2.1 Site descriptions and climate

The study sites are located within the Xishuangbanna National Nature Reserve, Yunnan, Southwest China. We selected five permanent tropical forest plots, each representing different successional statuses and forest types (Table 1 and Supporting Information: Table S7). TKF exhibited the highest tree density, whereas TRF had the highest number of tree species and the greatest canopy height (Table 1 and Supporting Information: Table S7). The basal area was highest in TKF and lowest in TSF-2. Among the top 10 dominant species in each plot, TSF-3 and TSF-2 had the highest number of pioneer species (Supporting Information: Table S7). Human disturbances in the studied forest plots are minimal, primarily involving light interventions by local communities for the collection of wild vegetables and tourism activities. Moreover, there is no artificial felling of forest trees within the five forest plots.

The Xishuangbanna region is characterized by a typical tropical monsoon climate with distinct dry and wet seasons. The dry season lasts from November to April, while the wet season occurs from May to October. The wet season is marked by heavy summer monsoon rainfall, which accounts for over 80% of the total annual precipitation (Cao et al., 2006; Supporting Information: Figure S2). Interannual variations in mean annual temperature are mild, ranging between 21.6°C and 23°C. In contrast, total annual precipitation varies greatly from year to year, ranging between 947 mm and 1799 mm from 2004 to 2020 (Figure 1e). The annual precipitation and relative humidity have been decreasing significantly, while maximum temperatures have shown a significant increase in recent decades (Figure 1a–c). In 2010, a historical drought event occurred in the study area (Figure 1d,f) (Li et al., 2019).

2.2 Forest inventory survey

A series of forest inventory surveys were conducted within the five forest plots from 2004 to 2020. Surveys were conducted annually before 2010, with a transition to a 5-year interval implemented thereafter. During each survey, all individual trees with a minimum DBH (diameter at breast height at 1.3 m above ground) of at least 2 cm were assessed. The survey recorded species names, tree height, and DBH, assigned unique identification tags to newly recruited individuals, and noted any dead trees.

2.3 Environmental conditions

All the tropical forest plots studied in this research share the same climatic conditions as they are located near each other within the same valley (within ~15 km distance). Therefore, we used the same climate data from the Xishuangbanna Station for Tropical Rain Forest Ecosystem Studies (XSTRFES) for all plots. The climate station of XSTRFES is situated within the Xishuangbanna Tropical Botanical Garden (21°54′ N, 101°46′ E, 580 m a.s.l.), and all five forest plots are within 10 km of this climate station. Soil water content (SWC) across the plots varies due to differences in topo-geology and forest successional history. SWC was measured at a depth of 20 cm using gravimetric methods. SWC measurements have been conducted every 2 months (Jan, Mar, May, Jul, Sep, and Nov) each year since 2006 by staff from XSTRFES. However, SWC data for the tropical secondary forest-3 (TSF-3) plot only covers 2017–2020; hence, we excluded SWC data from this plot for further analysis. During the study periods, the tropical secondary forest-1 (TSF-1) and tropical secondary forest-2 (TSF-2) plots showed higher SWC than other plots (Supporting Information: Figure S1). Monthly climate and SWC data were averaged for further analysis according to the yearly forest inventory survey period (census interval) rather than on an annual basis.

2.4 Demographic rates

The calculation involved stem diameter measurements at two censuses, denoted as DBH_t (diameter at the second census) and DBH_i (diameter at the first census). The difference was then divided by the time interval (∆t).

2.5 Net primary productivity (NPP_stem)

2.6 Statistical analyses

We considered all the assumptions of GLMMs and LMMs (Zuur et al., 2010), all the variables were scaled to converge models, and the standard estimates of effects were compared. Furthermore, before constructing LMMs and GLMMs, we checked for multicollinearity of environmental variables using the R-package “car” (Fox et al., 2007). Multicollinearity was determined by evaluating the Variance Inflation Factor (VIF). All of our statistical models (LMMs and GLMMs) showed VIF values ≤ 5.0 for Tmax, Precipitation, and SWC.

