Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Global meta-analysis shows pervasive phosphorus limitation of aboveground plant production in natural terrestrial ecosystems

Global meta-analysis shows pervasive phosphorus limitation of aboveground plant production in... ARTICLE https://doi.org/10.1038/s41467-020-14492-w OPEN Global meta-analysis shows pervasive phosphorus limitation of aboveground plant production in natural terrestrial ecosystems 1,2,3 3 1,2 4 1,2 3 Enqing Hou *, Yiqi Luo , Yuanwen Kuang , Chengrong Chen , Xiankai Lu , Lifen Jiang , 1,2 1,2 Xianzhen Luo & Dazhi Wen * Phosphorus (P) limitation of aboveground plant production is usually assumed to occur in tropical regions but rarely elsewhere. Here we report that such P limitation is more wide- spread and much stronger than previously estimated. In our global meta-analysis, almost half (46.2%) of 652 P-addition field experiments reveal a significant P limitation on aboveground plant production. Globally, P additions increase aboveground plant production by 34.9% in natural terrestrial ecosystems, which is 7.0–15.9% higher than previously suggested. In croplands, by contrast, P additions increase aboveground plant production by only 13.9%, probably because of historical fertilizations. The magnitude of P limitation also differs among climate zones and regions, and is driven by climate, ecosystem properties, and fertilization regimes. In addition to confirming that P limitation is widespread in tropical regions, our study demonstrates that P limitation often occurs in other regions. This suggests that previous studies have underestimated the importance of altered P supply on aboveground plant production in natural terrestrial ecosystems. Key Laboratory of Vegetation Restoration and Management of Degraded Ecosystems, South China Botanical Garden, Chinese Academy of Sciences, 2 3 Guangzhou 510650, China. Center of Plant Ecology, Core Botanical Gardens, Chinese Academy of Sciences, Guangzhou 510650, China. Center for Ecosystem Science and Society, Northern Arizona University, Flagstaff, AZ 86011, USA. Australian Rivers Institute, School of Environment and Science, Griffith University, Nathan, QLD 4111, Australia. *email: [email protected]; [email protected] NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w utrient limitation of aboveground plant production has from tropical to arctic regions, spanning over 131 in latitude 1–5 o o been widely acknowledged . In terrestrial ecosystems, (54.8 S–76.5 N), and occurring on all continents except Antarc- Nnitrogen (N) has been considered as the most important tica, where no data were available (Figs. 1–3). Phosphorus lim- 3,6 limiting nutrient of aboveground plant production ; phosphorus itation of aboveground plant production occurred on all studied (P) has also been viewed as important, but mainly in lowland continents (Figs. 1–3), although the proportion of P limitation 4,7 tropical regions where soils are generally strongly weathered . instances differed among ecosystem types (Supplementary This prevalent view, however, has been challenged by an Table 3). Globally, 301 of the 652 experiments (46.2% of all increasing number of significant P limitation cases in areas other experiments; 45.0% of experiments in natural terrestrial ecosys- 1,8,9 than the lowland tropical regions (e.g., tundra regions) . For tems; and 48.6% of experiments in croplands) revealed significant example, significant P limitation on aboveground plant produc- P limitation of aboveground plant production (Fig. 1). These tion has also been found in some temperate areas with strongly findings provide convincing evidence that P limitation of above- 10–12 weathered soils . Despite the important role of P in above- ground plant production in terrestrial ecosystems is a worldwide ground plant production, we still lack a clear understanding of phenomenon. where, to what degree, and under what conditions P limits The conventional notion that P limits aboveground plant 5,8,13 aboveground plant production over the global land surface . production mainly in tropical regions is based on the following As a consequence, none of the tens of models in the fifth phase of patterns: relative to temperate regions, tropical regions generally 2 21,22 the Coupled Model Intercomparison Project (CMIP5) archive have older and more weathered soils , higher plant N:P ratios , 22,23 represents terrestrial P biogeochemistry, which causes substantial higher plant P use efficiencies , and lower plant and soil P 2,24 uncertainty in estimates of strength of the terrestrial carbon (C) concentrations . All of these latitudinal patterns, however, can 13,14 sink through the 21st century . only indicate the relative magnitude rather than the actual Here we report the distribution, magnitude, and drivers of P magnitude of nutrient limitation across regions. The actual limitation of aboveground plant production in terrestrial ecosys- magnitude of nutrient limitation is determined most reliably by tems worldwide. To accomplish this, we use a global database of experiments in which the response of aboveground plant produc- 3,25 652 P-addition field experiments compiled from 285 papers tion to nutrient addition is quantified . Our meta-analysis of P- published between 1955 and 2017 (Supplementary Figs. 1a and 2). fertilization field experiments shows that P significantly limits The database includes P-addition experiments in all major types of aboveground plant production across tropical, subtropical, tempe- terrestrial ecosystems, including both natural terrestrial ecosys- rate, and (sub)artic regions, although both the magnitude of P tems (436 experiments in forests, grasslands, tundras, or wetlands) limitation and the percentage of P limitation instances were greater and croplands (216 experiments) (Supplementary Tables 1 and 2). in tropical and subtropical regions than in temperate and (sub)artic The number of P-addition experiments in natural terrestrial regions (Fig. 3 and Supplementary Table 3). ecosystems in this study is 3.8–8.8 times greater than the number The worldwide occurrence of P limitation of terrestrial 1,8,9,15 in the previous meta-analyses (N= 50–117) . In addition, aboveground plant production may be