Monitoring the Responses of Deciduous Forest Phenology to 2000–2018 Climatic Anomalies in the Northern Hemisphere

: Monitoring the phenological responses of deciduous forests to climate is important, due to the increasing frequency and intensity of extreme climatic events associated with climate change and global warming, which will in turn affect vegetation seasonality. We investigated the spatiotemporal patterns of the response of deciduous forests to climatic anomalies in the Northern Hemisphere, using satellite-derived phenological metrics from the Copernicus Global Land Service Leaf Area Index, and multisource climatic datasets for 2000–2018 at resolutions of 0.1 ◦ . Thereafter, we assessed the impact of extreme heatwaves and droughts on this deciduous forest phenology. We assumed that changes in the deciduous forest phenology in the Northern Hemisphere for the period 2000–2018 were monotonic, and that temperature and precipitation were the main climatic drivers. Analyses of partial correlations of phenological metrics with the timing of the start of the season (SoS), end of the season (EoS), and climatic variables indicated that changes in preseason temperature played a stronger role than precipitation in affecting the interannual variability of SoS anomalies: the higher the temperature, the earlier the SoS in most deciduous forests in the Northern Hemisphere (mean correlation coefﬁcient of − 0.31). Correlations between the SoS and temperature were signiﬁcantly negative in 57% of the forests, and signiﬁcantly positive in 15% of the forests ( p < 0.05). Both temperature and precipitation contributed to the advance and delay of the EoS. A later EoS was signiﬁcantly correlated with a positive Standardized Precipitation Evapotranspiration Index (SPEI) at the regional scale (~30% of deciduous forests). The timings of the EoS and SoS shifted by >20 d in response to heatwaves throughout most of Europe in 2003, and in the United States of America in 2012. This study contributes to improve our understanding of the phenological responses of deciduous forests in the Northern Hemisphere to climate change and extreme climate events.


Introduction
Interest in understanding the interactions between phenology and climate has increased over the past few decades [1], because vegetation phenology plays an important role in balancing biogeochemical cycles, such as the exchange of water, energy, and carbon [2][3][4][5]. Changes in the timing of phenology provide the first signals of adjustments in the responses of species to climatic anomalies [6].
Changes in the pattern of distribution of temperatures and precipitation as a consequence of global climate change, and the interactions with other cues-such as photoperiod-could strongly alter vegetation phenology [7][8][9][10][11]. Investigating the interactive effects of temperature and precipitation on phenology, in order to understand and anticipate the effects of climate change on vegetation, is therefore crucial [12]. Many previous studies have investigated the changes in vegetation phenology as a result of climate change and the associated global warming [13][14][15][16]. In this sense, the analysis of the sensitivity of vegetation to hydroclimatic anomalies is increasingly studied [1,17]-especially the effects of temperature on

Land-cover/vegetation Map
Because we focused on deciduous forests, we used a land-cover (LC) map that identified this type of vegetation: the Climate Change Initiative (CCI-LC) map series for 2015, at a spatial resolution of 300 m, available from the European Space Agency (ESA) [72]. CCI-LC discriminates 38 classes of land cover. We resampled the map to 0.1°, and analyzed only pixels containing deciduous forest.

Vegetation Phenology from SPOT VEGETATION and PROBA-V Data
The SoS and EoS during the 2000-2018 period were estimated using the time series of Copernicus Global Land Service LAI 1 km version 2, derived from SPOT VEGETA-TION (VGT) and PROBA-V data [73] (https://land.copernicus.eu/global/themes/vegetation (accessed on 13 June 2021)). These metrics were based on previous protocols and research [57,58,64] that used dynamic thresholds. This method is based on the percentage of the LAI amplitude in each pixel, in which the SoS is defined as the day of the year (DoY) when the LAI exceeds the 30% threshold, and the EoS is defined as the DoY when the LAI overpasses the 40% threshold after the growing season.

