Long-term trend in vegetation gross primary production, phenology and their relationships inferred from the FLUXNET data

Journal of Environmental Management 246 (2019) 605–616

Contents lists available at ScienceDirect

Journal of Environmental Management journal homepage: www.elsevier.com/locate/jenvman

Research article

Long-term trend in vegetation gross primary production, phenology and their relationships inferred from the FLUXNET data

T

Xiaojun Xua,b,c,∗, Huaqiang Dua,b,c, Weiliang Fana,b,c, Junguo Hub,∗∗, Fangjie Maoa,b,c, Hao Donga,b,c a

State Key Laboratory of Subtropical Silviculture, Zhejiang A & F University, Lin'an, 311300, Zhejiang, China Key Laboratory of Carbon Cycling in Forest Ecosystems and Carbon Sequestration of Zhejiang Province, Zhejiang A & F University, Lin'an, 311300, Zhejiang, China c School of Environmental and Resources Science, Zhejiang A & F University, Lin'an, 311300, Zhejiang, China b

A R T I C LE I N FO

A B S T R A C T

Keywords: Gross primary production Phenology Physiology Long-term trend Prediction

Climate-induced changes in plant phenology and physiology plays an important role in control of carbon exchange between terrestrial ecosystems and the atmosphere. Based on dataset during 1997–2014 from 41 flux tower sites (440 site-years) across the northern hemisphere, relationships between long-term trends in start of growing season (SOS), end of growing season (EOS), length of growing season (LOS), maximal gross primary production (GPPmax), and seasonal and annual gross primary production (GPP) were analyzed. Statistical Models of Integrated Phenology and Physiology (SMIPP) were built for predicting the long-term trends in annual GPP. Results showed that SOS advanced and EOS delayed for forest sites, while both SOS and EOS for grassland (GRA) sites delayed. Long-term trends in SOS and EOS of evergreen needle-leaf forests (ENF) sites were greater than those of deciduous broadleaf forests (DBF) sites. Seasonal and annual GPP for forest sites increased, among which long-term trend in annual GPP of ENF sites was the largest. Spring GPP of GRA sites decreased, but annual GPP increased. Strong relationships between long-term trends in phenological and physiological indicators and seasonal GPP were found. Long-term trend in GPPmax had the highest relationship with long-term trend in annual GPP for forest sites, but long-term trend in SOS was the most related to long-term trend in annual GPP for GRA sites. Increases in spring and autumn GPP due to a one-day advance in SOS and delay in EOS for DBF sites were greater than ENF sites. Delay in EOS resulted in more carbon sequestration than advance in SOS for forest sites, while advance in SOS significantly increased spring GPP for GRA sites. The SMIPP model driven by longterm trends in LOS and GPPmax had stronger explanatory power for predicting long-term trend in annual GPP than the SMIPP model driven by long-term trends in SOS, EOS, and GPPmax. Long-term trend in annual GPP was accurately predicted by using the SMIPP model, while long-term trend in annual GPP for GRA sites was more difficult to be captured than the forest sites. Drought and disturbance effects on phenology and physiology were major factors for model uncertainty. This study is helpful to understand changes in phenology and carbon uptake and their differences among different vegetation types and provides a potential way for predicting annual rate of change in carbon uptake through vegetation photosynthesis at a global scale.

1. Introduction Interactions between the forest and atmosphere through physical, chemical, and biological processes can reduce or amplify anthropogenic climate change (Bonan, 2008; Nolan et al., 2018). Increase in carbon sequestration by temperate forests through warming-induced changes in phenology plays an important role in reducing the rate of temperature rise (Keenan et al., 2014). Conversely, carbon emissions from vegetative ecosystems caused by extreme climate events (Angert et al.,

∗

2005; Zhao et al., 2010) and land use/cover change (Christidis et al., 2013; Huang et al., 2018; Pavani et al., 2018) can amplify climate change. An increasing trend of temperature and the corresponding change in the range of the annual average temperature caused by carbon loss in vegetative ecosystems was discovered (Angert et al., 2005; Zhao et al., 2010). Hence, understanding long-term trends in carbon sequestration by vegetative ecosystems is of great importance to evaluate the contribution of vegetation to climate change mitigation and predict the future climate-carbon cycle feedback.

Corresponding author. Zhejiang A & F University, 666# Wusu street, Lin'an, Zhejiang Province, 311300, China. Corresponding author. E-mail addresses: [email protected] (X. Xu), [email protected] (J. Hu).

∗∗

https://doi.org/10.1016/j.jenvman.2019.06.023 Received 27 January 2019; Received in revised form 4 June 2019; Accepted 6 June 2019 Available online 14 June 2019 0301-4797/ © 2019 Elsevier Ltd. All rights reserved.

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

2. Materials and methods

Interannual variability of the terrestrial carbon sequestration, which is strongly related to vegetation phenology and physiological processes (Xia et al., 2015), is large in a range from about 0.5 Gt carbon source in 1987 to 4.0 Gt carbon sink in 2011 (Le Quéré et al., 2013). Phenology and physiology together explained over 90% of interannual variability of annual gross primary production (GPP) (Xia et al., 2015; Zhou et al., 2016). The seasonal carbon anomalies were also strongly related to respective phenological and physiological indicators (Keenan et al., 2014; Zhou et al., 2016), such as spring GPP anomaly vs. start of growing season (SOS) anomaly, autumn GPP anomaly vs. end of growing season (EOS) anomaly, and summer GPP anomaly vs. the maximal GPP (GPPmax) anomaly. Among three indicators, the GPPmax played the dominant role in control of interannual variability of GPP (Zhou et al., 2017). However, the combined explanatory ability of the three indicators to interannual variability of annual GPP varied among different ecosystem types (Zhou et al., 2016). The three indicators better the interannual variability of annual GPP for non-forest ecosystems than deciduous broadleaf forests (DBF) and evergreen needle-leaf forests (ENF) (Zhou et al., 2016). The interannual variability of annual GPP for ENF sites was largely explained by the physiological indicator, but the phenological and physiological indicators contributed substantially to interannual variability of annual GPP at DBF and nonforest sites (Zhou et al., 2016). Sensitivities of interannual variability of seasonal and annual GPP for DBF sites to the phenological indicators were greater than that for ENF sites (Richardson et al., 2010; Keenan et al., 2014; Zhou et al., 2016). Compared with interannual variability of GPP, studies on relationships between plant phenology and physiology and long-term trends in annual GPP were relatively rare (Zhou et al., 2017). Significant temporal changes in SOS, EOS, GPPmax and annual GPP have been presented in many studies (Tao et al., 2006; Keenan et al., 2014; Wu et al., 2016). Advances in SOS and delay in EOS for the period 2001–2012 were detected based on remotely sensed daily greenness indices and the long-term ground observations (Keenan et al., 2014). A general trend of increasing GPP during both spring and autumn was also discovered across sites (Keenan et al., 2014). Previous studies showed that the long-term trend in carbon uptake varied during different periods and in different places due to effects of extreme climate events (Angert et al., 2005; Zhao et al., 2010; Munger et al., 2012). The long-term trend in annual GPP was mainly determined by the GPPmax, whereas the contribution of GPPmax to the long-term trend in annual GPP is lower than that to interannual variability of GPP (Zhou et al., 2017), which implied that responses of the three indicators to the long-term trend in annual GPP and interannual variability of GPP may be different. Overall, the ability of long-term trends in the three indicators in explaining the longterm trend in annual GPP was still unclear among different places, vegetation types and temporal phases. It is necessary to further analyze relationships between long-term trends in the three indicators and annual GPP and to build prediction models for the long-term trend in annual GPP. In this study, long-term trends in seasonal and annual GPP, phenological and physiological indicators for DBF, ENF, mixed forest (MF) and grassland (GRA) sites were calculated using data from 41 flux tower sites (440 site-years) across the Europe, Asia, and North America. The objectives were: (1) to analyze long-term trends in seasonal and annual GPP, phenological and physiological indicators for different plant functional types (PFTs); (2) to discover relationships and responses of long-term trends in seasonal and annual GPP to phenological and physiological indicators for different PFTs; (3) to build the prediction models for long-term trend in annual GPP using phenological and physiological indicators and test extrapolation ability of the models.

