Long-Term Vegetation Phenology Changes and Responses to Preseason Temperature and Precipitation in Northern China

: Due to the complex coupling between phenology and climatic factors, the inﬂuence mechanism of climate, especially preseason temperature and preseason precipitation, on vegetation phenology is still unclear. In the present study, we explored the long-term trends of phenological parameters of different vegetation types in China north of 30 ◦ N from 1982 to 2014 and their comprehensive responses to preseason temperature and precipitation. Simultaneously, annual double-season phenological stages were considered. Results show that the satellite-based phenological data were corresponding with the ground-based phenological data. Our analyses conﬁrmed that the preseason temperature has a strong controlling effect on vegetation phenology. The start date of the growing season (SOS) had a signiﬁcant advanced trend for 13.5% of the study area, and the end date of the growing season (EOS) showed a signiﬁcant delayed trend for 23.1% of the study area. The impact of preseason precipitation on EOS was overall stronger than that on SOS, and different vegetation types had different responses. Compared with other vegetation types, SOS and EOS of crops were greatly affected by human activities while the preseason precipitation had less impact. This study will help us to make a scientiﬁc decision to tackle global climate change and regulate ecological engineering. article presents a comprehensive analysis of SOS/EOS response patterns of various vegetation types to preseason temperature and preseason precipitation in the mid-latitude region of China from 1982 to 2014. We have discussed the inﬂuence of water and heat conditions on vegetation phenology and further revealed the relationship between vegetation phenology and climate factors.


Introduction
Vegetation phenology, the periodicity of growth and development of vegetation, plays a prominent role in regulating the carbon balance of terrestrial ecosystems [1,2]. Climate change, in turn, affects vegetation phenology [3]. Numerous studies have been conducted to investigate complex causal relationships between vegetation phenology and various climatic factors [4][5][6]. Studies have also attempted to reveal potential climate change signals from vegetation phenology by analyzing comprehensive trends across different vegetation types, phenology stages, and regions [7][8][9]. However, owing to the complex coupling between phenology and climate factors [10][11][12], understanding the internal forcing mechanism of climate-phenology interactions is not complete. As global climate change is projected to intensify in the future [13][14][15][16][17], it is imperative to continue studying the complex interacting mechanism of climate change and phenology to understand future ecosystem dynamics [18].
Among the various climate factors, temperature and precipitation prior to the growing season (hereinafter referred to as the "preseason") play an important role in regulating vegetation phenology. The start date (SOS) and end date (EOS) of the growing season are essential indicators of vegetation phenology. They are directly associated with climatic factors [19]. For example, the plant dormancy period is temperature-controlled vegetation phenology [20]. In regions with a cold season, vegetation must be exposed to sufficient heat accumulation in the growing season in order to break its dormancy and start growing [19]. Therefore, temperature can be considered the most important factor affecting vegetation phenology in many regions [21,22]. Additionally, precipitation is also a key factor that regulates the spring phenology of vegetation in some areas and affects vegetation phenology to a greater extent than the temperature in some areas [23][24][25]. Autumn phenology also plays a prominent role in regulating carbon balance and biomass regionally and globally [26,27]. Studies have shown that temperature has a strong controlling effect on the autumn phenology [28,29]. Higher temperatures tend to delay vegetation EOS [28,30]. However, the effect of temperature on EOS appears to be weaker in the Tibetan Plateau [31]. In the Greater Khingan Mountains, it has also been observed that precipitation, relative to temperature, has a stronger control on EOS [32]. In short, the responses of phenology to preseason temperature and precipitation vary in magnitude and direction, between species and regions, and even between the same species in various regions [33]. These complex response patterns have led to considerable spatial heterogeneity in the relationship between phenology and climate. Although many studies have discussed the relationship between preseason temperature and precipitation and vegetation phenology [24,[34][35][36][37][38][39], these conclusions cannot be applied to all regions or all vegetation. Therefore, it is necessary to conduct a comprehensive study on the relationship between vegetation phenology and climatic factors.
In addition, the cultivation methods in China are complex and extremely vulnerable to human activities. This leads to it being more challenging to explore the response patterns of different vegetation types to climate factors, and the complex response patterns of phenology to preseason climate are still unclear for different vegetation types. In the temperate zone north of 30 • N in China, vegetation types are abundant and seasonal, and the satellite-derived vegetation index is less affected by solar zenith angle [40]. Compared with south of 30 • N in China, the agricultural planting pattern is more simple, with single-season or double-season crops, which is convenient for phenological data extraction. Therefore, we selected the temperate zone north of 30 • N in China as the main study area to investigate the relationships between vegetation phenology and preseason temperature and precipitation.
The main purpose of this study is to provide a more macroscopic and comprehensive perspective to explore the response patterns of different vegetation types SOS and EOS to preseason temperature and precipitation. Specifically, we used the GIMMS-NDVI3g (Normalized Vegetation Index for the third generation Global Inventory Monitoring and Modeling System from the AVHRR sensor) dataset to extract vegetation phenology parameters (SOS and EOS) validated by ground observation phenology data. In particular, we considered the phenology of double-cropping crops, which is often ignored in many studies on the relationship between vegetation phenology and climate. Then, we investigated the integrated response of vegetation SOS and EOS to preseason temperature and precipitation from 1982 to 2014. The relationships between the phenology of various vegetation types (including deciduous broadleaf forest, deciduous needle leaf forest, mixed forest, grassland, savanna, wheat, maize, and rice) and climatic factors were also discussed.

