Contrasting temperature effects on the velocity of early- versus late-stage vegetation green-up in the Northern Hemisphere

1 INTRODUCTION

Global greening, indicated by the satellite-derived vegetation indicators, for example, Normalized Difference Vegetation Index (NDVI) and leaf area index (LAI), is one of the highly credible evidence for anthropogenic climate change (Chen, Park, et al., 2019; Piao et al., 2020; Zhu et al., 2016). This greening phenomenon indicates a major response of terrestrial ecosystems to climate change (Zhu et al., 2016), which also provides multiple feedbacks to Earth systems through biophysical and biogeochemical processes (Cheng et al., 2017; Keenan et al., 2016; Zeng et al., 2017). For example, it has been widely confirmed that greening enhances vegetation photosynthesis (Chen, Ju, et al., 2019; Winkler et al., 2019; Zhang et al., 2018), as well as terrestrial ecosystem carbon storage (Shevliakova et al., 2013). During the past decades, greening plays an important role in mitigating global warming, offsetting almost one third of anthropogenic CO₂ emission (Friedlingstein et al., 2020). Moreover, greening also significantly impacts hydrological cycle (Lian et al., 2020; Zeng et al., 2018) and alters the land surface energy balance through changes in evapotranspiration, albedo, and other land surface biophysical properties (Arora & Montenegro, 2011; Zeng et al., 2017). Considering large increases in the vegetation greenness during GUP and the widely observed changes in plant phenology, the impacts and feedbacks of the vegetation changes are more predominate during GUP (Buermann et al., 2018; Gonsamo et al., 2017). Therefore, it is important to understand how climate change regulates the interannual changes of vegetation green-up velocity.

For the temperate ecosystems, vegetation greening (i.e., the interannual increase of the annual integral of NDVI) indicates the enhanced magnitude of green-up (A_green-up) during GUP, which can be attributed to the lengthening of GUP (T_green-up) and/or the interannual increase of the velocity of leaf green-up (V_green-up, interannual increase of V_green-up was defined as green-up acceleration), given that A_green-up = T_green-up × V_green-up (Wang et al., 2018). On the one hand, climate warming induces the advance of the start of the growing season (SOS; Piao et al., 2006; Zhou et al., 2001), which would lengthen T_green-up if the peak of growing season (POS) remains relatively stable or has a relatively slower advancing speed. On the other hand, climate change may accelerate V_green-up through enhancing activities of enzyme (Kardol et al., 2010) and increasing water and nutrient availabilities (Finger et al., 2016; Keuper et al., 2012; Nemani et al., 2003). Previous studies have generally confirmed the advance of SOS in the past decades (Cleland et al., 2007; Cong et al., 2013; Piao et al., 2019) with an average rate of 2.1 days per decade from 1982 to 2011. Meanwhile, significant advance of POS in northern hemisphere was observed in 19.4% area from 1982 to 2012, with 0.61 days per year in these areas (Xu et al., 2016). These results explored the potential changes in T_green-up, but the changes in V_green-up in the past decades still need further investigation.

Wang et al. (2018) found that global average V_green-up increased by 0.003 m² m⁻² year⁻¹ from 1982 to 2015 (i.e., green-up acceleration). This acceleration of vegetation green-up was observed in most ecosystems except for closed evergreen coniferous forests, lichens and mosses, and sparse vegetation (Wang et al., 2018). However, the overall increase of V_green-up does not fully characterize the dynamic of vegetation because vegetation responses to climate change may be different during each month of the GUP (Zhang et al., 2020; Zhou et al., 2001). The limiting factor of vegetation green-up during different months may vary and changes of vegetation will further feedback to the climate (Lian et al., 2020; Lian et al., 2022), which are important for future climate projection. In addition, the changes of V_green-up in each month of GUP may show high spatial heterogeneity because of the large differences in biome distribution and their responses to the climate variations (Lian et al., 2020; Piao et al., 2019; Zhang et al., 2020; Zhou et al., 2001). Therefore, exploring the spatiotemporal changes of the velocity of vegetation green-up and identifying the key stage of green-up that mainly influences vegetation productivity are crucial for understanding the climate-vegetation feedbacks and even the biosphere–atmosphere interactions within the Earth system.