2.1. Flux tower data Eddy covariance measurements used in this study were collected from the FLUXNET2015 dataset, which can be downloaded from the FLUXNET-Fluxdata website (https://fluxnet. fluxdata.org/). The FLUXNET2015 dataset has been quality-controlled and gap-filled according to the standard protocols. First, the half-hourly net ecosystem exchange (NEE) is corrected for storage and de-spiked (Papale et al., 2006). Second, NEE is filtered with variable friction velocity threshold for each year (Barr et al., 2013; Papale et al., 2006), then gapfilled with the marginal distribution sampling method (Reichstein et al., 2005). Third, NEE was partitioned in the GPP and Ecosystem Respiration using a daytime based approach (Lasslop et al., 2010). The daily GPP in the FLUXNET2015 dataset was used in this study. For each year, the seasonal and annual values of GPP were calculated. Spring is defined as the months of March–May; summer is defined as the months of June–August; autumn is defined as the months of September–November. In total, 41 sites covering 440 site-years of data were used (Table S1). The data records ranged in length from 6 to 18 years. The vegetation at these sites was classified into four PFTs, with 12 DBF sites (n = 126), 17 ENF sites (n = 189), 7 GRA sites (n = 64), and 5 MF sites (n = 61), respectively. The 440 site-years of data were selected according to the following two criteria. (1)The sites with at least 6 siteyears of data and locating in the northern hemisphere were selected. (2)The site-years with more than 80% of the daily GPP data, which were measured or good quality gapfilled data with high confidence according to the quality flag, were selected (Zhou et al., 2016). The quality flag used in this study indicates percentage of measured and good quality gapfilled data. The definition of good quality was shown in Reichstein et al. (2005). Detailed description of each site is shown in supplementary Table S1. 2.2. Determination of phenological and physiological indicators Time series of the daily GPP were smoothed by using a SavitzkyGolay filter based on the Matlab R2013a. Savitzky-Golay filter has been successfully used for reconstructing a high-quality normalized difference vegetation index time series dataset (Chen et al., 2004). Filter coefficients of the Savitzky-Golay filter were determined by an unweighted linear least-squares regression and a second-order polynomial model. The span for the moving average is 91 days in order to produce relatively smooth and stable daily GPP values (Chen et al., 2004; Wu et al., 2017). The GPPmax equals to the maximum value of the smoothed daily GPP. The SOS and EOS were determined using a fixed threshold approach, as the date at which the smoothed daily GPP reached 25% of the mean amplitude for all years for that site (Melaas et al., 2013; Keenan et al., 2014). The amplitude was calculated as the GPPmax minus to the minimum of smoothed daily GPP values. The minimum of smoothed daily GPP values were set to zero if smoothed daily GPP values were negative. The length of season (LOS) was calculated as EOS minus to SOS. An example for determination of phenological and physiological indicators was shown in Fig. 1. 2.3. Determination of long-term trends in phenological date, GPPmax, and GPP A linear regression model was used to determine long-term trends in phenological date, GPPmax, and seasonal and annual GPP, which has been widely used to determine time trends of changes in phenology (Schwartz et al., 2006; Tao et al., 2006; Piao et al., 2010; Keenan et al., 2014), temperature and carbon uptake (Angert et al., 2005). The model is formulated as:

Yi, t = αi + βi × t + εi, t 606

(1)

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

Fig. 1. Example showing how start of growing season (SOS), end of growing season (EOS), and maximal gross primary production (GPPmax) were derived from daily gross primary production (GPP) time series; multi-years averaged flux data are indicated with black circle, and smoothing spline by solid red line; 25% amplitude threshold dates are indicated with vertical lines; minimum and maximum of smoothed GPP are indicated with horizontal lines. Data are averaged daily GPP from 2000 to 2014 for the FLUXNET site US- UMB (deciduous broadleaf forest). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

of GRA sites increased by a rate of 0.127 d y−1 (Fig. 2). EOS of all PFTs increased with rates of 0.111 d y−1, 0.289 d y−1, 0.186 d y−1, and 0.933 d y−1 for DBF, ENF, all forest, and GRA sites, respectively. Both long-term trends in SOS and EOS of ENF were significantly greater than those of DBF. Long-term trend in EOS of GRA sites was obviously greater than that of forest sites. LOS of GRA sites increased by a rate of 0.806 d y−1, followed by the ENF (0.581 d y−1), all forest (0.386 d y−1), and DBF (0.150 d y−1) sites. GPPmax of GRA sites increased with a rate of 0.008 g C m−2 y−1 and was significantly smaller than all forest sites (0.045 g C m−2 y−1). The GPPmax of DBF (0.111 g C m−2 y−1) sites increased the greatest, followed by the ENF (0.070 g C m−2 y−1) sites. The spring GPP increased by rates of 0.22 g C m−2 y−1, 2.64 g C m−2 y−1, and 2.16 g C m−2 y−1 for DBF, ENF and all forest sites, respectively, while the spring GPP for GRA (−0.76 g C m−2 y−1) sites decreased (Fig. 3). The summer GPP, autumn GPP and annual GPP for all PFTs sites increased. The long-term trend in annual GPP for GRA (6.42 g C m−2 y−1) sties was greater than DBF (3.56 g C m−2 y−1) sites, but was smaller than ENF (13.06 g C m−2 y−1) and all forest (8.41 g C m−2 y−1) sites. The long-term trend in annual GPP for ENF sites increased the greatest during this study period. Results indicated that the carbon uptake by photosynthesis increased during the early fifteen years of this century (1997–2014).

where Yi, t is the variable of interest at the ith site in year t; βi is an estimate of the annual rate of changes in variable of interest at the ith site; εi, t is a random error term; αi is a vector of estimated initial values (t = 0) of the variable of interest. 2.4. Statistical models for predicting long-term trends in annual GPP Two Statistical Models of Integrated Phenology and Physiology (SMIPP) were used to predict long-term trends in annual GPP in this study. The first model, proposed by Xia et al. (2015), includes the longterm trends in LOS and GPPmax and is referred to as the SMIPPx model (Equation (2)). The second model, proposed by Zhou et al. (2016), includes the long-term trends in SOS, EOS, and GPPmax, and is referred to as the SMIPPz model (Equation (3)).

LTGPPa = a × LTLOS + b × LTGPPmax + c

(2)

LTGPPa = d × LTSOS + e × LTEOS + f × LTGPPmax + g

(3)

Where a, b, c, d, e, f, and g are parameter coefficients of the statistical models. The LTSOS, LTEOS, LTLOS, LTGPPmax, and LTGPPa are long-term trends in SOS, EOS, LOS, GPPmax, and annual GPP, respectively. These variables calculated by using long time series data from 1997 to 2014 for different PFTs sites were used to calibrate the parameters of the models. Then, the long-term trends in SOS, EOS, LOS, GPPmax, and annual GPP calculated by using long time series data from 1997 to N were used to validate the prediction accuracy and extrapolation ability of the models. The goodness of fit of the models was measured by calculating the coefficient of determination (R2) and root mean squared error (RMSE) for the calibration and validation data.