Study Area
The fluctuations in normalized difference vegetation index (NDVI) are not obvious south of 30 • N in China, as a result of the most areas being covered by evergreen forests. In addition, the crops in this area have a complicated planting pattern of three or more seasons a year, which makes it difficult to identify. Therefore, our research focused on the temperate regions of China (north of 30 • N; Figure 1a), where the vegetation index retrieved by satellites is also less affected by the solar zenith angle [40]. Considering that the spatial resolution of GIMMS data is 0.08333 • , we chose MOD12C1 classification data with a spatial resolution similar to that of GIMMS for the purpose of avoiding the influence of excessive interpolation and resampling on data accuracy. Therefore, based on land use  Figure 1b). Specifically, we subtracted the 2001 land-use classification data from the 2014 land-use classification data. If the pixel value was equal to 0, this meant that the land use type had remained unchanged, and vice versa. Additionally, non-vegetation land-use types (including Barren, Water Bodies, Urban and Built-up Lands, Permanent Snow and Ice, and Permanent Wetlands) and land-use types of less than 100 pixels were also excluded. Thus, the remaining area in this study was 4.54 million km 2 (nearly half of the total land area of China). Vegetation types in the selected study area were classified as listed in Table 1.
seasons a year, which makes it difficult to identify. Therefore, our research focused on the temperate regions of China (north of 30°N; Figure 1a), where the vegetation index retrieved by satellites is also less affected by the solar zenith angle [40]. Considering that the spatial resolution of GIMMS data is 0.08333°, we chose MOD12C1 classification data with a spatial resolution similar to that of GIMMS for the purpose of avoiding the influence of excessive interpolation and resampling on data accuracy. Therefore, based on land use type classification data MCD12C1 (Land Cover Climate Modeling Grid product) from the MODIS (Moderate Resolution Imaging Spectroradiometer) satellite, we excluded the pixels with changes in land use/cover from 2001 to 2014 (data for land use/cover before 2001 were not available) to eliminate errors ( Figure 1b). Specifically, we subtracted the 2001 land-use classification data from the 2014 land-use classification data. If the pixel value was equal to 0, this meant that the land use type had remained unchanged, and vice versa. Additionally, non-vegetation land-use types (including Barren, Water Bodies, Urban and Built-up Lands, Permanent Snow and Ice, and Permanent Wetlands) and land-use types of less than 100 pixels were also excluded. Thus, the remaining area in this study was 4.54 million km 2 (nearly half of the total land area of China). Vegetation types in the selected study area were classified as listed in Table 1.

