## Introduction

Terrestrial carbon (C) dynamics primarily depend on plant carbon economics1,2,3. Plants allocate carbon to their above- and belowground organs, often studied as a ratio between root and shoot biomass (R/S)4. In general, the R/S reflects an optimal allocation of resources by plants and is a crucial parameter for estimating terrestrial C storage5. When global change factors induce alteration of resource supply for plants, their optimal allocation patterns may shift, resulting in deviations in R/S6. For example, drought and elevated CO2 generally enhance the biomass allocation of roots over shoots for acquiring water and nutrients7, whereas increased precipitation and nitrogen deposition usually elevate the biomass allocation to shoots for an increased light competition8,9. Climate warming has also been shown to affect net primary production in terrestrial ecosystems10,11,12, such as by increasing biomass production. If this triggers soil nutrient deficits, plant carbon economics are likely to change13,14,15. Accordingly, warming effects on R/S may depend on temperature effects on resource supply, which makes it challenging to predict warming effects on plants’ biomass allocations, particularly across different environmental conditions16,17.

The antecedent climatic conditions (e.g., mean annual temperature and precipitation) of a given area are often considered as the significant predictors of R/S variation across terrestrial plants4. However, whether the long-term correlation between R/S and climatic conditions conform to immediate R/S response to warmed environments remains unknown18,19. Apart from local climate, belowground properties of ecosystems (e.g., nutrients, root traits, and root-soil interface) can further influence the response of R/S to warming12,20,21. For instance, the mutualistic symbiosis between root and mycorrhizal fungi (MF), which is universally present in terrestrial ecosystems22, might affect patterns of plant carbon allocation depending on different nutrient foraging strategies of mycorrhizal root systems15,23,24. Plants with the same mycorrhizal association might respond in (ecologically) similar ways to global change factors, especially to alteration of resource availability in soils25. As two major types of mycorrhizas, arbuscular mycorrhizal fungi (AMF)’s hyphae penetrate the cell wall and invaginate the cell membrane within the host plant root, whereas the hyphae of ectomycorrhizal fungi (EMF) do not penetrate individual root cell. Moreover, AMF can increase the host-plant’s uptake of inorganic nutrients, while EMF provides their host plants with greater access to soil organic nutrient pools26,27. Studies have shown that the dominance of a given mycorrhizal fungi type (MFT) could vary across terrestrial biomes, which might explain the variable biomass allocations reported in different biomes28,29. Consequently, we suspect that warming effects on R/S across terrestrial biomes could also be regulated by biome-specific MFT dominance.

Here, we propose three potential hypotheses to better understand how biomass allocation in response to warming would alter the distribution of terrestrial carbon at a global scale30. First, biomass allocation between above- and belowground compartments of plants may differ due to warming. Using a meta-analysis, we test this hypothesis by examining how the intercept between R/S at warmed temperature and R/S at ambient temperature deviates from 0—we refer to it as a shift in vertical biomass allocation in terrestrial plants (Fig. 1a, b). For instance, when this intercept >0, it would imply that greater biomass is allocated downwards (i.e., vertically), meaning that there will be a greater R/S under warming than that at ambient temperature (Fig. 1a, b). Second, warming can homogenize R/S across biomes. To test this, we examine how the slope between R/S at warmed temperature and R/S at ambient temperature (as a proxy of biomes) deviates from 1—we refer to it as a shift in horizontal variability of R/S in terrestrial plants across biomes (Fig. 1c, d). For instance, when this slope < 1, it would imply that warming decreases the horizontal variability of R/S and thereby causes the homogenization of R/S among diverse biomes (Fig. 1c, d). Third, the mean annual precipitation (MAP) would be a key factor determining the response of R/S due to the direct dependency of belowground resources, such as nutrients and soil water, on MAP. To test these hypotheses, we collected results from 322 warming experiments (Supplementary Fig. 1) with 94 pairs of observations to explore the effects of warming on R/S (Fig. 1f). Using a meta-analytic technique, we show that, with a mean magnitude of 2.50 °C and the median of 2 °C (Supplementary Table 1), climate warming increases R/S (intercept > 0), whereas it decreases the variability in R/S among terrestrial biomes (slope < 1, Fig. 1e). These results suggest that plants with greater investment to shoot in ambient condition shift their allocation strategy by allocating more to root biomass in warmed conditions rather than to shoot biomass, and vice versa. This subsequently triggers warming-induced homogenization of R/S among diverse biomes (Fig. 1e). We further show that MAP and mycorrhizal association with plants correlate strongly with these patterns of R/S in terrestrial plants.

