Climate Controls on the Spatial Variability of Vegetation Greenup Rate across Ecosystems in Northern Hemisphere

Variations in individual phenological events in response to global change have received considerable attentions. However, the development of phenological stages is relatively neglected, especially based on in situ observation data. In this study, the rate of vegetation greenup (Vgreenup) across the Northern Hemisphere was examined for different plant functional types (PFTs) by using eddy covariance flux data from 40 sites (417 site-years). Then, the controls of climatic variables on the spatial distribution of Vgreenup across PFTs were further investigated. The mean Vgreenup was 0.22 ± 0.11 g C m−2 day−2 across all sites, with the largest and lowest values observed in cropland and evergreen needle-leaf forest, respectively. A strong latitude dependence by Vgreenup was observed in both Europe and North America. The spatial variations of Vgreenup were jointly regulated by the duration of greenup (Dgreenup) and the amplitude of greenup (Agreenup). However, the predominant factor was Dgreenup in Europe, which changed to Agreenup in North America. Spring climatic factors exerted significant influences on the spatial distribution of Vgreenup across PFTs. Specifically, increasing temperature tended to shorten Dgreenup and promote Agreenup simultaneously, resulting in an acceleration of Vgreenup. Dryness had a depression effect on Vgreenup for the whole study area, as exhibited by a lower Vgreenup with increasing vapor pressure deficit or decreasing soil moisture. However, Vgreenup in North America was only significantly and positively correlated with temperature. Without the limitation of other climatic factors, the temperature sensitivity of Vgreenup was higher in North America (0.021 g C m−2 day−2 °C−1) than in Europe (0.015 g C m−2 day−2 °C−1). This study provides new cognitions for Vgreenup dynamics from in situ observations in complement to satellite observations, which can improve our understanding of terrestrial carbon cycles.


Introduction
Vegetation phenology is the study of the timing of recurring plant life cycle events that are driven by environmental factors [1][2][3]. It not only is a key indicator of climate change but also exerts important influence on terrestrial ecosystem by altering carbon, nitrogen, and water cycles [4,5]. Many studies have addressed the relationship between the timing of individual phenological events and their drivers [6][7][8]. Under global warming, both remote sensing and ground-based observations reveal the substantial advancement of greenup, delay of leaf senescence, and extension of the growing season [9][10][11][12][13]. However, compared with individual phenological event, the development of the phenological stages has received inadequate attention.
The rate of greenup (Vgreenup) was proposed to facilitate the understanding and description of vegetation greenup development [14,15], which is calculated as the ratio of the duration of greenup (Dgreenup) to the amplitude of greenup (Agreenup). The variation in Vgreenup has strong linkage with vegetation growth and carbon sequestration

Measures of Vgreenup among PFTs
Vgreenup ranged from 0.06 to 0.54 g C m −2 day −2 across the 40 eddy covariance sites (Table 1), with a mean of 0.22 ± 0.11 g C m −2 day −2 . The difference in Vgreenup was also significant among PFTs (F = 123.64, p < 0.001) ( Figure 1). Cropland (CRO) exhibited the largest Vgreenup, followed by deciduous broadleaf forest (DBF), grassland (GRA), wetland (WET), mixed forest (MF), and evergreen needle-leaf forest (ENF) (Figure 1). Besides, Vgreenup varied more in CRO than in the other PFTs, as shown by a greater standard deviation. Vgreenup in GRA was comparable to that in DBF. In addition, there was no significant difference in Vgreenup between MF and WET.

Spatial Distribution of Vgreenup
Vgreenup had a clear spatial pattern, which showed a strong dependence on latitude across all sites but no dependence on longitude and altitude (Figure 2a−c). Vgreenup decreased 0.007 g C m −2 day −2 with an increase in latitude per degree, which was different Different letters indicate significant difference (p < 0.05) in one-way ANOVA using LSD multiple comparisons among PTFs. DBF, deciduous broadleaf forest; ENF, evergreen needle-leaf forest; MF, mixed forest; GRA, grassland; CRO, cropland; WET, wetland.