Datasets and Preprocessing
We used the NDVI product of the National Oceanic and Atmospheric Administration The GIMMS NDVI (version 3g.v1) dataset, which has a spatial resolution of 0.0833333 • and a temporal resolution of twice a month, can be downloaded from https://iridl.ldeo. columbia.edu/SOURCES (accessed on 19 January 2021). The ground-based data for crops were derived from the 10-day value dataset of crop growth of China Meteorological Data Network (http://data.cma.cn/ (accessed on 23 January 2021)). It contains information on the development period, plant height, soil moisture, etc., of different crops, observed at 778 sites from 1992 to 2013 and is widely used in Crop Phenology research [41,42]. We finally selected crop growth records of wheat, corn, and rice at 671 sites from 1992 to 2013 (some sites do not contain phenological information of the required crop types). The turning-green period of maize, the emergence period of wheat, and the transplanting period of rice in the ground-based dataset were defined as SOS, and the maturity period of maize, wheat, and rice in the ground-based dataset corresponded to EOS [43]. The meteorological data covering 33 years (1982-2014) of daily precipitation and temperature from 786 monitoring stations were derived from the China Meteorological Data Network (CMDN) (http://data.cma.cn/ (accessed on 18 January 2021)). We used the spline interpolation method to generate raster data with a spatial resolution of 0.83333 • to match the phenological data in the study area [44]. MCD12C1 datasets contain 3 sets of land use classification schemes, among which the IGBP (International Geosphere Biosphere Program Version 6) classification scheme has been widely used in various vegetation studies [45][46][47]. Therefore, we used the IGBP (International Geosphere Biosphere Program Version 6) classification data from MODIS MCD12C1 in 2001 and 2014 for vegetation classification. The original data, with a spatial resolution of 0.05 • , were resampled to generate a new dataset with a resolution of 0.83333 • (with the nearest neighbor method). In addition, the IGBP classification system divides landuse types into 17 categories, including land-use types without vegetation cover. Therefore, according to the research needs of this study, we eliminated the following pixels: • Pixels that showed changes in land-use type between 2001 and 2014; • Minor vegetation types (less than 100 pixels) in the study area; • Non-vegetated land-use types, such as bare land and urban built-up areas.

Calculation and Validation of Phenological Data
We used the Savizky-Golay (SG) filter to reconstruct the NDVI time series curve for each pixel. The SG filter can retain data information to the greatest extent while removing data noise, and it is more suitable for extracting double-season crop information [48]. The formula of SG filtering is given as: where F i and f i are the ith value of the filtered data and the original data, respectively, C j is the jth weight value of the filtering window and is calculated by polynomial fitting with the polynomial order of 4, and m is half the size of the filtering window (=6). Finally, a quadratic spline was used to interpolate the data to daily values [49].
After reconstructing the NDVI time series curve, we selected pixels with one or two maxima (the pixels with two maxima were the candidates for double-season crops) and calculated the second derivative. Two local maximum (inflexion) points in the first half of growth season were regarded as the SOS and the second half as the EOS [50]. To extract the phenological parameters of different crops accurately, we determined the limited time window to be 30 days before and after the average value of the phenological parameters (SOS and EOS) of various crops in different provinces based on ground-measured phenological data [43]. If the estimated SOS and EOS meet the restricted time window of a certain crop (i.e., maize, wheat, and rice), then the corresponding pixel is associated with the crop. If the SOS and EOS of the pixel with a double-season crop meet the limited time windows of both crops, then the pixel is associated with a double-season crop.
where F and F are the predicted phenological sequence data and their mean values, and f and f are the phenological series data and their mean values observed on the ground.