Here, we investigated the spatiotemporal change in the velocity of vegetation green-up and its climate drivers in middle-high latitude of the Northern Hemisphere (>30°N). Using NDVI data from Global Inventory Modeling and Mapping Studies (GIMMS), we defined V_NDVI (the NDVI increase between two consecutive months during GUP) to indicate the velocity of vegetation green-up at monthly scale (Figure 1). This method was firstly developed by Piao et al. (2022) and successfully used in temperate China to explore the speed of canopy development and senescence. Consequently, it allowed us to test our first hypothesis: The inter-annual trends in V_NDVI would vary among different months of GUP. Via performing partial correlation analysis between V_NDVI and climate factors (temperature, soil moisture and radiation), we further tested the second hypothesis: The different trends in V_NDVI in each month of GUP could result from their different responses to climate change. Furthermore, we classified the study areas according to the interannual changes of V_NDVI at different months of GUP to explore the key stage that mainly affects the trends in annual peak NDVI.

2 MATERIALS AND METHODS

3 RESULTS

3.1 Spatiotemporal changes in the velocity of vegetation green-up

An overall greening trend was observed in the study region from 1980s to 2010s (Figure 2a). Specifically, annual maximum NDVI (NDVI_max) showed positive trends in 67.41% areas, with 39.09% being significant (p < .05, Figure 2b). The increase in NDVI_max could result from the enhanced V_NDVI in different months of GUP, which was further explored below.

As shown in Figure 3a, areas with longer GUP usually presented smaller V_NDVI, with smaller variation of V_NDVI among different months, indicating a slow green-up process in these areas, for example, Europe and USA (Figure 4a–f). The highest velocity of green-up generally occurred in the second month of GUP (Figure 3a). In contrast, for the regions with shorter GUP (e.g., Siberia and the northernmost area of North America), monthly V_NDVI was much larger and showed greater variation among different months (Figures 3a and 4a–f), suggesting a rapid green-up. The largest V_NDVI was generally observed in the first month of GUP for these areas.

We further explored the interannual change in monthly V_NDVI and found divergent trends across different months of GUP (Figures 3b,c and 4g–r). For example, positive trends in V_NDVI were generally observed in Europe during March (Figure 4g). The positive trends spread to higher latitudes and inland regions in April (Figure 4h), with the onset of spring phenology in more areas (Figure S1a). Meanwhile, significantly positive trends in V_NDVI were also observed in the central North America (Figure 4h). During April, acceleration of vegetation green-up (positive interannual trend in V_NDVI) reached the peak and occurred in 71.46% areas. Due to the late SOS, positive trends in V_NDVI was observed in Siberia in May (Figure 4i) and in the northernmost region in June (Figure 4j). These results suggest that the velocity of green-up generally accelerated in early GUP. The overall results of trends in V_NDVI confirms this phenomenon (Figure 3b): The largest positive trends in V_NDVI were always observed in the first month regardless of the length of GUP. However, the trends in V_NDVI turned to negative in May for Europe and USA (Figure 4i), where spring phenology started relatively earlier. The negative trends in V_NDVI persisted to June and July in Eurasia (Figure 4j,k). These results suggested a general deceleration (negative interannual trend in V_NDVI) in late GUP. The robustness of above results was further confirmed by analyses using MODIS EVI (Figures S4 and S5).

The divergent trends in V_NDVI regulate the trends in monthly NDVI (Figure S7). Due to the acceleration of green-up in early GUP, we observed significant greening trends (positive interannual trends in monthly NDVI) during March and April in Europe (Figures S7a,b). However, the greening trends became weaker in May (Figure S7c), which resulted from the significantly negative trends in V_NDVI during May. In other words, although green-up decelerated in late GUP, the enhanced velocity of green-up in early GUP still resulted in an overall greening trend in Europe. In contrast, we did not observe widespread greening trends in North America during March to May, and some areas even exhibited significantly negative trends in monthly NDVI, which was due to the weak positive (or even negative) trends in V_NDVI.

