τ _e,C, τ _e,N andτ _e,P are important ecosystem properties, and are subject of extensive theoretical (Sierra et al. 2017; Lu et al. 2018) and observation-based studies (Carvalhais et al. 2014; He et al. 2016). Previous studies identified the spatial pattern and its dominant control of the individual MRT in different compartments of the terrestrial ecosystems (Helfenstein et al. 2020; Shiet al. 2020). Our study based on extensive field observations across 127 mature forests in China assessed how climate, vegetation, soil and terrain affected τ _e,C,τ _e,N and τ _e,P through their direct effects and interactions. Our study identified not only some patterns similar to previous studies but also some new patterns that are seemingly inconsistent with the existing theory of terrestrial biogeochemistry.

Climate is generally considered to exert a dominant control on terrestrial C dynamics (e.g., Carvalhais et al. (2014), Chenet al. (2021)). Consistent with the previous studies, this study found that climate variables together not only explained the largest portion of variance in the estimated τ _e,C, but also for τ _e,N and τ _e,P. In addition, we found that more than half of the climatic influences (τ _e,C: 52%; τ _e,N: 75%;τ _e,P: 57%) resulted from the interactions between climate and soil, vegetation, or terrain. Different fromτ _e,C and τ _e,N, the influences of soil (23%) and terrain (21%) onτ _e,P were comparable with that of climate (28%). This pattern is consistent with our understanding of the biogeochemical cycles of C, N and P. In undisturbed natural ecosystems, MRT of soil C is jointly controlled by the C input (both quality and quantity) and stabilization (Chen et al. 2021), whereas N input is dominated by N fixation in many undisturbed forests, and rate of N fixation is strongly controlled by available C (Wang et al. 2007), therefore vegetation plays an important role in τ _e,C andτ _e,N (see Fig. 3). Unlikeτ _e,C and τ _e,N, soil and terrain had much greater influences on τ _e,P than vegetation, as these two factors were related to soil weathering status, and an increasing fraction of soil P became occluded during pedogenesis (Walker & Syers 1976).

The breakpoint in the relationships between τ _e,Pand T _min further reinforces the significant role of P limitation on ecosystem dynamics. The breakpoint around 0 ℃ in our study coincided with the direction of changes of external P input withT _min (see Fig.4c). Our data indicates, in the mature forests located in the subtropical regions withT _min > 0 ℃ (see Fig. 1), vegetation evolved to make use of increasingly favourable water and temperature conditions (see Figure S1b-c) while P input decreased withT _min, for example, through increasing ecosystem P use efficiency (PUE, defined as the ratio of NPP and P uptake by plants) (see Fig. 6a, before vs. after breakpoints) and increasing P reportion and uptake (see Fig. 6b-c, before vs. after breakpoints). Previous studies reported different strategies of vegetation to maintain high productivity in mature forest ecosystem from T _min> 0 ℃ regions under P-limitation (most subtropical and tropical area) (Turner et al. 2018; Mo et al. 2019). For example, some plant can maintain high productivity in low P environment through reducing P demand for metabolism (Mo et al. 2019), replacing phospholipids with non-phospholipds during leaf development (Lambers et al. 2012), and absorbing soil organic P compounds (Turner 2008). Under P-limited conditions, through long-term natural selection and adaptation, vegetation evolved to minimize P losses as indicated by increasing τ _e,P with decreasing external P inputs (Vitousek et al. 2010). As a considerable portion of plant usable P in soil is bonded in organic matter, the conservation of P in ecosystems also means the retention of C and N because of the narrow range of variation of C﹕N﹕P ratio of soil organic matter (Tipping et al. 2016). Our study revealed the similar breakpoint pattern in τ _e,C andτ _e,N, despite the regressions are not statistically significant for τ _e,N (p> 0.05).

By analysing the direct and indirect (through interactions) effects of different climate variables that may contribute to the segmentation patterns, we confirmed that interactions among climate variables were not the primary drivers. Precipitation was identified as another important factor influencing τ _e,C by Carvalhaiset al. (2014). As shown in Figure S1b, precipitation increased linearly with T _min among the 127 forest sites, and correlations between precipitation and τ _e,C,τ _e,N and τ _e,P were not significant when T _min > 0 ℃ (see Fig. 5), therefore increase in precipitation withT _min will unlikely explain the estimated increase in τ _e,C with T _min whenT _min > 0 ℃. This is consistent with the result of Kramer and Chadwick (2018).

Instead, the interplay among climate, vegetation, and soil (geochemical factors) are important in shaping the segmentation patterns identified in this study. Consistent with previous studies, we found faster turnover rates of C, N or P under higher temperature in cold regions (T _min < 0 ℃) where climate was the primary control. High temperature favours accelerated biological activities, which increased plant turnover, soil organic matter decomposition and nutrient transfers (Leiros et al. 1999; Conantet al. 2011; Larsen et al. 2011; Carvalhais et al.2014; Bloom et al. 2016). In the region withT _min > 0 ℃ where subtropical forests are located, soils generally are highly weathered, and geochemical constraints (parent material and soil status) significantly influenced the ecosystem response to climate (Vitousek et al. 2010). We showed that P limitation and vegetation’s adaptation to the low P environment were most plausible mechanisms driven the change in the relationship between temperature and τ _e,C andτ _e,P between T _min< 0 ℃ and T _min > 0 ℃. In addition to P, we also found a high soil clay content and an increasing trend of soil clay content with T _min whenT _min > 0 ℃ (See Figure S3). Clay content is an important indicator of mineral protection organic matter that favours soil C retention (Kindler et al. 2011; Hemingwayet al. 2019; Chen et al. 2021), especially in forest soil (Six et al. 2002). This high clay content, associated with the strong weathering of soils, likely contributed significantly to the segmented relationship between T _min andτ _e,C and τ _e,P. It also is possible that other soil properties, such as metal oxide concentration and clay mineralogy also played a significant role in the positive relationship between τ _e,C andτ _e,P and T _min, whereasT _min didn’t directly affectτ _e,C and τ _e,P, but their variations were strongly correlated with other physical and geochemical properties in the subtropical and tropical regions (Yu et al.2019; Chen et al. 2021).

The segmented relationship suggests that the responses ofτ _e,C and τ _e,P to future warming may be significantly different between the subtropical forests and temperate forests in China. As clay content and other geochemical properties that are important to P cycle will not change significantly under warming from decades to century, the direct responses ofτ _e,C, τ _e,N andτ _e,P to future warming in the subtropical regions of China will be quite small, whereas τ _e,C,τ _e,N and τ _e,P of temperate forests are very sensitive to warming at a rate of -1 year/℃ forτ _e,C , -23 year/℃ for τ _e,Nand -216 year/℃ for τ _e,P (see Fig. 4g-i). Global models used for simulating terrestrial biogeochemical cycles under future climate change often applied one temperature-dependence function of soil organic C decomposition globally (Exbrayat et al. 2013), these models may significantly overestimate the temperature sensitivity and underestimate the soil C accumulation under future warming in the subtropical and tropical forests in China. With increasing evidence to support the significant role of soil geochemical properties on soil C stabilization (Kramer & Chadwick 2018; Basile-Doelsch et al. 2020), it is important to include the dependence of C or nutrient stabilization on soil geochemical in global land models for modelling soil C, N and P cycles (Rasmussen et al. 2018; Wang & Goll 2021).