Statistical Analysis
The Theil-Sen (TS) method was used to estimate the long-term trend in EOS/SOS for each pixel from 1982 to 2014 [55,56]. It is a robust method for monitoring trends in time series and is only slightly affected by outliers [57]. The Mann-Kendall (MK) method was used to test the significance of the long-term trend in EOS/SOS [58,59]. The TS and MK methods have been widely used in detecting long-term phenological trends [60,61] and have been shown to exhibit strong stability [57,[62][63][64].
To explore the lag effects of temperature and precipitation on vegetation phenology, we further calculated preseason lengths (in days) for SOS and EOS in each pixel. Here, the preseason length is defined as the n days before the 33-year average SOS/EOS dates for each pixel (1 ≤ n ≤ 180) (similar to "time window durations" discussed in Ren, Li, and Peichl [22]). Preseason temperature and precipitation are the average daily temperature and precipitation for different preseason lengths [65]. We used the partial correlation analysis and T-test to investigate the correlation between SOS/EOS and the preseason precipitation/temperature at each pixel for different preseason lengths (1-180 days). As a result, we obtained 4 × 180 partial correlation coefficients per pixel ( Figure 2). Finally, the preseason length of the highest absolute value of the partial correlation coefficient was selected as the optimal time scale of each category (similar to "best-performing time window durations" in [22]). For example, as shown in Figure 2, the highest partial correlation coefficient between the preseason temperature and SOS of the pixel in row i and column j is −0.60. Therefore, −0.60 was defined as the final partial correlation coefficient Remote Sens. 2022, 14, 1396 6 of 21 between the preseason temperature of the pixel and SOS, and the corresponding preseason length 36 was the optimal time scale between the preseason temperature and SOS in this pixel. Finally, we analyzed the partial correlation between EOS/SOS and preseason temperature/precipitation for each vegetation type from 1982 to 2014 to further study the response patterns of SOS and EOS of the various vegetation types to preseason temperature and precipitation.
ysis and T-test to investigate the correlation between SOS/EOS and the preseason precip-itation/temperature at each pixel for different preseason lengths (1-180 days). As a result, we obtained 4 × 180 partial correlation coefficients per pixel ( Figure 2). Finally, the preseason length of the highest absolute value of the partial correlation coefficient was selected as the optimal time scale of each category (similar to "best-performing time window durations" in [22]). For example, as shown in Figure 2, the highest partial correlation coefficient between the preseason temperature and SOS of the pixel in row i and column j is −0.60. Therefore, −0.60 was defined as the final partial correlation coefficient between the preseason temperature of the pixel and SOS, and the corresponding preseason length 36 was the optimal time scale between the preseason temperature and SOS in this pixel. Finally, we analyzed the partial correlation between EOS/SOS and preseason temperature/precipitation for each vegetation type from 1982 to 2014 to further study the response patterns of SOS and EOS of the various vegetation types to preseason temperature and precipitation. Figure 2. The partial correlation coefficient matrix for pixel in row i, column j (*: p-value<0.05, **: pvalue < 0.01, ***: p-value < 0.001.). T and P stand for preseason temperature and preseason precipitation. The red dotted circle represents the maximum value of each row, and its corresponding preseason length is the optimal time scale.

Validation of Satellite-Based Phenology Data
The phenological dates of ground observations for crops from 1992 to 2013 were compared to the NDVI-based dates. The verification results were shown in Figure 3 and Table  2. The crop phenology data were in good agreement with the ground observation data, with R 2 of 0.968, NRMSE of 10.2%, and PBIAS of −5.3%. The R 2 of three crop types (wheat, maize, and rice) were all ≥0.97. Except for wheat (the NRMSE equals 11.540), the NRMSE of the other two crops were all <10%. In general, the extracted phenology data met the data accuracy requirements. The partial correlation coefficient matrix for pixel in row i, column j (*: p-value<0.05, **: p-value < 0.01, ***: p-value < 0.001.). T and P stand for preseason temperature and preseason precipitation. The red dotted circle represents the maximum value of each row, and its corresponding preseason length is the optimal time scale.

Validation of Satellite-Based Phenology Data
The phenological dates of ground observations for crops from 1992 to 2013 were compared to the NDVI-based dates. The verification results were shown in Figure 3 and Table 2. The crop phenology data were in good agreement with the ground observation data, with R 2 of 0.968, NRMSE of 10.2%, and PBIAS of −5.3%. The R 2 of three crop types (wheat, maize, and rice) were all ≥0.97. Except for wheat (the NRMSE equals 11.540), the NRMSE of the other two crops were all <10%. In general, the extracted phenology data met the data accuracy requirements.

Long-Term Change in Vegetation Phenology
There was an overall advanced SOS trend for a large portion of pixels from 1982 to 2014 over the whole study area (Figure 4a,c). A total of 19.3% of the study area (about 875,766 km 2 ) had a significant change of SOS from 1982 to 2014 (p-value < 0.05), of which 13.5% (about 612,446 km 2 ) showed an advanced trend and 5.8% showed a delayed trend. On average, the SOS within this 19.3% of the study area was advanced by 0.15 days/yr from 1982 to 2014 (Figure 4a,c). The spatial distribution shows that the delayed SOS was relatively scattered over the study area, while advanced SOS was relatively concen- As for the EOS, there was an overall delayed trend for a large portion of the study area (Figure 5a,c). There was a significant change of EOS in 27.5% (about 1.25 million km 2 ) of the study area (p-value < 0.05), of which 23.1% (about 1.05 million km 2 ) showed a significantly delayed trend and only 4.4% (about 201,122 km 2 ) showed a significant advanced trend. Overall, EOS was delayed by 0.19 days/yr in 27.5% of the study area. Compared with SOS, the geographical distribution of EOS with a significant delay trend was more concentrated for savanna in the northeast and cropland in the east (Figure 1a vs. Figure 5a,c). The EOS of grassland and rice had the largest number of days' delay, about 0.33 days/yr and 0.31 days/yr, respectively (p-value < 0.05), while the EOS of savanna and mixed forest had the smallest number of days' delay, with values of 0.06 days/yr and 0.03 days/yr, respectively (p-value < 0.05) (Figure 5e-l).

