Seasonal divergence in the interannual responses of Northern Hemisphere vegetation activity to variations in diurnal climate

Seasonal asymmetry in the interannual variations in the daytime and nighttime climate in the Northern Hemisphere (NH) is well documented, but its consequences for vegetation activity remain poorly understood. Here, we investigate the interannual responses of vegetation activity to variations of seasonal mean daytime and nighttime climate in NH (>30 °N) during the past decades using remote sensing retrievals, FLUXNET and tree ring data. Despite a generally significant and positive response of vegetation activity to seasonal mean maximum temperature () in ~22–25% of the boreal (>50 °N) NH between spring and autumn, spring-summer progressive water limitations appear to decouple vegetation activity from the mean summer , particularly in climate zones with dry summers. Drought alleviation during autumn results in vegetation recovery from the marked warming-induced drought limitations observed in spring and summer across 24–26% of the temperate NH. Vegetation activity exhibits a pervasively negative correlation with the autumn mean minimum temperature, which is in contrast to the ambiguous patterns observed in spring and summer. Our findings provide new insights into how seasonal asymmetry in the interannual variations in the mean daytime and nighttime climate interacts with water limitations to produce spatiotemporally variable responses of vegetation growth.

The vegetation activity in temperate and cold regions is strongly controlled by climate, which is causing rapid changes in the vegetation activity in the Northern Hemisphere (NH) as the Earth warms [1][2][3][4] . Satellite observations, Earth system models and atmospheric inversions suggest that growing-season warming has increased the photosynthetic activity of the NH terrestrial vegetation during past decades 1,2,4 to a greater extent than other environmental processes (e.g., CO 2 and nitrogen fertilization) 5 . However, the responses of vegetation to climate warming are not uniform in space or time; covariation between the growing season temperature and vegetation activity varies regionally within the NH and appears to be weakening in many regions 6 . Simultaneously, mean daytime and nighttime warming during the growing-season produce different responses in vegetation productivity in the NH 3 . As spatial-temporal patterns of daytime and nighttime warming interact with extant seasonal hydrology, energy, and phenology cycles, stark differences in the vegetation responses are generated. However, such observed vegetation growth effects due to seasonal asymmetry in the interannual variations in the mean daytime and nighttime climate (VDNC) [7][8][9] and the underlying mechanisms that cause these effects to vary spatially remain poorly understood. Therefore, the effects of seasonal asymmetry in interannual variations in the mean diurnal climate on vegetation activity in the NH need to be better understood, especially because this asymmetry is predicted to increase 10 .
Asymmetry in the seasonal VDNC 11,12 , together with other factors (e.g., land use changes 13 ), could dramatically modify the seasonal thermo-hydrological patterns that govern land surface-climate feedbacks 14,15 . At middle and high latitudes in the NH, this asymmetry could hypothetically even modify the sign of biosphere-atmosphere feedbacks, which are known to be globally significant [15][16][17] . Therefore, in this study, we quantified the general patterns in the interannual responses of active vegetation growth (AVG, i.e., vegetation growth during the growing season) (for more detail, see Methods) at middle and high northern latitudes (>30°N) using 1) the satellite-derived Normalized Difference Vegetation Index (NDVI) 18,19 and the Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) 20 , 2) FLUXNET-based up-scaled gross primary productivity (GPP) 21 , and 3) tree growth increments to the interannual variability (IAV) of seasonal mean maximum temperature (T max ), mean minimum temperature (T min ) and water availability index (WAI) during past decades (for more detail, see Methods).