Rainfall and Temperature Datasets
The data for temperature and precipitation were collected from the ERA5 hourly gridded datasets from 2000 to 2018, with a spatial resolution of 0.25° [74]. To achieve higher resolution, we interpolated all climatological data from the spatial resolution of 0.25 to 0.1°, using a cubic convolution that calculated the value of each pixel by fitting a smooth curve based on the surrounding 16 pixels. Temperature was the air temperature at a height of 2 m, and precipitation was the accumulated amount of liquid or frozen water that fell, calculated as the sum of large-scale and convective precipitation, in millimeters.
We used the mean temperature and accumulated precipitation for the preseason, using time lags of 1, 3, and 6 months prior to the timings of the SoS and EoS.

Land-Cover/vegetation Map
Because we focused on deciduous forests, we used a land-cover (LC) map that identified this type of vegetation: the Climate Change Initiative (CCI-LC) map series for 2015, at a spatial resolution of 300 m, available from the European Space Agency (ESA) [72]. CCI-LC discriminates 38 classes of land cover. We resampled the map to 0.1 • , and analyzed only pixels containing deciduous forest.

Vegetation Phenology from SPOT VEGETATION and PROBA-V Data
The SoS and EoS during the 2000-2018 period were estimated using the time series of Copernicus Global Land Service LAI 1 km version 2, derived from SPOT VEGETATION (VGT) and PROBA-V data [73] (https://land.copernicus.eu/global/themes/vegetation (accessed on 13 June 2021)). These metrics were based on previous protocols and research [57,58,64] that used dynamic thresholds. This method is based on the percentage of the LAI amplitude in each pixel, in which the SoS is defined as the day of the year (DoY) when the LAI exceeds the 30% threshold, and the EoS is defined as the DoY when the LAI overpasses the 40% threshold after the growing season.

Rainfall and Temperature Datasets
The data for temperature and precipitation were collected from the ERA5 hourly gridded datasets from 2000 to 2018, with a spatial resolution of 0.25 • [74]. To achieve higher resolution, we interpolated all climatological data from the spatial resolution of 0.25 to 0.1 • , using a cubic convolution that calculated the value of each pixel by fitting a smooth curve based on the surrounding 16 pixels. Temperature was the air temperature at a height of 2 m, and precipitation was the accumulated amount of liquid or frozen water that fell, calculated as the sum of large-scale and convective precipitation, in millimeters.
We used the mean temperature and accumulated precipitation for the preseason, using time lags of 1, 3, and 6 months prior to the timings of the SoS and EoS.

SPEI
Drought indices have become useful tools for analyzing, assessing, and estimating the dry and humid periods that may have an impact on phenology [36]. Various drought indices are available, such as the Standardized Precipitation Index (SPI) or the Palmer Drought Severity Index [75] (PDSI), and several recent studies have analyzed drought conditions using the standardized precipitation evapotranspiration index (SPEI) [76,77]. The SPEI is based on precipitation and potential evapotranspiration, which includes the Remote Sens. 2021, 13, 2806 4 of 21 role of temperature in drought severity [78][79][80]. The SPEI considers drought timescales, which represent the cumulative water balance over the previous 1-48 months. It is based on the Standardized Precipitation Index (SPI) calculation method, but with improvements to include the potential evapotranspiration (PET). The SPEI uses the weekly (or monthly) difference between precipitation and PET (based on Thornthwaite's method), and it is calculated as a standardized variable, which allows the comparison with other SPEI values and climatic variables over time and space [79]. Unlike the SPI, the SPEI includes the effects of temperature variability on drought estimation and, therefore, it takes into account the effects of warming processes on drought severity [79,80].
SPEI dataset version 2.6 was downloaded from the Global SPEI database [81] at a spatial resolution of 0.5 • , and was resampled to 0.1 • for the analysis. This dataset is based on the United Nations Food and Agriculture Organization (FAO)-56 Penman-Monteith estimation of potential evapotranspiration. The complete procedure for calculating the SPEI is provided by Vicente-Serrano et al. [79].
We determined drought severity for the timings of the SoS and EoS using different SPEI timescales (1, 3, 6, and 12 months), representing the cumulative water balance for the timings of preseason and presenescence. Positive SPEI values indicate that humidity is higher than the historical median, while negative values indicate a water deficit [79,80]. The SPEI data were classified into seven categories (Table 1) based on the World Atlas of Desertification [82] for analysis of conditions of wetness or dryness.

