Yunnan province (Figure 1), situated in the southwestern region of China, is renowned for its rich plant diversity and endemism. It encompasses a wide range of vegetation types, ranging from tropical rainforests to alpine shrublands. The distinct geographic location of Yunnan, where the Himalayan mountains and the Southeast Asian peninsula converge, makes it a critical area for studying the impacts of climate change on vegetation phenology. Firstly, Yunnan plays a vital role in the global carbon cycle, given its vast expanse of forests and other vegetation acting as carbon sinks (Yuan et al., 2017; Ding et al., 2021). Secondly, Yunnan exhibits a high sensitivity to variations in temperature and precipitation, which can profoundly impact vegetation phenology and ecosystem function (Ge et al., 2021). Thirdly, Yunnan province is renowned for its remarkable diversity in phenological patterns (Liu et al., 2022). The major vegetation types in Yunnan include tropical rainforest, monsoon rainforest, evergreen broad-leaved forest, coniferous forest, subalpine meadow, etc., each exhibiting distinct phenological characteristics. Changes in vegetation phenology and carbon uptake in Yunnan can therefore have significant implications for regional and global C cycling and climate. Overall, Yunnan’s rich plant diversity, sensitivity to climate variability, and ecological importance make it an ideal study object for advancing our understanding of the impacts of vegetation phenology on ecosystem function and service. Moreover, the unique characteristics of Yunnan can also provide insights into the broader global patterns of plant response and adaptation to climate change.

Based on the research objectives of this paper and existing studies (Lu et al., 2021; Zhang et al., 2021; Chen and Zhang, 2023), we selected representative influencing factors of climate, phenology and vegetation growth conditions, respectively. The three meteorological variables were precipitation (PREC), temperature (TEMP) and solar radiation (SRAD). The three phenology variables were the start (SOS), end (EOS) and length (LOS) of the growing season. The normalized vegetation index (NDVI) was used to characterize the natural growth of vegetation. Due to the drastic changes in vertical elevation in Yunnan Province, this study specifically included elevation above sea level (ELEV) in the analysis. The researchers downloaded time series data (e.g., remote sensing and meteorological data) and non-time series data (e.g., elevation data) and pre-processed them, including coordinate alignment, projection transformation and resolution adjustment. Among them, the meteorological data (1979–2018, spatial resolution 0.1°) were obtained from the National Meteorological Data Center (https://data.cma.cn/) and the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/). MODIS satellite data (2001–2019, spatial resolution 500 m) MODIS satellite data (2001–2019, spatial resolution 500 m), including MOD09A1-(NDVI), MCD12Q2-(LOS/EOS/SOS) and MOD17A3-(NPP). Digital elevation data (90 m spatial resolution) were downloaded from the USGS (https://lpdaac.usgs.gov/data/). Among the data products, the Terra and Aqua combination Moderate Resolution Imaging Spectroradiometer (MODIS) Land Cover Dynamics (MCD12Q2) version 6 data product provides global land surface phenology metrics for each year from 2001 through 2019. The MCD12Q2 Version 6 data product is derived from time series of the 2-band Enhanced Vegetation Index (EVI2) calculated from MODIS Nadir Bidirectional Reflectance Distribution Function (BRDF)-Adjusted Reflectance (NBAR). Vegetation phenology metrics at 500 meter spatial resolution are identified for up to two detected growing cycles per year. Each MCD12Q2 file (multilayer data file HDF4 format) provides information layers of detected phenological indicators on an annual basis. For example, total number of vegetation cycles, green start (i.e., SOS), green rise midpoint, maturity, green peak, senescence, green fall midpoint, and dormancy (i.e., EOS). And LOS is obtained by subtracting SOS from EOS. Due to the limitations of MODIS data and meteorological data time series, the time span used in this project is 2001–2018. The temporal resolution of all time series data is in years. Additionally, given the large number of SEM models, the long computation time, and the spatial resolution of all the data, we used the “cubic” method to resample the data and increase/decrease the spatial resolution to 5000 m to improve computational efficiency.