This divergent contributions of V_NDVI in different months to vegetation greenness were further confirmed as shown by Figure S8. We showed the month with maximum V_NDVI (M_mv) in Figure S8a and the month with the maximum trend in V_NDVI (M_mtv) in Figure S8b. M_mv showed a clear latitudinal pattern, changing from April in low latitude to June in high latitude (Figure S8a). M_mv usually occurred in the SOS month (M_SOS) in areas with short GUP but the month following M_SOS in areas with long GUP (Figure S8d), which again suggested different green-up paces. M_mtv also showed latitudinal pattern (Figure S8b), which occurred at SOS month for 46.25% areas (Figure S8e). Comparing with M_mv, M_mtv occurred earlier in 22.33% areas, simultaneously in 43.97% areas, and later in 33.70% areas (Figure S8g).

Furthermore, given the different V_NDVI in each month of GUP, we also explored the relative trends ((interannual trend in V_NDVI)/(multiyear average V_NDVI)) in V_NDVI (Figures 3c and 4m–r). Interestingly, although the trend in V_NDVI during the late GUP was much smaller than that in early GUP (Figure 3b), the relative trends of them were comparable (Figure 3c). Consequently, the month with maximum relative trend in V_NDVI (M_mrt) was not always same to M_mtv, which was later in 35.74% areas (earlier in 5.01% areas, simultaneously in 59.25% areas, Figure S8i). It indicates although V_NDVI in late GUP is relatively smaller than that in early stage, it is also sensitive to climate change.

3.2 Climate drivers of the interannual changes in the velocity of vegetation green-up

We further preformed partial correlation analysis to explore how climate factors (temperature, soil moisture and radiation) drove the changes in the velocity of vegetation green-up. Figure 5 shows that temperature played a key role in regulating the interannual trends in V_NDVI, but it presented opposite impacts between the early and late GUP. In March, positive correlation between temperature and V_NDVI was observed in 94.02% area, with 58.07% being significant (Figure 5a). This significant and positive effect of temperature on V_NDVI spread to higher latitudes and inland regions in April with the onset of spring phenology in more areas (Figure 5b, positive correlation in 85.41% area, with 53.39% being significant). However, much of the positive effect reversed into negative in these areas during May (Figure 5c). At the same time, soil moisture showed positive effect on V_NDVI, especially for the inland of Eurasia (Figure 5i), where vegetation growth was more constrained by water availability. In the high-latitude areas like Siberia, where SOS occurred in May, temperature still showed positive effect on V_NDVI (Figure 5c). In June, the negative effects of temperature were observed in most area (Figure 5d, negative correlation in 64.78% area, with 24.93% being significant). The same results were observed in July (Figure 5e), during which temperature presented negative effect on V_NDVI in 78.08% areas, with 28.07% being significant. Such divergent effects of temperature on V_NDVI were clearer as presented in Figure 5s, where positive effects were observed in the first month of GUP. In contrast, for the following months, negative effects of temperature on V_NDVI were generally observed. It is noteworthy that soil moisture still positively affected V_NDVI in the inland of Eurasia and North America during May and June (Figure 5i,j), suggesting the essential role of water supply in regulating vegetation growth in arid and semiarid ecosystems. In comparison with temperature, radiation showed relatively weaker impacts on V_NDVI over the entire GUP, and no clear spatial pattern of its impact was observed (Figure 5m–r). Analyses based on EVI also supports the above findings (Figure S6).

Consistent with V_NDVI, the temporal dynamics of monthly NDVI were also mainly regulated by temperature (Figure S9), but the positive effects of temperature on NDVI were stronger and lasted for a longer period of time than on V_NDVI (Figure 5t). The positive correlation between temperature and NDVI were observed in 95.33%, 94.95%, 91.77% and 86.06% areas in March, April, May and June, respectively (Figure S9a–d). Negative effects of temperature on NDVI were mainly observed in arid and semiarid areas in the inland of Eurasia in June and Siberia in July (Figure S9d,e). Soil moisture also showed positive effect on monthly NDVI in the inland of Eurasia, which were stronger than that on V_NDVI (Figure S9i–l). Similar to V_NDVI, the impacts of radiation on NDVI were weak and showed high spatial heterogeneity (Figure S9m–r).