3.2. Relationships between long-term trends in phenological date, GPPmax and GPP For all PFTs sites, long-term trend in SOS had the highest relationship with long term trend in spring GPP, long-term trend in EOS had the highest relationship with long term trend in autumn GPP, and long term trend in GPPmax were the most related to long term trend in summer GPP (Table 1). Long term trend in SOS was negatively correlated (p < 0.05) with long-term trend in EOS for DBF sites and all forest sites, which implied that advancing in SOS was accompanied with delay of EOS. Linear relationship between long-term trends in SOS and spring GPP was closer than that between long-term trends in EOS and autumn GPP. For forest sites, long-term trends in SOS and EOS did not explain

3. Result 3.1. Long-term trends in phenological date, GPPmax and GPP Analysis of long-term trends showed that SOS of DBF, ENF, and all forest (DBF + ENF + MF, the same below) sites decreased by rates of −0.040 d y−1, -0.291 d y−1, and -0.200 d y−1, respectively, while SOS 607

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

Fig. 2. Distributions of long-term trends in start of growing season (LTSOS), end of growing season (LTEOS), length of growing season (LTLOS), and maximal gross primary production (LTGPPmax), respectively, for the three plant functional types sties and all forest sites. The mean values of these indicators are shown in asterisks. DBF-deciduous broadleaf forests; ENF-evergreen needle-leaf forests; GRA-Grassland; All Forest-DBF + ENF + mixed forest. In total, 41 sites located in the Europe, Asia, North America were used, including 12 DBF sites (n = 126), 17 ENF sites (n = 189), 7 GRA sites (n = 64), and 5 MF sites (n = 61), respectively.

among DBF, ENF, and GRA sites (Fig. 4 (a), (b)). Advance in long-term trend in SOS and delay in long-term trend in EOS led to more carbon assimilation in spring and autumn for DBF sites than ENF sites, where a one-day advance in long-term trend in SOS and a one-day delay in longterm trend in EOS increased long-term trends in spring GPP and autumn GPP by 9.33 g C m−2 d−1 and 12.34 g C m−2 d−1 for DBF sites, respectively. However, a one-day advance in long-term trend in SOS and a one-day delay in long-term trend in EOS only increased long-term trends in spring GPP and autumn GPP by 5.42 g C m−2 d−1 and 8.34 g C m−2 d−1 for ENF sites, respectively. For GRA sites, a one-day change in long-term trend in SOS caused the largest long-term trend in spring GPP changes (14.9 g C m−2 d−1), but a one-day change in long-term trend in EOS caused the smallest long-term trend in autumn GPP changes (4.9 g C m−2 d−1) among the three PFTs. Responses of long-term trends in SOS to spring GPP were significantly different from responses of long-term trends in EOS to autumn GPP between forest sites and GRA sites. Carbon assimilation in autumn due to delay in long-term trend in EOS was greater than carbon assimilation in spring due to advance in long-term trends in SOS for both DBF and ENF sites, which was opposite from GRA sites. The slope between long-term trends in GPPmax and summer GPP was similar between DBF (86.75 g C m−2) and ENF (89.58 g C m−2) sites, and both

long-term trends in spring GPP and autumn GPP variations as well as long-term trend in GPPmax explained long-term trend in summer GPP variations. Besides, long-term trends in SOS and EOS did not explain long-term trends in annual GPP variations as well as long-term trend in GPPmax explained long-term trends in annual GPP variations, which implied that dominant role of long-term trend in GPPmax in controlling long-term trends in annual GPP for forest sites. The linear relationship between long-term trends in SOS and spring GPP was stronger for DBF sites than for ENF sites. Long-term trend in GPPmax explained longterm trend in summer GPP variations less well for GRA sites than for forest sites. Long-term trend in SOS for GRA sites had stronger linear relationship with long-term trends in annual GPP than long-term trend in GPPmax, which implied that important role of long-term trend in SOSin controlling long-term trends in annual GPP for GRA sites. The linear relationships between long-term trend in EOS and long-term trends in autumn GPP and annual GPP were not significant.

3.3. Responses of long-term trends in seasonal GPP to phenology and GPPmax There were obviously different between responses of long-term trends in SOS to spring GPP and long-term trends in EOS to autumn GPP 608

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

Fig. 3. Distributions of long-term trends in spring gross primary production (GPP) (LTGPPsp), summer GPP (LTGPPsu), autumn GPP (LTGPPau), and annual GPP (LTGPPa), respectively, for the three plant functional types sties and all forest sites. The mean values of these indicators are shown in asterisks. DBF-deciduous broadleaf forests; ENF-evergreen needle-leaf forests; GRA-Grassland; All Forest-DBF + ENF + mixed forest. In total, 41 sites located in the Europe, Asia, North America were used, including 12 DBF sites (n = 126), 17 ENF sites (n = 189), 7 GRA sites (n = 64), and 5 MF sites (n = 61), respectively.

were higher than that in GRA (69.18 g C m−2) sites (Fig. 4(c)).

site for GRA) did not well be captured by the SMIPPx model (Fig. 5). The contribution of long-term trends in EOS to annual GPP was negative (parameter coefficient for long-term trend in EOS was negative) and not significant (p > 0.05) in the SMIPPz model for the DBF sites, which implied that delay in EOS will cause decrease in long-term trend in annual GPP. The contribution of long-term trends in SOS to annual GPP was positive (parameter coefficient for long-term trend in SOS was positive) in the SMIPPz model for the ENF sites, which implied that advance in SOS will cause decrease in long-term trend in annual GPP. Above phenomena cannot be explained by the relationships between long-term trends in annual GPP and phenology, which showed that there were negative relationship between long-term trends in SOS and annual GPP and positive relationship between long-term trends in EOS and annual GPP (Table 2). Surprisingly, the validation results for the statistical models showed that the SMIPPx model was more superior to the SMIPPz model in predicting long-term trend in annual GPP for all the PTFs (Fig. 6). A high prediction ability of the SMIPPx model for predicting a wide range of variation in long-term trend in annual GPP of the DBF, ENF and all

3.4. Prediction of long-term trends in annual GPP using phenology and GPPmax Coefficients and RMSE of multiple regression models for predicting long-term trends in annual GPP of different PFTs are shown in Table 2. The results of the statistical model including long-term trends in LOS and GPPmax (the SMIPPx model) were compared with the statistical model including long-term trends in SOS, EOS, and GPPmax (the SMIPPz model). The SMIPPz model was shown to be lower RMSE than the SMIPPx model for all the PFTs for the training dataset, especially for the ENF sites and GRA sites. The long-term trend in annual GPP estimates from both the SMIPPx model and the SMIPPz model had significant (R2 > 0.87, p < 0.01) relationship with observed long-term trend in annual GPP. The R2 between estimates and observations for the SMIPPx model was slightly less than that of the SMIPPz model for the ENF sites and GRA sites because the long-term trend in annual GPP variability for several sites (FR-LBr and US-GLE sites for ENF; US-IB2 609