3.1 Demographics rates and NPP_stem

Secondary forest plots in early successional stages showed faster growth rates compared to late-successional and climax forest plots. The highest diameter growth rate (DGR, mm yr⁻¹) was observed in TSF-3, while the lowest DGR occurred in TKF (Figure 2a). Similarly, relative mortality rates (RMR, % yr⁻¹) were higher in early successional secondary forest plots (TSF-2 and TSF-3) than in late successional (TSF-1) and primary forests plots (TKF and TRF) (Figure 2b). TSF-3 showed the highest relative recruitment rates (RRR, % yr⁻¹), whereas TKF exhibited the lowest RRR among the five forest plots (Figure 2c). The highest stem net primary productivity (NPP_stem) was observed in TSF-3 (2.218 Mg C ha⁻¹ yr⁻¹), with lowest NPP_stem in TSF-2 (0.874 Mg C ha⁻¹ yr⁻¹) (Figure 2d).

The overall temporal dynamics of NPP_stem (Mg C ha⁻¹ yr⁻¹) and demographic rates revealed notable adverse effects of drought during 2009–2010 (Figure 3a–d). We recorded the lowest diameter growth rate, as well as lower relative recruitment rate and NPP_stem for most tropical forest plots during this period (Figure 3a,c,d). Similarly, relative mortality rates in 2010 were distinctly higher for early successional secondary forest plots compared to primary forest plots (Figure 3b), indicating that early successional secondary forests are more vulnerable to extreme drought and warming climates. LMMs showed significant declining trends in diameter growth rate and relative recruitment rate across the five tropical forest plots over the past 16 years (Supporting Information: Table S6). NPP_stem also showed a declining trend, although the trend was marginally significant (Supporting Information: Table S6). Conversely, we observed a marginally significant increasing trend over time for relative mortality rate (Supporting Information: Table S6).

3.2 Effects of environmental variables on forest demography and productivity

We observed a significant negative effect of maximum temperature (Tmax) and a positive effect of precipitation on DGR (mm yr⁻¹) across all five forest plots (Figure 4 and Supporting Information: Table S1). GLMMs showed a significant positive effect of Tmax on tree mortality in all five forest plots (Figure 4 and Supporting Information: Table S2), while the negative effect of precipitation on tree mortality was only significant for the primary forest-TRF and late successional secondary forests (TSF-1; Figure 4 and Supporting Information: Table S2). Tree recruitment was mainly driven by precipitation except for the TSF 3 plot (Figure 4 and Supporting Information: Table S3). However, Tmax showed significantly a negative effect on tree recruitment in TRF and TSF-1, and a positive effect in TSF-3 (Figure 4 and Supporting Information: Table S3). Soil water content (SWC) was significantly and negatively linked to DGR (mm yr⁻¹) in TKF, TRF, TSF-1, and TSF-2 (Figure 4 and Supporting Information: Table S1). SWC also showed a negative effect on tree mortality in TKF, TRF, and TSF-2, but a positive effect in TSF-1 (Figure 4 and Supporting Information: Table S2). Furthermore, SWC was negatively associated solely with tree recruitment in TSF-2 (Figure 4 and Supporting Information: Table S3).

Results from the linear mixed-effects model further indicated significant negative effects of Tmax on NPP_stem across all five tropical plots (Figure 5a and Supporting Information: Table S4). However, the effects of SWC and precipitation on NPP_stem were not statistically significant.

4.1 Long-term variation in forest demographic rates and productivity

We observed that early successional secondary forest plots (TSF-2 and TSF-3) typically showed higher diameter growth rates than primary forest plots (TKF and TRF) and the late successional secondary forest plot (TSF-1) (Figure 2a). Generally, the early successional vegetation in tropical forests shows rapid growth rates due to the availability of soil nutrients and light competition (Matsuo et al., 2024). In addition, secondary forests in tropical regions experience rapid initial growth dominated by pioneer species (Supporting Information: Table S7), leading to higher short-term growth rates compared to primary forests (Van Breugel et al., 2007). However, as these forests progress through successional stages, growth rates may decrease due to factors such as increased competition and resource-use efficiency (Küppers, 1992). Similarly, early successional forest plots also had higher relative mortality rates (TSF-2 and TSF-3) (Figure 2b). The mortality rate in secondary forests generally depends on their competitive strategy, as pioneer tree species in secondary forests tend to invest more in rapid growth but are less resilient to environmental stress (Brown & Lugo, 1990; Guariguata & Ostertag, 2001). The TSF-3 plot featured the highest relative recruitment rate among the studied forest plots (Figure 2c). The higher relative recruitment rate in TSF-3 is likely related to its high relative mortality rate, as low survival is commonly associated with high recruitment (Kambach et al., 2022). Generally, tree mortality shapes forest dynamics by influencing recruitment processes (Lutz et al., 2014). Mortality reduces competition among established trees, making more resources such as light, water, and nutrients available for regeneration (Gessler et al., 2017). When trees die, they create gaps in the canopy, allowing increased sunlight penetration to the forest floor, which stimulates the growth of understory plants and new tree establishment (Pedersen & Howard, 2004). Additionally, decomposing trees release nutrients into the soil, enhancing fertility and supporting the growth of young trees (Ludovici & Kress, 2006). Furthermore, the results revealed that among the early successional forest plots, TSF-3 exhibited the highest NPP_stem, whereas TSF-2, within the same successional stage, displayed the lowest NPP_stem (Figure 2d). Secondary forests, undergoing regeneration after disturbances, have been recognized as significant for carbon sequestration (Heinrich et al., 2021; Kenzo et al., 2010; Medlyn et al., 2000; Taylor et al., 2014), which aligns with the highest NPP_stem observed in TSF-3. TSF-2 consists of a lower proportion of bigger trees (Supporting Information: Figure S4), thus contributing to lower basal area (Table 1), which may result in lower biomass (Supporting Information: Figure S3) and lower productivity (Figure 2d).