3.3 Portfolio of changes in the velocity of vegetation green-up

To further dissect the changes in green-up paces, we classified the study areas into seven groups based on the interannual trends in V_NDVI in different months of GUP (Figure 6). Type “+ −” indicates vegetation green-up accelerated in early GUP but decelerated in late GUP, which accounted for 45% areas (Figure 6b). This type mainly distributed in Europe, Siberia, and northwest of North America (Figure 6a), where significant increase of peak NDVI was observed (Figure 2a). Another two groups including “+” (acceleration in all months) and “+ − +” (acceleration in early GUP, then deceleration, finally acceleration) accounted for 11% and 15%, respectively (Figure 6b). They mainly distributed in central Eurasia and central North America (Figure 6a), where peak NDVI also significantly increased in past decades (Figure 2a). In contrast, the types indicating deceleration of green-up in early GUP including “−” (deceleration in all months), “− +” (deceleration in early GUP but acceleration in late GUP), and “− + −” (deceleration in early GUP, then acceleration, finally deceleration in late GUP) generally distributed in areas where we observed negative trend in peak NDVI (Figures 2a and 6a). Typically, type “− +” mainly distributed in Canada, which accounted for 13% of study areas (Figure 6b). The remaining three groups including “−”, “− + −” and “other” accounted for 3%, 9% and 4% areas, respectively (Figure 6b).

We further compared the dynamics of NDVI_max between group with increasing V_NDVI in early GUP (“+”, “+ −” and “+ − +)” and group with decreasing V_NDVI in early GUP (“−”, “− +” and “− + −” (Figure 6c). For the group with increasing V_NDVI in early GUP, we observed positive trend in NDVI_max in 72.67% (44.41% being significant) areas, with an average trend of 0.0010 year⁻¹. In contrast, for the group with decreasing V_NDVI in early GUP, we only observed positive trend in NDVI_max in 53.89% (26.09% being significant) areas. The average trend in NDVI_max in this group was 0.0002 year⁻¹, significantly smaller than that of group with increasing V_NDVI in early GUP (p < .001, one-way analysis of variance). Comparison between type “+ −” and “− +” also confirmed such difference (Figure 6c).

Such patterns become more evident if we converted the calendar month to phenological month (Figure S10). The first month of GUP (i.e., SOS month) showed the most substantial interannual trend (Figure S10a), which exhibited similar spatial pattern to the trend in NDVI_max (Figure 2a). Positive trend in V_NDVI for the SOS month was observed in 72.97% areas, with 52.75% having increasing NDVI_max. That is, for areas showing acceleration of vegetation green-up in the SOS month, 72.29% of them showed increasing NDVI_max. The above results indicate the acceleration of vegetation green-up in early GUP, especially the first month, is the major driver of increasing NDVI in summer.

4 DISCUSSION

Many previous studies have explored the changes in NDVI induced by climate change and land use management (Chen, Park, et al., 2019; Piao et al., 2020; Zhu et al., 2016), but studies focusing on the spatiotemporal changes in V_NDVI are rather limited (Wang et al., 2018). In this study, we observed an inconsistency between the changes in NDVI and V_NDVI: NDVI generally showed positive inter-annual trends in most months during GUP, while the trends in V_NDVI varied across different months. Specifically, vegetation green-up generally accelerated in early GUP but decelerated in late GUP, which confirms our first hypothesis. The different inter-annual trends between NDVI and V_NDVI is mainly due to that NDVI represents the state of canopy, but V_NDVI represents the process of canopy development, which is directly related to the photosynthetic carbon assimilation and partitioning. Therefore, V_NDVI could be more sensitive to the climate change at monthly scale and can be a better indicator of sub-seasonal vegetation dynamics than NDVI. NDVI, on the other hand, depends not only on vegetation responses to recent climate (V_NDVI), but also its status of the previous timestep. Further analyses confirmed above views. Although both NDVI and V_NDVI were dominantly regulated by temperature, comparing with the overall positive response of NDVI to increasing temperature, V_NDVI showed divergent responses to climate warming in early versus late GUP. In early GUP, climate warming generally accelerated vegetation green-up, but in late GUP, warming decelerated green-up paces, which confirms our second hypothesis. The observed declining temperature dependency of NDVI may be explained by a strong positive temperature effect on V_NDVI during the early GUP, whose carryover effect is progressively enervated by the negative temperature effect on V_NDVI for later GUP.