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

Table 1 Relationships between long-term trends in phenology, maximal gross primary production (GPPmax), and seasonal and annual gross primary production (GPP). PFTplant functional type; DBF-deciduous broadleaf forest; ENF-evergreen needle-leaf forest; GRA-grassland; All Forest-DBF + ENF + mixed forest. LTSOS, LTEOS, and LTGPPmax are long-term trends in start of growing season (SOS), end of growing season (EOS), and GPPmax, respectively. LTGPPsp, LTGPPsu, LTGPPau, and LTGPPa are long-term trends in spring, summer, autumn and annual GPP, respectively. In total, 41 sites located in the Europe, Asia, North America were used, including 12 DBF sites (n = 126), 17 ENF sites (n = 189), 7 GRA sites (n = 64), and 5 MF sites (n = 61), respectively. PFT DBF

ENF

GRA

All Forest

LTSOS LTEOS LTGPPmax LTGPPa LTSOS LTEOS LTGPPmax LTGPPa LTSOS LTEOS LTGPPmax LTGPPa LTSOS LTEOS LTGPPmax LTGPPa

LTEOS

LTGPPmax

LTGPPsp

LTGPPsu

LTGPPau

LTGPPa

−0.69∗

−0.55 0.62∗

−0.78∗∗ 0.54∗

0.26

−0.82∗ −0.33

−0.53∗∗

−0.72∗∗ 0.65∗∗

−0.45 0.46 0.97∗∗ 0.93∗∗ −0.76∗∗ 0.54∗ 0.99∗∗ 0.93∗∗ −0.75 0.04 0.91∗ 0.88∗ −0.68∗∗ 0.62∗∗ 0.98∗∗ 0.94∗∗

−0.62∗ 0.76∗∗ 0.80∗∗ 0.87∗∗ −0.39 0.78∗∗ 0.71∗∗ 0.88∗∗ −0.36 0.76 0.33 0.57 −0.49∗∗ 0.81∗∗ 0.76∗∗ 0.89∗∗

−0.69∗ 0.66∗ 0.97∗∗

−0.45

−0.93∗∗ 0.75∗∗ 0.68∗ 0.81∗∗ −0.84∗∗ 0.69∗∗ 0.77∗∗ 0.87∗∗ −0.98∗∗ −0.31 0.89∗ 0.94∗∗ −0.84∗∗ 0.69∗∗ 0.74∗∗ 0.83∗∗

−0.72∗∗ 0.76∗∗ 0.93∗∗ −0.94∗∗ −0.01 0.82∗ −0.72∗∗ 0.79∗∗ 0.94∗∗

Note: * is significant at p = 0.05; ** is significant at p = 0.01. Fig. 4. Responses of long-term trends in seasonal gross primary production (GPP) to phenology and physiology for the three plant functional types, (a) long-term trends in spring GPP (LTGPPsp) vs. start of growing season (LTSOS), (b) long-term trends in autumn GPP (LTGPPau) vs. end of growing season (LTEOS), and (c) long-term trends in summer GPP (LTGPPsu) vs. maximal gross primary production (LTGPPmax). DBF-deciduous broadleaf forests; ENFevergreen needle-leaf forests; GRA-Grassland. The US-Wkg (GRA) site was excluded. In total, 41 sites located in the Europe, Asia, North America were used, including 12 DBF sites (n = 126), 17 ENF sites (n = 189), 7 GRA sites (n = 64), and 5 MF sites (n = 61), respectively.

forest sites were shown in Fig. 6 and their RMSEs were 8.8 g C m−2 d−1, 9.3 g C m−2 d−1, and 8.1 g C m−2 d−1, respectively, when comparing estimates with observations. The prediction ability of the SMIPPz model for predicting the long-term trend in annual GPP of the DBF, ENF and all forest sites was poorer than the SMIPPx model based on higher RMSEs (14.3 g C m−2 d−1, 11.0 g C m−2 d−1, and 9.7 g C m−2 d−1, respectively) compared to those from the SMIPPx model. An outlier (ITRo2 site) was occurred in DBF sites, in where the physiology and phenology of DBF were affected by the 2003 European drought. The estimates for the IT-Ro2 site from both the SMIPPx and SMIPPz models had huge bias compared with observations, which implied that drought

effects on physiology and phenology reduced the explanatory ability of the models. Interestingly, both of the SMIPPx and SMIPPz models were constructed based on dataset from all forest sites can increase accuracy of long-term trend in annual GPP estimates and reduce drought effects on model performance and can be used to predict long-term trend in annual GPP of forest ecosystems, because the parameter coefficients for the models became more reasonable and stable as the dataset increases (Table 2). Obvious difference in the prediction ability between the SMIPPx model and the SMIPPz model for the GRA sites was shown (Fig. 7). The RMSE of estimates from the SMIPPx model for the GRA sites was 610

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

Table 2 Parameter coefficients, RMSE and R2 of the Statistical Models of Integrated Phenology and Physiology driven by long-term trends in length of growing season (LOS) and maximal gross primary production (GPPmax) (SMIPPx) and driven by long-term trends in start of growing season (SOS), end of growing season (EOS), and GPPmax (SMIPPz) for estimating the long-term trend in annual GPP (LTGPPa). PFT-plant functional type; DBF-deciduous broadleaf forest; ENF-evergreen needle-leaf forest; GRA-grassland; All Forest-DBF + ENF + mixed forest. Con means coefficient of constant term. LTSOS, LTEOS, and LTGPPmax are long-term trends in SOS, EOS, and GPPmax, respectively. In total, 41 sites located in the Europe, Asia, North America were used, including 12 DBF sites (n = 126), 17 ENF sites (n = 189), 7 GRA sites (n = 64), and 5 MF sites (n = 61), respectively. PFT

DBF ENF GRA All forest

SMIPPx model

SMIPPz model 2

Con

LTLOS

LTGPPmax

RMSE

R

Con

LTSOS

LTEOS

LTGPPmax

RMSE

R2

−0.42 2.56 −3.26 1.51

4.56 6.01 10.78 5.57

122.31 111.34 124.68 113.32

3.66 7.85 9.08 6.34

0.96 0.90 0.87 0.93

0.14 −1.69 3.39 0.54

−8.09 1.18 −32.76 −1.67

−2.62 14.91 7.34 10.62

125.93 126.20 43.04 114.49

3.16 4.91 4.23 5.49

0.97 0.96 0.97 0.95

Fig. 5. The relationships between observed longterm trend in annual GPP (LTGPPa) and estimated LTGPPa from the Statistical Models of Integrated Phenology and Physiology driven by long-term trends in length of growing season and maximal gross primary production (GPPmax) (SMIPPx) and driven by long-term trends in start of growing season, end of growing season, and GPPmax (SMIPPz), respectively. DBF sites-deciduous broadleaf forest sites; ENF sitesevergreen needle-leaf forest sites; All forest sitesDBF + ENF + mixed forest sites; GRA sitesGrassland sites. The US-Wkg (GRA) site was excluded. In total, 41 sites located in the Europe, Asia, North America were used, including 12 DBF sites (n = 126), 17 ENF sites (n = 189), 7 GRA sites (n = 64), and 5 MF sites (n = 61), respectively.

25.9 g C m−2 d−1, which was significantly smaller than that from the SMIPPz model (RMSE = 83.6 g C m−2 d−1). If the US-Wkg site, where was disturbed by grazing, was excluded, the RMSEs of estimates from the SMIPPx model and the SMIPPz model for the GRA sites obviously decreased, with values of 16.2 g C m−2 d−1 and 34.3 g C m−2 d−1, respectively. The SMIPPz model had the problem in underestimation of negative long-term trend in annual GPP and overestimation of positive long-term trend in annual GPP. The long-term trends in annual GPP of GRA sites were more difficult to be predicted and the estimates had higher RMSE than those of forest sites. The validation results confirmed that the SMIPPx model had good and stable performance in predicting long-term trend in annual GPP of forest sites and GRA sites. However, drought and human disturbance had a huge influence on decrease in prediction accuracy of long-term trend in annual GPP, especially for the SMIPPz model.