Methodology and Statistical Analysis
We first estimated the significant trends in the time series of the estimates (SoS and EoS). Secondly, we explored the impacts of anomalies in hydroclimatic variables-such as temperature, precipitation, and drought-on deciduous phenology throughout the Northern Hemisphere for 2000-2018. We applied spatiotemporal response analysis to determine the relationships between phenology and climate variables. Moreover, we investigated the spatial pattern of the sensitivity of phenology to climate, and its relationship with preseason and presenescence temperature, precipitation, and drought. We thereafter focused on some of the extreme events-including high temperatures, low temperatures, and severe drought-that have occurred in the past two decades.
We used different software to calculate and evaluate the effects of climatic variables on vegetation phenology. We first used Google Earth Engine [83] to download and process the time series of the hydroclimatic variables, and for estimating land surface phenological metrics in the Northern Hemisphere [57,58]. We then used RStudio, XLSAT, and Idrisi TerrSet for processing and statistically analyzing all data. Finally, we used ESRI ArcGIS 10.5 for generating the graphs and maps.

Trend Analysis
We calculated the trends in the estimated time series of the SoS and EoS before analyzing the relationships between phenology and climate. Temporal trends in the datasets were calculated by applying the Theil-Sen (TS) median slope trend analysis, which is an effective method for analyzing the rate of change in observations over a period of time (2000-2018 in our study), and is similar to linear least squares regression. TS analysis is based on nonparametric statistics (Mann-Kendall), and is independent of the assumptions of linear regression. Medians are used to calculate the trend, which is consequently less susceptible to noise and outliers [29,84,85]. The equation used to estimate the TS slope is: where x j and x i are values in years i and j, respectively. We estimated the significance of the TS slope using a nonparametric test (Mann-Kendall significance test), which provided a standardized Z and the corresponding probability (p). Positive and negative slopes indicated that the SoS and EoS had delayed and advanced trends, respectively, during the study period. The Mann-Kendall significance (Z and P) was calculated as: and: where: We also used the nonparametric Pettitt test method [86] to detect the possible abrupt change points in the phenological time series. This test allows us to identify shifts in the average, and their significance. The null hypothesis of the Pettitt test is the absence of a change point. The empirical significance level (p-value) was computed using XLSTAT statistical and data analysis software 2021 v.3.1, at a significance level of 5%. The nonparametric Pettitt statistical test is defined as: where: where t is the period length and n is the number of data in the statistical series. The p-value and an interval around the p-value were evaluated using a Monte Carlo method.

Standardized Anomalies
We calculated standardized anomalies in the interannual time series (2000-2018) of phenology, temperature, and precipitation. Standardized anomalies were calculated by dividing anomalies by the standard deviation. The equation is [87]: where Z is a standardized anomaly, X is the annual value, and µ and σ are the interannual mean and standard deviation, respectively, for each variable analyzed (i.e., the timings of the SoS and EoS, the mean preseason or presenescence temperature, or the accumulated preseason or presenescence precipitation).
Note that we estimated the anomalies in the time series of temperature and precipitation using the mean temperature and the accumulated precipitation for different timescales (1, 3, and 6 months) before estimating the timings of the SoS and EoS (see Section 2.1.3).