The divergent responses of V_NDVI to climate warming may be due to different resource limitations of vegetation growth in early and late GUP. During the early GUP, climate warming could accelerate vegetation green-up via following mechanisms: First, climate warming induced the advance of SOS (Piao et al., 2006; Zhou et al., 2001). Earlier start of vegetation photosynthesis would enhance leaf C accumulation and thus form a positive feedback (Chapin, 1991), resulting in the increasing V_NDVI in early GUP. Second, vegetation growth is usually limited by energy in spring because of the low temperature (Chapin, 1991; Chapin et al., 2002), hence warming would alleviate the energy restriction, then increase vegetation productivity and accelerate vegetation greening (Zhu et al., 2016). Third, increasing temperature in spring could increase soil moisture and nutrient availability through enhancing the thawing of frozen soils in tundra ecosystems (Finger et al., 2016; Keuper et al., 2012), further accelerates vegetation greening.

However, theses mechanisms cannot persist to the late GUP when vegetation growth is no longer limited by heat. On the contrary, climate warming can directly inhibit photosynthesis if the temperature exceeds the temperature optima of vegetation productivity, and consequently decelerated vegetation green-up (Chen et al., 2021; Huang et al., 2019). In addition, the acceleration of vegetation green-up in early GUP may lead to premature depletion of soil moisture and nutrient (Lian et al., 2020; Zhang et al., 2021), and thus result in soil drought and nutrient deficiency, slowing down the velocity of vegetation green-up in late GUP. This effect would be further enhanced by increasing temperature, which increases atmospheric vapor pressure deficit and intensifies the water stress (Yuan et al., 2019). On this occasion, vegetation would allocate less carbon to organs consuming water (i.e., leaves) but more carbon to organs for water acquisition and transportation (e.g., roots and stems), which also results in the deceleration of vegetation green-up (Chapin et al., 2002). Indeed, the negative effects of warming on V_NDVI were observed earlier in the water-limited ecosystems of the inland areas, where soil moisture showed strong and positive effects on vegetation green-up (Figure 5). Interestingly, such divergent responses of vegetation green-up to climate warming were also found among different regions, especially for the temperature-constrained ecosystems vs. arid and semiarid ecosystems (Wang et al., 2018). Combining our study with the results in Wang et al. (2018), it is likely that climate warming accelerates vegetation green-up in the temperature limited regions and/or stages, but decelerates it where/when temperature is not major limiting factor.

Besides the shifting of limiting factors, the observed divergent changes of V_NDVI in different months may also reflect the movement of vegetation growth cycle (Park et al., 2015). Annual cycle of vegetation growth is impacted not only by environmental resource supplies but also by the longevity of leaf structure (Reich et al., 1992) and programmed cell-death in plants (Lim et al., 2007). Therefore, earlier leaf unfolding would also lead to the advance of programmed leaf life cycle and further result in different changes of V_NDVI in early and late GUP. As our analyses were conducted based on grid NDVI product, the mixed pixel effect would make the changes of V_NDVI more complicated (Chen et al., 2018). For instance, it is well known that increasing temperature can accelerate plant growth via temperature-dependent enzymatic catalytic reactions if temperature is lower than temperature optima (Atkinson & Porter, 1996). However, the optimal temperature of different species in a same pixel may be various. The different temperature optima and asynchronous movements of growth cycles among different species would generate a nonlinear response of V_NDVI to climate change at landscape or larger levels (Chen et al., 2018).