The overall temperature in Yunnan Province is high in the south and low in the north, consistent with the latitudinal distribution. Northwestern Yunnan has lower temperatures due to the effect of higher elevation. On the other hand, some low-lying areas, such as dams and river valleys (Figure 1 Elevation), have higher temperatures (Figure 1 TEMP). Rainfall in Yunnan Province is influenced by the monsoon climate zone in which it is located, with a general distribution of high in the southwest and low in the northeast. Also influenced by the monsoon, southeastern Yunnan has relatively high rainfall. Moreover, due to the influence of altitude, rainfall in the upland region is lower than that in the lowland region (Figure 1 PREC). In particular, under the combined effects of location, monsoon, topography and sunshine hours, solar radiation in Yunnan Province shows a characteristic mid-high and two-side-low distribution form. At the same time, the area with higher radiation in the middle shows a distribution form of high in the north and south and low in the middle due to the topography (Figure 1 SRAD). From Figure 2, it can be seen that the climate of Yunnan Province has a certain trend over the study period. Temporally, temperature showed a significant upward trend (slope = 0.032, P = 0.035), whereas neither rainfall nor solar radiation changes were significant (Figure 2D). Spatially (based on each pixel), the trend in rainfall exhibited a minor magnitude (slope ranging from −0.002 to 0.005) and indicated a slight decrease in the northwestern region (Figure 2A). Conversely, the trend of solar radiation displayed the largest magnitude (slope ranging from −0.89 to 1.68) and was more variable in the study area. Specifically, the outer regions of Yunnan Province were dominated by decreasing trends, while the central part was dominated by increasing trends (Figure 2B). Temperature trends were moderate and relatively consistent within the study area, except for a significant decreasing trend in the western margin and a significant increasing trend in the northwestern part (Figure 2C).

To analyze the relationship between the climate and NPP and phenology over time, we performed a correlation analysis for the 2001–2018 time series (Figure 3). The p-values for all correlations were less than 0.05. Among them, NPP was mainly positively correlated with PREC, LOS and NDVI, with correlation coefficients of 0.68, 0.67 and 0.76, respectively. LOS was mainly positively correlated with TEMP, PREC, EOS and NDVI, and negatively correlated with ELEV, with correlation coefficients of 0.54, 0.52, 0.71, 0.67 and −0.59, respectively. SOS was mainly negatively correlated with NDVI, with a correlation coefficient of −0.5. EOS was mainly positively correlated with TEMP, and negatively correlated with ELEV, with correlation coefficients of 0.6 and −0.61, respectively. The climate factors that have the greatest impact on NPP, EOS/LOS, SOS and NDVI are PREC, TEMP, SRAD and PREC, respectively.

SOS showed a certain trend of advancement over time, although the trend was not very significant and only satisfied a significance of 0.1 (Figure 4A). On the other hand, EOS did not show a significant linear relationship over time. However, it satisfied a significant cubic relationship, forming a fluctuating change of advancement, followed by postponement and then advancement again (Figure 4B). While both SOS and EOS did not show significant linear changes over time, LOS showed a significant linear growth trend (Figure 4C). The SOS and EOS show a clear quadratic U-shaped relationship (Figure 4D). As shown in the figure, when the SOS is less than 106 day of year, a delayed SOS will advance the EOS, resulting in a shorter LOS. Conversely, when the SOS is greater than 106 day of year, the EOS will also be delayed if the SOS is delayed. However, since the delay in the EOS is greater than the delay in the SOS, it leads to an increase in the LOS. Consequently, the non-linear relationship between SOS and EOS leads instead to a very significant linear growth trend in LOS. Furthermore, it is evident from the figure that the NPP exhibits a notable decrease with a delay in the SOS. Specifically, for each day of SOS delay, the NPP decreases by 4.58 gC/m² yr. LOS is jointly determined by SOS and EOS. Specifically, LOS exhibits a highly significant positive correlation with EOS across years. For every day of delay in EOS, LOS will lengthen by 0.66 days/year (Figure 4E). Unlike EOS, SOS has a segmented effect on LOS. When SOS is less than 106 day of year, SOS has a significant impact on LOS. Every day of delay in SOS reduces the LOS by 2.97 days (Figure 4F). However, when SOS is greater than 106 day of year, the effect of SOS change on LOS is significantly attenuated and the change in LOS is mainly determined by EOS (Figures 4E, F).