Our study also detected a decline in diameter growth rates during the extreme drought period (2009–2010 census) across the studied forest plots, except TSF-3 (Figure 3a). In 2010, an extreme drought event occurred in Southwest China, leading to a significant reduction in vegetation growth (Li et al., 2019). The period after 2010, characterized by several extremely dry and hot years (Figures 1d–f), resulted in lower diameter growth rates compared to the pre-drought period, especially in primary and late successional secondary forest plots (Figure 3a). This suggests that early successional secondary forest plots (TSF-2 and TSF-3) can recover faster after drought than other plots. We also found that during the hot-dry year of 2010, both TSF-2 and TSF-3 showed higher relative mortality rates compared to other forest plots (Figure 3b), suggesting that trees in early successional forest types are more susceptible to drought. This susceptibility may result from the faster growth rate characteristic of TSF-2 and TSF-3, which is often associated with lower wood density trees, rendering them more vulnerable to drought conditions (Phillips et al., 2010). A uniform decline in relative recruitment rates was observed during the hot-dry period of the 2009–2010 census (Figure 3c). Similarly, NPP_stem also declined during that hot-dry period across all forest plots, except TSF-2 (Figure 3d). Thus, the carbon sink of tropical forest ecosystems might be greatly impacted by climate warming and the increasing frequency of extreme drought events (Frank et al., 2015).

4.2 Environmental influences on tropical tree demography and productivity

Our investigation revealed that diameter growth rates were negatively affected by Tmax and positively influenced by precipitation, indicating that the diameter growth rates among studied forest plots were primarily moisture-limited (Figure 4). This finding is consistent with other studies highlighting the adverse impact of high temperatures and drought on tropical forest growth (Altman et al., 2017; Hájek et al., 2021; Liang et al., 2023; Walker & Johnstone, 2014; Zuidema et al., 2022), as well as the significant positive association between precipitation and tree growth (Brienen & Zuidema, 2005; Wagner et al., 2014). High temperatures pose a significant threat to plant growth primarily by inducing heat stress (Wahid et al., 2007), which can reduce growth rates by disrupting photosynthesis and elevating respiration rates (Bita & Gerats, 2013). Additionally, higher temperatures exacerbate water stress and impede nutrient uptake and metabolism (Krasensky & Jonak, 2012). We expect that the growth rate of trees in tropical forests is likely to decrease due to climate warming and frequent extreme drought. Indeed, our study showed a significant decline in the 16-year trend of diameter growth rates across our five forest plots (Supporting Information: Table S6). Therefore, assuming the continuation of current climate trends (Figure 1a–c), it is expected that the growth of tropical trees will decline in the coming years.

Furthermore, we found that Tmax is a substantial driver impacting tree mortality across all studied plots (Figure 4). Previous studies found that rising temperatures lead to increased mortality rates in moist tropical forests (Adams et al., 2009; McDowell et al., 2018). Park Williams et al. (2013) and Vlam et al. (2014) further highlighted the negative effect of temperature on tree growth and forest vigor, supporting the link between high temperatures and tree mortality. These universal relationships underscore the importance of temperature dynamics in understanding and predicting tree mortality patterns. Additionally, precipitation showed a negative effect on tree mortality in TRF and TSF-1. Consistent with the effect of precipitation, SWC also showed a significant negative effect on tree mortality in TKF, TRF, and TSF-2 (Figure 4). These results align with previous studies indicating that the probability of mortality increases substantially with decreasing soil water availability (Caspersen & Kobe, 2001). However, TSF-1 exhibited a notable positive effect of SWC on tree mortality (Figure 4). This plot also had the highest soil water content among the five forest plots (Supporting Information: Figure S1), suggesting that soil water deficit might not be a limiting factor for tree mortality in this plot. Instead, the impact of Tmax on tree mortality is more evident (Figure 4). We anticipate a substantial rise in tree mortality within these forest plots in the future, with early successional secondary forests likely to experience the greatest impact.