explained by the 1,26 41.7% of the experiments in our database were published after biochemical machinery shared by autotrophs , the great 2007 and few of these were included in previous syntheses dedi- variability of plant characteristics and environmental conditions cated to N–P interactions (Supplementary Fig. 2). The collected within regions , and the multiple pathways resulting in P experiments are located on all continents except Antarctica limitation of aboveground plant production . Researchers have (Supplementary Fig. 1a) and have wide ranges of mean annual proposed that the P and N demands of core biochemical −1 precipitation (MAP, 80–5302 mm yr ) and mean annual tem- machinery (mainly concerning rRNA and proteins) shared by all perature (MAT, −12.1 to 27.5 C) (Supplementary Table 1). photoautotrophs may cause plant growth to be limited by P and 1,26 Compared to previous datasets, this up-to-date dataset better N to a similar degree . Although ecosystem properties such as captures Earth’s diverse terrestrial habitats and thereby provides a soil P availability are key drivers of P limitation of aboveground 8,27 much clearer understanding of the role of P supply on above- plant production in terrestrial ecosystems , ecosystem proper- ground plant production. ties vary greatly across sites . For example, the soil P supply rate −2 −1 To explore the global distribution of P limitation, we first ranges from ~1 to > 10000 g m yr in both tropical and estimate a threshold value of P limitation, i.e., a critical P effect temperate regions . P limitation in tropical regions is often size that best corresponds to a critical z-score at P = 0.05, based attributed to the occlusion of P in soil by chemical or physical on the statistical results provided in the 285 papers that com- mechanisms, the chronic loss of dissolved inorganic and organic prised our database (see “Methods” section; Supplementary P by leaching, and/or the exhaustion of soil primary minerals Fig. 3). We then map the global distribution of significant and during long-term soil development . However, there are also non-significant P limitation cases. We quantify the magnitude of other pathways that can cause P limitation in different regions P limitation at the global scale as well as in various groups of and at different timescales (from years to millions of years), ecosystems using a meta-analysis approach typically used in including the formation of soil layers (e.g., iron pans) that ecological studies, i.e., the natural logarithm transformed physically prevent/inhibit access by roots to potentially available response ratio (Ln(RR)) of aboveground plant production to P P, transactional limitations in which the input of P by weathering additions weighted by the inverse variance (details in “Methods” is less than the input of other resources, low-P parent material, 16–19 section) . Finally, we explore the effects of climate, ecosystem sinks that reduce P levels, and anthropogenic increases in the properties, and fertilization regimes and their relative importance supply of other resources and especially N and atmospheric CO . in predicting the P effect size using a boosted regression tree Permafrost, for example, can isolate plants from deeper portions 20 4,28 method . In general, we show a more widespread and much of the soil profile in cold regions . Low-P parent materials stronger P limitation of aboveground plant production in natural explain P limitation in some temperate regions . These and other 1,8,9,15 terrestrial ecosystems than previously suggested . pathways (e.g. the precipitation of P with Ca in arid soils) can cause P limitation in many temperate and (sub)arctic regions. Results and discussion Globally distributed P limitation. Our synthesis revealed that P The magnitude of P limitation. In natural terrestrial ecosystems, limitation of aboveground plant production is globally distributed, P additions increased aboveground plant production over controls 2 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE P limitation in natural terrestrial ecosystems Significant (196) Non-significant (240) P limitation in croplands Significant (105) Non-significant (111) Fig. 1 Locations of the 652 experiments in which the effect of P addition on aboveground plant production was assessed. a Natural terrestrial ecosystems. b Croplands. Experiments were determined to have significant P limitation based on the Ln(Response Ratio). If the Ln(Response Ratio) was higher than a threshold value (0.23 for natural terrestrial ecosystems and 0.09 for croplands), it was considered a significant case (Z test, P < 0.05) of P limitation. Determination of the threshold values is described in the “Methods” section and is supported by the Supplementary Fig. 3. Numbers in brackets are the number of experiments in the indicated group. Source data are provided as a Source Data file. by an average of 34.9%, with a 95% confidence interval of to 41.1% (N = 44) and 43.0% (N = 46) of the 107 terrestrial P- 30.0%–40.1% (N = 436; Table 1). The estimates were robust given addition experiments in the study by Elser et al. were performed our large sample size, as suggested by our sensitivity tests (not in Europe and North America, respectively (Supplementary sensitive to outliers; Supplementary Fig. 4), publication bias tests Fig. 1b). The P effect size in North America (36.9%) in the (no significant publication bias; Supplementary Fig. 5a) and current study was close to the global average, but the P effect size temporal change test (a significant but minor change in effect size in Europe (21.7%) was much smaller than the global average and with publication year; Supplementary Fig. 6a). Our average the averages in Australia (50.6%), Asia (40.4%), and South (34.9%) was about two times greater than the average reported in America (37.7%) (Fig. 3). Therefore, the global magnitude of P a recent global meta-analysis (17.7%) that used the same meta- limitation is much larger than that based on data mainly from analysis method but with a much smaller sample size (N = 60). To Europe and North America. The experiments in Augusto et al. compare our meta-analysis with three other previous meta- were spread quite evenly over the global land surface (Supple- 1,8,9 analyses , we also calculated the magnitude of P limitation in mentary Fig. 1c). Their relatively low estimates might be partly the natural terrestrial ecosystems by weighting the Ln(RR) uni- explained by their use of a slightly different way (relative to our formly or by weighting the RR with the inverse variance (see study) to remove pseudo-replications, i.e., the latest measurement details in “Methods” section). The estimated averages (36.3% and for forests but the earliest measurement for other ecosystems. 