As shown in Figure 5, the explanatory power of ELEV and climate factors for NPP is 0.6 (SEM1). The explanatory power for NPP reaches 0.8 (SEM2) when NDVI is added to the model SEM1. However, introducing SOS-EOS or LOS as phenological factors only yields an explanatory power for NPP ranging from 0.7 to 0.74 in SEM3. When both NDVI and phenological factors SOS-EOS or LOS are included in the model (SEM4), the explanatory power for NPP demonstrates a minimal increase of 0.01–0.02 to reach 0.81–0.82 compared to that in SEM2. In addition, the direct effects of PREC and TEMP on NPP were significantly reduced by adding NDVI and phenological factors and then decomposed into indirect effects, while the direct effects of SRAD on NPP were increased. For instance, (1) When we introduced the EOS-SOS or LOS factor (SEM3), the indirect influence of PREC on NPP through phenology was 0.134 (EOS-SOS) and 0.132 (LOS), splitting out 22.4% and 22.1% of the total effect, respectively. The indirect effects of TEMP were 0.053 and 0.065, splitting 45.3% and 55.6% of the total effect, respectively. (2) When we introduced the NDVI factor (SEM2), the indirect effect of PREC was 0.262, accounting for 43.9% of the total effect, while the indirect effect of TEMP was 0.024, splitting 20.5% of the total effect. (3) When both phenology (EOS-SOS/LOS) and NDVI factors were introduced (SEM4), it combined the partitioning effect of both types of factors on PREC and TEMP, although the total effect on NPP did not increase. That is, the indirect effect of PREC on NPP was 0.267 (EOS-SOS&NDVI) and 0.279 (LOS&NDVI), splitting 44.7% and 46.7% of the total effect, respectively. The indirect effects of TEMP on NPP were 0.041 and 0.047, splitting 35% and 40.2% of the total effect, respectively. (4) The direct effect of SRAD increases from 0.29 in SEM1 to 0.36 in SEM4b. It is worth noting that although the SOS-EOS model exhibits slightly higher ultimate explanatory power for NPP than the LOS model, the latter demonstrates a lower Akaike Information Criterion (AIC) value.

Among the climate factors, TEMP mainly affects EOS directly (Figure 6 EOS), SRAD mainly affects SOS directly, and PREC mainly affects SOS indirectly (Figure 6 SOS), while PREC and TEMP have no significant direct effect on SOS. Similarly, Chen and Zhang (2023) found that in Yunnan, the temperature was significantly positively correlated with EOS and SRAD was significantly positively correlated with SOS. Zhang et al. (2023) also found that temperature had a greater effect on autumn phenology than radiation and precipitation in China. In contrast to our study, Wu et al. (2021) found that SOS was significantly negatively correlated with temperature in most of the study area, and Shen et al. (2022) also found that SOS in the study area was more sensitive to mean temperature and warming in both time and space, while Yu et al. (2022) found that the main trends of SOS and EOS in the study area were closely related to changes in temperature and precipitation.