Spatial Distribution of Vgreenup
Vgreenup had a clear spatial pattern, which showed a strong dependence on latitude across all sites but no dependence on longitude and altitude (Figure 2a-c). Vgreenup decreased 0.007 g C m −2 day −2 with an increase in latitude per degree, which was different from the increasing Vgreenup trend with latitude observed by remote sensing in a previous study [21]. Besides, the strong dependence of Vgreenup on latitude was also observed in North America as well as in Europe (Figure 2d,g). Vgreenup decreased 0.008 and 0.011 g C m −2 day −2 per degree latitude increase in Europe and North America, respectively. There was no evident dependence of Vgreenup on longitude and altitude on the two continents separately.

Spatial Relationships between Vgreenup and Its Components
Vgreenup anomalies showed a significantly negative correlation with Dgreenup anomalies and a significantly positive correlation with Agreenup anomalies across all sites in the whole study area (Figure 3a−b). This was different from the findings of a previous satellite-based study [19], in which Vgreenup was positively correlated with both Dgreenup and Agreenup at a global scale. Across all sites in this study, the linear spatial correlation between Vgreenup anomalies and Dgreenup anomalies (R 2 = 0.61, p < 0.001) was relatively weaker than that between Vgreenup anomalies and Agreenup anomalies (R 2 = 0.83, p < 0.001), which implies the predominant role of Agreenup in the whole study area. The relative contributions of Dgreenup and Agreenup to the spatial variation of Vgreenup were different between Europe and North America (Figure 3c−f). Dgreenup anomalies exhibited a higher correlation (R 2 = 0.67, p < 0.001) in explaining the spatial variation of Vgreenup anomalies in comparison to Agreenup anomalies (R 2 = 0.59, p < 0.001) in Europe (Figure 3c−d). However, Vgreenup anomalies correlated more strongly with Agreenup anomalies (R 2 = 0.95, p < 0.001) than with Dgreenup anomalies (R 2 = 0.58, p < 0.001) in North America (Figure 3e−f), which indicated that the spatial variation of Vgreenup was more predominated by Agreenup.

Spatial Relationships between Vgreenup and Its Components
Vgreenup anomalies showed a significantly negative correlation with Dgreenup anomalies and a significantly positive correlation with Agreenup anomalies across all sites in the whole study area (Figure 3a,b). This was different from the findings of a previous satellite-based study [19], in which Vgreenup was positively correlated with both Dgreenup and Agreenup at a global scale. Across all sites in this study, the linear spatial correlation between Vgreenup anomalies and Dgreenup anomalies (R 2 = 0.61, p < 0.001) was relatively weaker than that between Vgreenup anomalies and Agreenup anomalies (R 2 = 0.83, p < 0.001), which implies the predominant role of Agreenup in the whole study area. The relative contributions of Dgreenup and Agreenup to the spatial variation of Vgreenup were different between Europe and North America (Figure 3c

Spatial Relationships between Vgreenup and Climatic Factors
The spatial pattern of Vgreenup was regulated by multiple spring climatic factors across all sites in the whole study area (Figure 4a). Temperature (TA) played a predominant role in controlling the spatial variation in Vgreenup across all sites, with a significantly positive partial correlation (R = 0.66, p < 0.001), followed by soil moisture (SM) (R = 0.45, p < 0.01), vapor pressure deficit (VPD) (R = −0.37, p < 0.05), shortwave radiation (SW) (R = 0.35, p < 0.05), and wind speed (WS) (R = 0.35, p < 0.05). However, no significant relationship was found between precipitation (PRE) and Vgreenup. TA was significantly negatively correlated with Dgreenup (R = −0.51, p < 0.01) and was positively correlated with Agreenup (R = 0.71, p < 0.001), which indicated that increasing TA would shorten Dgreenup and inhibit Agreenup. Meanwhile, SM and VPD showed contrasting effects on Dgreenup as well as on Agreenup. SM exhibited a negative correlation with Dgreenup (R = −0.46, p < 0.01) and a positive correlation with Agreenup (R = 0.41, p < 0.05), while VPD was positively correlated with Dgreenup (R = 0.41, p < 0.05) and negatively correlated with