47.6%, respectively) were again higher than those reported in the Moreover, all of the previous syntheses had a much smaller 8 1 9 previous meta-analyses (23.4% –29.3% and 31.7% , respectively, sample size (Table 1) and therefore a relatively poorer N= 50–117; Table 1). Moreover, we found that the P effect size representation of global natural terrestrial ecosystems than our increased with quantity of P added and with the experimental study (Supplementary Figs. 1, 4). duration (Fig. 3). Therefore, the magnitude of P limitation in the The P effect size was much smaller in croplands than in natural natural terrestrial ecosystems was even larger after the Ln(RR) was terrestrial ecosystems (Table 1) and was even smaller after it was weighted by the quantity of P added (40.5%) or experimental adjusted with the trim-and-fill method (4.1%; Supplementary duration (48.4%) (Table 1). Fig. 5b). The pattern holds true for most continents and climate The estimate for natural terrestrial ecosystems was lower in zones (Fig. 3), under all fertilization regimes (Fig. 3), on all types Elser et al. than in our study (Table 1). This perhaps because up of soils (Supplementary Fig. 7), and for all meta-analytic methods NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w Table 1 The magnitude of P limitation in natural terrestrial ecosystems is larger than previously estimated. Method This study Previous studies Effect size [Lower CI, Number of Effect size [Lower CI, Number of Upper CI] (%) experiments Upper CI] (%) experiments Natural terrestrial ecosystems Ln(RR) weighted by the inverse variance 34.9 [30.0, 40.1] 436 17.7 [11.1, 24.8] 60 Ln(RR) weighted uniformly 36.3 [31.0, 41.9] 436 29.3 [14.9, 45.4] 107 23.4 [16.8, 30.4] 117 RR weighted by the inverse variance 47.6 [34.5, 60.8] 436 31.7 [21.6, 41.8] 50 Ln(RR) weighted by P addition amount 40.5 [27.5, 54.8] 436 Ln(RR) weighted by experimental duration 48.4 [33.5, 64.9] 436 Croplands Ln(RR) weighted by the inverse variance 13.9 [11.1, 16.8] 216 Ln(RR) weighted uniformly 16.3 [13.0, 19.7] 216 RR weighted by the inverse variance 15.4 [7.0, 23.7] 216 Ln(RR) weighted by P addition amount 24.7 [11.2, 39.8] 216 Ln(RR) weighted by experimental duration 25.9 [12.0, 41.6] 216 The magnitude of P limitation in the natural terrestrial ecosystems and also in the croplands was calculated using the five methods listed in the table. CI indicates confidence interval. Ln(RR) indicates ln transformed response ratio. (Table 1). The pattern can be at least partly explained by the fertilization regimes, soil properties, and plant properties each higher availability of soil P in croplands than in natural terrestrial explained some percentage (9.1%–40.0%) of the total explained ecosystems (Supplementary Fig. 8). Before P-addition experi- variation in P effect size in both the natural terrestrial ecosystems 2 2 ments were performed, some of the croplands likely received P (R = 0.59) and the croplands (R = 0.79) (Fig. 4 and Supple- fertilizers that increased soil P availability, although no pre- mentary Figs. 9−11. That the P effect size is regulated by multiple experiment fertilization was recorded in the source literature (see factors rather than by any single factor once again explains why P “Methods” section for the selection criteria). In croplands in limitation is widespread on the globe. For example, P limitation France, for example, an average of 82% of soil P was estimated to can occur in the tropical and subtropical natural ecosystems due have originated from former fertilization applications . Our to the high temperatures and precipitation that drive plant P results, therefore, suggest that P limitation in croplands has been demand and to the low soil extractable P concentration that limits 30,31 24 largely alleviated by historical fertilizations and that a soil P supply (Supplementary Fig. 10a, c, d). In contrast, the reduced amount of P fertilizer is needed to increase crop occurrence of P limitation in the temperate and (sub)arctic nat- production in the future. It follows that to accurately predict the ural ecosystems may be attributed to their generally high soil future fertilizer effect on crop production, models require organic matter contents and pH values (Supplementary Fig. 10e, fertilization history. Our results concerning differences between h). High soil organic matter content can reduce soil P availability croplands and natural terrestrial ecosystems may also be related by occluding P in the organic forms and by enhancing microbial 25,27,34 to the lower soil organic matter contents and shorter experi- immobilization of P in the soil . High soil pH can reduce mental durations in croplands (Fig. 3 and Supplementary soil P sorption capacity and thus increase the use efficiency of P Table 1). fertilizer by plants (i.e., increase the P effect size). Moreover, A positive asymmetric distribution of the P effect size was both high soil organic matter content and moderate soil pH can observed in the croplands (Supplementary Fig. 5b). This does not enhance the availability of nutrients such as N, potassium, and 35,36 necessarily indicate a publication bias (i.e., the tendency of journals calcium in soils , which may exaggerate the response of plant 32 9,31,37 to favor the publication of statistically significant results) . growth to P addition . Fertilization experiments in croplands typically have multiple As our study did for P, the study of LeBauer and Treseder nutrient (e.g., P, N, and K) treatments, and P addition is only one examined plant responses to N additions alone; they found a of the multiple treatments. Therefore, there is no apparent global distribution of N limitation on terrestrial primary 1 37 tendency of journals