Results
Covariance between vegetation activity and the seasonal mean T max . Partial correlation analyses reveal a clear spatial pattern in the interannual responses of the mean growing-season (April-October) NDVI (NDVI GS ) to the IAV of the seasonal mean T max during 1982-2008 ( Fig. 1). NDVI GS generally exhibits a positive correlation with T max in spring (April-May) (T max sp ), summer (June-August) (T max su ) and autumn (September-October) ( ) T max au in most (~60-70%) of the boreal NH (Figs 1 and 2a, Supplementary Table S1, Figs S1-S3), with statistically significant (p < 0.10) partial correlation coefficients (R > 0.30) in ~22-25% of boreal NH despite the great spatial variations in the responses (Fig. 1, Supplementary Table S2 and Figs S1, S3). Such response patterns of the AVG to the IAV of the seasonal mean T max in the boreal NH are also supported by the analyses of mean growing-season FAPAR (FAPAR GS ) (Supplementary Figs S4a-c and S5a) and the results of independent ridge regression analyses (for more detail, see the supplementary text) of the relationships among , NDVI GS FAPAR GS and the seasonal mean T max during 1982-2008 (Supplementary Figs S6a-c and S7a-c). Coincidently, the ridge regression analyses reveal a generally positive interannual sensitivity of total growing-season GPP (GPP GS ) to seasonal mean T max (γ GPP , indicated by the coefficient of ridge regression) from spring to autumn in most (~51-89%) of the boreal NH (with statistical significance in ~20-62% of the boreal NH) despite large spatial variations (Fig. 3a, Supplementary Fig. S8a-c), with γ GPP ranging from 36.5 ± 62.4 to 51.3 ± 88.7 g C m −2 yr −1 °C −1 between seasons (Fig. 3a, Supplementary Fig. S8a-c).
By contrast, NDVI GS exhibits a significant (p < 0.10) negative correlation (R < − 0.30) with the IAV of mean T max sp and T max su in ~13-16% of the temperate NH (< 50 °N) during 1982-2008 ( Fig. 1, Supplementary Table S1 and Fig. S1). The most prominent negative correlations are observed in the arid climate zones of the northern U.S.A., southern Eurasia and parts of the Mediterranean regions ( Fig. 1  and S8a-c). Partial correlation analyses reveal a generally negative correlation of TRI with mean T max sp and T max su in the temperate NH during 1950-2008 (if available) (with statistical significance for ~15% of samples) compared with a generally positive correlation in the boreal NH (with statistical significance for ~13% of samples) (Fig. 1, S1). Similarly, ridge regression reveals that GPP GS in arid regions of the temperate NH (i.e., RegAR) exhibits a pervasively (~83-85% of RegAR) negative sensitivity to the IAV in T max sp and T max su during 1982-2008, with mean γ GPP values of -24.9 ± 29.7 and -24.33 ± 28.5 g C m −2 yr −1 °C −1 , respectively (Fig. 3a, Supplementary Table  S3).
Both partial correlation and ridge regression analyses of the relationships among NDVI GS , FAPAR GS , GPP GS , TRI and the seasonal mean T max consistently reveal a divergent response of the AVG to the IAV in the seasonal mean T max for portions of the middle and high latitudes of the NH in two aspects. First, the AVG in parts of the boreal NH, especially regions with dry summer (i.e., RegAS) and warm temperate regions (e.g., the western and southeastern U.S.A.), responds differently to mean T max sp and T max su (Fig. 1, Supplementary Table S2, Table S3, Figs S1, S3, S4a,b and S6-S7a,b): The pervasive (~31%) significant positive correlation (R > 0.30) between NDVI GS and T max sp in RegAS tends to weaken (~9.1%) in summer ( Fig. 1, Supplementary Table S2, Figs S1 and S3), whereas the NDVI GS maintains a significant positive correlation (R > 0.30) with the WAI (for ~15-17% of the samples) in both spring and summer ( Fig. 1, Supplementary Table S2, Figs S1 and S2c). Eddy covariance observations in these regions support this notion of seasonal water limitation, indicating that GPP GS responds positively to WAI in spring and summer during 1982-2008 (with spring and summer γ GPP values of 1.2 ± 1.7 and 0.9 ± 1.8 g C m −2 yr −1 mm −1 , respectively, in RegAS) (Fig. 3c, Supplementary Table S3, Figs S8g,h,S9g,h and S10). In addition to the regions with dry summers in the boreal NH, GPP GS in certain parts of the temperate NH also exhibits divergent Figure 2. PDFs of the partial correlation coefficients between the mean growing-season (April-October) NDVI and the seasonal VDNC. PDFs of the partial correlation coefficients (P R ) between the mean growing-season NDVI and spring (green lines), summer (blue lines) and autumn (brown lines) maximum temperature (a), minimum temperature (b) and water availability index (c) in temperate (< 50 °N, dotted lines) and boreal (≥ 50 °N, solid lines) regions of the NH. The percentages of the pixels that exhibit significant (p < 0.10) positive (green bars) and negative (red bars) correlations between the mean growing-season NDVI and spring (SP), summer (SU) and autumn (AU) maximum temperature, minimum temperature and WAI in temperate (light color) and boreal (dark color) regions are specified in the insets. The Spearman correlation coefficients R ± 0.51 and ± 0.30 in the x axis tick label correspond to the 1% and 10% significance levels of Student's t-test, respectively.