Spatial Relationships between Vgreenup and Climatic Factors
The spatial pattern of Vgreenup was regulated by multiple spring climatic factors across all sites in the whole study area (Figure 4a). Temperature (TA) played a predominant role in controlling the spatial variation in Vgreenup across all sites, with a significantly positive partial correlation (R = 0.66, p < 0.001), followed by soil moisture (SM) (R = 0.45, p < 0.01), vapor pressure deficit (VPD) (R = −0.37, p < 0.05), shortwave radiation (SW) (R = 0.35, p < 0.05), and wind speed (WS) (R = 0.35, p < 0.05). However, no significant relationship was found between precipitation (PRE) and Vgreenup. TA was significantly negatively correlated with Dgreenup (R = −0.51, p < 0.01) and was positively correlated with Agreenup (R = 0.71, p < 0.001), which indicated that increasing TA would shorten Dgreenup and promote Agreenup. Meanwhile, SM and VPD showed contrasting effects on Dgreenup as well as on Agreenup. SM exhibited a negative correlation with Dgreenup (R = −0.46, p < 0.01) and a positive correlation with Agreenup (R = 0.41, p < 0.05), while VPD was positively correlated with Dgreenup (R = 0.41, p < 0.05) and negatively correlated with Agreenup (R = −0.33, p > 0.05). This indicated that increasing VPD and decreasing SM would lengthen Dgreenup and lower Agreenup. Besides, a negative effect of SW on Dgreenup (R = −0.45, p < 0.01) and a positive effect of WS on Agreenup were also observed (R = 0.49, p < 0.01). It could also be observed that PRE had a minor influence on Dgreenup and Vgreenup.
In Europe, the effects of spring climatic factors on Vgreenup and its components were similar to those in the whole study area. Except for PRE, all spring climatic factors showed significant correlations with either Dgreenup or Agreenup in Europe (Figure 4b). A significantly positive effect of TA (R = 0.77, p < 0.01) and SW (R = 0.60, p < 0.05) on Vgreenup and a significantly negative effect of VPD on Vgreenup (R = −0.78, p < 0.01) could be found. However, no significant relationship was found between Vgreenup and spring climatic factors except for TA (R = 0.51, p < 0.05) in North America (Figure 4c). TA also showed significant relationships with Dgreenup (R = −0.54, p < 0.05) and Agreenup (R = 0.56, p < 0.05) in North America. Besides, SM was significantly and negatively correlated with Dgreenup (R = −0.61, p < 0.01) but exerted no significant influence on Agreenup or Vgreenup in North America (Figure 4c).   This study also analyzed the temperature sensitivity of Vgreenup and its components ( Figure 5). In response to a 1 • C increase in TA, Dgreenup would shorten by 0.87 days across all sites of the whole study area, while Agreenup and Vgreenup would increase by 0.41 g C m −2 day −1 and 0.019 g C m −2 day −2 , respectively (Figure 5a-c). The temperature sensitivity amplitudes in relation to Vgreenup and its components were different between Europe and North America (Figure 5d-i). The temperature sensitivity of Dgreenup was slightly greater in Europe (−0.90 days • C −1 ) than in North America (−0.85 days • C −1 ) (Figure 5d,g). However, the effect of TA on Agreenup was much stronger in North America than in Europe. With a 1 • C increase in TA, Agreenup would increase by 0.20 and 0.53 g C m −2 day −1 in Europe and North America, respectively (Figure 5e,h). Therefore, a greater temperature sensitivity of Vgreenup was observed in North America (0.021 g C m −2 day −2 • C −1 ) than in Europe (0.015 g C m −2 day −2 • C −1 ) (Figure 5f,i). respectively. *, ** and *** indicate significant correlation at p < 0.05, 0.01 and 0.001 levels, respectively.
This study also analyzed the temperature sensitivity of Vgreenup and its components (Error! Reference source not found.). In response to a 1 °C increase in TA, Dgreenup would shorten by 0.87 days across all sites of the whole study area, while Agreenup and Vgreenup would increase by 0.41 g C m −2 day −1 and 0.019 g C m −2 day −2 , respectively (Figure 5a−c). The temperature sensitivity amplitudes in relation to Vgreenup and its components were different between Europe and North America (Figure 5d−i). The temperature sensitivity of Dgreenup was slightly greater in Europe (−0.90 days °C −1 ) than in North America (−0.85 days °C −1 ) (Figure 5d, g). However, the effect of TA on Agreenup was much stronger in North America than in Europe. With a 1 °C increase in TA, Agreenup would increase by 0.20 and 0.53 g C m −2 day −1 in Europe and North America, respectively (Figure 5e, h). Therefore, a greater temperature sensitivity of Vgreenup was observed in North America (0.021 g C m −2 day −2 °C −1 ) than in Europe (0.015 g C m −2 day −2 °C −1 ) (Figure  5f, i).

