Material and Methods

Study sites along elevations in 10 mountains

This study was conducted in 10 mountains of the BEST (Biodiversity along Elevational gradients: Shifts and Transitions research network (https://BEST-mountains.org). This network was designed to use standard field methods (e.g., plot design, plant survey, and sample collection) to demonstrate biogeographic patterns and ecological processes across regions. The 10 mountains belong to subtropical and tropical regions (Fig. 1a) where the climate is heavily buffered by dense canopies compared to temperate and boreal forests (De Frenne et al. 2019). The geographical distribution of the studied region spanned from 98˚42′ – 119˚26′ in longitude and 19˚4′ – 31˚9′ in latitude. Elevation ranges from 252 to 3835 m a.s.l.. The mean soil temperature of the growth season (June to August) ranges from ca . 13 ℃ – 22 ℃, with the coolest climate in Jade Dragon Snow Mountain (YMT) and the warmest climate in the tropical mountain Bawangling (BWL). Along elevational gradients, species composition often shifts sequentially, i.e., evergreen-broadleaved species, mixed evergreen-broadleaved and deciduous species, deciduous and conifer species, and high mountain shrubs. In each mountain, 20 m × 20 m permanent sample plots were established from the mountain bottom to the top at a distance ofca . 100 m difference. In total, 142 plots were included in our study (Table S1). Within each plot, all tree individuals with stems > 1 cm were identified to species level. Tree species richness (using “Tree” when referred as environmental factor) was calculated and used as the biotic factor in decomposition processes.

Standardized teabag experiment

We used the tea bag index (TBI) to investigate litter decomposition processes along the elevation of each mountain by following the protocol from Keuskamp et al. (2013). In this approach, two types of teabags (green tea: EAN8722700055525 and rooibos tea: EAN8722700188438) were used as standard leaf litter bags, which can be used globally and across biomes to generate comparable results (Keuskamp et al. 2013, Djukic et al. 2018). The material of the teabag is made of polypropylene and has a mesh size of 0.25 mm allowing the access of microfauna, microbes, and very fine roots (Fig 1c, d).
Within each plot, we selected four sampling sites according to the criteria below: 1) understory species composition at each site is representative and similar to neighbor sites; 2) homogeneous in microhabitat between sites, e.g., slope and canopy cover; 3) no clear rocks, big tree roots, anthropogenic disturbances; 4) four sites were as evenly distributed as possible within a plot. Each site was labeled with a PVC tube (Fig. 1b). Before the start of the field experiment, all teabags were oven-dried at 70 °C for 24 h. Each teabag was identified and buried in the upper 8 cm of the top soil layer for three months. At each site, we buried 12 teabags (6 red vs. 6 green). In total, 6,864 teabags were buried (6 replications × 2 tea types × 4 sites × 142 plots), and 87% (5,996) of the bags were retrieved. With intensive cooperation from each region, this field experiment was conducted nearly simultaneously across all mountains, buried between late May and early June 2021 and retrieved between late August and early September 2021. This decomposing period was designed to capture energy flow in peak growing season in subtropical and tropical forests.
At the end of the decomposition period, we retrieved all tea bags and transported them immediately in ice-isolated boxes to the Lab at East China Normal University, Shanghai. During the process of retrieval and transportation, sterilized gloves and sampling bags were used to avoid any contamination. Soil particles adhered on the surface of the litter bag were removed once received and dried at 70 ℃ for 48 h. Dried litter bags were again cleaned by hand carefully to avoid ash falls. Then we recorded the extent of damage to each tea bag according to the number and size of holes. Undamaged bags with one or several minor holes (< 1 mm) were used for weighing the remaining materials. According to the decomposition protocol, we calculated TBI, which includes the decomposition rate constant (\(k\)) and stabilization factor (S ). In short, mass loss (\(W\)) at time t is a double-exponential function of decomposed fraction (\(a\)) with the constant k (eqn. 1). The k represents how fast the labile carbon is about to be decomposed under a certain environment, reflecting the velocity of carbon loss. S indicates the fraction of the recalcitrant which stabilized from a theoretically hydrolysable fraction (\(H\)) due to environmental constraints (eqn. 2):
\(W\left(t\right)=\text{ae}^{-kt}+\left(1-a\right)\ \) eqn. 1
\(S=1-\frac{a}{H}\) eqn. 2
Meanwhile, we measured three microhabitat factors (Cover: canopy cover, Thick: ground litter thickness, and slope) which can potentially affect decomposition directly and indirectly. Canopy cover was measured at 50 cm above the soil surface during the sunny day (avoid solar radiation at noon) at each site three times by using a fish-lance (238˚, wide-angle view). Canopy pictures were analyzed using the software Gap Light Analyzer (Frazer et al. 1999). Ground litter thickness was the layer of dead leaves and debris that covered the soil surface. The slope of each site was measured with consistent orientation, i.e., standing the downhill side and facing the mountain slope.