to favor publication of statistically significant production. Elser et al. and Harpole et al. , however, reported P effects. Instead, the asymmetric distribution may mainly result a prevalent co-limitation of terrestrial primary production by N from a true heterogeneity . The growth of plants in some regions and P. Although the results seem conflicting (i.e., globally is known to be strongly limited by P supply (e.g., Australia and distributed P limitation, N limitation, or N and P co-limitation), 4,7 27,37 lowland tropical areas with strongly weathered soils) .However, they can be reconciled by the multiple limitation hypothesis . we are unaware of any large, negative P effect on plant growth at A prevalent co-limitation of terrestrial primary production by N the community level, although there are rare reports of P toxicity and P suggests a generally balanced N and P limitation in global 1,37 symptoms in some plants that have adapted to low soil P terrestrial ecosystems , while a globally distributed N limitation 33 3,6 availability and of P-driven limitation of plant growth by N via indicate widespread N limitation in terrestrial ecosystems . soil microbes in some N-limited ecosystems .Therefore,the Given that both are reasonable, a worldwide occurrence of P detection of many more positive P effects than negative P effects is limitation in terrestrial ecosystems is expected, as observed in the reasonable (Supplementary Fig. 5). present study; the absence of a worldwide occurrence of P limitation would either imply an imbalance of N and P limitation, which would counter the finding of prevalent N and P co- Predictors of the magnitude of P limitation. Although our limitation (if widespread N limitation is true), or imply a less results show that P limitation is a global phenomenon, the widespread N limitation, which would counter the finding of magnitude of P limitation did vary greatly among experiments, globally distributed N limitation (if prevalent N and P co- with the Ln(RR) ranging from −0.48 to 2.44 (Fig. 2). Climate, limitation is true). Furthermore, the widespread P limitation 4 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE 2.5 ab Forest (52.5%) Grassland (36.8%) 2.0 Tundra (43.8%) Wetland (52.9%) 1.5 1.0 0.5 0.0 –0.5 020 40 60 80 0 20 40 60 80 100 1.5 cd Cropland (48.6%) 1.0 0.5 0.0 –0.5 0 20 40 60 80 0 20 40 60 80 100 Absolute latitude (°) Accumulated proportion (%) Fig. 2 Consistent occurrence of significant P limitation in all types of ecosystems. a Significant P limitation occurred in the natural terrestrial ecosystems at almost all latitudes. The magnitude of P limitation decreased with latitude in wetlands (meta-regression, R = 0.35, P < 0.05, N = 85) but not in any other type of natural terrestrial ecosystem (meta-regression, P > 0.05). b Significant P limitation occurred in all types of natural terrestrial ecosystems. Values in the brackets indicate the percentage of significant P limitation cases in the total sample size in each type of natural terrestrial ecosystem. ° ° c Significant P limitation occurred in croplands at all explored latitudes (absolute latitude between 0.1 and 56.5 ), and the magnitude of P limitation decreased with absolute latitude (meta-regression, R = 0.16, P < 0.05, N = 216). d Significant P limitation was found in 48.6% of the P-fertilization experiments in the croplands. As in Fig. 1, statistical significance of P limitation was assessed based on the Ln(Response Ratio). Dashed lines in all plots (0.23 in (a) and (b), 0.09 in (c) and (d)) indicate where the magnitude of P limitation approximates a 0.05 significance level (Z test). In (a) and (c), the size of each point is proportional to the weight used for meta-regression analysis. Source data are provided as a Source Data file. identified in this study and the widespread N limitation reported plants in natural terrestrial ecosystems (Fig. 3 and Supplementary in LeBauer and Treseder together imply that P limitation and N Fig. 10). This may lead to an underestimation of the global limitation are largely independent of each other . This possibility average of P limitation in natural terrestrial ecosystems. More- is supported by another synthesis study, which reported that the over, additional uncertainties can be introduced by the statistical effects of N supply and P supply on aboveground plant analyses (Table 1), missing variances of aboveground plant production are additive in most terrestrial ecosystems . Taken production (Supplementary Table 4), and the various measures of together, the evidence suggests the worldwide occurrence of both aboveground plant production used in different experiments P limitation and N limitation in terrestrial ecosystems , which (Supplementary Table 5). Missing measurements of ecosystem supports the multiple limitation hypothesis and challenges properties such as soil extractable P concentration and pH 27,37 Liebig’s Law of the Minimum . (Supplementary Table 1) can lead to an underestimation of their Although our dataset is much larger than those in previous relative importance in predicting the magnitude of P limitation. syntheses, there is still a large uncertainty in our estimate of the Third and last, given the long span of time of the datasets global magnitude of P limitation in terrestrial ecosystems. There (Supplementary Fig. 2), the nature of nutrient limitation has are three sources of this uncertainty. First, while ecosystems in likely changed over much of the land covered in this analysis, due, Australia, East Asia, and West Asia are better represented in our for example, to the changes in atmospheric N deposition and to dataset than in the previous ones, ecosystems in North Asia and changes in fertilization practices in the croplands . In spite of the tropics are still largely underrepresented (Supplementary these uncertainties, our large dataset of P addition experiments Fig. 1a). More experiments from North Asia may lower the global provides a much clearer pattern of the global distribution of P averages of P limitation, while more experiments from the tropics limitation and a more robust estimate of the global magnitude of would likely increase the global averages. Mature mixed forests P limitation in terrestrial