Overall SOS Response and Spatial Distribution
Spatial distribution of the correlations between SOS and preseason temperature for the first and second season of the double-season crop as well as single-season vegetation are shown in Figure 6a,b. The SOS changing rate per preseason temperature and the significance level of the correlations are shown in Figure 6c-f. We found that about 65.4% (approximately 1.57 million km 2 ) of the study area showed a negative correlation between SOS and preseason temperature, with 42.3% (about 1.02 million km 2 ) having a significant negative correlation (p-value < 0.05, Figure 6g). About 16.4% (about 395,443 km 2 ) of the study area had a significant positive correlation (p-value < 0.05; Figure 6g). For every 1 • C increased in preseason daily mean temperature at the optimal time scale, the SOS advanced 0.62 days (p-value < 0.05) (Figure 6h).
Spatial distribution of the correlations between SOS and preseason precipitation for the first and second season of the double-season crop as well as single-season vegetation are shown in Figure 6i,j. The SOS changing rate per preseason precipitation and the significance level of the correlations are also shown in Figure 6k-n. The relationship between vegetation SOS and preseason precipitation was not as significant as the preseason temperature for the study area. We found that the SOS was negatively correlated with preseason precipitation (p-value < 0.05) in 19.4% (about 465,440 km 2 ) of the study area, whereas the area in which SOS was positively correlated with preseason precipitation was only 10.8% (about 260,502 km 2 ) (p-value < 0.05) (Figure 6o). Apparently, preseason temperature and precipitation showed varying degrees of influence on SOS in the study area. Although the influence of preseason precipitation on SOS was not obvious, preseason precipitation explained variations in the SOS for areas where preseason temperature could not, such as the border area between Inner Mongolia and Liaoning Province (Figure 6e,m).  As for the EOS, there was an overall delayed trend for a large portion of the study area (Figure 5a,c). There was a significant change of EOS in 27.5% (about 1.25 million km 2 ) of the study area (p-value < 0.05), of which 23.1% (about 1.05 million km 2 ) showed a significantly delayed trend and only 4.4% (about 201,122 km 2 ) showed a significant advanced trend. Overall, EOS was delayed by 0.19 days/yr in 27.5% of the study area. Compared with SOS, the geographical distribution of EOS with a significant delay trend was more concentrated for savanna in the northeast and cropland in the east (Figure 1a vs. Figure  5a,c). The EOS of grassland and rice had the largest number of days' delay, about 0.33 days/yr and 0.31 days/yr, respectively (p-value < 0.05), while the EOS of savanna and

SOS Responses of Different Vegetation Types
The partial correlation between preseason temperature/precipitation and SOS for the six vegetation types in the study area is shown in Figure 7a-l. The partial correlation coefficient between preseason temperature and SOS were mostly negative (ranging from −0.01 to −0.70), indicating similar response patterns for the six vegetation types. Among the six vegetation types, the partial correlations were stronger for the deciduous broadleaf forest, deciduous needle leaf forest, mixed forest, and savanna, with R values of ≤−0.40 (p-value < 0.05), while for grassland and cropland, the partial correlations were relatively weaker, with R values of −0.01 and −0.32, respectively. We found that the SOS advanced for more than 2 days for every 1 • C increase in preseason temperature for deciduous needle leaf forest, deciduous broadleaf forest, mixed forest, and savanna, indicating higher sensitivity. However, there was no significant partial correlation between the preseason precipitation and the SOS of each vegetation type.