Scientific RepoRts | 6:19000 | DOI: 10.1038/srep19000 interannual sensitivity to T max sp and T max su (Fig. 3a, S9a,b and S10a), with γ GPP values of 8.9 ± 22.1 in spring and − 20.9 ± 43.6 g C m −2 yr −1 °C −1 in summer in RegTH (Fig. 3a, Supplementary Fig. S8a,b). At the same time, the TRI values in the temperate NH exhibit pervasive positive correlations with T WAI sp and T WAI su (with statistical significance at p < 0.05 level in ~14% and ~40% of samples, respectively) during 1950-2008 (Figs 1 and S1). Second, the interannual responses of the AVG to T max au diverge from the responses to T max sp and T max su in the majority of the temperate NH, especially in regions with drought limitations (e.g., RegAR and RegTA) (Supplementary Table S2). The NDVI GS exhibits a significant positive correlation (R > 0.30) with mean T max au in ~15-16% of the temperate NH but exhibits a significant positive correlation with T max sp and T max su in only ~< 8% of that area (Figs 1 and 2a, Table S2, Fig. S1 and S2a).

Covariance between vegetation activity and the seasonal mean T min . The interannual responses
of the AVG in the middle and high latitudes of the NH to the mean spring and summer minimum temperatures (T min sp and T min su , respectively) are relatively ambiguous (Fig. 1, Supplementary Fig. S1, S2b, S4d,e, S5b and S6-S9d,e), yet the trends in NDVI GS , GPP GS and TRI indicate that the AVG responds positively to T min sp in certain parts of the temperate NH, particularly in arid regions (primarily in the central U.S.A. and southern Eurasia). The correlation with NDVI GS is statistically significant (R > 0.30, p < 0.10) in ~17.5% of these regions (Fig. 1, Supplementary  Table S2, Fig. S1). Interestingly, such positive responses of the AVG to mean T min tend to weaken from spring to summer in arid and temperate arid regions (Figs 1 and 3d,e, Supplementary Fig. S1) and instead become predominant in the high-latitude NH. Significant positive responses of NDVI GS and GPP GS to T min su are observed in ~12-20% and ~23-64%, respectively, of cold regions and tundra (RegAH, RegAS and RegET) (Figs 1 and 3b, Supplementary Fig. S1, S3 and Table S2). Coincidently, the TRI exhibits a more pervasive positive correlation with T min sp (~15.8%) than with T min su (~7.9%) in the temperate NH (Fig. 1). Contrary to the ambiguous relationships between AVG and T min sp and T min su , both NDVI GS and GPP GS exhibit pervasively (60-63%) negative responses to T min au in most of the temperate and cold NH except in certain mountainous regions (e.g., the Rocky Mountains in western North America). Statistical significance is observed for NDVI GS in ~16% and ~18% of the temperate and boreal NH, respectively (Figs 1,2b and 3b, Supplementary Table  S1, Table S3, Fig. S1 and Fig. S3). However, TRI data indicate a persistently positive rather than negative correlation with T min au during 1950-2008 (Fig. 1, Supplementary Fig. S1). Additionally, the observed divergent response patterns between AVG and seasonal VDNC are not susceptible to the arbitrary definitions of active growing seasons ( Supplementary Fig. S11-S12) and selection of analysis periods (see Methods for details) ( Supplementary  Fig. S13).

