water stress will reduce the carbon exchange between the ecosystem and atmosphere physiologically as a result of the combination of stomatal, mesophyll conductance and biochemical limitations [8,78,79]. Such results can provide a theoretical basis for the development of carbon flux estimation and crop yield prediction models in water-limited environments.

3.3. Comparison of Relative Effects of Critical Factors on ELUE by Path Analysis

The path analysis showed that the controlling variables considered in this study can account for 73.7%, 85.3%, 75.3% and 65.5% of the total ELUE variation (Figure 10), while relative effects of unconsidered variables on ELUE (i.e., nutrition content in soil/leaf, CO₂ concentration, etc.) still accounted for 26.3%, 14.7%, 24.7% and 34.5% in the rice, soybean, maize and winter wheat fields, respectively.

The quantified standardized total effects of the controlling variables on ELUE in four agroecosystems are shown in Figure 11. NDVI is the most critical controlling variables for ELUE in all the fields. Previous researchers also demonstrated that ELUE variation patterns were mainly determined by vegetation index (LAI, NDVI, EVI etc.) [4,7]. Campoe et al. (2013) reported that the seasonal variations of GPP were mainly exerted by LAI on carbon assimilation [80]. Analysis done by a process-based model also showed that vegetation indices had significant positive effects on GPP [81], as canopy development can influence the biophysical properties of vegetation, carbon and the absorption of energy as discussed in Section 3.2.1 [59,82,83].

In addition to NDVI, there was one variable (T_a) in the rice fields, three variables (EF, g_c, and VPD) in the soybean fields, four variables (EF, g_c, T_a and R_n) in the maize fields and two variables (EF and g_c) in the wheat fields that significantly affected ELUE (p < 0.01, Figure 11). Crop growth sometimes suffered adverse environmental conditions, such as high T_a, low water availability, etc. T_a was generally considered as an essential driver of GPP when analyzing biomes and increasing importance in the coldest and energy-limited areas [7,84]. Therefore, the warming climate may increase GPP and ELUE in an energy-limited area. However, we found that the total effects of EF on ELUE were higher than that of T_a among the typical crops, except for paddy rice, which indicated that water availability was more important than T_a in controlling agroecosystem ELUE. A similar conclusion was also found in the Mediterranean environment, where water stress significantly affected ELUE inter-annual variations, while T_a marginally affected ELUE [85]. Our analysis supports the idea that ELUE is constrained by water availability more than temperature at the typical crop sites, except for paddy rice. Water availability affects nearly all aspects of plant growth and most physiological processes. Meanwhile, the stress response depends on intensity, rate, duration and the stage of plant growth [7,71,73]. Plants close their stomates, and thereby lower g_c, to prevent further water loss by transpiration in the early drought stages [86]. Consequently, the supply of CO₂ for the carboxylation through stomatal is reduced and thereof GPP. Studies also demonstrated the decrease in APAR was likely caused by photoprotective mechanisms such as changes in leaf inclination/leaf rolling as the cumulative deficit on soil water [10,87]. However, the long-term effect of soil water deficit on canopy assimilation is a limitation in canopy structure development (i.e., LAI, NDVI), even shortening the growing season, and thus reduces the seasonal cumulative ecosystem productivity and the consequent ELUE [88]. However, those effects of water stress seldom or never occurred in long-term flooded paddy rice croplands. Therefore, T_a might exert more effect on ELUE than water availability in paddy rice.

4. Conclusions

Based on the 35 site-years CO₂ flux measurements from 11 cropland sites globally, we investigated the variation characteristics of multi-timescales ELUE as well as the effects of the environmental factors (i.e., T_a, R_n, PAR, EF, VPD, and P) and vegetation physiological factors (i.e., NDVI, g_c) on ELUE variation across the four typical agroecosystems that are widespread on earth, including paddy rice, soybean, maize and winter wheat. Specifically, a trough of hourly ELUE was found across the four typical agroecosystems. The variability in daily ELUE exhibited apparent seasonal dynamics with crop growth, and the summer maize croplands possessed a higher ELUE than the other three C₃ agroecosystems. The crop-specific variation is in control of ELUE; overall, the variations in ELUE and GPP were positively related to the variations of NDVI, EF (except for that in paddy rice croplands), g_c, T_a and R_n, while the opposite relationship was found with VPD. No significant consistent relationship was found between total growing season precipitation and ELUE or GPP. The ELUE variation trend is approximately proportional to the square root of inverse time after a precipitation event across the different crop sites. The results showed that NDVI is the most confident estimator for ELUE in the four ecosystems. Water availability plays a secondary role controlling ELUE, and vegetation productivity is more constrained by water availability than T_a in summer maize, soybean and winter wheat. This study improves our understanding of the influence of bio-physiology and environmental factors on ELUE.