Correlation and Partial Correlation Analyses
We calculated the coefficients for the correlations between the estimated time series of phenology (SoS and EoS), temperature, precipitation, and SPEI for different preseason lengths. We used two types of correlation analysis to quantify the response of vegetation to climatic drivers. We first calculated the linear correlation (Pearson's correlation) between the interannual anomalies of phenology (y) and hydroclimatic variables (x) at a significance level of 95%: where r xy is the Pearson's correlation coefficient between x and y, n is the number of observations, xi is the value of x for observation i, and yi is the value of y for observation i.
A multivariate linear regression model was used for multivariate cases. We correlated SoS and EoS anomalies with the climatic variables (temperature and precipitation) for timescales of 1, 3, and 6 months prior to the dates of the SoS and EoS, and summarized the climatic variables for the preseason periods in which the correlation was highest. For the SPEI, we used timescales from 1 to 12 months [80].
We then applied a partial correlation analysis of the SoS and EoS using the preseason and presenescence climatic variables for the three timescales (1, 3, and 6 months). The precipitation data for the same timescale were used as a constant factor for calculating the partial correlation between phenology and temperature. Similarly, the influence of temperature was considered to be a constant for calculating the correlation with accumulated precipitation. The partial correlation coefficient between vegetation and mean temperature, and its significance, were tested as: where v is the vegetation phenology, t is temperature, p is precipitation, and r vt and r vp are the simple correlation coefficients of the phenology (SoS or EoS) with the mean preseason or presenescence temperature and accumulated precipitation, respectively. Equation (8) is also valid for calculating the partial correlation between phenology and precipitation by changing the order in which the data are entered. The significance of the partial correlation coefficients at the 95% level was evaluated using Student's t-test. The Pearson's correlation and partial correlation analysis produced similar maps, so the results obtained via partial correlation will be taken into account in subsequent sections for analyzing the correlations between phenology, temperature, and precipitation.

Sensitivity Analysis
We used multiple linear regression and sensitivity analysis to further investigate the interactions of phenology with temperature, precipitation, and SPEI. The responses of the sensitivity of vegetation phenology to the climatic variables corresponded to the slopes of the linear regressions between the phenological metrics and the climatic variables, representing the unit change in phenological date divided by the unit change in temperature, precipitation, or SPEI.

Trends in the Time Series of Estimated Phenology
The Pettitt test confirmed that phenological time series were monotonic, and any statistically significant change points were detected (p-values of 0.74 and 0.86 for SoS and EoS, respectively). The temporal trends in SoS for 2000-2018 ( Figure 2a) were significant (p < 0.05) for 20.5% of the deciduous forests in the Northern Hemisphere. Negative trends, representing an advance in the timing of the SoS, and positive trends, representing a delay in the timing of SoS, accounted for 61.5 and 38.54% of these pixels, respectively (Table S1). Northeastern Europe mainly had negative SoS trends, while Russia and North America mainly had positive trends (Figure 1a). The EoS changed significantly in 23.8% of the deciduous forests, with 40.5 and 59.49% of these forests having a delayed and advanced EoS, respectively (Table S1). Our results indicated that the SoS and EoS advanced by 0.08 and 0.1 d/y, respectively. Positive significant trends were mainly in northeastern Europe, and negative trends were mainly in eastern North America (Figure 5a).