## Results

Based on 94 pairs of observations of root:shoot ratio (R/S), average warming of 2.5 °C increased the R/S by 6.1 to 8.8% (with a mean increase of 8.1%), which was significantly larger than 0 (p < 0.01, Fig. 1f). The intercept between R/S at warmed temperature and that at ambient temperature ranged from 0.044 to 0.107 (CI > 0, p < 0.05, Fig. 1e), suggesting higher biomass allocation to roots in vertical dimension in warmed environments (Fig. 1a). Meanwhile, the slope between R/S at warmed temperature and that at ambient temperature ranged from 0.908 to 0.997 (CI < 1, p < 0.05, Fig. 1e), indicating a horizontal homogenization of R/S among diverse biomes under warmed conditions (Fig. 1e). The biomes with higher R/S in ambient condition had greater variation in R/S in response to warming (Fig. 1g and Supplementary Fig. 2).

On the global scale, 164 species of vascular plants belonging to angiosperm (144) and gymnosperm taxa (20) were included in this study. Relative to annual species (23), perennial ones (141) were more common, whereas the numbers of woody (77) and herb species (87) were comparable in our meta-analysis (Fig. 2). Although the responses of R/S to warming across all the vascular plants displayed a significant phylogenetic signal (Blomberg’s K > 1, Supplementary Table 2), mean annual precipitation (MAP) was the most critical factor in determining the warming effects on R/S [i.e., response ratio (RR) of R/S, RR(R/S)] compared to all other predictors used in our analysis (Fig. 3). We found that MAP was negatively correlated with RR(R/S) (R2 = 0.104, Figs. 3 and 4, Supplementary Table 3). Moreover, the weighted RR(R/S) reversed from positive to negative in sites with MAP higher than 900 mm (Fig. 4).

Mycorrhizal fungi type (MFT: 121 for AMF and 43 for EMF, Fig. 2) was the second most important factor explaining warming effects on plant’s R/S (Fig. 3a). Besides MFT, no significant differences were found in any plant biomass variables (including R/S) in response to warming between evergreen vs. deciduous trees, broad vs. coniferous leaf trees, annual vs. perennial plants, and angiosperms vs. gymnosperms (Supplementary Table 4). Although there was a significant interaction between MFT and biomes (Supplementary Table 5), MFT still explained a greater proportion of variance for RR(R/S) relative to biome types (i.e., forest, grassland, cropland, tundra, and wetland, Fig. 3b). The sensitivity of RR(R/S) to MAP (i.e., the slopes in Fig. 4b) in biomes with ectomycorrhizal fungi (EMF) was smaller than in biomes with arbuscular mycorrhizal fungi (AMF). Moreover, both warming magnitude and warming duration didn’t affect the RR(R/S) across warming experiments (Fig. 3a and Supplementary Table 6).

Overall, warming enhanced plant total biomass (TB, +10.0%) mainly by increasing belowground biomass (BGB, +13.1%) even when it decreased aboveground biomass of plants (AGB, −8.9%, p > 0.05, Fig. 5). Warming did not change TB or BGB of plants associated with AMF but decreased their AGB by 13.8% and subsequently increased R/S by 9.9%. Warming stimulated TB, AGB, and BGB in biomes with EMF and with mixed AMF and EMF (AM-EMF), but only enhanced R/S in biomes with EMF-associated plants (Fig. 5). MFTs were significantly associated with the warming response of AGB. In contrast, the effects of biomes and plant functional types (PFTs) were true only for the response of BGB (Supplementary Table 5). In our path model, we further found that MFTS (i.e., AMF, EMF, and AM-EMF) explained 22.9% variation in AGB’s response to warming, whereas PFTS (i.e., herbs and woody plants) and MAP were the major significant pathways contributing to 25.4% of variation in BGB’s response (Fig. 6). Both responses of AGB and BGB to warming directly contributed (AGB negatively and BGB positively) to shifts in RR(R/S) in our path model, explaining a total of 46.3% of variation in RR(R/S) across studies (Fig. 6). Moreover, RR(R/S) was correlated with BGB responses to warming [RR(BGB)] (R2 = 0.31, p < 0.001), but not with AGB responses (R2 = 0.01, p = 0.392, Supplementary Fig. 2). The importance of PFTs on RR(R/S) was only true in studies with a warming-induced inhibition of TB (Supplementary Fig. 4). In addition, other experimental factors (e.g., methods of warming and belowground biomass measurement, Supplementary Table 7) and classifications of studies (e.g., field or laboratory experiments, Supplementary Table 4) did not explain any variation in RR(R/S).

