Mean residence time (MRT) of an element is the average time taken by an element from entry to exit in a system and can be estimated as the ratio of pool size and input flux at the steady state (Sierra et al.2017). MRTs of ecosystem carbon (C), nitrogen (N) or phosphorus (P) (hereafter τ _e,C, τ _e,N andτ _e,P, respectively) are key ecosystem properties influencing many ecosystem functions. For example,τ _e,C is critically important for ecosystem C sequestration potential and response of land C uptake to global change (Luo et al. 2017). Friend et al. (2014) found thatτ _e,C dominated the uncertainty in the modelled land C response to future climate change and atmospheric CO₂ concentration. τ _e,N andτ _e,P are closely related to nutrient use efficiency, which can be calculated as the product of net primary production (NPP) per unit nutrient and nutrient residence time (Berendse & Aerts 1987). Longer τ _e,N orτ _e,P contribute to more efficient N or P conservation within the ecosystem to support plant productivity under nutrient-limited conditions (Wang et al. 2018b). Understanding drivers of τ _e,N and τ _e,Pis helpful in both studying the biogeochemical cycles of N and P, and revealing mechanisms regulating τ _e,C.

Variations ofτ _e,C, τ _e,N andτ _e,P are complex and depend on multifaceted interplay among climate, vegetation, soil, terrain, land use and disturbance history. As an integrated outcome of multiple ecological processes (e.g., plant respiration, phenology, plant C allocation, plant mortality, organic matter decomposition and stabilization),τ _e,C can vary from a few decades in the tropical ecosystems to over 1000 years in wetlands at the high latitudes (Fanet al. 2020). Among various regulatory factors, how climate affects τ _e,C has been extensively studied at global scale, partly due to the relative availability of high resolution climate datasets. Mean annual temperature (MAT) was identified as the most dominant control on the spatial pattern of globalτ _e,C (Carvalhais et al. 2014; Fan et al. 2020). The relationship between τ _e,C (or the MRT of soil C) and MAT was found to be negatively linear or nonlinear, with a change in the slope of the negative relationship towards the high end of MAT (Figure 1 of Koven et al. (2017)) or around tropical/subtropical regions (Figure 7 of Fan et al. (2020)). Using extensive field observations in China, Wang et al. (2018a) identified both MAT and mean annual precipitation (MAP) as the dominant drivers of the spatial variation of τ _e,C. But Wang et al.(2018a) did not quantify the influences of land use and disturbance history, which were known to significantly influenceτ _e,C (Sanderman et al. 2017). At regional scales, vegetation, soil and terrain were also identified as important drivers of τ _e,C or components ofτ _e,C. For example, a study along a 4000-km natural transect in south America found that climate and soil geochemical properties co-dominated the spatial variation of soil C storage and turnover (Doetterl et al. 2015).

There are only limited studies focusing on τ _e,Nor τ _e,P especially at regional and global scales. In general, τ _e,P >τ _e,N > τ _e,Cbecause of the resorption by plants and immobilization by soil microbes (Wang et al. 2010). Given the greater fraction of total ecosystem C and nutrients in soil than in plants (Fan et al. 2020), the relatively narrow range of C﹕N ratio of soil organic matter and strong coupling of C and N cycles, spatial variations and key drivers ofτ _e,C and τ _e,N were expected to be similar (Post et al. 1982; Post et al.1985). For τ _e,P , the dominant drivers are likely to be different because soil P is also strongly influenced by soil age and geochemical properties (Walker & Syers 1976). Indeed, a recent study by He et al. (2021) found that soil P concentration down to 1 m soil depth was strongly influenced by parental material, MAT and soil texture globally, and also by topography at regional scale. A recent study also found that the spatial patterns of ecosystem total N and P down to 1 m soil depth were quite different, and that the younger and slighter-weathered temperate soils had lower N﹕P ratio than the subtropical and tropical soils in China (Zhang et al. 2021). So far no studies have been on the dominant drivers of the regional or global variations of both τ _e,N andτ _e,P.

Lack of reliable observations across a wide range of climate-, vegetation-, soil- and terrain- related factors also limits the quantifications of the patterns and dominant drivers ofτ _e,C, τ _e,N andτ _e,P. Earlier studies either focused on one category of drivers (Carvalhais et al. 2014) or individual residence time only. Even for the relatively well studiedτ _e,C, previous studies were subject to large uncertainties with limited observations, such as poor information about disturbance history (Wang et al. 2018b; Chen et al. 2020). Estimate of MRT as the ratio of pool and input or output flux may have large errors if the system is not at steady state (Lu et al.2018), which is true for most recently disturbed ecosystems.

In this study, we compiled the measurements of C, N and P pools of plant biomass, surface litter and soil down to 1m depth, and site-specific 30 variables related to climate, vegetation, soil and terrain (see Methods) of 127 mature and undisturbed natural forests in China. These intact forests are fundamental in understanding ecological processes but have been in sharp decline and are becoming rarer (Potapov et al.2017), which reflects the importance of this work. These forests span from 22.24°N to 52.42°N, cover a wide range of climate and soil conditions and represent all major forest types in China (see Methods). This dataset allow us to thoroughly address the following questions: (1) how climate, vegetation, soil and terrain influenceτ _e,C, τ _e,N andτ _e,P? (2) whether a threshold exists in the response of τ _e,C, τ _e,N andτ _e,P to air temperature? If so, what is the likely mechanism?