ecosystems than previous datasets. were also underrepresented (Supplementary Table 3), and the Our findings have important implications for understanding inclusion of an increased number of forests may lower the global the role of P supply in controlling aboveground plant production average of the natural terrestrial ecosystems (Supplementary in terrestrial ecosystems. The results show a more widespread and Fig. 7). Second, most experiments were performed for ≤ 10 yrs much stronger limitation of aboveground plant production by P −1 and with a cumulative P addition ≤ 500 kg ha , which may not in natural terrestrial ecosystems than previously thought. The be long enough or high enough to fully stimulate the growth of results confirm the necessity of incorporating P limitation in NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 5 Ln (Response ratio) Ln (Response ratio) Vegetation Soil weathered P addition amount Experimental Fertilizer type –1 (kg ha ) duration (year) ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w Forest Australia Grass Asia Tundra South America Wetland North America Europe Strongly Africa Intermediately Slightly ≤1000 m >1000 m >500 200–500 50–200 Tropic ≤50 Subtropic Temperate >10 (Sub)arctic 5–10 1–5 Humid ≤1 Sub-humid Dry subhumid SSP Semi-arid TSP Arid Others 020 40 60 0 20 40 60 80 100 Effect size (%) Effect size (%) Fig. 3 P limitation was significant in all regions and major types of ecosystems and under various fertilization regimes. Exceptions are the non- significant P limitation in two groups of experiments in the croplands that had a small sample size (N = 8). Natural terrestrial ecosystems are shown in green (total N= 436), and croplands are shown in yellow (total N = 216). Values represent effect sizes ± 95% confidence intervals. The size of each point is proportional to the sample size (sample sizes are listed in Supplementary Table 3). The dashed lines indicate the no-fertilization effect. SSP is single superphosphate, and TSP is triple superphosphate. Source data are provided as a Source Data file. Natural terrestrial ecosystems Croplands 25 25 20 18.5 20 17.8 16.5 14.6 15.1 15 15 12.7 12.0 12.0 11.9 11.5 10.2 9.2 10 10 8.6 8.9 6.2 5.0 5 5 3.8 2.7 2.0 1.0 0 0 Fig. 4 Relative influence of climate, fertilization regimes, and ecosystem properties on the magnitude of P limitation. a Natural terrestrial ecosystems. b Croplands. The number above each bar indicates the percentage of the total explained variation accounted for by the variable. Fertilization regimes are in pink, climate factors are in blue, soil properties are in gray, and vegetation properties are in green. Source data are provided as a Source Data file. 6 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications Mean annual temperature P addition amount Mean annual precipitation Soil available P Soil organic C Experimental duration Vegetation type Soil pH Fertilizer type N fixation vegetation Soil available P Mean annual temperature P addition amount Crop species Soil organic C Soil pH Mean annual precipitation Fertilizer type Experimental duration N fixation crop Aridity Climate zone Altitude Continent Relative influence (%) NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE 13,14 earth system models . The results also show a much smaller P treatments (concentrations of available P, organic C, and total N; pH in water; and particle size), and parent material type. For each experiment in forest ecosystems, fertilizer effect in croplands than in natural terrestrial ecosystems, forest composition (i.e., pure or mixed forest) and the average forest age during the which suggests that P limitation in croplands has been generally experiment were also recorded. alleviated by historical fertilizations. Finally, the co-regulation of plant response to altered P supply by climate, ecosystem Data preparation. In cases where the referenced studies did not report the latitude properties, and fertilization regimes highlights the importance or longitude of the P-addition experiment (52% of the studies did not report both of taking a systems approach to study how nutrient supply affects latitude and longitude), the approximate latitude and longitude were derived by geocoding the name of the location in Google Earth 7.0 (the free version). In cases aboveground plant production. where the referenced studies did not report MAT (76%), MAP (54%), or altitude (65%), the values were derived from WorldClim using site geographic location (i.e. latitude and longitude). The aridity index (AI) of each site was obtained from Methods 40 CGIAR-CSI using data from WorldClim ; the AI value decreases as aridity Data collection. With the aim of constructing a comprehensive database of the increases. experimentally determined effects of P additions on aboveground plant production Soil type was classified according to the U.S. Department of Agriculture soil in global terrestrial ecosystems, we collected as many experiments that fulfilled our classification system . Soils were grouped based their degree of weathering 42,43 criteria (described below) as possible. Relevant studies were identified by searching according to previous studies : Andisols, Histosols, Entisols, and Inceptisols ISI Web of Knowledge, Google Scholar, and China Knowledge Resource Integrated were considered to be slightly weathered soils; Aridsols, Vertisols, Mollisols, and Database using combinations of keywords such as “phosph* addition”, “phosph* Alfisols were considered to be intermediately weathered soils; and Spodosols, fertili*”, “phosph* enrich*”, “aboveground biomass”, “primary product*”, “crop Ultisols, and Oxisols were considered to be strongly weathered soils. Parent yield”, and “grain yield”. Our survey also included studies summarized in pre- material types were grouped into four geological classes according to a previous viously published syntheses and the subsequent relevant studies citing those study : acid, intermediate, mafic, and calcareous rocks. syntheses. A PRISMA flow diagram (Supplementary Fig. 12) shows the procedure For comparison of P effect sizes among regions, experiments in the database we used for the selection of studies. were grouped in four different ways. First, experiments were grouped according