## Discussion

Our meta-analysis shows that climate warming enhances plants’ biomass allocation to belowground but mainly so in drier habitats, agreeing with the hypothesis of warming-induced vertical downward biomass allocation (Fig. 1)31. The contingency of warming effects on plant biomass via soil moisture (Supplementary Fig. 5) might be one of the primary reasons for the upregulation of R/S in many study sites, except for the tundra and alpine biomes where soils are often frozen17,32. Even in the studies with decreased total plant biomass in response to warming, we found that belowground biomass allocations were higher than aboveground ones (Supplementary Figs. 6 and 7)33,34. Such a plant strategy can have important implications for soil carbon dynamics, potentially due to a lower turnover rate of root-associated carbon (i.e., carbon in hyphae and root per se) compared to shoot-associated carbon33,34. Biomes with lower original R/S at ambient conditions showed greater allocation to belowground biomass when warming most likely had induced greater demand for resources35, agreeing with our hypothesis of horizontal homogenization of R/S across biomes in a warmer world (Fig. 1). We suspect that the upregulated R/S due to warming might alter the carbon transfer between atmosphere and soil21,36.

More specifically, our results suggest that in areas with mean annual precipitation (MAP) lower than 900 mm, warmed air and soil surface could induce greater belowground biomass allocation potentially for plants to exploit resources in deeper soil layers, where water is more readily available, and temperature is cooler (Fig. 7a)37. The long-term adaptation of plants to water deficiency, e.g., a higher water use efficiency, ensures more biomass accumulation under warmed conditions, thereby triggering an upregulation of R/S (Fig. 7a, and prolonging the C residence time in plants38. In areas with MAP greater than 900 mm, adequate soil moisture could enable positive effects of warming on nutrient turnover and availability, e.g., soil NH4+ and NO3 for AMF and EMF, respectively (Fig. 7b)15,39. Indeed, the allocation to belowground biomass in these areas declined consequently in our meta-analysis, but only for terrestrial plants associated with AMF (Fig. 7b)40.

The regulation of MAP on R/S under warmed conditions was mainly through the adjustment of belowground biomass accumulation (Supplementary Fig. 3), which responded to variable effects of warming on soil moisture among plant functional types (PFTs, Supplementary Fig. 6)12,41. Significant changes in belowground biomass (BGB) in response to warming indeed varied among PFTs compared to other categories used in our meta-analysis (Supplementary Table 5). Woody plants, which are characterized by deeper and evolutionarily mature roots42, can redistribute soil water by hydraulic lift, supporting an increment in BGB under warmed conditions (Fig. 5)43. Herbaceous plants, which are characterized by evolutionarily younger roots relative to woody plants, can potentially adjust root traits (e.g., specific root length) more readily than biomass12,41. These warming-induced changes in root traits may have contributed to their non-significant responses in terms of BGB under warmed conditions (Fig. 5). In addition to MAP, due to interaction between temperature and soil moisture, antecedent mean annual temperature (MAT) was another important background climatic factor influencing the response of R/S in our meta-analysis44. For instance, MAT displayed a negative relationship with R/S due to a warming-induced significant increment of aboveground biomass (AGB) in most of the biomes with MAT < 13 °C and plants associated with EMF (Supplementary Table 3)45.