Discussion
Many previous studies have reported the distribution patterns along latitude, longitude, and altitude for vegetation phenology and productivity in adaptation to climate changes [27][28][29]. This study revealed the spatial pattern of Vgreenup across PFTs in the northern temperate and boreal regions. A strong dependence on latitude was observed for Vgreenup across all sites in both Europe and North America (Figure 2a), with a decreasing trend with rising latitude. This means there was more rapid canopy development in lower latitudes than in higher latitudes. However, Vgreenup demonstrated no clear patterns with longitude and altitude (Figure 2b,c). Furthermore, the above patterns were also confirmed when only considering sites in Europe or North America (Figure 2d-i). A previous study based on remote sensing observation found the latitude dependence of Vgreenup on both continents, but an opposite trend with latitude was observed in this study [21]. The discrepancies might be caused by the difference in the target scale between the satellite and field observations. The spatial variation of Vgreenup was associated with the decoupling of the spatial patterns of its two components [19,22]. In this study, both Dgreenup and Agreenup were significantly correlated with Vgreenup along a spatial dimension, indicating that the spatial variations of Vgreenup were jointly regulated by Dgreenup and Agreenup (Figure 3a,b). In the whole study area, Vgreenup was more correlated with Agreenup (R 2 = 0.83) than with Dgreenup (R 2 = 0.61), which implies a more predominant effect of Agreenup variability compared with Dgreenup variability. This result is in line with the finding of a previous satellite-based study at a global scale [19]. Nevertheless, the variations of Dgreenup and Agreenup made different contributions to Vgreenup variation between continents. According to the correlation coefficients between Vgreenup and its components (Figure 3c-f), the spatial pattern of Vgreenup was more determined by Dgreenup in Europe but by Agreenup in North America.
This study further investigated the controls of the main spring climatic factors on the distributions of Vgreenup and its components (Dgreenup and Agreenup), by using the partial correlation. TA showed significant effects on Dgreenup (R = −0.51, p < 0.01), Agreenup (R = 0.71, p < 0.001) and Vgreenup (R = 0.66, p < 0.001) in the whole study area (Figure 4a). Specifically, increasing TA could shorten Dgreenup and promote Agreenup simultaneously, leading to an acceleration of Vgreenup. TA was reported to be the major driver for spring phenology based on long-term FLUXNET measurements [30]. Advanced SOG and the peak of the growing season (closing to EOG) in response to climate warming had been widely observed [10,[31][32][33], but their non-uniform response magnitude would alter Agreenup. Under global warming, global average Agreenup decreased by 1.4 days during 1982-2015, according to the satellite LAI product [19]. Besides, TA had a stronger link with Agreenup (R = 0.71, p < 0.001) than with Dgreenup (R = −0.51, p < 0.01) in this study. The boosting effect of TA on Agreenup was consistent with the global greening and increasing productivity in recent decades due to climate warming [17,34,35]. Zhang et al., also reported the positive effect of increasing TA on the seasonal maximal gross primary productivity (GPP max ) based on eddy covariance at an alpine meadow site [36]. Rising TA can promote plant growth by enhancing enzymatic activity when TA is lower than the optimum temperature [37]. With the positive effect of spring TA on Vgreenup, increasing spring TA could enhance spring GPP, which had been previously confirmed across 21 FLUXNET sites in temperate and boreal forests [38]. In this study, increasing SW also contributed to shorter Dgreenup and higher Vgreenup, similar to effects of TA (Figure 4a). SW is highly related to the heat requirement for leaf unfolding [39]. Lower solar radiation can significantly delay spring bud development by decreasing the accumulation of growing degree days [40]. Therefore, SW can influence Dgreenup and Vgreenup by altering the temperature sensitivity of SOG and EOG [41].
VPD and SM are commonly used indicators for atmosphere and soil dryness, respectively [42][43][44]. A significant effect by drought on vegetation greenup development was observed in this study. By increasing VPD or decreasing SM, Dgreenup and Agreenup tended to become longer and lower, respectively, resulting in the deceleration of Vgreenup for the whole study area (Figure 4a). Previous studies based on eddy covariance observations also demonstrated the reduction in photosynthesis caused by dryness [45,46]. GPP max has also been reported to be enhanced by summer PRE, based on data from 24 AmeriFlux sites [47]. However, no significant relationship was found between spring PRE and Vgreenup in this study. This might be caused by the fact that vegetation growth was more directly affected by SM and VPD compared with PRE. Vegetation tends to lower photosynthesis by closing stomata in a high VPD condition [48,49]. Different from VPD, SM influences vegetation photosynthesis by controlling the absorption of soil nutrients. Increasing SM will facilitate plant nutrient absorption, leading to the stimulation of Vgreenup [50].
Besides, this study found that there were positive effects by wind speed on Agreenup and Vgreenup for the entire study area (Figure 4a). This might be explained by the fact that moderate wind speed would increase the gas exchange rate of leaf stomata, accelerating transpiration and intercellular CO 2 exchange, which results in a higher intercellular CO 2 concentration and improved photosynthesis capacity [51].
In this study, the climatic controls on the spatial variations of Vgreenup were also compared between Europe and North America. The temperature sensitivities of Agreenup and Vgreenup were greater in North America than in Europe ( Figure 5). This difference in temperature sensitivity could be caused by the effects of climatic factors other than TA. The spatial variations of Vgreenup were jointly controlled by TA, SW, and VPD in Europe but only affected by TA in North America (Figure 5b,c). Without the limitations of solar radiation and water supply, the vegetation may have developed greater temperature sensitivities to maximize the thermal benefits [52][53][54].
ENF sites were included to investigate Vgreenup in this study. It is difficult to extract the phenological metrics from ENF by using NDVI or LAI, due to their narrow seasonal variations in vegetation greenness [55,56]. In comparison, GPP derived from eddy covariance has a great advantage in extracting the phenological metrics from ENF, which can help improve the estimation of Vgreenup for ENF. Besides, solar-induced chlorophyll fluorescence can reflect vegetation photosynthesis directly [57], which can be used to investigate the phenological metrics and Vgreenup for ENF in the future. VPD and SM are often coupled through land-atmosphere interactions, hindering the ability to predict ecosystem responses to dryness [58,59]. Several methods have been proposed to distinguish the effects of VPD and SM on vegetation growth [60][61][62]. Multi-methods may need to be synthesized for quantifying the relative contributions of VPD and SM to Vgreenup in future work.