Discussion
The results of both the partial correlation and ridge regression analyses demonstrate that the AVG in the boreal NH is generally temperature-limited: a warmer mean daytime temperature (i.e., T max ) during the active growing season can intuitively enhance the vegetation activity 1,2,4,22,23 . However, we have identified an obvious seasonally divergent response of the AVG to the IAV in seasonal mean T max in parts of the high-latitude NH, especially in regions with dry summers. The AVG responds differently to T max sp and T max su (Fig. 1, Supplementary Table S2, Figs S1, S3, S4a,b and S5-S7a,b), and a pervasive (~31%) significant positive correlation is observed between NDVI GS and T max sp shading in summer (Fig. 1, Supplementary Table S2 and Fig. S3). In fact, the AVG in cold regions with dry summers is vulnerable to water stress under daytime warming, thereby resulting in seasonal drought 16,24,25 , as confirmed by the close (~65.8-66%) positive coupling between the IAV in GPP GS and the WAI during spring and summer. In addition to regions with dry summers in the boreal NH, the AVG in certain parts of the temperate NH also responds differently to mean T max sp and T max su (Fig. 1 Supplementary Figs S3, S6-S9a,b and S10a). The observed divergent responses of the AVG to T max sp and T max su in these regions may be partially attributed to the daytime warming-induced progressive drought from spring to summer, as revealed by a water balance equation and the relationships between the seasonal soil moisture and the VDNC during 1982-2008 (Fig. 4, Supplementary Figs S14-S16). TRI data reveal that tree growth in the temperate NH experiences continuous water limitations during spring and summer, as revealed by pervasive positive correlations of the TRI with T WAI sp and T WAI su . In fact, warmer daytime temperatures in spring can lead to an earlier spring phenology and greater photosynthesis 26,27 , thereby enhancing vegetation growth as long as the soil water and atmospheric vapor pressure are sufficient. Conversely, if the increased warming rates during the early growing season are unmatched by water resources, the warmer temperatures can trigger a prolonged warming-induced drought, especially in regions that suffer seasonal drought limitations (e.g., RegAS and most regions in the temperate NH) (Fig. 4). Such cases of excessive soil water stress or, more commonly, vapor pressure deficits (Supplementary Figs S14-S16, ref. Liu et al., 2013 28 ) can dramatically reduce vegetation growth and even negate springtime ecosystem carbon gains 16,25,29 .
In addition to the divergent responses of the AVG to mean T max sp and T max su in parts of the mid-and high-latitude NH, the AVG responds differently to T max au than to T max sp and T max su across most of the temperate NH. Significant positive correlations between AVG and T max au are observed in nearly twice as many areas as those of T max sp and T max su . Early and mid-growing season (e.g., spring and early summer) warming tends to trigger regional droughts ( Supplementary Fig. S14-S16), which can exert strong control on vegetation growth and even cause tree mortality in the temperate NH and some parts of the cold NH 28,[30][31][32] . Nevertheless, the emergence of a positive response of NDVI GS to T max in autumn may reflect alleviation of water stress on photosynthetic activity in that period (Figs 1and 4, Supplementary Figs S1, S14) 33 , which is confirmed by a generally negative correlation between NDVI GS and the autumn WAI (Fig. 1). Consequently, alleviation of water limitations in autumn (Fig. 4,  Supplementary Fig. S14) results in a generally positive correlation between the TRI and T max au in ~49% of the temperate NH except in portions of arid regions (e.g., the central U.S.A.). By contrast, a predominantly (~60-70%) negative correlation exists between the TRI and T max sp and T max su (Fig. 1, Supplementary Figs S1, S13). These factors also lead to a conversion from a generally (~55-81%) negative correlation between γ GPP and T max sp and T max su to a positive correlation (in ~57-67% of the temperate NH) between γ GPP and T max au in the temperate NH ( Supplementary Fig. S8a-c).