Correlation and Sensitivity of Phenology with Climatic Variables
The correlation between SoS anomalies and preseason temperature (Figure S1a and Figure 3) indicated that 72.13% of all pixels had significant correlations at p < 0.05, with negative correlations accounting for 57% of all pixels ( Table 2). An earlier SoS tended to be associated with higher temperatures in 35% of the pixels, and a later SoS was associated with lower temperatures in 21% of the pixels (Table S2). Temperature and SoS were negatively correlated in Eurasia and eastern North America, with correlation coefficients (r) of~−0.7 and −0.6, respectively ( Figure S1).
The spatial pattern of the distribution of the partial correlations between SoS and precipitation was more heterogeneous than the correlation between SoS and temperature. The correlation between SoS and precipitation was positive in most pixels, accounting for 42% of them (Table 2; Figure 3b and Figure S1b), which led to significant positive correlation advance or delay (20.9% and 21.4%) in the SoS with precipitation decrease or increase, respectively (Table S2).
The mean sensitivity of the SoS to temperature and precipitation was −2.45 d/ • C and 0.8 d/10 mm, respectively ( Figure 4). The sensitivity to temperature was highest (~−5 d/ • C) in Eurasia and southeastern North America (Figure 2b), while the sensitivity to precipitation was highest (3 d/10 mm) at latitudes > 60 • N ( Figure 2c).
The mean sensitivity of the SoS to temperature and precipitation was −2.45 d/°C and 0.8 d/10 mm, respectively ( Figure 4). The sensitivity to temperature was highest (~−5 d/°C) in Eurasia and southeastern North America (Figure 2b), while the sensitivity to precipitation was highest (3 d/10 mm) at latitudes > 60° N (Figure 2c).    The sensitivity analysis (Figures 4 and 5) indicated that the response of phenology to climatic anomalies was lower for the timing of the EoS than for the timing of the SoS. The pattern of temperature sensitivity was very heterogeneous, with symmetric distributions of positive and negative correlations between temperature and EoS in 23.57 and 20.75% of the pixels, respectively (Table 2, Figure 3c). EoS advanced with temperature by an average of ~0.5 d/°C (Figure 5b). The sensitivity of EoS to precipitation had the opposite pattern: the timing of EoS was delayed by an average of ~0.5 d/10 mm. Correlations were significantly positive (i.e. a delay in the timing of EoS with an increase in precipitation) in 34.44% of the study area ( Figure S2), with sensitivities highest in southern and south-    The sensitivity analysis (Figures 4 and 5) indicated that the response of phenology to climatic anomalies was lower for the timing of the EoS than for the timing of the SoS. The pattern of temperature sensitivity was very heterogeneous, with symmetric distributions of positive and negative correlations between temperature and EoS in 23.57 and 20.75% of the pixels, respectively (Table 2, Figure 3c). EoS advanced with temperature by an average of ~0.5 d/°C (Figure 5b). The sensitivity of EoS to precipitation had the opposite pattern: the timing of EoS was delayed by an average of ~0.5 d/10 mm. Correlations were significantly positive (i.e. a delay in the timing of EoS with an increase in precipitation) in 34.44% of the study area ( Figure S2), with sensitivities highest in southern and south- The sensitivity analysis (Figures 4 and 5) indicated that the response of phenology to climatic anomalies was lower for the timing of the EoS than for the timing of the SoS. The pattern of temperature sensitivity was very heterogeneous, with symmetric distributions of positive and negative correlations between temperature and EoS in 23.57 and 20.75% of the pixels, respectively (Table 2, Figure 3c). EoS advanced with temperature by an average of 0.5 d/ • C (Figure 5b). The sensitivity of EoS to precipitation had the opposite pattern: the timing of EoS was delayed by an average of~0.5 d/10 mm. Correlations were significantly positive (i.e. a delay in the timing of EoS with an increase in precipitation) in 34.44% of the study area ( Figure S2), with sensitivities highest in southern and southwestern Europe (~3 d/10 mm). In contrast, 28.42% of the pixels had negative correlations (i.e., an advance in the timing of the EoS with an increase in precipitation)-mainly in northeastern Europe and areas of Russia (Figure 5c). western Europe (~3 d/10 mm). In contrast, 28.42% of the pixels had negative correlations (i.e., an advance in the timing of the EoS with an increase in precipitation)-mainly in northeastern Europe and areas of Russia (Figure 5c).

Response of Vegetation Phenology to Drought Using the SPEI
When analyzing the influence of drought on vegetation, we found that SPEI calculated using timescales between 1 and 3 months was the best correlated with the timings of the SoS and EoS in >50% of the pixels (Table S3).
The Spearman correlations between the SoS and the SPEI were positive in 52.3% of the Northern Hemisphere, and were significant (p < 0.05) in 17.5% of the pixels (Table 2)-mostly >60 • N in northeastern Europe and North America (Figure 6a). The correlations were negative (47.6% of the pixels, significant in 9.8%) in northeastern Europe (between 50 • and 60 • ) in areas where the sensitivity of the SoS to the SPEI was highest ( Figure S3a).