to To be included in our database, published experiments were required to satisfy their continental locations: Australia, Asia, Africa, Europe, North America, and the following criteria: (1) the P-addition experiment was conducted in the field and South America. Second, experiments were grouped based on absolute latitude into o o o included P-addition and control treatments within the same ecosystem under the four latitude belts or regions: tropic (23.4 S–23.4 N), subtropic (23.4–35 Sor same environmental conditions, and also included measures of aboveground plant o o o o o N), temperate (35–50 Sor N), and (sub)arctic (>50 Sor N). Third, production in both P-addition and control treatments; (2) no fertilization was experiments were grouped according to altitude into low-altitude experiments recorded in the control treatment either before the start of the experiment or (< 1000 m a.s.l.) and high-altitude experiments (≥1000 m a.s.l.). Finally, during the experiment; (3) the P treatment received a P fertilizer that did not experiments were divided based on site aridity level into five groups: arid (AI ≤ contain N so as to avoid the effect of N; as a result experiments with application of 0.20), semi-arid (0.20 < AI ≤ 0.50), dry subhumid (0.50 < AI ≤ 0.65), sub-humid ammonium phosphate, manure or other fertilizers were excluded. (0.65 < AI ≤ 1.0), and humid (1.0 < AI). The complete dataset is available at To be considered an experiment in our analysis, a reported experiment had to Figshare . be temporally and spatially distinct and had to have internally consistent controls. Multiple experiments could be reported by one publication; for instance, the Phosphorus limitation threshold. One major objective of our study was to map application of the same experimental treatments was considered to represent the global distribution of experiments in which P significantly limited aboveground multiple experiments if the treatments were applied at several sites with different plant production. To do this, we had to define a threshold value that separated vegetation types. When multiple measures were reported over time at a single experiments that did or did not find significant P limitation. We estimated the experimental site, we used the latest measure. When multiple levels of P fertilizer threshold value separately for the natural terrestrial ecosystems and the croplands, treatments were reported, we used the measure with the highest amount of P using a method described in a recent study . In general, we first collected the addition. Choosing the latest measure and the highest P addition amount increased reported statistical responses of aboveground plant production to P additions from the likelihood that P additions fulfill plant demand and overcome the sorption of P 25,27 the source references. We then investigated the distribution of the Ln(RR) values. fertilizer by soils and soil microbial competition for P fertilizer . When multiple Finally, we identified the threshold value of the Ln(RR) that optimizes the dis- forms of P fertilizers were tested, we chose the treatment of single superphosphate tinction between statistically significant positive P effects and statistically non- or triple superphosphate if available. significant P effects. Of the 128 experiments in the natural terrestrial ecosystems We included only experiments that reported the response of community-level that reported a significant P limitation, 84% had an Ln(RR) value ≥ 0.23 (Supple- aboveground plant production to P additions. Single-species responses were not mentary Fig. 3a). Similarly, of the 162 experiments in the natural terrestrial eco- included unless drawn from a mono-dominant community. If several species from systems that reported a non-significant P effect, 85% had an Ln(RR) < 0.23 a community were individually assayed, an average across all species was used. (Supplementary Fig. 3b). When the two groups were combined, the maximum Experiments in forest or savanna ecosystems that only reported the response of percentage (84%) of correct classification (i.e., a significant positive effect was understory or herbaceous response to P additions were not included. Experiments classified as a significant case and a non-significant effect was classified as a non- with only stand biomass responses were excluded unless the stand biomass data significant case) was obtained with an Ln(RR) value of 0.23 (Supplementary could be used to calculate aboveground plant biomass production. Fig. 3c). Therefore, 0.23 was used as the threshold Ln(RR) value to distinguish Ecosystems were classified as forest, grassland, tundra, wetland, or cropland; significant from non-significant P limitation in natural terrestrial ecosystems. This natural forests, plantations, shrublands, and savannas were all classified as forest. In threshold value is close to the one used in a previous study (Ln(RR) of 0.20) .A forest ecosystems, beside aboveground plant biomass production (N = 33), we also similar approach was applied to the P-addition experiments in the croplands, such accepted proxy variables that are known to be correlated with aboveground plant that 0.09 was used as the threshold Ln(RR) value for croplands (Supplementary biomass production, such as litterfall production (1) and the rate of increase in tree Fig. 3d–f). diameter (34), stem volume (25), basal area (25), or height (16) (Supplementary Table 5). We showed that the weighted Ln(RR) did not differ significantly among the various variables used (Supplementary Table 5). In croplands, beside Meta-analysis. We quantified the magnitude of P limitation at the global scale and aboveground plant biomass production (N = 85), we also accepted marketable in various groups of ecosystems by weighting the Ln(RR) with the inverse variance yield (131), because we found that marketable yield responded similarly to P 16–19 and a random-effect model . To do this, we extracted means, standard additions as aboveground plant production in the croplands based on studies that deviations (SDs), and sample sizes (n) from the published studies. If standard error reported both measures (Supplementary Fig. 13). In tundras, beside aboveground (SE) rather than SD was reported, SD was calculated: biomass production (N = 10), we also included leaf mass per tiller (4), tiller pffiffiffi SD ¼ SE n ð1Þ biomass (1), and plot level NDVI (1) (Supplementary Table 5). In wetlands, beside aboveground biomass production (N = 72), we also included height increase (5), If neither SD nor SE was reported, we approximated the missing SD by multiplying leaf area index (3), the production of whole plants (3), and chamber based gross the reported mean by the average coefficient of variance of our complete dataset. If primary production (2) (Supplementary Table 5). sample size was not reported, we assigned sample sizes as the median sample size of In total, we collected data from 652 P addition experiments reported in 285 our complete dataset. We approximated the SDs and the sample sizes separately for published papers, including 436 experiments from natural terrestrial ecosystems the natural terrestrial ecosystems and the croplands and also separately for the (including forests, grasslands, tundras, and wetlands) and 216 experiments from control and the P-addition treatments (see details in Supplementary Table 4). croplands (see experimental locations in Supplementary Fig. 1a). Beside The Ln(RR) of an experiment was calculated as follows: aboveground plant production measures, our database also included site characteristics and fertilization regimes, which were used to explain the variation in LnðÞ RR ¼ Ln ¼ LnðÞ X  LnðÞ X ð2Þ Ln(RR). Site characteristics included site location (latitude and longitude), climate t c variables (MAT and MAP), topographical conditions (altitude and slope), plant characteristics (vegetation type, and symbiotic N fixation), soil type, soil where X . and X . are mean aboveground plant production in the P treatment and t c physiochemical properties before the experiments began or from the control control, respectively. NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w The weighted mean response ratio (Ln(RR )) of a group of ecosys was follows: fertilization regimes in predicting the Ln(RR) in the natural terrestrial ecosystems and in the croplands. Before BRT analyses, variable selections were made to avoid w ´ LnðRR Þ i¼1 i i high correlations among predictors. Specifically, (1) Soil organic C concentration Ln RR ¼ ð3Þ i¼1 i was included as an indicator of soil organic matter content, while soil total N concentration was not included in the BRT models due to its high correlation with where m is the number of experiments in the group (e.g., a region), and w . is the soil organic C concentration (Natural terrestrial ecosystems: r = 0.85, P < 0.001, weighting factor of the ith experiment in the group. The w . was calculated as N = 92; Croplands: r = 0.95, P < 0.001, N = 54). (2) Soil available P concentration follows: was included as an indicator of soil P availability, while soil total P concentration was not included. Ecosystem properties such as soil particle size and parent w ¼ ð4Þ v material type were not included in our BRT analyses due to the very large pro- portions of missing data (Supplementary Table 1), including which can bias the where v . is the variance of study (i) in the group. The v . was calculated as follows: i i estimate of their relative importance. Parameter values used for the BRT analyses generally followed the v ¼ v þ T ð5Þ i i recommendation of a previous study , i.e. bag fraction as 0.75, the number of where v is the within-study variance of study (i), and T is the between-studies cross validation as 10, and tree complexity as 2. Learning rate was set at a small variance. The v was calculated as follows: value (i.e., 0.005) to include a large number (>1000) of regression trees in the 2 2 models. Because Ln(RR) is a continuous numerical variable, a Gaussian S S t c v ¼ þ ð6Þ distribution of errors was used. The relative importance of each predictor 2 2 n X n X t t c represented a percentage of the total variation explained by the models. The BRT analyses were performed with the “gbm” package version 2.1.5 plus the custom where n and n are the sample sizes for the P treatment and the control groups, t c respectively, and S and S are the standard deviations for the P treatment and the code of another study in R version 3.3.1. For evaluation of the spatial structure of t c control groups, respectively, of study (i). The calculation of T can be seen in the BRT residuals, the global Moran’s I statistic was applied to determine the Borenstein et al. . significance using the “spdep” package version 0.7.7 . The standard error of the Ln(RR ) was calculated as: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Reporting summary. Further information on research design is available in s Ln RR ¼ P ð7Þ the Nature Research Reporting Summary linked to this article. þ m i¼1 i The 95% confidence interval (CI) for the Ln(RR ) was calculated as follows: + Data availability All data used in this study are available at Figshare (https://doi.org/10.6084/m9. 95%CI ¼ Ln RR ±1:96s Ln RR ð8Þ þ þ figshare.8969963). The source data underlying all Figures except Supplementary Figs. 5, If the 95% CI did not overlap with zero, the overall P addition effect in the group of 10, and 11 are also available on the above web page. Supplementary Figs. 5, 10, and 11 are experiments was considered significant. The percentage change in aboveground directly created using R functions, as described in Methods. plant production induced by P addition (i.e., the effect size) in a group of ecosystems was measured as follows: Code availability Effect sizeðÞ % ¼ðexp Ln RR 1Þ100% ð9Þ þ The code used in this study is available at Figshare (https://doi.org/10.6084/m9. figshare.8969963). The meta-analyses were performed using “meta” package in R version 3.3.1 . 