Carbon Fluxes and Climatic Data
Daily GPP data and climatic data were retrieved from the FLUXNET 2015 Tier 1 dataset (https://fluxnet.org/data/fluxnet2015-dataset/, accessed on 10 May 2022). The FLUXNET 2015 dataset provides in situ estimates of carbon, water, and heat fluxes between ecosystems and atmosphere of 212 sites across the globe, which are determined using the eddy covariance technique. The eddy covariance data for all the sites were processed by consistent and uniform procedures including quality control, gap-filling, and net ecosystem exchange (NEE) partitioning [63]. NEE measurements were partitioned into GPP and ecosystem respiration (ER). GPP based on nighttime partitioning method (GPP_NT_VUT_REF) was used in this study [64]. To avoid spurious effects caused by low-quality data, sites having at least 5 years with high-quality carbon flux data (NEE_VUT_REF_QC > 0.75 in a year) above 30 • N were selected. In addition, only sites with mean start of greenup (SOG, defined below) occurring during spring (April-June) were included in this study. At last, 40 sites (417 site-year data) were selected to be analyzed, including 14 DBF sites, 10 ENF sites, 3 MF sites, 3 GRA sites, 8 CRO sites, and 2 WET sites ( Figure 6, Table 1).
analyzed, including 14 DBF sites, 10 ENF sites, 3 MF sites, 3 GRA sites, 8 CRO sites, and 2 WET sites ( Figure 6, Table 1). To analyze the relationship between Vgreenup and climatic variables, this study also collected the daily climatic data including TA, PRE, SW, WS, and VPD provided by the FLUXNET sites. Considering the unavailability of SM at most sites, the daily root-zone SM was extracted from Global Land Evaporation Amsterdam Model (GLEAM) v3.5a to reflect the soil water condition in root zone of each site (https://www.gleam.eu/, accessed on 15 May 2022) [65,66]. The depth of the root zone is different for low vegetation (0-100 cm) and tall vegetation (0-250 cm). The GLEAM v3.5a dataset was produced by combining satellite and reanalysis data, which provides temporally/spatially continuous SM at spatial resolution of 0.25° (~25 km), spanning from 1980 to 2020. The overall performance accuracy of GLEAM SM products is considerably high [43]. The mean values in spring were calculated for all climatic factors.