Plant’s association with mycorrhizal fungi further regulated biomass allocation patterns of terrestrial plants in response to warming. The response of R/S to warming among 164 vascular host plants displayed a significant phylogenetic signal (Fig. 2 and Supplementary Table 2), whereas significant differences were not found between evergreen vs. deciduous trees, broad vs. coniferous trees, annual vs. perennial plants, and angiosperms vs. gymnosperms (Supplementary Table 4). Although both arbuscular mycorrhizal fungi (AMF-) and ectomycorrhizal fungi (EMF-) associated plants exhibited an overall increase in R/S, the underlying mechanisms are likely to be different (Figs. 5, 7). Benefiting from hydrolytic and oxidative extracellular enzymes produced by the hyphae of EMF46, EMF-associated plants are capable of growing better under nutrient-deficient conditions relative to AMF-associated plants13,45. EMF-associated plants showed an increasing trend of AGB, BGB, and R/S under warming (Fig. 5), likely due to hyphal-facilitated soil organic matter (SOM) degradation and root nutrient absorption47. By contrast, AMF-associated plants are often more abundant in warmer sites (e.g., in temperate grasslands and deciduous forests) with drier soils48,49,50,51. Although AMF association correlated with greater water use efficiency in plants under warmed conditions (WUE, Supplementary Fig. 7), AMF-associated plants seemed to slow down the shoot biomass accumulation due to warming-induced soil water deficiency (mainly in grasses, Supplementary Fig. 9)52 and the potential increment of plant-derived carbon to AMF53. Interestingly, the regulation of mycorrhizal fungi types on responses of R/S to warming were mainly related to shoot biomass but not to root biomass (Fig. 6), implying the important roles of mycorrhizal associations on aboveground biomass allocation in terrestrial plants in a warmer world24,54,55.

As for biomes with mixed AMF and EMF (AM-EMF, Fig. 5, Supplementary Fig. 8), the complementary resource use (e.g., water and nutrients) between plants with arbuscular and ectomycorrhizal fungi provides a plausible explanation for the observed stability of R/S and higher productivity in response to warming (Fig. 4)56,57. As we did not find any significant differences in plant biomass allocation (R/S) or biomass itself (AGB and BGB) to warming between evergreen vs. deciduous trees, broad vs. coniferous trees, annual vs. perennial plants, or angiosperms vs. gymnosperms (Supplementary Table 4) when we ignored the types of mycorrhizal fungi in our analysis, it further confirms the importance of plants’ association with mycorrhizal fungi in understanding their biomass responses to warming compared to many other plant biotic characteristics.

We suspect that a stronger association of warming effects on biomass allocation with MAP and mycorrhizal associations could potentially lead to a redistribution of R/S in the horizontal dimension with more homogenization and lower variability of biomass allocation patterns among diverse biomes (Fig. 1)58. Although we found some clear signals in how warming effects were observed through a greater allocation of root biomass in terrestrial plants relative to shoot biomass, we caution readers about the variability in effect sizes, which could be caused by the distinct effect of warming magnitude and warming duration in a given biome (Supplementary Table 3). For example, warming magnitude might negatively affect the response of R/S for biomes with EMF, while warming duration could induce negative and positive effects for biomes with woody and herbaceous plants, respectively, as revealed in our analysis (Supplementary Table 3).

In summary, climate warming with a mean magnitude of 2.5 °C can enhance belowground biomass allocation as a whole, but mainly so in drier habitats. The warming of 2.5 °C also resembles the predicted IPCC warming (2.4 °C rise of global mean temperature) under a high greenhouse gas emissions scenario (SSP5-8.5) in the following decades (2041–2060)59, making our results further relevant for the future of terrestrial carbon dynamics. Our results highlight that habitat dryness (or wetness), and types of mycorrhizal association are crucial factors to consider when we aim to predict biomass allocation strategies in terrestrial plants in response to warming across biomes.