4. Discussion 4.1. Relationships between phenological and physiological indicators and GPP in forest sites In this study, long-term trend in SOS was negatively correlated with long-term trends in EOS and GPPmax, implying that a long-term trend toward earlier SOS may generally be accompanied with a long-term trend toward greater GPPmax and a long-term trend toward later EOS and vice versa. This is different with previous reports, which showed that EOS was positively correlated with SOS (Wu et al., 2016). Like relationships between the three indicators and their respective seasonal GPP variability (Zhou et al., 2016), the long-term trends in seasonal GPP were also well captured by their respective indicators. Therefore, long-term trends in the three indicators can be used to predict longterm trends in annual GPP. It can become a new way to predict annual rate of change in GPP in the future using long-term trends in the three indicators. 611

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

Fig. 6. The relationships between observed long-term trend in annual GPP (LTGPPa) and predicted LTGPPa from the Statistical Models of Integrated Phenology and Physiology driven by long-term trends in length of growing season and maximal gross primary production (GPPmax) (SMIPPx) and driven by long-term trends in start of growing season, end of growing season, and GPPmax (SMIPPz), respectively. DBF sites-deciduous broadleaf forest sites; ENF sites-evergreen needle-leaf forest sites; All forest sites- DBF + ENF + mixed forest sites. The 1997-N means LTGPPa calculated from 1997 to N. N increases from 2002 to 2013 with a step by one year.

of seasonal GPP than phenology in spring for forest sites, consistent with an earlier study which reported the mean increase in GPP per day for late autumn being larger than that of early spring (Keenan et al., 2014). Previous research also showed that autumn phenology played a more significant role in regulating interannual variability of annual carbon uptake in forests than spring phenology (Wu et al., 2013a, b). Autumn warming led to a later EOS (Archetti et al., 2013) and the mean temperature over the growing season also triggered a later EOS because of positive relationship between growing season and autumn temperatures (Liu et al., 2018). Both climate and spring phenology had influence on the delayed autumn phenology in the Northern Hemisphere (Liu et al., 2016). Delay in EOS induced by these factors may contribute to enhance net forest carbon uptake in the future because

The long-term trend in GPPmax was the most related to long-term trends in annual GPP and played a much stronger role in control of long-term trends in annual GPP than the long-term trends in SOS and EOS for forest sites, which is consistent with the vast majority of literature on plant phenology and physiology control of trend and variability of GPP (Xia et al., 2015; Zhou et al., 2016, 2017; Fu et al., 2017a; Niu et al., 2017). The main reason for it probably is the direct link between photosynthetic physiology and carbon assimilation (Zhou et al., 2017). 4.2. Sensitivities of seasonal GPP to phenology Phenology in autumn had stronger influence on the long-term trend 612

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

Fig. 7. The relationships between observed long-term trend in annual GPP (LTGPPa) and predicted LTGPPa from the Statistical Models of Integrated Phenology and Physiology driven by long-term trends in length of growing season and maximal gross primary production (GPPmax) (SMIPPx) and driven by long-term trends in start of growing season, end of growing season, and GPPmax (SMIPPz) for the grassland (GRA) sites, respectively. (a) excluding the US-Wkg site, (b) including the US-Wkg site. The 1997-N means LTGPPa calculated from 1997 to N. N increases from 2007 to 2013 with a step by one year.

and the SMIPPz models were compared in this study. The simulation accuracy of the SMIPPz model was higher than that of the SMIPPx model, but the predictive ability of the SMIPPx model was more superior to that of the SMIPPz model, which was different with the results focused on predicting annual GPP variability by using the SMIPPx and the SMIPPz models (Zhou et al., 2016). One reason for this is that there was a multicollinearity problem caused by significant relationships among phenological and physiological indicators, such as a strong negative correlation between longterm trends in SOS and EOS in DBF sites, between long-term trends in EOS and GPPmax in ENF sites, and between long-term trends in SOS and GPPmax in GRA sites. Significant relationship between SOS and EOS was also found in other research (Wu et al., 2016). The first issue with multicollinearity is that parameter estimates have no specific meaning and even their sign change (Belsley et al., 1991). For example, the parameter estimate of long-term trend in EOS for DBF sites was negative and did not conform to positive correlation between long-term trends in EOS and annual GPP because long-term trend in EOS had significant multicollinearity with long-term trends in SOS and GPPmax. The second issue with multicollinearity is that the resulting out-ofsample predictions will be imprecise due to large errors in parameter estimates (Chatterjee et al., 2000). The parameter estimate of long-term trend in SOS for GRA sites probably was imprecise resulting in large changes in long-term trend in annual GPP caused by small changes in the long-term trend in SOS. The other reason is extreme climate events (such as drought) and disturbance (such as grazing) effects on the prediction accuracy of the models, which was proven by high errors in estimates for the IT-Ro2

increase in autumn photosynthesis was larger than increase in autumn respiration (Keenan et al., 2014). Temperature response of autumn phenology was significantly larger than that of spring phenology, which indicated a possible greater contribution of delay in EOS than of advance in SOS to extending the LOS and further increasing carbon sequestration under future warmer conditions (Fu et al., 2018). Autumn phenology and its associated environmental drives remain poorly understood and a relatively neglected aspect in climate change research (Liu et al., 2016, 2018; Fu et al., 2018). More and more attention should be paid to deep understanding of the importance of autumn phenology on vegetation carbon uptake. The difference in responses of long-term trends in spring GPP and autumn GPP to phenology between DBF and ENF sites was significant, but the relationships between long-term trend in summer GPP and physiology between DBF and ENF sites were similar. The long-term trends in spring GPP and autumn GPP of DBF sites were more sensitive to long-term trend in phenology than ENF sites due to stronger spring and autumn variations in leaf area index in DBF than ENF, which was similar with the results for large variations in seasonal GPP to phonological anomalies for DBF sites (Richardson et al., 2010; Keenan et al., 2014) and implied that seasonal carbon uptake of DBF was more sensitive to climate change effects on phenology than ENF. 4.3. Predictive ability of the models for long-term trends in annual GPP The predictive ability of both phenological and physiological indicators and the combination of them have been tested in existing studies (Xia et al., 2015; Zhou et al., 2016). The results of the SMIPPx 613