The positive responses of the AVG to variations in the mean nighttime temperatures T min sp and T min su in parts of NH are most likely attributable to the reported compensation effects of nocturnal warming 34 . However, the shifting responses of the AVG to T min sp and T min su in arid and temperate arid regions (Fig. 1, Supplementary Fig. S1, S3 and Table S2) imply that the compensation effects of nocturnal warming on vegetation growth might be seasonally and spatially dependent and are likely modified by other bio-physical factors, such as water conditions (Fig. 4) and plant functional types 35 . In contrast to the ambiguous relationships between the AVG and T min sp and T min su , there is a pervasive negative response of the AVG, from the perspective of NDVI GS and GPP GS , to T min au across most of the temperate and cold NH except in certain mountainous regions (e.g., the Rocky Mountains in western North America). The dominant effect of autumn nocturnal warming on NH vegetation growth may be enhanced plant respiration, which consequently leads to autumn carbon loss 36 . However, the TRI data indicate a persistent positive rather than negative correlation with T min au (Fig. 1). In addition to enhanced whole-plant respiration, warmer autumn (both day and night) temperatures may reactivate cambial activity 37 . Combined with alleviation of water limitations (Fig. 4, Supplementary Figs S14-S17), these process may partially explain the observed widespread positive correlation between the TRI and T min au in the temperate NH. Overall, the observed divergent responses of AVG to seasonal T min in portions of the NH are to some degree attributable to the water-mediated counterbalancing effects of T min on vegetation growth and respiration 34,38,39 .
In summary, remote sensing retrievals, empirical up-scaling measurements from FLUXNET, and tree growth data consistently indicate a possible water-mediated divergent response of vegetation activity to seasonal VDNC, especially in temperate regions and boreal regions with dry summers in the NH. Our research provides new insights into the seasonally divergent responses of the AVG to the seasonal asymmetry in the VDNC and reinforces the idea that the projected intensification in the asymmetry in the seasonal VDNC under continued global warming and increasing climate extremes 40 may have strong, spatially variable effects on vegetation growth and carbon dynamics in the NH, with associated changes in regional ecosystem services, the terrestrial carbon cycle, and global climate feedbacks.

Methods
Remote sensing data and up-scaled GPP. We used three types of data to investigate how AVG (i.e., vegetation growth during the growing season) responded to the mean seasonal VDNC: the satellite-derived AVHRR/ GIMMS Normalized Difference Vegetation Index (NDVI, specifically version NDVI3g) 18,19 , the satellite-derived Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) 20 , and FLUXNET-based up-scaled gross primary production (GPP) data 21 . The data featured a spatial resolution of 0.5° × 0.5° and a monthly temporal resolution for 1982-2008. Pixels with a multi-year mean annual NDVI value < 0.10 were masked in our study.
Tree ring data. In total, 149 standard TRI chronologies from the mid-and high-latitude NH were used to investigate the interannual responses of tree growth to the seasonal VDNC. Standard TRI chronologies were built following the standard procedures of dendrochronology 41 , including fitting a curve to the raw ring-width series and then dividing each ring-width value by the corresponding curve value to generate a series of growth indices. Of these chronologies, 136 were obtained from the International Tree-Ring Data Bank (ITRDB, http://www. ncdc.noaa.gov/data-access/paleoclimatology-data/datasets/tree-ring) based on four criteria: 1) intact records of latitude, longitude, elevation, species and sample depth; 2) the chronologies cover at least 1950-1998 or extend to more recent times (i.e., early 21st century); 3) the sample depth for each site-year is greater than 10; and 4) the altitude of the sample site is less than the 90 th percentile of the elevation distribution of the surrounding 0.5° × 0.5° region based on the GTOPO30 dataset (with ~1 km resolution, available from U.S. Geological Survey, https://lta.cr.usgs.gov/GTOPO30) to match the regional vegetation growth patterns and avoid site-specific representations. Our research group provided another 13 standard chronologies from northern China. Our chronologies satisfy the four criteria listed above (see Table S4 for details).