Overall SOS Response and Spatial Distribution
Spatial distribution of the correlations between SOS and preseason temperature for the first and second season of the double-season crop as well as single-season vegetation are shown in Figure 6a,b. The SOS changing rate per preseason temperature and the significance level of the correlations are shown in Figure 6c-f. We found that about 65.4% (approximately 1.57 million km 2 ) of the study area showed a negative correlation between SOS and preseason temperature, with 42.3% (about 1.02 million km 2 ) having a significant negative correlation (p-value < 0.05, Figure 6g). About 16.4% (about 395,443 km 2 ) of the study area had a significant positive correlation (p-value < 0.05; Figure 6g). For every 1 °C increased in preseason daily mean temperature at the optimal time scale, the SOS advanced 0.62 days (p-value < 0.05) (Figure 6h).  The frequency distribution of partial correlation coefficients between preseason temperature/precipitation and SOS for the six vegetation types is shown in Figure 7m-x. For deciduous needle leaf forest, deciduous broadleaf forest, mixed forest, and savanna, partial correlation coefficients between preseason temperature and SOS were mostly negative (98.8%, 87.8%, 49.8%, and 93.8%, respectively; p-value < 0.05). The effect of temperature on vegetation spring phenology is stronger than that of precipitation on vegetation spring phenology. Of this, about 31.1% and 24.6% of the study area showed a significant positive and negative correlation, respectively, between EOS and preseason precipitation (Figure 8o). On average, an increase in daily mean precipitation of 1 mm in the optimal time scale caused a delay in EOS by 1.86 days (Figure 8p). In general, vegetation EOS in most of the area has a significant positive correlation with preseason temperature, while the correlation with preseason precipitation is subtle. However, preseason precipitation can explain changes in EOS that cannot be explained by preseason temperature in some regions, such as the border area of Inner Mongolia and Heilongjiang Province reflecting the complex internal impact mechanism of preseason precipitation on EOS (Figure 8e,m).

EOS Responses of Different Vegetation Types
The partial correlation between preseason temperature/precipitation and EOS for the six vegetation types in the study area is shown in Figure 9a-l. In contrast with the response pattern of the SOS to preseason temperature, the partial correlations between preseason temperature and EOS were mostly positive (ranging from 0.36 to 0.55; Figure 9g-l). Greater correlation coefficients were found for deciduous needle leaf forest and savanna (0.55 and 0.52, respectively; p-value < 0.05). The EOS for mixed forest and savanna showed greater sensitivity to preseason temperature, with more than 2 days delay per 1 • C increased (Figure 9g,k). As opposed to the response pattern of the SOS/EOS to preseason temperature, the partial correlation between preseason precipitation and EOS was inconsistent across various vegetation types. For example, partial correlation coefficients between preseason precipitation and EOS were mostly negative for deciduous needle leaf forest, deciduous broadleaf forest, mixed forest, and savanna (−0.46, −0.42, −0.54, and −0.45, respectively; p-value < 0.05), which was opposite to the partial correlation pattern between preseason temperature and EOS. Among them, the EOS of the mixed forest was the most sensitive to preseason precipitation, with more than 4 days' delay per 1 mm increase in precipitation. Remote Sens. 2022, 14, x FOR PEER REVIEW 14 of 25   The frequency distribution of correlation coefficients between preseason temperature/precipitation and EOS for the six vegetation types is shown in Figure 9m-x. For deciduous needle leaf forest, deciduous broadleaf forest, mixed forest, and savanna, the partial correlation coefficients between EOS and preseason temperature were mostly positive (95.9%, 44.5%, 39.2%, and 69.7%, respectively; p-value < 0.05), while the partial correlation coefficients between EOS and preseason precipitation were mostly negative (84.6%, 42.5%, 56.9%, and 61.1%; p-value < 0.05). For grassland, the partial correlation coefficients between preseason precipitation and SOS were mainly negative (with the proportion of 21.6%), whereas the partial correlation coefficients with EOS were mainly positive (with the proportion of 41.0%). The partial correlations between preseason precipitation and EOS for grassland and cropland were not as significant as for the other vegetation types.