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

ENF (Fu et al., 2017b). In semi-arid grasslands, summer drought obviously depressed carbon uptake amplitude during the mid-growing season (Niu et al., 2017). Increase of summer or autumn precipitation obviously delayed the EOS of GRA (Fu et al., 2017b). Besides, the responses of carbon fluxes, phenological and physiological indicators for GRA to local environmental factors are more complicated than for forest ecosystems. The SOS of GRA in moist environments was sensitive to variation of temperature, but the SOS of GRA in dry areas was limited by soil water content and often initiated by precipitation events (Xin et al., 2015). The relationships between spring carbon uptake anomalies and carbon uptake anomalies in other seasons were insignificant for GRA sites (Fu et al., 2017b), which indirectly implied that climate event and human disturbance effects on seasonal carbon uptake of GRA should not be overlooked and exceed the legacy effects of spring carbon uptake on later seasons. However, GRA is more vulnerable to human disturbances than forest ecosystems. Human disturbances may reduce the relationships between carbon fluxes and phenological and physiological indicators. Man-made changes in structural and functional characterization of GRA, such as grazing, are not only influenced the contribution of physiological and ecological processes to carbon flux variability, but also reduced the strength of flux-climate correlations (Hwayne et al., 2008). Previous studies showed that all the spring environmental factors totally explained 37% and 16% of the variation in SOS in DBF and ENF, respectively, but explained only 3% in GRA (Fu et al., 2017b). Such low explanatory of environmental factors to spring phenology was probably caused by human disturbance effects on GRA. An earlier study proposed that the obvious seasonal changes made it easier to capture the changes in vegetation phenology and physiology, and hence change in annual GPP at the non-forest sites including GRA sites (Zhou et al., 2016). However, it seemed that the distinct seasonal changes made it more difficult to track the long-term trend in annual GPP of GRA sites than forest sites in this study. The large fluctuation in phenological and physiological indicators and seasonal GPP in GRA sites caused by climate and human disturbances may lead to unstable long-term trends in phenological and physiological indicators and seasonal GPP, which reduces the prediction ability of the models for GRA sites. Besides, a previous study has suggested that the spatiotemporal variation of vegetation phenology for GRA was difficult to be captured using satellite observations (Xin et al., 2015). The C3 grasses starts growth early in the spring and may have a second growing peak when temperature cools down in the fall (Wang et al., 2013), which also make it difficult to estimate the phenological and physiological indicators. Given that GRA is a key component in terrestrial biomes, more attention should be paid to improve the models for estimating physiology, phenology and carbon fluxes of GRA (Xin et al., 2015) and to further discover climate event and human disturbance effects on phenology and carbon dynamic of GRA.

(DBF) and US-Wkg (GRA) sites which suffered from drought. Previous study also showed that drought effects decreased the prediction accuracy of annual GPP variability in forest ecosystems (Zhou et al., 2016). GPPmax did not fully capture the drought effect on plant physiology because GPPmax emerged early, declined slightly and even increased during drought years (Zhou et al., 2016). The explanatory power of the SMIPP model became weak due to the time lag between GPPmax and drought induced GPP decline (Zhou et al., 2016). The carbon uptake enhancement induced by warmer spring canceled out a drought-induced reduction in summer photosynthesis (Angert et al., 2005), which may reduce relationship between GPPmax and annual GPP. Such complicate effects of drought on carbon dynamic probably reduce the relationship between long-term trends in GPPmax and annual GPP and further lead to low explanatory power of the SMIPP model in predicting long-term trend in annual GPP. The SMIPP model cannot be used for predicting long-term trend in annual GPP unless GPPmax and LOS can be obtained from other data sources, such as remote sensing data, which was one of the limitations of the model (Xia et al., 2015). Fortunately, more and more research has proposed a lot of feasible methods to accurately estimate the phenological and physiological indicators (Melaas et al., 2016; Peng et al., 2018; Richardson et al., 2018a), which enables the SMIPP model based on Fluxnet sites data to be extended to predict long-term trend in annual GPP at a regional or global scale (Zhou et al., 2016). Strong relationships between remotely sensed land surface phenology and canopy photosynthesis phenology from carbon flux measurements were discovered (Xia et al., 2015; D'Odorico et al., 2015; Melaas et al., 2016; Peng et al., 2017). The PhenoCam network including over 130 digital red-green-blue cameras are now starting to provide high accurate land surface phenology observations (Richardson et al., 2007, 2018a; Sonnentag et al., 2012), which increases our confidence in the ability of remote sensing techniques to accurately characterize land surface phenology and predict the long-term trends in annual GPP at a global scale based on land surface phenology (Peng et al., 2018; Richardson et al., 2018b). 4.4. Differences between GRA and forest ecosystems In contrast with forest sites, phenology in spring had stronger influence on long-term trend in spring GPP than phenology in autumn effect on long-term trend in autumn GPP for GRA sites. The long-term trend in SOS for GRA sites played the most important role in control of long-term trend in annual GPP, which was also different with forest sites. An earlier study also demonstrated that an increase in autumn NEE of GRA per extra day of EOS was smaller than increase in spring NEE per day advance of SOS (Fu et al., 2017b). A higher correlation coefficient between SOS and spring GPP for non-forest sites (including GRA sites) than forest sites was also found (Zhou et al., 2016). The sensitive of long-term trend in summer GPP for GRA sites to physiology was weaker than forest sites, which is in consistence with the slope of relationship between maximum NEE and NEE for GRA sites smaller than that for forest sites (Fu et al., 2017b). Accuracy of the models for predicting long-term trend in annual GPP of GRA sites was lower than that of forest sites, implying that the long-term trend in annual GPP of GRA sites was more difficult to be captured by using the phenological and physiological indicators than forest sites. In comparison with forest ecosystems, changes in SOS, GPPmax, and EOS for GRA ecosystems were more distinct (Zhou et al., 2016) because GRA ecosystems generally had poor self-regulation abilities under environmental stresses (Niinemets, 2010; Teuling et al., 2010) and responded more quickly and intensively to environmental changes (Zhang et al., 2016). On the one hand, carbon fluxes, phenological and physiological indicators of GRA were more sensitivity to climate factors than forest ecosystems. For example, variation of carbon uptake amplitude in GRA was more depended on summer climate factors than those in DBF and

5. Conclusions The long-term trends in SOS, EOS, LOS, GPPmax, and seasonal and annual GPP were calculated, and their relationships were analyzed based on dataset from 41 flux tower sites across the Europe, Asia, and North America. Carbon uptake by forest and GRA ecosystems through photosynthesis increased during the early fifteen years of this century and played an important role in mitigating climate change. The ENF ecosystem had the largest long-term trend in annual GPP than DBF and GRA ecosystems, but the sensitivities of seasonal carbon to phenology for DBF sites were stronger than ENF sites. The EOS played a more important role in carbon uptake than SOS for forest sites, which was already discovered by previous researches. Spring carbon uptake of GRA sites was decreasing, but reasons for it were still unclear in this study. More attention should be paid to these potential questions in the future, such as high long-term trend in annual GPP of ENF, long-term trend in annual GPP of DBF responses to phenology under climate 614

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

change, EOS effects on carbon uptake by forest, and decrease of spring carbon uptake due to delay in SOS in GRA sites. Statistical models driven by the phenological and physiological indicators can be used to accurately predict the long-term trend in annual GPP of forest and GRA ecosystems, although the models existed uncertainty caused by extreme climate event and disturbance effects. The models proposed in this study will be a feasible way to predict annual rate of change in carbon uptake through vegetation photosynthesis at a global scale when the phenological and physiological indicators are accurately derived from remote sensing technique.