Furthermore, we found that the overall effect of SRAD on SOS and EOS was the largest among the climate factors (Figure 6). Stronger SRAD is usually accompanied by higher temperature and sufficient sunshine intensity (Yang et al., 2017), so in addition to its role in photosynthesis, it can provide valuable insights into photoperiod estimation (Yang et al., 2017) and indicate the duration of sunlight at sunrise. Stronger SRAD is associated with higher sunshine intensity, higher temperature and longer sunshine duration (Calle et al., 2010). This suggests that SRAD is the most important climatic factor determining SOS and EOS in Yunnan Province. These results are consistent with those of Chen and Zhang (2023). In addition, although TEMP may not have as much influence on SOS as PREC and SRAD, it still plays an important role. Specifically, TEMP determines the onset of photoperiodic signal capture. Since autumn and spring have the same photoperiod length, plants must first experience low temperatures and then receive signals that winter has passed before they can begin to receive photoperiodic signals (Körner and Basler, 2010). Photoperiod is defined as the duration of light availability in a 24-hour period (Piao et al., 2019), and changes in photoperiod are relatively more accurate and stable than changes in temperature. Yunnan Province has a complex and diverse topography, and temperature varies dramatically. If the vegetation SOS is only sensitive to temperature, it is easy to make “mistakes”. Photoperiod sensitivity can, therefore, protect plants from the potentially fatal consequences of tracking temperatures at the “wrong” time of year (Körner and Basler, 2010). As a result of global warming, plants may sense temperature increases earlier and thus emit SOS earlier; however, future prolonged and rapid warming may increase the thermal requirements for SOS by reducing the number of cold days or slowing the realization of cold requirements, potentially slowing the trend of SOS progression (Shen et al., 2022). As the SOS photoperiod threshold (set by genes) is approached, internal controls will increasingly limit the progressive trend of SOS. When the photoperiod threshold is reached, the SOS will stop advancing, regardless of increasing temperature or solar radiation, due to internal genetic limitations. Breaking this limit will require several generations of trees to renew the mutation, which could take hundreds of years. This phenomenon may provide some insight into the slowing of phenological change in most vegetation (Körner and Basler, 2010).

In addition, the negative correlation between PREC and SOS may also be caused by increased precipitation leading to increased cloudiness, which in turn reduces SRAD and indirectly affects SOS. It is also worth noting that the photoperiodic sensitivity of spring phenological events may vary considerably between species and latitudes, and the influence of climate on phenology may also vary considerably between regions (Piao et al., 2019). As the exact mechanisms of the effects of temperature, precipitation and their interactions on phenology are incompletely understood (Yu et al., 2022), and the responses of different ecosystems to different levels of climate change are far from clear, the main reasons for these differences need to be further investigated and analyzed, especially in the Chinese region (Yang et al., 2017).