Optimal Time Scale for Preseason Temperature and Precipitation
Area percentage distribution of optimal time scale for preseason temperature and precipitation impacts on SOS for different vegetation types are shown in Figure 10. In general, the preseason length where the temperature had the greatest influence on SOS was mainly concentrated at the range of 0-60 days. The optimal time scales for deciduous needle leaf forest, deciduous broadleaf forest, and savanna were mostly concentrated in 20-40 days (79.3%; p-value < 0.05), 0-50 days (75.1%; p-value < 0.05), and 0-75 days (90.0%; p-value<0.05), respectively. Please note that, as the preseason length increased, the influence of preseason temperature on the SOS gradually diminished. In contrast, the distribution of optimal time scale for the influence of preseason temperature on vegetation EOS was more scattered. For example, although the optimal time scale of most pixels was concentrated in 0-30 days, there are still about 17.8% (deciduous needle leaf forest), 8.4% (deciduous broadleaf forest), 9.8% (mixed forest), 6.6% (grassland), 33.0% (savanna), 9.2% (maize), 6.4% (wheat), and 17.7% (rice) of pixels with the optimal time scale concentrated in 150-180 days (p-value < 0.05). Therefore, preseason temperature showed a more profound impact on autumn vegetation phenology of vegetation in the study area. Area percentage distributions of optimal time scale for preseason temperature and precipitation impacts on EOS for different vegetation types are shown in Figure 11. Although the optimal time scale pixels were mostly concentrated in 0-30 days, their distributions were not concentrated for the SOS (Figure 11a-h), showing that the impact of precipitation on vegetation SOS was not as significant as temperature. In contrast, the impact of precipitation on vegetation EOS was stronger than the impact on SOS. The optimal time Area percentage distributions of optimal time scale for preseason temperature and precipitation impacts on EOS for different vegetation types are shown in Figure 11. Although the optimal time scale pixels were mostly concentrated in 0-30 days, their distributions were not concentrated for the SOS (Figure 11a-h), showing that the impact of precipitation on vegetation SOS was not as significant as temperature. In contrast, the impact of precipitation on vegetation EOS was stronger than the impact on SOS. The optimal time scale was also concentrated in 0-60 days. At the same time, the rainfall in the 60-100 days of many pixel seasons also had a strong control effect on the vegetation EOS. The impact of preseason precipitation on EOS gradually diminished as the preseason time increased (Figure 11i-p).

Effects of Preseason Temperature on Vegetation Phenology
In most temperate regions of China, we observed a significant negative correlation between the spring phenology of vegetation and preseason temperature, as well as a significant positive correlation between autumn phenology and preseason temperature (Figures 6 and 8), which is consistent with observations of a previous study [25]. The results suggest that higher temperatures will advance SOS and delay EOS in most temperate regions of China. Sufficient heat accumulation is required for the onset of vegetation growth [66]. Generally, the temperature in the study area needs to reach 0-5 °C in order to end the dormant period and meet the necessary weather conditions for leaf onset [19]. Higher temperatures can accelerate heat accumulation, thereby promoting the greening of vegetation. This is why the optimal time scale of the preseason temperature was concentrated in the 0-60 days prior to the mean SOS, which is also known as the dormant period.
Preseason temperature also has a significant controlling effect on autumn phenology (Figure 8). According to reports, higher temperatures in summer and autumn will increase the activity of photosynthetic enzymes and reduce the degradation rate of chlorophyll, Figure 11. (a-h) Histograms of the optimal time scale of different land-use types for SOS vs. preseason precipitation, and (i-p) for EOS vs. preseason precipitation. Red part means p-value < 0.05, while gray means p-value ≥ 0.05.

Effects of Preseason Temperature on Vegetation Phenology
In most temperate regions of China, we observed a significant negative correlation between the spring phenology of vegetation and preseason temperature, as well as a significant positive correlation between autumn phenology and preseason temperature (Figures 6 and 8), which is consistent with observations of a previous study [25]. The results suggest that higher temperatures will advance SOS and delay EOS in most temperate regions of China. Sufficient heat accumulation is required for the onset of vegetation growth [66]. Generally, the temperature in the study area needs to reach 0-5 • C in order to end the dormant period and meet the necessary weather conditions for leaf onset [19]. Higher temperatures can accelerate heat accumulation, thereby promoting the greening of vegetation. This is why the optimal time scale of the preseason temperature was concentrated in the 0-60 days prior to the mean SOS, which is also known as the dormant period.
Preseason temperature also has a significant controlling effect on autumn phenology (Figure 8). According to reports, higher temperatures in summer and autumn will increase the activity of photosynthetic enzymes and reduce the degradation rate of chlorophyll, leading to the delay of EOS [32,67]. At the same time, higher preseason temperature is also a sign of warm winter, which means that extreme weather conditions such as frost are less harmful to vegetation [68]. In short, the senescence of vegetation is a complex and slow process involving many non-biological factors [69,70], which are beyond the scope of this article, except for preseason temperature and precipitation.
Although the optimal time scale of temperature for the EOS was mainly concentrated at 0-60 days in the study area, there were also many areas where the optimal time scale of temperature was concentrated at 120-180 days. This may be attributed to the fact that a higher temperature in spring would quickly lead to the accumulation of the heat required for vegetation growth, giving rise to the rapid growth of vegetation and promoting the early start of autumn phenology [66]. In addition, the correlation between the SOS of vegetation (grassland and cropland) and preseason temperature was weaker than that for the other four vegetation types, as has also been reported in other studies [71].

Effects of Preseason Precipitation on Vegetation Phenology
Compared with temperature, the responses of vegetation phenology to preseason precipitation in temperate regions of China were more complex. For deciduous needle leaf forest, deciduous broadleaf forest, mixed forest, and savanna, the correlations between preseason precipitation and SOS were not significant. This may be due to the well-developed root system of the woodland, the stronger soil water storage capacity, and therefore the stronger resistance to drought [34]. Furthermore, temperate forests have relatively low evapotranspiration and high precipitation. They can retain water for the next rainy season, which reduces the impact of drought on vegetation [72]. Therefore, the change of preseason precipitation has a minor influence on the spring phenology of forests.
Interestingly, we found that the EOS of deciduous coniferous forest, deciduous broadleaved forest, mixed forest, and savanna were mainly negatively correlated with preseason precipitation. This may be related to the average precipitation in the region [71]. Excessive precipitation may cause the vegetation growing season to end prematurely in areas with abundant precipitation. Studies have shown that soil moisture affects the growth of vegetation [73][74][75][76][77]. The increase in soil moisture may promote the photosynthesis of vegetation, which in turn prompts the early end of the vegetation growing season [32,78]. Other climatic factors, such as snowmelt, may also replenish soil moisture, leading to an earlier fall of vegetation phenology. In addition, due to the influence of permafrost in some forests in cold regions, high soil moisture can easily limit the absorption of soil nutrients by vegetation, thereby ending growth prematurely [32,79].
The impact of precipitation on vegetation EOS was greater than that on vegetation SOS. For many vegetation types, the effect of precipitation on EOS was stronger and more complex than that of temperature. The responses of vegetation phenology to precipitation in various vegetation types and regions are different. Excessive rainfall surplus or deficit can affect vegetation growth. In short, vegetation senescence is a complicated process and the understanding of its relationship with climatic factors needs to be further investigated in detail. There is no dispute that temperature has a strong control on EOS. However, the effect of precipitation on vegetation phenology is more complex and is related to multiple other factors. Therefore, the influence of extreme conditions and composite factors should be further considered in future research on the relationship between vegetation phenology and precipitation.

Impacts, Limitations, and Future Work
This study showed that vegetation phenology was significantly correlated with preseason temperature and preseason precipitation. Generally, there was a negative correlation between temperature and vegetation SOS, and a positive correlation between temperature and vegetation EOS in the temperate regions of China. The effect of precipitation on vegetation phenology was much more complicated, with various patterns of influence depending on the region and vegetation type. Overall, this study comprehensively explored the relationship between SOS/EOS of different vegetation types and preseason temperature/precipitation, therefore contributing to the study of the complex influencing mechanisms between vegetation phenology and climate.
Our study focused on the different response patterns between various vegetation types. However, many studies have shown that phenology of the same vegetation type may respond differently to climatic factors depending on other factors such as geographical conditions [21]. This topic will be considered in future work. In addition, owing to limitations in spatial resolution of used datasets, we only limited our study to five natural vegetation types and three crops. With the help of higher-spatial-resolution data, it will be necessary in future to explore the phenology variations of more vegetation types in response to preseason temperature and precipitation.

Conclusions
In this study, we investigated long-term phenological trends in the mid-latitude region of China from 1982 to 2014. In particular, we analyzed the SOS/EOS response patterns of various vegetation types to preseason temperature and preseason precipitation. The preseason length of optimal influence was also examined. Moreover, the complex interaction mechanisms between vegetation phenology and climate were explored.
The results show that the SOS of vegetation in most regions within the study area had an advanced trend, whereas the EOS had a delayed trend from 1982 to 2014. The phenology of various vegetation types had the same response pattern to preseason temperature, which exerted a strong controlling effect. In contrast, the phenology of various vegetation types had different response patterns to preseason precipitation, which showed a stronger effect on EOS than on SOS in many areas. The impact of precipitation on vegetation types was relatively more complex.
In summary, this article presents a comprehensive analysis of SOS/EOS response patterns of various vegetation types to preseason temperature and preseason precipitation in the mid-latitude region of China from 1982 to 2014. We have discussed the influence of water and heat conditions on vegetation phenology and further revealed the relationship between vegetation phenology and climate factors.