mixed-grass prairie. Glob. Chang. Biol. 14, 1620–1632. Keenan, T.F., Gray, J., Friedl, M.A., Toomey, M., Bohrer, G., Hollinger, D.Y., Munger, J.W., O'Keefe, J., Schmid, H.P., Wing, I.S., Yang, B., Richardson, A.D., 2014. Net carbon uptake has increased through warming-induced changes in temperate forest phenology. Nat. Clim. Change 4, 598–604. Lasslop, G., Reichstein, M., Papale, D., Richardson, A.D., Arneth, A., Barr, A., Stoy, P., Wohlfahrt, G., 2010. Separation of net ecosystem exchange into assimilation and respiration using a light response curve approach: critical issues and global evaluation. Glob. Chang. Biol. 16, 187–208. Le Quéré, C., Andres, R.J., Boden, T., Conway, T., Houghton, R.A., House, J.I., Marland, G., Peters, G.P., et al., 2013. The global CO2 budget 1959 – 2011. Earth Syst. Sci. Data 5, 165–185. Liu, G., Chen, X., Zhang, Q., Lang, W., Delpierre, N., 2018. Antagonistic effects of growing season and autumn temperatures on the timing of leaf coloration in winter deciduous trees. Glob. Chang. Biol. 24, 3537–3545. Liu, Q., Fu, Y.H., Zhu, Z., Liu, Y., Liu, Z., Huang, M., Janssens, I.A., Piao, S., 2016. Delayed autumn phenology in the northern hemisphere is related to change in both climate and spring phenology. Glob. Chang. Biol. 22, 3702–3711. Melaas, E.K., Richardson, A.D., Friedl, M.A., Dragoni, D., Gough, C.M., Herbst, M., Montagnani, L., Moors, E., 2013. Using fluxnet data to improve models of springtime vegetation activity onset in forest ecosystems. Agric. For. Meteorol. 171–172, 46–56. Melaas, E.K., Sulla-Menashe, D., Gray, J.M., Black, T.A., Morin, T.H., Richardson, A.D., Friedl, M.A., 2016. Multisite analysis of land surface phenology in North American temperate and boreal deciduous forests from landsat. Remote Sens. Environ. 186, 452–464. Munger, J.W., Wofsy, S.C., Foster, D.R., 2012. Detection of Long-Term Trends in Carbon Accumulation by Forests in Northeastern U. S. And Determination of Causal Factors: Final Report. United States: N. p., 2012. . https://doi.org/10.2172/1033926. Niinemets, Ü., 2010. Responses of forest trees to single and multiple environmental stresses from seedlings to mature plants: past stress history, stress interactions, tolerance and acclimation. For. Ecol. Manage. 260, 1623–1639. Niu, S., Fu, Z., Luo, Y., Stoy, P.C., Keenan, T.F., Poulter, B., Zhang, L.M., Piao, S.L., Zhou, X.H., Zheng, H., Han, J.Y., Wang, Q.F., Yu, G.R., 2017. Interannual variability of ecosystem carbon exchange: from observation to prediction. Glob. Ecol. Biogeogr. 26, 1225–1237. Nolan, R.H., Sinclair, J., Eldridge, D.J., Ramp, D., 2018. Biophysical risks to carbon sequestration and storage in Australian drylands. J. Environ. Manag. 208, 102–111. Papale, D., Reichstein, M., Aubinet, M., Canfora, E., Bernhofer, C., Kutsch, W.L., Longdoz, B., Rambal, S., Valentini, R., Vesala, T., Yakir, D., 2006. Towards a standardized processing of net ecosystem exchange measured with eddy covariance technique: algorithms and uncertainty estimation. Biogeosciences 3, 571–583. Pavani, B.F., Sousa Júnior, W.C., Inouye, C.E.N., Vieira, S.A., Mello, A.Y.I., 2018. Estimating and valuing the carbon release in scenarios of land-use and climate changes in a Brazilian coastal area. J. Environ. Manag. 226, 416–427. Peng, D., Wu, C., Zhang, X., Yu, L., Huete, A.R., Wang, F., Luo, S.Z., Liu, X.J., Zhang, H., 2018. Scaling up spring phenology derived from remote sensing images. Agric. For. Meteorol. 256–257, 207–219. Peng, D., Zhang, X., Wu, C., Huang, W., Gonsamo, A., Huete, A.R., Didan, K., Tan, B., Zhang, B., 2017. Intercomparison and evaluation of spring phenology products using national phenology network and Ameriflux observations in the contiguous United States. Agric. For. Meteorol. 242, 33–46. Piao, S.L., Fang, J.Y., Zhou, L.M., Ciais, P., Zhu, B., 2010. Variations in satellite-derived phenology in China's temperate vegetation. Glob. Chang. Biol. 12, 672–685. Reichstein, M., Falge, E., Baldocchi, D., Papale, D., Aubinet, M., Berbigier, P., Bernhofer, C., Buchmann, N., Gilmanov, T., Granier, A., Grünwald, T., Havránková, K., Ilvesniemi, H., Janous, D., Knohl, A., Laurila, T., Lohila, A., Loustau, D., Matteucci, G., Meyers, T., Miglietta, F., Ourcival, J.M., Pumpanen, J., Rambal, S., Rotenberg, E., Sanz, M., Tenhunen, J., Seufert, G., Vaccari, F., Vesala, T., Yakir, D., Valentini, R., 2005. On the separation of net ecosystem exchange into assimilation and ecosystem respiration: review and improved algorithm. Glob. Chang. Biol. 11, 1424–1439. Richardson, A.D., Black, T.A., Ciais, P., Delbart, N., Friedl, M.A., Gobron, N., Hollinger, D.Y., Kutsch, W.L., Longdoz, B., Luyssaert, S., Migliavacca, M., Montagnani, L., Munger, J.W., Moors, E., Piao, S.L., Rebmann, C., Reichstein, M., Saigusa, N., Tomelleri, E., Vargas, R., Varlagin, A., 2010. Influence of spring and autumn phenological transitions on forest ecosystem productivity. Philos. Trans. R. Soc. B Biol. Sci. 365, 3227–3246. Richardson, A.D., Hufkens, K., Milliman, T., Aubrecht, D.M., Chen, M., Gray, J.M., Johnston, M.R., Keenan, T.F., Klosterman, S.T., Kosmala, M., Melaas, E.K., Friedl, M.A., Frolking, S., 2018a. Tracking vegetation phenology across diverse North American biomes using Phenocam imagery. Sci. Data 5, 180028. Richardson, A.D., Hufkens, K., Milliman, T., Frolking, S., 2018b. Intercomparison of phenological transition dates derived from the Phenocam dataset v1.0 and MODIS satellite remote sensing. Sci. Rep. 8, 5679. Richardson, A.D., Jenkins, J.P., Braswell, B.H., Hollinger, D.Y., Ollinger, S.V., Smith, M.L., 2007. Use of digital webcam images to track spring green-up in a deciduous broadleaf forest. Oecologia 152, 323–334. Schwartz, M.D., Ahas, R., Aasa, A., 2006. Onset of spring starting earlier across the northern hemisphere. Glob. Chang. Biol. 12, 343–351. Sonnentag, O., Hufkens, K., Teshera-Sterne, C., Young, A.M., Friedl, M., Braswell, B.H., Milliman, T., O'Keefe, J., Richardson, A.D., 2012. Digital repeat photography for phenological research in forest ecosystems. Agric. For. Meteorol. 152, 159–177. Tao, F., Yokozawa, M., Xu, Y., Hayashi, Y., Zhang, Z., 2006. Climate changes and trends in phenology and yields of field crops in China, 1981–2000. Agric. For. Meteorol. 138, 82–92. Teuling, A.J., Seneviratne, S.I., Stöckli, R., Reichstein, M., Moors, E., Ciais, P., Luyssaert, S., van den Hurk, B., Ammann, C., Bernhofer, C., Dellwik, E., Gianelle, D., Gielen, B.,