Calculation of Vgreenup
To remove outliers of GPP data, a Savitzky-Golay filter was adopted to smooth the daily GPP time series [67]. The width of the moving window was set to be 91 days, and the iteration time was set to be 1 according to a previous study [68]. Then, a modified double logistic function was used to fit the daily GPP for each year [32,69]: where f(t) is the daily GPP at day of year (DOY) t; is the background GPP; and represent the amplitude relative to the background for the early summer plateau and the late summer plateau, respectively; and are the transition curvature parameters (normalized slope coefficients); and and are the midpoints in DOYs of these transitions for greenup and senescence, respectively.
The relative threshold method is a simple but practical method to extract phenological metrics [70]. In this study, the start and end of greenup period (SOG and EOG) were defined as the dates when the fitted GPP curve first reach 20% and 80% the seasonal amplitude, respectively [19]. Dgreenup was defined as the difference between EOG and SOG. Agreenup was calculated as the difference in GPP values at dates of EOG and SOG. At last, Vgreenup was calculated as the ratio of Agreenup to Dgreenup [19]. The process for determining Dgreenup, Agreenup, and Vgreenup is illustrated in Figure 7. To analyze the relationship between Vgreenup and climatic variables, this study also collected the daily climatic data including TA, PRE, SW, WS, and VPD provided by the FLUXNET sites. Considering the unavailability of SM at most sites, the daily root-zone SM was extracted from Global Land Evaporation Amsterdam Model (GLEAM) v3.5a to reflect the soil water condition in root zone of each site (https://www.gleam.eu/, accessed on 15 May 2022) [65,66]. The depth of the root zone is different for low vegetation (0-100 cm) and tall vegetation (0-250 cm). The GLEAM v3.5a dataset was produced by combining satellite and reanalysis data, which provides temporally/spatially continuous SM at spatial resolution of 0.25 • (~25 km), spanning from 1980 to 2020. The overall performance accuracy of GLEAM SM products is considerably high [43]. The mean values in spring were calculated for all climatic factors.

Calculation of Vgreenup
To remove outliers of GPP data, a Savitzky-Golay filter was adopted to smooth the daily GPP time series [67]. The width of the moving window was set to be 91 days, and the iteration time was set to be 1 according to a previous study [68]. Then, a modified double logistic function was used to fit the daily GPP for each year [32,69]: where f (t) is the daily GPP at day of year (DOY) t; α 1 is the background GPP; α 2 − α 1 and α 3 − α 1 represent the amplitude relative to the background for the early summer plateau and the late summer plateau, respectively; ∂ 1 and ∂ 2 are the transition curvature parameters (normalized slope coefficients); and β 1 and β 2 are the midpoints in DOYs of these transitions for greenup and senescence, respectively. The relative threshold method is a simple but practical method to extract phenological metrics [70]. In this study, the start and end of greenup period (SOG and EOG) were defined as the dates when the fitted GPP curve first reach 20% and 80% the seasonal amplitude, respectively [19]. Dgreenup was defined as the difference between EOG and SOG. Agreenup was calculated as the difference in GPP values at dates of EOG and SOG. At last, Vgreenup was calculated as the ratio of Agreenup to Dgreenup [19]. The process for determining Dgreenup, Agreenup, and Vgreenup is illustrated in Figure 7.