Response of vegetation phenology to drought using the SPEI
When analyzing the influence of drought on vegetation, we found that SPEI calculated using timescales between 1 and 3 months was the best correlated with the timings of the SoS and EoS in > 50% of the pixels (Table S3).
The Spearman correlations between the SoS and the SPEI were positive in 52.3% of the Northern Hemisphere, and were significant (p < 0.05) in 17.5% of the pixels (Table  2)-mostly > 60° N in northeastern Europe and North America (Figure 6a). The correlations were negative (47.6% of the pixels, significant in 9.8%) in northeastern Europe (between 50° and 60°) in areas where the sensitivity of the SoS to the SPEI was highest ( Figure S3a).
Correlations between the timings of the EoS and the SPEI were positive in 58.8% of the study area (Table 2), with significant correlations for 28.8% of the pixels (p < 0.05). The correlations were particularly strong in southwestern Europe and northeastern North America (Figure 6b), in areas where the sensitivity of the EoS to the SPEI was highest ( Figure S3b). Drought weakly affected phenology at high northern latitudes (>60° N), where temperature (Figure 5b), not precipitation (Figure 5c), was the main variable limiting phenology. The correlations between the EoS and the SPEI were thus weak.

Phenological Responses to Recent Climatic Extremes
We analyzed the effects of three heat-and cold waves on vegetation phenology in Europe and North America. Correlations between the timings of the EoS and the SPEI were positive in 58.8% of the study area (Table 2), with significant correlations for 28.8% of the pixels (p < 0.05). The correlations were particularly strong in southwestern Europe and northeastern North America (Figure 6b), in areas where the sensitivity of the EoS to the SPEI was highest ( Figure S3b). Drought weakly affected phenology at high northern latitudes (>60 • N), where temperature (Figure 5b), not precipitation (Figure 5c), was the main variable limiting phenology. The correlations between the EoS and the SPEI were thus weak.

Phenological Responses to Recent Climatic Extremes
We analyzed the effects of three heat-and cold waves on vegetation phenology in Europe and North America.

Effect of the 2003 Summer Heatwave in Western Europe
The year 2003 was one of the driest and warmest years recorded in the past 30 years in most of Central Europe [88,89]. The effects of this extreme episode were represented by negative anomalies in the timing of the EoS throughout most of Western Europe (Figure 7a). We highlight a region of southern France and southwestern Germany where an unstandardized anomaly for the timing of the EoS had a mean of −22 d.