## Methods

### Data collection

Using the ISI Web of Science, peer-reviewed journal articles (1950–2020) related to plant biomass, growth, carbon cycling, and/or soil nitrogen under experimental warming were searched with the following specific keywords: (warming OR temperature OR heat* OR greenhouse) AND (biomass OR plant OR allocation OR root OR leaf OR stem OR photosynthe* OR growth OR aboveground OR belowground OR respiration OR soil OR carbon OR nitrogen OR microb*). To avoid bias in publication selection, we used the following criteria to select the studies for our meta-analysis (Supplementary Data 1): (i) Warming treatments were conducted in terrestrial biomes and at least one of the following variables was measured: plant total biomass (TB), aboveground biomass (AGB), belowground biomass (or root biomass, BGB), root:shoot ratio (R/S), water use efficiency (WUE), soil inorganic nitrogen (SIN), soil NH4+, soil NO3, microbial biomass (MB), and/or microbial biomass C/N (MB C/N). (ii) The experimental temperature, warming methods (e.g., open-top chamber, infrared heater, or soil heating cable and greenhouse), and dominant plant species were indicated clearly in both warming and control groups. (iii) Apart from the difference in experimental temperature, other initial environmental conditions and plant species compositions were the same between the control and warming groups. (iv) The duration of the warming experiment was at least longer than one growing season. (v) The mean, standard deviations/errors, and sample sizes of these variables could be extracted from the figures, tables, or the text directly.

Using the above-mentioned five selection criteria, 322 papers associated with the field (280) or laboratory (42) warming experiments were selected (Supplementary Table 4, Notes and References). The methods of air and/or soil warming in selected studies included open-top chamber, infrared heater, soil heating cable, and greenhouse (Supplementary Table 6). The observations (e.g., TB, AGB, BGB, or R/S, WUE, SIN, soil NH4+, soil NO3, MB, and/or MB C/N) in both control and warming groups were extracted using software GetData (version 2.22) as well as treatment and environmental information [e.g., warming duration (DUR) and magnitude (WM), latitude (LAT), mean annual precipitation (MAP) and temperature (MAT)]. If the local climate condition was not reported in a paper, the WorldClim dataset (http://www.worldclim.org/) was used to obtain climate variables based on latitude and longitude for that given study60. Roots or belowground biomass measurements across studies included direct harvest, soil core, and ingrowth mesh bags (Supplementary Table 6). In order to analyze the effects of climate (only field studies) and treatment variable (e.g., warming magnitude) on plant biomass responses to warming, we extracted MAP, MAT, DUR, and WM as well as the soil properties (0–30 cm depth), i.e., bulk density (BD), clay content (CLAY), and soil organic carbon (SOC) at each site from Global Gridded Surfaces of Selected Soil Characteristics (BD from IGBP-DIS)61 and the Harmonized World Soil Database (CLAY and SOC, from version 1.2, https://daac.ornl.gov/SOILS/guides/HWSD.html, Supplementary Table 3). The sites of the selected studies were mainly located in East Asia, North America, and Europe (Supplementary Fig. 1). The MAP and MAT of sites in field experiments ranged from 27 mm to 2400 mm (median 636.2 mm), and −20 °C to 30 °C (median 7.4 °C), respectively. Experimental warming duration and magnitude ranged from one growing season to 25 years (median 2 years), and 0.26 °C to 12 °C (median 2 °C), respectively (Supplementary Fig. 1 and Table 1).

The biome types (including cropland, desert, forest, grassland, tundra, and wetland, Supplementary Table 5), plant functional types (PFTs, including woody plant vs. herb, tree vs. shrub, grass vs. forb, Supplementary Fig. 1), other taxonomy categories (evergreen vs. deciduous tree, broad vs. coniferous leaf tree, annual vs. perennial plant, angiosperm vs. gymnosperm, monoculture vs. mixed plant community, Figs. 2 and 3, Supplementary Table 4) and mycorrhizal fungi types (MFTs, Supplementary Fig. 1 and Supplementary Table 5) for dominant plants in each case study were confirmed based on the original publications and the latest FungalRoot database (https://nt.ars-grin.gov/fungaldatabases/)45. The MFTs of biomes were labeled as AMF, EMF, or AM-EMF if the root symbiosis of dominant plants were arbuscular mycorrhizal fungi (AMF), ectomycorrhizal fungi (EMF), or mixed AMF and EMF (AM-EMF). Most analyses in this study focused on AMF, EMF, and AM-EMF, as biomes with non-mycorrhizal fungi (NMF) were too few to conduct any meta-analysis (Supplementary Fig. 1).