The spatial pattern of EOS trends is more complicate than that of SOS (Piao et al., 2019). EOS is positively correlated with TEMP and shows significant cubic polynomial fluctuations over time due to the very drastic temperature changes in Yunnan Province. Consistent with our findings, Shen et al. (2022) also observed that higher T_min (Minimum Temperature) substantially delays the EOS by slowing the physiological processes of leaf senescence. However, Wu et al. (2021) found that the autumn phenology of typical vegetation types in the study area was advanced significantly, even more than the spring phenology. In addition, recent studies found a positive intercorrelation between spring and autumn phenology (Piao et al., 2019; Chen and Zhang, 2023). Nonetheless, our study found a significant quadratic U-shaped relationship between SOS and EOS (Figure 4D). When SOS is greater than 106 days, the direction of change of SOS and EOS is primarily synchronous, i.e. if SOS is postponed, EOS is also delayed. Combined with Figures 4D–F, it becomes evident that once SOS is above 106 days, there is no longer a significant relationship between SOS and LOS, and the prolongation of LOS is mainly dominated by EOS. Therefore, with the aging of the plant leaves and increased respiration due to delayed EOS, NPP decreased instead. However, when SOS was less than 106 days, the changes in SOS and EOS became opposite, i.e. SOS advanced, EOS delayed, and LOS prolonged, resulting in an increase in NPP. Shen et al. (2022) also showed that EOS was affected by SOS, which was manifested in the form of EOS advancing (delayed) as SOS advanced (delayed). Warmer temperatures and increased precipitation were shown to be the main drivers of SOS advancement and EOS delay, but insufficient moisture led to SOS delay and EOS advancement. Shen et al. (2022) found that although the direct effect of temperature on plant phenology is stronger than that of moisture conditions, the deterioration of moisture conditions due to warming may weaken (or even reverse) the effect of temperature on phenology. Coincidentally, the mean value of SOS over the study period was also about 106 days. This means that the mean value of SOS (106 days) is likely to be the dividing line between changes in NPP (increase or decrease) due to changes in EOS (advance or shift) in the later period, which represents a transition point in the biological growth and maintenance process in Yunnan Province. In addition, the SOS in Yunnan Province maintained a continuous advancement from 2012 to 2015, and the changes in SOS and EOS became opposite after 2014 (Figures 4A, B). This change may be related to the consecutive province-wide drought disasters in Yunnan from 2009 to 2013. Annual precipitation was more than 5% below normal in each of these five years, with three consecutive years of 14% below normal in 2010, 2011 and 2012. There is some evidence that warmer or earlier springs may cause earlier autumn senescence because of the fixed lifespans of leaves or adversely affect plant productivity later in the season through the build-up of water deficits (Buermann et al., 2018; Liu et al., 2022). Shen et al. (2022) also found that the EOS tends to advance under higher T_max (Maximum Temperature), likely owing to confounding water limitation. Because earlier spring phenology may increase soil water loss in the early stages of the growing season, thereby increasing the prevalence of summer drought that may subsequently result in earlier leaf senescence (Piao et al., 2019). Li et al. (2023) also found that Spring phenology strongly governs vegetation growth and soil water consumption in the growing season. Vegetation phenology interacts with drought timing and thus determines subsequent vegetation growth. Furthermore, the interaction between spring and autumn phenological events is likely to modify phenological responses to ongoing climate warming (Piao et al., 2019). When the water deficit due to the advancement of SOS reaches the threshold of vegetation water or nutrient tolerance, EOS no longer postpones but starts to advance with the advancement of SOS (Shen et al., 2022). Possible underlying mechanisms for such interseasonal phenological correlations may be directly related to leaf traits and may be ascribed to the indirect effects of environmental factors as well (Piao et al., 2019). Because the SOS is involved in plants’ life rhythm and carbon storage, it could modify the matching between environmental conditions and life-cycle stages (Shen et al., 2022). However, the relative importance between biological factors and environmental factors on phenological responses to climate warming needs to be further investigated (Piao et al., 2019). Likewise, Xiong et al. (2023) found that spring phenology trends in northern hemisphere mid- and high-latitude vegetation showed an “early to late” reversal during the warming stagnation period, and the average rate of change around the turning point was small. The non-linear nature of this change invalidates the linear trend analysis often used in existing studies, leading to important differences in the analysis of spring phenology trends in the northern hemisphere mid- and high-latitude vegetation during the warming stagnation (Xiong et al., 2023). The timing of plant phenology events is determined by a variety of biological and environmental factors. However, the extent to which these factors influence plant phenology depends largely on the different developmental stages of phenological events and specific differences in plant life history strategies. Therefore, the relative importance of biological and environmental factors on the phenological response to climate warming requires further research in the future (Piao et al., 2019).

As seen in Figure 4, while neither SOS nor EOS showed a significant linear change, LOS maintained a significant linear growth trend during the study period. On the one hand, although there is advancement in EOS, the advancement trend of SOS is stronger and there is an enhanced influence of SOS on LOS. Piao et al. (2019) also found that the increase in LOS is mainly driven by the advancement of SOS in Eurasia. On the other hand, because SOS has not yet reached the photoperiodic threshold, there is still room for SOS advancement. It can be seen that LOS in Yunnan is mainly dominated by both SOS and EOS in the early stage and by SOS in the later stage. Additionally, although NDVI and TEMP are the main influences on LOS, PREC has a significant nonlinear effect.