Effect of the 2003 Summer Heatwave in Western Europe
The year 2003 was one of the driest and warmest years recorded in the past 30 years in most of Central Europe [88,89]. The effects of this extreme episode were represented by negative anomalies in the timing of the EoS throughout most of Western Europe ( Figure  7a). We highlight a region of southern France and southwestern Germany where an unstandardized anomaly for the timing of the EoS had a mean of -22 d. An intense drought also occurred in 2003, which we identified using the SPEI (Figure 7b, Table 3). More than 80% of the pixels indicated an intense drought prior to the timing of the EoS. Fischer et al. [88] observed that an early EoS and stress from the lack of soil moisture contributed greatly to the suppression of evapotranspiration after the summer, and that this interaction may have amplified the temperature anomalies by locally increasing the flux of sensible heat. Figure 7c shows the standardized anomalies between phenology and the climatic variables, which allows for the visualization of the positive temperature anomalies for the timing of the EoS and the precipitation deficit for the timings of both the SoS and EoS. An intense drought also occurred in 2003, which we identified using the SPEI (Figure 7b, Table 3). More than 80% of the pixels indicated an intense drought prior to the timing of the EoS. Fischer et al. [88] observed that an early EoS and stress from the lack of soil moisture contributed greatly to the suppression of evapotranspiration after the summer, and that this interaction may have amplified the temperature anomalies by locally increasing the flux of sensible heat. Figure 7c shows the standardized anomalies between phenology and the climatic variables, which allows for the visualization of the positive temperature anomalies for the timing of the EoS and the precipitation deficit for the timings of both the SoS and EoS. We analyzed the early spring for 2012 in North America, focusing on the southeastern United States (US). We calculated the interannual anomalies for the phenological metrics and climatic variables for each pixel, in order to assess the spatial patterns of the phenological responses to the climatic extremes. Previous studies [55,56] considered the spring SoS anomaly in 2012 to be the earliest spring recorded since 1900 across North America (Figure 8a). Karl et al. [56] reported that the anomaly was driven by a strong and stable high-pressure anticyclone that remained over much of the northeast from late February to April, causing record high temperatures (Figure 8b,c) and phenological advancement. Unlike the two previous analyses, this third case refers to a delayed SoS due to a negative temperature anomaly for the timing of the preseason. Figure 9a,b shows the SoS anomaly and preseason mean temperature for 2005, respectively. The positive anomalies of the SoS affected much of Central and Eastern Europe-especially the northern Balkan  (Figure 8c). The anomalies for the timing of the SoS were negative in most of the pixels (96.2%) (Figure 8a). Figure 8c shows the dominant role of temperature in the phenological advancement of the SoS in 2012, with mean standardized anomalies > 2 • C. Precipitation was slightly negatively anomalous before the timing of the SoS in 2012, but with no water stress, and the majority of the pixels (73%) indicated a normal SPEI between −1 and 1 (Table 3).

Effect of the Late 2005 Cold Wave in the Balkans
Unlike the two previous analyses, this third case refers to a delayed SoS due to a negative temperature anomaly for the timing of the preseason. Figure 9a (Table 3).

Discussion
Climatic projections indicate a likely increase in temperatures in much of the world, especially at the higher latitudes of the Northern Hemisphere [9,10]. Recent studies [12,90] have also found that extreme climatic events have increased in frequency, inten-