Acknowledgments This research was supported by the National Natural Science Foundation of China (Grant Nos. 31870619, 31670644, 31570629). This work used eddy covariance data acquired and shared by the FLUXNET community, including these networks: AmeriFlux, AfriFlux, AsiaFlux, CarboAfrica, CarboEuropeIP, CarboItaly, CarboMont, ChinaFlux, Fluxnet-Canada, GreenGrass, ICOS, KoFlux, LBA, NECC, OzFlux-TERN, TCOS-Siberia, and USCCC. The ERA-Interim reanalysis data are provided by ECMWF and processed by LSCE. The FLUXNET eddy covariance data processing and harmonization was carried out by the European Fluxes Database Cluster, AmeriFlux Management Project, and Fluxdata project of FLUXNET, with the support of CDIAC and ICOS Ecosystem Thematic Center, and the OzFlux, ChinaFlux and AsiaFlux offices. Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.jenvman.2019.06.023. References Angert, A., Biraud, S., Bonfils, C., Henning, C.C., Buermann, W., Pinzon, J., Tucker, C.J., Fung, I., 2005. Drier summers cancel out the CO2 uptake enhancement induced by warmer springs. Proc. Natl. Acad. Sci. Unit. States Am. 102, 10823–10827. Archetti, M., Richardson, A.D., O'Keefe, J., Delpierre, N., 2013. Predicting climate change impacts on the amount and duration of autumn colors in a new England forest. PLOS ONE 8, e57373. https://doi.org/10.1371/journal.pone.0057373. Barr, A.G., Richardson, A.D., Hollinger, D.Y., Papale, D., Arain, M.A., Black, T.A., Bohrer, G., Dragoni, D., Fischer, M.L., Gu, L., Law, B.E., Margolis, H.A., McCaughey, J.H., Munger, J.W., Oechel, W., Schaeffer, K., 2013. Use of change-point detection for friction–velocity threshold evaluation in eddy-covariance studies. Agric. For. Meteorol. 171–172, 31–45. Belsley, D.A., 1991. Conditioning Diagnostics: Collinearity and Weak Data in Regression. Wiley, New York. Bonan, G.B., 2008. Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science 320 (5882), 1444–1449. Chatterjee, S., Hadi, A.S., Price, B., 2000. Regression Analysis by Example, third ed. Wiley, New York. Chen, J., Jönsson, P., Tamura, M., Gu, Z., Matsushita, B., Eklundh, L., 2004. A simple method for reconstructing a high-quality NDVI time-series data set based on the Savitzky-Golay filter. Remote Sens. Environ. 91, 332–344. https://www. sciencedirect.com/science/article/pii/S003442570400080X. Christidis, N., Stott, P.A., Hegerl, G.C., Betts, R.A., 2013. The role of land use change in the recent warming of daily extreme temperatures. Geophys. Res. Lett. 40, 589–594. D'Odorico, P., Gonsamo, A., Gough, C.M., Bohrer, G., Morison, J., Wilkinson, M., Hanson, P.J., Gianelle, D., Fuentes, J.D., Buchmann, N., 2015. The match and mismatch between photosynthesis and land surface phenology of deciduous forests. Agric. For. Meteorol. 214–215, 25–38. Fu, Y.H., Piao, S., Delpierre, N., Hao, F., Hänninen, H., Liu, Y., Sun, W., Janssens, I.A., Campioli, M., 2018. Larger temperature response of autumn leaf senescence than spring leaf-out phenology. Glob. Chang. Biol. 24, 2159–2168. Fu, Z., Dong, J., Zhou, Y., Stoy, P.C., Niu, S., 2017a. Long term trend and interannual variability of land carbon uptake—the attribution and processes. Environ. Res. Lett. 12 014018. Fu, Z., Stoy, P.C., Luo, Y., Chen, J., Sun, J., Montagnani, L., Wohlfahrt, G., Rahman, A.F., Rambal, S., Bernhofer, C., Wang, J.S., Shirkey, G., Niu, S., 2017 bb. Climate controls over the net carbon uptake period and amplitude of net ecosystem production in temperate and boreal ecosystems. Agric. For. Meteorol. 243, 9–18. Huang, X., Zhou, Y., Zhang, Z., 2018. Carbon Sequestration Anticipation Response to land use change in a mountainous karst basin in China. J. Environ. Manag. 228, 40–46. Hwayne, P., Frank, A.B., Sanabria, J., Phillips, R.L., 2008. Interannual variability in carbon dioxide fluxes and flux-climate relationships on grazed and ungrazed northern

615

Journal of Environmental Management 246 (2019) 605–616

X. Xu, et al.

effects, covariation with climate and implications for productivity. Agric. For. Meteorol. 216, 177–187. Xia, J., Niu, S., Ciais, P., Janssens, I.A., Chen, J., Ammann, C., Arain, A., Blanken, P.D., et al., 2015. Joint control of terrestrial gross primary productivity by plant phenology and physiology. Proc. Natl. Acad. Sci. Unit. States Am. 112, 2788–2793. Xin, Q., Broich, M., Zhu, P., Gong, P., 2015. Modeling grassland spring onset across the western United States using climate variables and MODIS-derived phenology metrics. Remote Sens. Environ. 161, 63–77. Zhang, Y., Xiao, X., Zhou, S., Ciais, P., McCarthy, H., Luo, Y., 2016. Canopy and physiological limitation of GPP during drought and heat wave. Geophys. Res. Lett. 43, 3325–3333. Zhao, M., Running, S.W., 2010. Drought-induced reduction in global terrestrial net primary production from 2000 through 2009. Science 329, 940–943. Zhou, S., Zhang, Y., Caylor, K.K., Luo, Y., Xiao, X., Ciais, P., Huang, Y., Wang, G., 2016. Explaining inter-annual variability of gross primary productivity from plant phenology and physiology. Agric. For. Meteorol. 226–227, 246–256. Zhou, S., Zhang, Y., Ciais, P., Xiao, X., Luo, Y., Caylor, K.K., Huang, Y., Wang, G., 2017. Dominant role of plant physiology in trend and variability of gross primary productivity in North America. Sci. Rep. 7, 41366.

Grünwald, T., Klumpp, K., Montagnani, L., Moureaux, C., Sottocornola, M., Wohlfahrt, G., 2010. Contrasting response of European forest and grassland energy exchange to heatwaves. Nat. Geosci. 3, 722–727. Wang, C., Hunt, E.R., Zhang, L., Guo, H., 2013. Phenology-assisted classification of C-3 and C-4 grasses in the US great plains and their climate dependency with MODIS time series. Remote Sens. Environ. 138, 90–101. Wu, C.Y., Peng, D.L., Soudani, K., Siebicke, L., Gough, C.M., Arain, M.A., Bohrer, G., Lafleur, P.M., Peichl, M., Gonsamo, A., Xu, S.G., Fang, B., Ge, Q.S., 2017. Land surface phenology derived from normalized difference vegetation index (NDVI) at global fluxnet sites. Agric. For. Meteorol. 233, 171–182. Wu, C., Chen, J.M., Black, T.A., Price, D.T., Kurz, W.A., Desai, A.R., Gonsamo, A., Jassal, R.S., Gough, C.M., Bohrer, G., Dragoni, D., Herbst, M., Gielen, B., Berninger, F., Vesala, T., Mammarella, I., Pilegaard, K., Blanken, P.D., 2013a. Interannual variability of net ecosystem productivity in forests is explained by carbon flux phenology in autumn. Glob. Ecol. Biogeogr. 22, 994–1006. Wu, C., Gough, C.M., Chen, J.M., Gonsamo, A., 2013b. Evidence of autumn phenology control on annual net ecosystem productivity in two temperate deciduous forests. Ecol. Eng. 60, 88–95. Wu, C., Hou, X., Peng, D., Gonsamo, A., Xu, S., 2016. Land surface phenology of China's China's temperate ecosystems over 1999–2013: spatial–temporal patterns, interaction

616