Figure 7 shows that NDVI, SRAD, SOS, EOS, and LOS have mainly direct effects on NPP. ELEV has mainly indirect effects on NPP. PREC and TEMP have a mixture of direct and indirect effects on NPP, and the indirect effect occurs mainly due to the introduction of NDVI or/and phenological factors. As reported by Chen and Zhang (2023), the interaction of PREC and TEMP with other factors had a large effect on NPP. Moreover, in our study the effect of LOS on NPP was slightly lower than that of SOS-EOS on NPP, which differs from the results of most studies. LOS is jointly constrained by both SOS and EOS, and although LOS showed a significant and continuous increasing trend, the changes in SOS and EOS were more complex, as mentioned above (Figure 4). Therefore, SOS-EOS has a better explanatory power for the complex changes in NPP. Similar to our results, Yu et al. (2022) also found that SOS-EOS was more important than LOS. Although the delayed EOS led to a longer LOS, there was no increase in NPP due to the limited and reduced photosynthetic capacity coupled with an increase in respiration within the forest. Furthermore, besides SRAD being critical for phenology, Yang et al. (2017) found that the direct effect of solar radiation on NPP (0.31) was greater than precipitation (0.25) and temperature (0.07). Consistently, our findings also showed that the direct effect of SRAD (0.36) on NPP was greater than PREC (0.33) and TEMP (0.08). Notably, in the correlation analysis, the correlation between TEMP and NPP was much greater than that of SRAD. However, in the analysis of SEM, the direct effect of SRAD on NPP was much greater than that of TEMP. These results suggest that TEMP may not be the main direct limiting factor for vegetation NPP in Yunnan Province. Instead, the study found that SRAD may have a greater direct influence on NPP due to its contribution to photosynthesis. Similarly, the study of Chen and Zhang (2023) found that SRAD is one of the factors that has an important effect on NPP and should not be ignored. In addition, the relatively low effect of TEMP on NPP suggests that the direct effect of warming on NPP is relatively low. Nevertheless, other climate changes due to warming, like shifts in the timing, location, and intensity of extreme weather and precipitation, may considerably impact NPP (He et al., 2022).

Based on the SEM comparison (Figure 5), it was observed that the overall explanatory power of each climate factor on NPP remained unchanged (with PREC, TEMP, and SRAD exhibiting total influences on NPP of 0.597, 0.117, and 0.321, respectively). However, the introduction of phenology or NDVI altered the direct influence or explanatory power of the climate factors on NPP and resulted in the explanatory power of ELEV on NPP being almost completely dominated by indirect effects (Figure 7). Specifically, the direct influence of TEMP and PREC on NPP is weakened, but still greater than their indirect influence on NPP. On the other hand, the direct influence of SRAD is strengthened and becomes the dominant factor among the climate factors with direct influence on NPP. This result is mainly due to the fact that both TEMP and PREC influence NPP indirectly by affecting NDVI and phenology (Michaletz et al., 2014; Chu et al., 2016; Buermann et al., 2018; Piao et al., 2019). Among them, PREC affects NPP mainly by influencing the vegetation growth condition (NDVI) after SOS, because PREC has the largest influence on NDVI among the climatic and vegetation growth conditions (Figure 3). TEMP affects NPP mainly by changing LOS through EOS, because TEMP has the largest effect on EOS among the climatic and vegetation growth conditions (Figure 3). While SRAD directly affects SOS, the subsequent plant growth is mainly limited and affected by water, and the indirect effects of SRAD on NPP through phenology and NDVI have positive and negative effects that offset each other, so that the direct effects of SRAD on NPP are almost unaffected by the perturbation of phenology and NDVI. At the same time, because SRAD mainly affects SOS, and the effect of SOS on NPP is also largely dependent on subsequent PREC and NDVI, the indirect effect of SRAD on NPP is not significant. As SRAD is somewhat correlated with both TEMP and PREC (Figure 3), the weaker indirect effect of SRAD may have been included in the indirect effects of PREC and TEMP. Ultimately, we conclude that whether climate has an indirect effect on NPP is largely determined by the factors we have analyzed and the characteristics of the study area.