Across the mid-high-latitude Northern Hemisphere, areas with earlier starts of green-up usually showed slower green-up path (Figures 3 and 4). In other words, spatially, higher spring temperature leads to earlier start of green-up but negative effect on its velocity. This paradox was also found by previous studies (Clark et al., 2015; Seyednasrollah et al., 2018). Moreover, the temporal trends of V_NDVI and their relationships with climate change showed high spatial heterogeneity. Specifically, in areas with relatively shorter GUP, the overall temporal changes in V_NDVI were clear, that is, acceleration in the early stage but deceleration in the last stage (Figure 3). In contrast, taking the areas with longer GUP (>3 months) as a whole, V_NDVI showed positive trends in the early months but then negative trends in following months, and positive trends in the last months during the GUP, which may be resulted from more complicated feedbacks between vegetation and environment (climate and soil). Therefore, only focusing on the trends in vegetation greenness is far from enough to understand the response of vegetation to climate change because many important details are masked.

The portfolio of change in the velocity of vegetation green-up (Figure 6) can better describe the complicated vegetation dynamics during GUP. Our results indicate that the early GUP (especially the SOS month) is a key stage that regulates the peak NDVI in summer. Vegetation green-up acceleration in SOS month generally results in a higher vegetation greenness in summer, which can be widely observed in Eurasia. In contrast, in a large part of North America, vegetation green-up deceleration in SOS month resulted in vegetation browning (negative NDVI trend) in past decades. Such deceleration was likely caused by the decreasing temperature in early spring (Figure S11). These results may result from the vegetation carryover effect (Lian et al., 2021) and demonstrated that vegetation green-up acceleration in the first month of GUP was an important contributor to the recently observed annual greening trend. Like an old saying in Chinese “The whole year's work depends on a good start in spring”, our results highlight the importance of vegetation growth in early spring to peak vegetation productivity.

It is also important to highlight, however, that our results could involve some uncertainties. Given that our analyses were conducted at monthly level, the advance of leaf unfolding could also induce the lengthening of growth period in the first month, which also increases V_NDVI. Therefore, it is important to investigate green-up trajectories using data with higher temporal resolution in future studies. Nevertheless, the overall acceleration of green-up during the whole GUP (Wang et al., 2018) and the deceleration of green-up in the late stage again confirmed the robustness of green-up acceleration in early spring. It is also noteworthy that SOS derived from satellite data is generally much later than date of leaf unfolding (Gordo & Sanz, 2009), so the bias induced by changes in leaf unfolding date may be negligible.

In summary, our study revealed the divergent trends in the velocity of vegetation green-up in different months of GUP and their responses to climate change. These results suggest that the velocity of vegetation green-up can be more sensitive to climate change than vegetation greenness (e.g., NDVI, LAI) at monthly scale and thus can be used to understand the dynamics of vegetation growth at sub-seasonal or even finer scale. Moreover, our study also suggests that the first month of GUP is the key stage that determines vegetation productivity in summer and potentially in the whole year. Accordingly, green-up acceleration in the first month of GUP contributed largest to global greening trend over the past decades. Given the different magnitude of climate change and divergent responses of V_NDVI to climate change in different months of GUP, it is critically important to further explore the spatiotemporal heterogeneity in the velocity of vegetation green-up and its climate drivers through more long-term observation and more accurate modelling approach for understanding and predicting the climate-vegetation feedbacks.

AUTHOR CONTRIBUTIONS

Songbai Hong and Yao Zhang designed the research; Songbai Hong performed the data analysis; Songbai Hong and Yao Zhang contributed to the interpretation of the results; Songbai Hong drafted the manuscript; all authors participated in discussions and the editing of the manuscript.

CONFLICT OF INTEREST

The authors declare no competing interests.

DATA AVAILABILITY STATEMENT

All data used in this study are openly available from the following: GIMMS NDVI are available at http://poles.tpdc.ac.cn/en/data/9775f2b4-7370-4e5e-a537-3482c9a83d88/. CRU air temperature is available at https://crudata.uea.ac.uk/cru/data/hrg/cru_ts_4.05/cruts.2103051243.v4.05/; C3S soil moisture is available at https://cds.climate.copernicus.eu/cdsapp#!/dataset/satellite-soil-moisture?tab=overview; CRU-JRA v2.2 solar radiation is available at https://catalogue.ceda.ac.uk/uuid/7f785c0e80aa4df2b39d068ce7351bbb. MODIS EVI is available at https://lpdaac.usgs.gov/products/mod13c2v006/. All computer codes used in this study are available from the corresponding author upon reasonable request.