The results revealed that higher precipitation can enhance the likelihood of tree recruitment in all forest plots except for TSF-3 (Figure 4). These findings are consistent with previous studies that emphasize the key role of precipitation in natural regeneration processes. For instance, precipitation has been shown to enhance seed availability and establishment (Vieira & Scariot, 2006), and increase species diversity (Rito et al., 2017). Furthermore, we observed a significant negative relationship between Tmax and tree recruitment in TRF and TSF-1 forest plots (Figure 4). Consistent with our results, previous studies also found that higher temperatures were associated with decreased growth and recruitment rates in tropical trees (Rehm & Feeley, 2015; Vlam et al., 2014). An unexpected positive relationship between Tmax and tree recruitment was observed in the TSF-3 plot (Figure 4). We propose that higher recruitment in TSF-3 is associated with its higher mortality driven by high T_max, which further increases recruitment due to more open canopy and higher light availability for understory seedlings. The results also revealed that SWC had an adverse effect on tree recruitment in TSF-2 (Figure 4). We suggest that relying solely on soil water content as an edaphic factor may not adequately to explain tree recruitment because the recruitment of tropical tree species may be more closely associated with soil nutrients rather than soil water content alone (Holste & Kobe, 2017). Generally, tree recruitment is intricately tied to competitive interactions, seed dispersal mechanisms, soil nutrients, and light availability (Davis et al., 1999; Ibáñez & McCarthy-Neumann, 2014; Kraaij & Ward, 2006).

We observed a strong inverse relationship between T_max and NPP_stem, highlighting temperature extremes as a critical issue for carbon sink capacity in the study area (Figure 5a). While Linger et al. (2020) observed a positive correlation between seasonal precipitation and NPP in tropical seasonal rainforests (TRF), our study did not detect a significant effect of precipitation on NPP_stem. This discrepancy could be attributed to the fact that our analysis did not account for seasonal differentiation, potentially weakening the signal of precipitation on NPP_stem. Pau et al. (2018) proved that Gross Primary Productivity (GPP) was highly sensitive to temperature, which might ultimately impact NPP through its effects on photosynthesis and respiration. Additionally, previous studies have reported temperature as a critical predictor for NPP (Churkina et al., 1999; Ouyang et al., 2014; Reich et al., 2014; Vilanova et al., 2018), which were consistent with our findings. Based on the observed climate changes within the study area (Figure 1a–c), we anticipate a reduction in the carbon sink capacity of tropical forests in the future. Early successional forests are expected to face greater risks due to higher mortality rates (Figure 2b), which will impact net primary productivity (Searle et al., 2022). Our study focused solely on NPP_stem and did not account for litterfall production. The carbon allocated to wood can be sustained longer than carbon allocated to leaves or reproductive tissues, thus making NPP_stem the most important component of total forest NPP (Carvalhais et al., 2014).

4.3 Implications for forest conservation and restoration

The investigation revealed that T_max exerts a substantial negative influence on tree growth and productivity, and a positive influence on tree mortality across five forest plots, with early successional forests exhibiting the highest susceptibility due to their higher mortality rates. Furthermore, secondary forest plots generally showed high carbon sink capacity (Figure 2d) and have great potential for biodiversity conservation, climate change mitigation, and landscape restoration (Poorter et al., 2021). We anticipate that the prevalence of frequent drought events and rising temperatures in the study area will exacerbate challenges, leading to reduced tree growth, decreased recruitment, and increased mortality, ultimately impacting carbon accumulation within forest ecosystems. We advocate for effective conservation strategies that prioritize enhancing ecosystem restoration in secondary forests through measures such as promoting climate-tolerant species diversity and restoring natural hydrological regimes. These secondary forests are expanding rapidly in areas across the globe, driven by both anthropogenic activities and natural processes (Bongers et al., 2015). Leveraging the adaptive capacities of secondary forests while conserving the unique characteristics of primary forests is essential for fostering the restoration of forest landscapes amidst ongoing climate change.