Discussion
Climatic projections indicate a likely increase in temperatures in much of the world, especially at the higher latitudes of the Northern Hemisphere [9,10]. Recent studies [12,90] have also found that extreme climatic events have increased in frequency, intensity, and duration-consistent with IPCC projections-which will affect many ecosystems, and especially vegetation. Climatic variability adds uncertainty in analyzing and predicting the impacts of climate change on vegetation phenology [3,38,91].
Previous studies have attributed the recent shifts in phenology to climate change, and to the effects of variations in temperature and precipitation [15][16][17]92]. Assessing the pattern of distribution, the variability of phenology, and the correlations between phenology and climatic variables is crucial for understanding the potential effects of future climate change [93]. Remotely sensed data have been widely used to assess trends in phenological time series and their responses to climatic variability. Numerous studies have reported that the phenological trends from 2000 to 2018 were lower than the rates of change from 1980 to 1999 [14,[94][95][96]. De Beurs et al. [15] and Piao et al. [28] reported that the SoS advanced by 6.6 and 7.9 d/decade in North America and China, respectively, between 1982 and 1999. Jeong et al. [14] and Zeng et al. [95], however, found that phenological trends declined significantly after 2000, with an advancement of the timing of the SoS of 0.1-0.2 d/decade, respectively-consistent with our results. Some studies have even reported delays of 1 d/decade in the timing of the SoS after 2000, specifically in western North America [94,96].
In our study, we used land surface phenological metrics at the continental scale derived from LAI time series (2000-2018) of the SPOT-VGT and PROBA-V sensors [57,58]. Since the time series start in 2000, the detected trends are limited to reduced areas. For this reason, we mainly focused on analyzing the correlations between phenology and climatic variables, and analyzing the responses of phenological anomalies to recent climatic events.
Most previous studies have only emphasized the role of preseason temperature in determining the SoS, because an increase in the advance of the SoS is a consequence of global warming [8,28,29,[97][98][99]. Other studies focusing on the EoS have recorded a delay in senescence, which led to longer growing seasons in some areas of Eurasia and North America [14,100]. We focused on analyzing the responses of phenology to mean preseason and presenescence temperatures, accumulated precipitation, and drought across the deciduous forests in the Northern Hemisphere. Our results indicated that anomalies of temperature and precipitation controlled the changes in phenological metrics in the deciduous forests, consistent with previous studies [2,101]; the anomalies of SoS were particularly strongly associated with the changes in mean preseason temperature, so this climatic variable may have been the main cause of the advance or delay in the start of the growing season, affecting 72.1% of the study area (mean r of −0.31) (p < 0.05). Correlations were significantly negative in 57% of the pixels (p < 0.05), i.e., an advance or delay in the SoS due to higher or lower temperatures, respectively.
The effects of the relationships between the climatic variables and vegetation on the timing of the EoS were more complex. The timing of the end of the growing season in response to temperature was generally delayed in most pixels. Water stress associated with droughts during summer, however, advanced the EoS in some regions-such as Southern Europe ( Figure S2 and Figure 6)-with negative anomalies longer than −20 d for the timing of the EoS throughout most of Europe (in 2003), consistent with other studies [102][103][104][105].
The influence of drought on the timing of the EoS was lower in humid and cold regions (such as those pixels located between 55 and 80 • N) than in drier regions, such as areas with a Mediterranean climate ( Figure S3). The availability of water in Mediterranean areas was the primary limiting resource for the timing of the EoS, but temperature and other variables-such as photoperiod-may play a larger role at higher latitudes.
The distribution of the regression coefficients between the anomalies in phenology and the climatic variables was highly spatially heterogeneous, due to the spatial and latitudinal heterogeneity in climate, as well as the biological characteristics of the species, which could account for the variable responsiveness to climate [25,106]. The responses of phenology to climate-especially to climatic extremes-also vary with climatic gradients, type of event, and biome, and even among individuals of the same species [38,91].
These phenological changes may affect climate change via the feedback between vegetation and climate [1,3,[107][108][109]. For example, climatic anomalies or extremely high temperatures (such as during heatwaves) alter vegetation growth due to both the high temperatures and lower amounts of soil moisture [47,51,110]. Altered vegetation growth greatly affects the uptake of CO 2 , depending on the availability of soil water, regional characteristics, and plant species [3]. Deficits in soil moisture lead to lower water evaporation, which lowers the release of latent heat from the land, prevents the development of clouds, and may consequently intensify droughts because precipitation is reduced [111][112][113], which could also involve teleconnections between areas [109].

Conclusions
This study comprehensively analyzed the response of vegetation to climatic anomalies in the Northern Hemisphere, and assessed the impact of extreme climatic events on deciduous phenology. Our results suggest that deciduous phenology in the Northern Hemisphere is very sensitive to shifts in temperature-especially for the timing of the SoS-but also indicate the importance of water availability to the timing of the EoS, as the increase in drought stress contributes to its advance throughout some regions (e.g., Southern Europe). Our results also revealed that extreme climate events exert severe impacts on vegetation phenology for the timing of both the SoS and EoS. These findings highlight the need to develop strategies focused on mitigating the impact of future climate changes on vegetation, and their monitoring. The interactions of temperature, precipitation, drought, and phenology with other variables-such as solar radiation, soil type, soil moisture, and the coupling between variables and feedback between vegetation and climate-warrant future research.  Table S1: Areas with significant trends (p ≤ 0.05) in the time series (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) for the timing of the SoS and EoS. The slope of the regression line is also represented; Table S2: Percentages of pixels with positive and negative correlations (p ≤ 0.05) between phenology and climatic variables, indicating whether these correlations represent an advance or delay in phenology. The highest correlation values (positive or negative) are shown in bold.