Microclimatic and edaphic variables

Soil microclimate (temperature and humidity) was recorded at the centre of each plot at a depth of 8 cm relative to the soil surface at 15-minute intervals using temperature and moisture loggers (TMS4, TOMST Ltd.). Due to the high risk of monkey disturbance in EMS, the microclimate in this mountain was recorded using iButton (Maxim Integrated DS1925), which can fully be buried under the soil. Soil microclimate indices of the three months included 8 temperature indices and 7 moisture indices referring to the bioclimatic variables in WorldClim or CHELSA. Eventually, four microclimatic indices were retained after controlling the collinearity. These four indices reflect both the mean and variation of soil microclimate matching with decomposing period (growing season), including the monthly mean soil temperature (Temp; ℃), standard deviation of soil temperature (TempV), the monthly mean soil moisture (Mois), and standard deviation of soil moisture (MoisV).
Soil samples were collected at 0 – 10 cm using 5-cm augers. In each plot, five soil cores were taken randomly and mixed homogeneously, after the removal of visible roots, debris, and stones. Soil samples were air-dried and sieved through a 1-mm mesh. We measured soil pH and total phosphorus (P) following the standard protocol of Ma et al. (2019). We focused on the two soil factors because: 1) soil pH plays a major role in the structure of microbial community composition globally locally (Hendershot et al. 2017, Ma et al. 2022); 2) the large variation of soil total P across the studied regions implies its potential constraints on ecosystem processes (He et al. 2016).

Statistical analyses

We first tested the elevational pattern of decomposition rate (k ) and stabilization (S ) within each mountain using least square regression followed by model diagnostics of residuals normality and homoscedasticity using the function autoplot from the R packageggfortify (Tang et al. 2016). Before the test, we scaled the elevation of each mountain to 0 – 1 to avoid scale differences in elevation ranges between mountains. The regression was also used to examine the correlation between k and S within each mountain. The same correlation method was used for testing the correlation significance of environmental drivers (Fig. S1). Spearman correlation coefficients (< |0.7|) were used as a threshold to identify multicollinearity (Dormann et al. 2013) when multiple regression analyses were applied.
To quantify the dominant role (w ) of soil microclimate in decomposition processes, we applied all possible combinations of the ten environmental variables and fitted these component models with the least square regression for k and S in each mountain separately, and calculated the relative importance value (w ) as the weight of Akaike’s Information Criterion (AIC) based on model averaging. For each mountain and the response variable, we extracted the best model which contained the lowest AIC and was used to indicate the significant drivers in each mountain. To avoid multicollinearity, we calculated VIF values and refit the best model by excluding variables that had a VIF value > 4. This operation only occurred in the three best models (Temp in BWL when predicting k , Temp and TempV in TMS when predicting S , and TempV and Cover in EMS when predictingS ). As the extent of the significance matched well with the value of the weight (Table S2), we used the latter to represent the relative importance because full model averaging optimizes the uncertainties and biases from threshold-selected models (Burnham and Anderson 2002). Since the data of Tree, edaphic factors, Mois, and MoisV were unavailable in EMS, we removed EMS and focused on 9 mountains when comparing the relative importance of each factor in decomposition.
Finally, we assessed the specific effect of the four soil microclimatic factors and three non-climatic factors on k and S within each mountain and across all mountains. The three factors were tree diversity, soil pH, and slope, which were representative in terms of its profound direct and indirect influences on decomposition via litter quality, decomposers, and microhabitat, respectively. We used least square regression models and a linear mixed-effects models (mountain as a random effect) to test each factor’s influence on k and S within each mountain and across all mountains. The mixed-effects model was estimated using the lme4 package (Bates et al. 2015), and the values of explained variance (R 2) of the models were calculated using the function r.squaredGLMM in R packageMuMin (Bartoń 2023).
All statistical analyses were conducted in R 4.3.1 (R Core Team 2014).