Climate data. Monthly mean maximum (T max , averaging daily maximum temperature) and mean minimum temperature (T min , averaging daily minimum temperature) at a spatial resolution of 0.5° were obtained from the newly updated CRU TS 3.22 dataset (http://www.cru.uea.ac.uk/). The seasonal (in detail see section of sensitivity test) mean T max and T min were then calculated by averaging the monthly mean climate data. The seasonal mean T max and T min are used in this study to represent the general climate conditions of seasonal mean daytime and nighttime temperature 3 , respectively. Remotely sensed soil moisture data with a daily resolution for the period 1978-2013 were obtained from the Essential Climate Variable Soil Moisture dataset (ECV SM 02.0) (http://www. esa-soilmoisture-cci.org/). These data were produced following the method described by Liu et al. (2011) and Liu et al. (2012) 42,43 , representing surface soil moisture with a global coverage and spatial resolution of 0.25°. We then resampled this surface soil moisture data to a spatial resolution of 0.5° × 0.5°. Unfortunately, the soil moisture data were not continuous during 1982-2008, and only pixels with at least 19 years of data between 1982 and 2008 were retained.
Water deficit and WAI. The monthly water deficit was estimated using a simple water balance equation: and PET i are the water deficit (positive values indicate water surpluses), precipitation and potential evapotranspiration in month i. The potential evapotranspiration is calculated using Thornthwaite's equation 44 . Then, the total seasonal water deficit is calculated by summing the monthly WD values. Negative values express the degree of dryness; the more negative the value, the more intense the water limitation. In our study, the monthly WAI was calculated based on the water balance equation proposed by Kleidon and Heimann (1998) 45 and was used to represent the soil moisture conditions. Statistical analysis. Spearman partial correlation and ridge regression were performed to investigate the responses of vegetation activity in the mid-and high-latitude NH to the mean seasonal VDNC. The interannual sensitivity of vegetation growth to IAV in the seasonal VDNC was evaluated using the ridge regression coefficients. Ridge regression has been demonstrated to be an effective method for addressing collinearity in multivariate regression 46 . The probability density function (PDF) of the Spearman partial correlation coefficients between the seasonal VDNC and the IAV in the AVG was estimated for seven major climate zones (as defined by the Köppen-Geiger climate classification 47 ), including arid (RegAR, grouping of BWk, BWh, BSk, and BSh), warm temperate humid (RegTH, grouping of Cfa, Cfb and Cfc), warm temperate arid (RegTA, grouping of Csa, Csb, Csc, Cwa, Cwb, and Cwc), cold humid (RegAH, grouping of Dfa, Dfb, Dfc, and Dfd), cold summer dry (RegAS, grouping of Dsa, Dsb, Dsc, and Dsd), cold winter dry (RegAW, grouping of Dwa, Dwb, Dwc, Dwd) and polar tundra (RegET). Prior to performing the Spearman partial correlation and ridge regression analyses, all variables (i.e., NDVI, FAPAR, GPP, TRI and the climate variables) were detrended using a linear function. The statistical analyses and figure creation were performed using the MATLAB software package (R2012b).

Sensitivity test.
In this study, we attempted to define the seasons (spring, summer, and autumn) in a manner that is consistent across the mid-and high-latitude NH. Notably, three different definitions of seasons were used in our study to verify the robustness of our conclusions relative to the arbitrary definitions of vegetation growth seasons in the mid-and high-latitude NH. The three definitions are as follows: