Vegetation Phenological Differences Between Polar‐ and Equatorial‐Facing Slopes in the Three Rivers Source Region, Tibetan Plateau

Vegetation growth is influenced by the microclimate driven by aspects, as evident in the asymmetric vegetation greenness on polar‐facing slopes (PFS) and equatorial‐facing slopes (EFS). However, it remains uncertain whether aspects influence vegetation phenology. To address this question, we defined the aspect‐induced phenological differences between PFS and EFS from 2019 to 2022 within each 3 × 3 km2 grid, using average phenological metrics extracted from Sentinel‐2 data. We found that the start of the growing season (SOS) occurs earlier on EFS in cold and humid regions, but in arid areas, PFS has an earlier SOS. The end of the growing season (EOS) consistently occurred later on EFS due to radiation limitations in autumn phenology. Employing the space‐for‐time approach, the observed distribution of phenological differences within the climate space could potentially indicate the phenological trends of different slope orientations in the future. Our study provides valuable insights into topographic regulation on vegetation phenology.


Introduction
Changes in vegetation phenology can have a significant impact on the carbon budget of terrestrial ecosystems and the feedback between the biosphere and atmosphere (P.Li et al., 2018;X. Shen et al., 2022).Over the past few decades, there has been an increasing focus on how climate factors influence vegetation phenology (Bradley et al., 2011;M. Shen et al., 2015;Zhang et al., 2018).Previous studies have shown that temperature, precipitation, and radiation are the main drivers of vegetation phenology (Gao & Zhao, 2022;Piao et al., 2019;Zhu et al., 2017), with temperature dominating in high latitude regions, and precipitation influencing the temperature dependence of phenology in arid/semiarid regions (M.Shen et al., 2015;D. Wu et al., 2015).Additionally, radiation has a significant impact on vegetation photosynthesis in autumn (Zhang et al., 2018).However, it is unclear whether the response of phenology to climate factors will change with terrain in hillslope environments.
In fact, microclimate can be influenced by terrain, particularly on the opposite polar-facing slopes (PFS) and equatorial-facing slopes (EFS), where a radiation difference of up to 30% can lead to distinct micrometeorological conditions (Del-Toro-Guerrero et al., 2019;Kumari et al., 2020).Generally, EFS are xeric and warmer, while PFS are relatively colder and wetter.Prior research has indicated that the distribution of vegetation in mountainous ecosystems can be regulated by the opposite microclimate.For example, in high latitude regions where temperature is a limiting factor, vegetation benefits more from the environment on EFS compared to the PFS.Conversely, in arid and semi-arid regions where water is scarce, vegetation on the PFS tends to thrive due to their better water retention (Kumari et al., 2020;Yin et al., 2023;Zou et al., 2023).Aside from vegetation greenness, environmental differences between opposing aspects may result in phenological differences.Specifically, in arid and semi-arid regions, where EFS receive more solar radiation and endure higher temperatures, the vegetation in early spring may face water limitations, potentially leading to a delayed SOS on EFS during the spring season (Z.Liu et al., 2022).In autumn, radiation is a major limiting factor for vegetation growth (Descals et al., 2023), EFS offer more conducive environments for vegetation growth compared to PFS, potentially resulting in a delayed EOS on EFS.However, these hypotheses have not been tested by empirical evidence.
Prior research on aspect-controlled ecosystems has primarily relied on field observations, which, while providing detailed information, suffer from a limited number of sites, leading to disagreement on the aspect's impact on vegetation (Barbosa et al., 2006;Lawley et al., 2016).For instance, in Southwest China, J. Yang et al. (2020) reported higher vegetation coverage on PFS than EFS, while Zhou et al. (2019) reached the opposite conclusion for the same region.Remote sensing provides up-to-date information with consistent coverage on a regular basis, and satellite data sets with high spatiotemporal resolutions present an opportunity for detecting vegetation on different aspects, which to some extent overcome the limitations of field observations.However, previous research has paid little attention to investigate the vegetation status on different slope orientations based on remote sensing products, and even less attention has been paid to phenological differences between PFS and EFS.In response to this question, we utilized Sentinel-2 observations, which offer high spatial and temporal satellite data, to extract phenology information on PFS and EFS in decametric spatial resolution (Kollert et al., 2021).
We hypothesized that the phenology of PFS significantly differs from EFS and that this discrepancy relates to climate conditions.To test this hypothesis, we extracted SOS and EOS from Sentinel-2 in 2019-2022 to reveal aspect-induced phenological differences within each 3 × 3 km 2 grid in the Three Rivers Source Region (TRSR) on the Tibetan Plateau.We then analyzed the distribution of phenological differences in the climate space and attempted to unravel the underlying mechanism behind these differences.

Study Area
The TRSR, with an area of 3.36 × 10 5 km 2 , is located in the east of the Tibetan Plateau (Figure 1).It is known as the "Chinese Water Tower" because it is the source of the Yangtze River, the Yellow River, and the Lancang River.The TRSR is predominantly covered by grassland, making up 80% of the total area (Y.Yang et al., 2022).The widespread distribution of grasslands minimizes the potential variation in landcover types between PFS and EFS.Moreover, the region's pronounced climate heterogeneity (Figure S2 in Supporting Information S1) creates favorable conditions for exploring phenological disparities between PFS and EFS of grassland under different climate environments.

Sentinel-2 Data
Sentinel-2 provides atmospherically corrected remote sensing reflectance products with high spatial resolutions of 10 m, 20 m, and 60 m.These products are available publicly and free of charge, featuring a 5-day repeating cycle (Drusch et al., 2012).Although Sentinel-2 has been providing data since 2017, the data quality during the initial 2 years was less than ideal.To enhance the robustness of the results, we utilized the red and near-infrared (NIR) bands at a spatial scale of 10 m from Sentinel-2, encompassing the years 2019-2022, for computing the Normalized Difference Vegetation Index (NDVI).Subsequently, we resampled the NDVI to 30 m using bilinear interpolation to match the spatial resolution of the Digital Elevation Model (DEM).To ensure the accurate estimation of vegetation phenology and avoid the potential influence of snow cover, pixels identified as snow or ice by the quality indicator were replaced with the average NDVI values from the winter months (January-February and November-December).

Auxiliary Data
The NASA Shuttle Radar Topography Mission (SRTM) provides global-scale elevation data at the spatial scale of 30 m.In this study, we generated slopes and aspect maps based on DEM using the Google Earth Engine (GEE) platform.PFS were identified as pixels with aspect values between 315-360°and 0-45°, while EFS were defined as pixels with aspect values between 135 and 225°.Although calculating NDVI through ratios partially mitigates the impact of terrain shadows (Chen et al., 2020), the influence of terrain features on NDVI remains significant, particularly in areas with steep terrain.To minimize the effect of shading, we retained only pixels with slopes less than 25°.The threshold was determined based on 2 considerations: (a) Kumari et al. (2020) reported that the impact of shadows on NDVI could be minimized when the slopes are less than 25°; (b) the number of pixels with slopes higher than 25°was limited in each 3 × 3 km 2 grid (Figure S1 in Supporting Information S1), thus the removal of these pixels did not significantly affect the calculation of NDVI differences on opposing aspects.
Climate data used in this study were sourced from National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/ home), which provides the monthly air temperature and precipitation at the 1 km spatial scale.To detect the distribution of spring or autumn phenology in climate space, this study utilized only preseason temperature and precipitation data, specifically from March to May for spring (X.Liu et al., 2021) and from August to October for autumn.The temperature means the average value of preseason and the precipitation refers to the cumulative value.These data were resampled to a spatial resolution of 3 km using bilinear interpolation to match the spatial resolution of phenological differences between PFS and EFS.
Grassland in this study was delineated according to the European Space Agency (ESA) WorldCover 10 m land cover map, which is produced from the Sentinel-1 and Sentinel-2 data.

Methods
In this study, we employed the Maximum Separation (MS) Method, which is a variant of the threshold method proposed by Descals et al. (2020), to extract phenological parameters.Specifically, we defined a time series as a set of paired values [t, b], where t represents time and b denotes the corresponding NDVI value.The MS method defines a dynamic threshold u based on a percentage (p) of amplitude, as shown in Equation 1.In this study, we set p to 0.3.
Based on u, we converted the time series into binary sequences, designating "1" for values exceeding u and "0" for values below u.It is noteworthy that the "0" sequence was subdivided into two distinct parts in this process.Next, we identified SOS 1 as the DOY corresponding to the first non-null NDVI value in the "1" sequence, while SOS 0 represented the DOY corresponding to the last non-null NDVI value in the first "0" sequence.We determined SOS using the linear interpolation method described in Equation 2, where NDVI sos1 and NDVI sos0 are the NDVI values that correspond to SOS 1 and SOS 0 , respectively.The extraction method for EOS was similar to that for SOS, EOS 0 was defined as the DOY corresponding to the first non-null NDVI value in the second "0" sequence, and EOS 1 was the DOY of the last non-null NDVI value in the "1" sequence.EOS was defined using linear interpolation method shown in Equation 3, with NDVI eos1 and NDVI eos0 refer to the NDVI values corresponding to EOS 1 and EOS 0 , respectively.
To ensure the accuracy of phenological differences (ΔSOS, ΔEOS) calculations, we implemented stringent data filtering criteria.First, we excluded outliers that exceed the ninth and first decile values.Second, we assigned a null value to pixels with a count of less than 150 on either PFS or EFS within a 3 × 3 km 2 grid.Finally, we employed t-test to assess the significance of phenological differences between PFS and EFS within each 3 × 3 km 2 grid.We only considered pixels with significant differences (P < 0.05), which accounted for over 90% of all pixels for both ΔSOS and ΔEOS.
After data filtering, ΔSOS and ΔEOS were expressed by the average difference of 30 m phenology pixels on the PFS and EFS in each 3 × 3 km 2 grid.Positive values of ΔSOS indicate an earlier start of the growing season on EFS compared to PFS, while negative values suggest a later start on EFS.Similarly, positive (negative) ΔEOS values denote a later end of the growing season on the PFS (EFS).

The Variation of Phenology by Aspects
Figure 2 illustrates the variation in phenology with slope orientations.As shown in Figure 2a, the SOS is predominantly concentrated in May, with no discernible difference between PFS (north) and EFS (south).In Figure 2b, the EOS occurs in September.Notably, in contrast to the SOS, there is a substantial disparity in EOS between PFS and EFS, with the EOS on PFS happening approximately 5 days earlier than on EFS.
To provide a more intuitive analysis of the phenological differences between PFS and EFS, we presented a spatial distribution of the phenology differences in Figure 3.As shown in Figure 3a, regions with an earlier SOS in EFS covered 46.4% of the area, primarily located in the southeast of TRSR.In contrast, areas with an earlier SOS in PFS accounted for 53.6% of the regions, predominantly observed in the western regions.This indicates that, during the spring, the grassland benefits more from the environment on EFS in the southeast of TRSR, whereas in the west, the environment on PFS is more favorable for grassland growth.Regarding the EOS, regions with negative differences between PFS and EFS covered over 70% of TRSR, indicating that the EOS on EFS typically occurs later than on PFS across most of TRSR, especially in the southeast regions.However, in the western part of TRSR, the EOS on EFS concludes earlier than PFS.Therefore, the grass growing season on EFS is longer than that on PFS in the southeast regions of TRSR, while the grass growing season on PFS is longer in western regions.

Variation of Phenological Differences With Climate
In terms of climate space, we found that the phenological differences are co-determined by temperature and precipitation (Figure 4).However, in both spring and autumn, precipitation plays a dominant role in the phenological difference between the PFS and EFS, while the ΔSOS is more significantly influenced by temperature compared to ΔEOS.From Figure 4a, in the regions with relatively high precipitation and low temperature, the SOS on EFS tends to precede that on PFS.In most regions with relatively high temperature, the SOS occurs sightly earlier on PFS.However, an unexpected scenario unfolds in regions with the highest temperature and moderate precipitation, where the SOS on EFS precedes that on PFS.Regarding the EOS (Figure 4b), in regions characterized by low temperature and high precipitation, the EOS on PFS concludes earlier than EFS.But in low precipitation areas, the EOS on EFS marginally precedes that on PFS.Similar to the SOS distribution

Geophysical Research Letters
10.1029/2023GL107316 within the climate space, regions with the highest temperature deviate from the general pattern, with the EOS ending later on EFS than on PFS.
Analyzing the changes in ΔSOS and ΔEOS with climate variations, we found that ΔSOS shows a substantial decrease with rising temperature (R = 0.79, P < 0.01) and decreasing precipitation (R = 0.89, P < 0.01).
Additionally, the ΔEOS decreases as precipitation increases (R = 0.94, P < 0.01) and as temperature decreases (R = 0.60, P < 0.01).In the context of a fixed temperature range, we noted a shift in the direction of ΔSOS as precipitation increased.Particularly in regions with moderate temperatures, ΔSOS transitioned from negative to positive with an increase in precipitation.Similarly, within a certain precipitation interval, higher temperature led to a shift in ΔSOS from positive to negative.While the change in ΔEOS direction may not be as pronounced as that of ΔSOS, its value undergoes substantial changes when one variable is held constant while the other varies.

Discussion
The radiation differences induced by divergent slope orientations engender microclimatic conditions, resulting in pronounced environmental disparities between the PFS and EFS.EFS are generally warmer and drier, while PFS are relatively cooler and moister.Previous research has suggested that these aspects can cause inhomogeneous distribution of vegetation (Armesto & Martίnez, 1978;Kumari et al., 2020;Yin et al., 2023).However, it is still an open question whether aspects can cause phenological differences in vegetation.To address this question, we conducted an analysis of the distribution of phenological differences between PFS and EFS.Our results indicate that, like vegetation greenness, the phenology of vegetation is also regulated by aspects.Specifically, though the mean SOS values between PFS and EFS are roughly comparable, the spatial distribution of SOS differences between opposite aspects exhibits notable heterogeneity.The regions where EFS experience an earlier SOS are primarily concentrated in the southeast, whereas PFS show earlier SOS in the western part of the region.For the EOS, the environment on EFS appear to be more favorable for vegetation growth in autumn, particularly in the southeastern study area.Conversely, in the western region of the study area, the EOS concludes later in PFS.
By calculating the phenological differences between PFS and EFS in each 3 × 3 km 2 grid and analyzing their relationship with climate, we discovered that precipitation predominantly influences these phenological differences in spring and autumn.Additionally, temperature has a more pronounced effect in spring than in autumn.Putting the phenological differences in climate space, we observed that in regions characterized by high precipitation and comparatively lower temperatures, exemplified by the southeastern area of TRSR, the SOS on EFS occur earlier than on PFS.This can be elucidated by considering two aspects.On the one hand, in regions characterized by cold and humid conditions, temperature predominantly drives vegetation growth (M.Huang et al., 2017;X. Wu et al., 2017).The warmer conditions on EFS expedite early spring vegetation growth, resulting in an earlier SOS.On the other hand, lower temperatures on PFS lead to delayed snowmelt, potentially keeping the grassland snow-covered in early spring (Wang et al., 2017), thereby impeding photosynthesis and resulting in a later SOS on PFS compared to EFS.However, in regions with relatively high temperatures, such as the northeast of TRSR, SOS on PFS occurs slightly earlier than on EFS.This is mainly attributed to the limitation of soil moisture on vegetation growth.PFS, with their superior water retention, encourage earlier spring growth and an earlier SOS.In contrast, the vegetation on EFS experiences water stress in early spring, resulting in a later SOS in this situation.An exceptional situation is observed in regions with extremely high temperatures, where an earlier SOS is observed on EFS, particularly in the northeastern TRSR region (Figure S2a in Supporting Information S1).This area, situated at a lower elevation compared to adjacent regions (Figure S3 in Supporting Information S1), likely forms a valley characterized by lower radiation levels and enhanced water retention.Consequently, these conditions create a more conducive environment for vegetation growth on the EFS during early spring.Unlike the SOS, the EOS occurs later on EFS than PFS in most of TRSR.This suggests that vegetation benefits more from the EFS environment in autumn, possibly due to radiation being the most important significant limiting factor for EOS (Zhang et al., 2018).However, in both warmer and drier regions, the EOS on the PFS concludes later than on the EFS.This delay may be attributed to the water limitations impacting vegetation on the EFS during autumn.The unexpected EOS in the highest temperature region is attributed to limited radiation and strong water retention in valley areas.Thus, EFS, with its relatively higher radiation levels, offers a more suitable environment for vegetation in this context.
Additionally, it is evident that ΔSOS exhibits a positive correlation with precipitation and a negative correlation with temperatures.Conversely, ΔEOS shows a negative correlation with precipitation but a positive correlation with temperatures.When one variable varies while the other is held constant, especially for ΔSOS, the directions of phenological differences shift.More specifically, in regions with moderate temperatures, as the precipitation increases, grassland transition from early green on PFS to early green on EFS.If precipitation remains constant, grasslands transform from early green on EFS to early green on PFS as temperature increases.Using spatial variation instead of temporal analysis, the distribution of phenological differences in climate space partially reflects the trajectory of future phenological shifts.For instance, with temperature increases and an escalation of drought in the future, regions conducive to grassland prosperity on EFS will become increasingly scarce.Conversely, PFS, with its superior water retention, will become progressively more favorable for vegetation growth.Furthermore, prior studies have established that the soil quality in PFS surpasses that of EFS in arid/semiarid regions (Y.M. Huang et al., 2015;R. Li et al., 2018;Qin et al., 2016), and the porous and moist soil in PFS enhances nutrient absorption in vegetation.All of these findings collectively support the notion that vegetation on PFS will thrive more in the future compared to EFS.This conjecture corresponds with the findings of Xie et al. (2023) and Yin et al. (2023) to some extent.
Although our study provides valuable insights into the spatial distribution of phenological differences between PFS and EFS, some uncertainties remain.One potential limitation is the use of NDVI calculated from satellite products, which may be subject to distortion by topography.However, to minimize this issue, we only extracted vegetation phenology utilizing the pixels with gentle slopes (slope <25°) following the approach of Kumari et al. (2020).Additionally, the short time coverage of Sentinel 2 posed a challenge in depicting trends in phenological differences.Furthermore, we were unable to measure detailed phenology information between PFS and EFS without field measurements.Therefore, further field experiments on paired PFS and EFS are necessary to deepen our understanding of these phenological differences.Despite these limitations, our study also deepens the consciousness of the phenological differences between PFS and EFS.Our forthcoming studies will also focus on examining the growth dynamics of vegetation on eastern and western slopes, given the divergent receipt of solar radiation in the morning and afternoon between the two slopes.

Conclusions
In this study, we calculated the phenological differences between PFS and EFS from 2019 to 2022 within each 3 × 3 km 2 grid and mapped them onto a climate space.The results indicated that the microclimates influence the distribution of phenology, the SOS for EFS occurred earlier in both cold and humid regions, while in arid regions, the SOS for PFS was earlier than that for EFS.Different from SOS, the EOS for EFS is more consistently delayed, with over 70% of the study area experiencing a later EOS on EFS.Using the space-fortime approach, phenological differences between PFS and EFS in climate space may reflect future variations in phenology across different slope orientations.Specifically, the vegetation growth seasons on PFS might be longer as the rise of temperature and the intensification of drought.To our knowledge, this is the first study to evaluate phenological differences across various slope orientations using remote sensing products, thereby offering a significant reference for enhancing our understanding of phenological variation in diverse topographic settings.

Figure 1 .
Figure 1.The location and landcover types the Three Rivers Source Region based on the European Space Agency WorldCover product.

Figure 2 .
Figure 2. Variation of multi-year (2019-2022) average SOS (a) and EOS (b) with slope orientations, the blue points refer to the average value of phenology in different aspects.

Figure 3 .
Figure 3. Spatial distribution of the multi-year average (2019-2022) SOS differences (ΔSOS) (a) and EOS differences (ΔEOS) (b) between polar-facing slopes and equatorial-facing slopes in 3 × 3 km 2 grids, only the pixels with P < 0.05 were retained, the unit d means day.The histogram in the lower left corner represents proportion of positive and negative values to the total number of grids.

Figure 4 .
Figure 4. Distribution of the phenological differences between polar-facing slopes (PFS) and equatorial-facing slopes (EFS) in the climate space for Three Rivers Source Region in 2019-2022.Panel (a) refers the SOS difference (ΔSOS) between PFS and EFS in each bin of mean preseason temperature and cumulative preseason precipitation, (b) is the same as (a) but for EOS.The upper and right figures refer to the variation of phenological differences with precipitation and temperature, respectively.