Statistical Analysis
Considering the limited time series of GPP, this study only analyzed the spatial variations rather than the temporal variations for Vgreenup. The indicators of Dgreenup, Agreenup, and Vgreenup were derived from each site-year. Then, the multi-year indicators were averaged at each site to analyze the spatial pattern of Dgreenup, Agreenup, and Vgreenup. One-way analysis of variance (ANOVA) and least significant difference (LSD) multiple comparisons test were performed to determine significant differences in Vgreenup among PFTs. The significant level was set at α = 0.05. Simple linear regression was adopted to analyze the relationship between Vgreenup and geographical factors (latitude, longitude, and altitude). Correlations between spatial anomalies of Vgreenup and its components (Dgreenup and Agreenup) were analyzed to identify whether the spatial variations in Vgreenup were affected by Dgreenup and Agreenup. Partial correlation analysis was also conducted to evaluate the isolated effects of spring climatic factors (TA, PRE, SW, VPD, WS, and SM) on Dgreenup, Agreenup, and Vgreenup. The statistical significances of the regression and correlation coefficients were examined using the t-test. The p-values less than 0.05 were considered significant. The direction and magnitude regarding driving factors on spatial variations of Vgreenup were compared between Europe and North America. Besides, the results in this study were also compared with the relevant findings in previous studies.

Conclusions
Based on 417 site-years' flux data from 40 sites in the northern temperate and boreal regions, this study examined the characters of Vgreenup for different PFTs and further investigated the roles of climatic variables in controlling the spatial variation of Vgreenup. The results found that CRO and ENF had the largest and lowest Vgreenup among all PTFs, respectively. Vgreenup showed strong dependence on latitude but no obvious dependence on longitude or altitude. The spatial variations of Dgreenup and Agreenup jointly controlled the distribution of Vgreenup but made distinct contributions to Vgreenup variation between Europe and North America. Different controls of spring climatic factors on the spatial variation of Vgreenup were also observed between Europe and North America. In Europe, TA and SW had a positive effect on Vgreenup, while VPD Figure 7. Scheme for determination of Dgreenup, Agreenup, and Vgreenup using daily GPP data at IT-Col site (deciduous broadleaf forest) in 2014. GPP max and GPP min represent the maximum and minimum smoothed GPP during the growth period, respectively. A GPP represents the seasonal amplitude of GPP, which is the difference between GPP max and GPP min .

Statistical Analysis
Considering the limited time series of GPP, this study only analyzed the spatial variations rather than the temporal variations for Vgreenup. The indicators of Dgreenup, Agreenup, and Vgreenup were derived from each site-year. Then, the multi-year indicators were averaged at each site to analyze the spatial pattern of Dgreenup, Agreenup, and Vgreenup. One-way analysis of variance (ANOVA) and least significant difference (LSD) multiple comparisons test were performed to determine significant differences in Vgreenup among PFTs. The significant level was set at α = 0.05. Simple linear regression was adopted to analyze the relationship between Vgreenup and geographical factors (latitude, longitude, and altitude). Correlations between spatial anomalies of Vgreenup and its components (Dgreenup and Agreenup) were analyzed to identify whether the spatial variations in Vgreenup were affected by Dgreenup and Agreenup. Partial correlation analysis was also conducted to evaluate the isolated effects of spring climatic factors (TA, PRE, SW, VPD, WS, and SM) on Dgreenup, Agreenup, and Vgreenup. The statistical significances of the regression and correlation coefficients were examined using the t-test. The p-values less than 0.05 were considered significant. The direction and magnitude regarding driving factors on spatial variations of Vgreenup were compared between Europe and North America. Besides, the results in this study were also compared with the relevant findings in previous studies.

Conclusions
Based on 417 site-years' flux data from 40 sites in the northern temperate and boreal regions, this study examined the characters of Vgreenup for different PFTs and further investigated the roles of climatic variables in controlling the spatial variation of Vgreenup. The results found that CRO and ENF had the largest and lowest Vgreenup among all PTFs, respectively. Vgreenup showed strong dependence on latitude but no obvious dependence on longitude or altitude. The spatial variations of Dgreenup and Agreenup jointly controlled the distribution of Vgreenup but made distinct contributions to Vgreenup variation between Europe and North America. Different controls of spring climatic factors on the spatial variation of Vgreenup were also observed between Europe and North America. In Europe, TA and SW had a positive effect on Vgreenup, while VPD showed a negative influence. However, Vgreenup was only affected by TA in North America. Such a difference led to a higher temperature sensitivity for Vgreenup in North America than in Europe. This study can help us to better understand the development of vegetation greenup from the perspective of field observation, which is an important supplement and validation for satellite-based research. Besides, long-term field observation is needed to further investigate the temporal dynamics of Vgreenup, for a more comprehensive analysis in the future.