### Data analysis

In this study, we employed response ratio [RR, natural log (ln) of the ratio of the mean value in warming treatment ($$\bar{{X}_{t}}$$) to that in control ($$\bar{{X}_{{{{{{\rm{c}}}}}}}}$$), Eq. (1)] to reflect warming effects on various response variables62,63.

$$RR=\,{{{{\mathrm{ln}}}}}\left(\frac{\overline{{X}_{t}}}{\overline{{X}_{c}}}\right)=\,{{{{\mathrm{ln}}}}}(\overline{{X}_{t}})-\,{{{{\mathrm{ln}}}}}(\overline{{X}_{c}})$$
(1)

The weighted response ratio (RR++, Eq. (2)), and the standard error of RR++ [s(RR++), Eq. (3)] in each subgroup [e.g., different MFTs (AMF, EMF, or AM-EMF) or PFTs (woody plants or herbs)] were calculated using individual RR(RRij) and its weight (wij), which is the reciprocal of the variance (vij, Eq. (4), Fig. 4).

$$R{R}_{++ }=\frac{{\sum }_{i=1}^{m}{\sum }_{j=1}^{k}{{{{{{\rm{w}}}}}}}_{ij}R{R}_{ij}}{{\sum }_{i=1}^{m}{\sum }_{j=1}^{k}{{{{{{\rm{W}}}}}}}_{ij}}$$
(2)
$${{{{{\rm{s}}}}}}(R{R}_{++ })=\sqrt{\frac{1}{{\sum }_{i=1}^{m}{\sum }_{j=1}^{k}{w}_{ij}}}$$
(3)

where m is the number of subgroups, and k is the number of RR in the ith subgroup (i = 1, 2, …, m; j = 1, 2, …, k).

$${w}_{{ij}}=\frac{1}{{{{{{{\bf{V}}}}}}}_{{ij}}},\;v=\frac{{{S}_{t}}^{2}}{{n}_{t}{{\bar{X}}_{t}}^{2}}+\frac{{{S}_{c}}^{2}}{{n}_{c}{{\bar{X}}_{c}}^{2}}$$
(4)

The $${n}_{t}$$, $${n}_{c}$$, and $${S}_{t}$$, $${S}_{c}$$ are the numbers of replicates and standard deviations for a given variable in the warming and control group, respectively. The warming effect on a given response variable was considered statistically significant when the 95% CI of the effect size did not overlap with zero. The percentage change of a response variable under warmed conditions was estimated by [exp(RR++) − 1] × 100%.

The between-group heterogeneity (Qb) of effect sizes was estimated using the Q-statistic in MetaWin 2.1 (Supplementary Table 5)64. Stepwise linear regression was used to analyse the impacts of environmental conditions on the responses of root:shoot ratio [RR(R/S), Supplementary Table 4]. We used Blomberg’s K values to test the phylogenetic signals on the warming response of plant biomass variables using the “picante” package (version 1.8.2)65 in R (version 4.1.3, R Core Team, 2018, Supplementary Table 2). The effects of biomes, MFTs, PFTs, experimental methods (warming methods and methods to measure root or belowground biomass), warming magnitude (WM), and warming duration (DUR) were examined to explain the variation in response variables using the analysis of variance (ANOVA, Supplementary Tables 5 and 6). Both WM and DUR were sub-grouped into six classes according to the method explained in Lin et al. (Supplementary Table 6)12. The importance of each predictor for the response ratio of a given response variable was expressed as the sum of Akaike weights derived from the model selection using the “glmulti” package (version 1.0.8)66 in R (Fig. 3a, Supplementary Fig. 3). Nested analysis was performed using the “nlme” package (version 3.1-158)67 in R to estimate the variance explained by MFTs (Fig. 3b). We used the “metafor” package (version 3.4-0)68 in R for the linear regression analysis between MAP and RR(R/S) (Fig. 4). Structural equation model (SEM) was performed using the “lavaan” package (version 0.6-12)69 in R to examine the effects of PFTs, MFTs, background climate (MAT, MAP), and warming treatment (WM and DUR) on RR(R/S) via changes in AGB and BGB under warming (Fig. 6). The figures to depict correlations between specific response variables were drawn in SigmaPlot 12.5 (Systat Software, San Jose, CA, Supplementary Figs. 2 and 3).

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.