1,8,9 To compare our analyses with the previous meta-analyses , we also calculated the global magnitude of P limitation using two other methods: Received: 24 July 2019; Accepted: 14 January 2020; 1,8 (1) Ln(RR) weighted uniformly , where effect size only depends on the means of control and P treatment groups. (2) RR weighted by the inverse variance , where the RR rather than the Ln(RR) is weighted by the inverse variance. We also calculated the global magnitude of P limitation by considering the effects of the quantity of P added and experimental duration: References (3) Ln(RR) weighted by the quantity of P added. The weight of each experiment 1. Elser, J. J. et al. Global analysis of nitrogen and phosphorus limitation of was calculated as follows: primary producers in freshwater, marine and terrestrial ecosystems. Ecol. Lett. 10, 1135–1142 (2007). w ¼ P ð10Þ P 2. Aerts, R. & Chapin, F. S. The mineral nutrition of wild plants revisited: a re- i¼1 i evaluation of processes and patterns. Adv. Ecol. Res 30,1–67 (1999). −1 where P is the cumulative quantity of P added during the experiment (kg ha ), 3. Vitousek, P. M. & Howarth, R. W. Nitrogen limitation on land and in the sea - and M is the number of experiments in the natural terrestrial ecosystems (436) how can it occur? Biogeochemistry 13,87–115 (1991). or the croplands (216). 4. Vitousek, P. M., Porder, S., Houlton, B. Z. & Chadwick, O. A. Terrestrial (4) Ln(RR) weighted by experimental duration. The weight of each experiment phosphorus limitation: mechanisms, implications, and nitrogen-phosphorus was calculated as follows: interactions. Ecol. Appl. 20,5–15 (2010). 5. Turner, B. L., Brenes-Arguedas, T. & Condit, R. Pervasive phosphorus w ¼ P ð11Þ limitation of tree species but not communities in tropical forests. Nature 555, i¼1 367–370 (2018). 6. LeBauer, D. S. & Treseder, K. K. Nitrogen limitation of net primary where D is the duration of the experiment (yr). productivity in terrestrial ecosystems is globally distributed. Ecology 89,371–379 We checked the quality of our meta-analysis according to the checklist of 19 (2008). Koricheva and Gurevitch . Our meta-analysis fulfilled all of the quality criteria for 7. Vitousek, P. M. & Farrington, H. Nutrient limitation and soil development: a meta-analysis in plant ecology (Supplementary Table 6). We performed two experimental test of a biogeochemical theory. Biogeochemistry 37,63–75 sensitivity tests (Supplementary Fig. 4). One was the leave-one-out meta-analysis (1997). using the “metainf” function. The other was the cumulative meta-analysis using a 8. Augusto, L., Achat, D. L., Jonard, M., Vidal, D. & Ringeval, B. Soil parent random-effect model in the “metacum” function based on inverse-variance material - a major driver of plant nutrient limitations in terrestrial ecosystems. weighted Ln(RR). The cumulative meta-analysis was repeated 1000 times with Glob. Change Biol. 23, 3808–3824 (2017). random orders of experiments. We also ran a cumulative meta-analysis using the 9. Li, Y., Niu, S. & Yu, G. Aggravated phosphorus limitation on biomass “metacum” function based on the uniformly weighted Ln(RR). We created funnel plots to detect possible publication bias using the “funnel” function. The possible production under increasing nitrogen loading: a meta‐analysis. Glob. Change publication bias was statistically tested using the “metabias” function. If a Biol. 22, 934–943 (2016). publication bias was suggested, we further adjusted the P effect size by the trim- 10. Laliberté, E. et al. Experimental assessment of nutrient limitation along a 2‐ and-fill method using the “trimfill” function. Finally, we explored the relationships million‐year dune chronosequence in the south‐western Australia biodiversity between the Ln(RR) and the absolute value of latitude and publication year using hotspot. J. Ecol. 100, 631–642 (2012). the “metareg” function. All the above functions were from the “meta” package in R 11. Peltzer, D. A. et al. Understanding ecosystem retrogression. Ecol. Monogr. 80, version 3.3.1 . 509–529 (2010). 12. Wardle, D. A., Walker, L. R. & Bardgett, R. D. Ecosystem properties and forest decline in contrasting long-term chronosequences. Science 305,509–513 Boosted regression tree analysis. Boosted regression tree (BRT) analyses were (2004). conducted to quantify the relative importance of climate, ecosystem properties, and 8 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE 13. Reed, S. C., Yang, X. & Thornton, P. E. Incorporating phosphorus cycling into 41. Soil Survey Staff. Soil Taxonomy: A Basic System of Soil Classification for global modeling efforts: a worthwhile, tractable endeavor. N. Phytol. 208, Making and Interpreting Soil Surveys, 2nd edition. (US Department of 324–329 (2015). Agriculture, Soil Conservation Service, 1999). 14. Wieder, W. R., Cleveland, C. C., Smith, W. K. & Todd-Brown, K. Future 42. Yang, X. & Post, W. M. Phosphorus transformations as a function of productivity and carbon storage limited by terrestrial nutrient availability. pedogenesis: a synthesis of soil phosphorus data using Hedley fractionation Nat. Geosci. 8, 441–U435 (2015). method. Biogeosciences 8, 2907–2916 (2011). 15. Yue, K. et al. Influence of multiple global change drivers on terrestrial carbon 43. Smeck, N. E. Phosphorus dynamics in soils and landscapes. Geoderma 36, storage: additive effects are common. Ecol. Lett. 20, 663–672 (2017). 185–199 (1985). 16. Luo, Y., Hui, D. & Zhang, D. Elevated CO stimulates net accumulations of 44. Hou, E. et al. A global dataset of phosphorus addition experiments in carbon and nitrogen in land ecosystems: a meta-analysis. Ecology 87,53–63 terrestrial ecosystems. Figshare https://doi.org/10.6084/m9.figshare.8969963 (2006). (2019). 17. Gurevitch, J. & Hedges, L. V. Statistical issues in ecological meta-analyses. 45. Borenstein, M., Hedges, L. V., Higgins, J. P. T. & Rothstein, H. R. A basic Ecology 80, 1142–1149 (1999). introduction to fixed-effect and random-effects models for meta-analysis. Res. 18. Gurevitch, J., Koricheva, J., Nakagawa, S. & Stewart, G. Meta-analysis and the Synth. Methods 1,97–111 (2010). science of research synthesis. Nature 555, 175–182 (2018). 46. R Core Team. R: A Language and Environment for Statistical Computing. 19. Koricheva, J. & Gurevitch, J. Uses and misuses of meta‐analysis in plant http://www.R-project.org/. (R Foundation for Statistical Computing, Vienna, ecology. J. Ecol. 102, 828–844 (2014). Austria, 2017). 20. Elith, J., Leathwick, J. R. & Hastie, T. A working guide to boosted regression 47. Greenwell, B., Boehmke, B., Cunningham, J. & GBM Developers. gbm: trees. J. Anim. Ecol. 77, 802–813 (2008). Generalized Boosted Regression Models. R package version 2.1.5. https:// 21. Reich, P. B. & Oleksyn, J. Global patterns of plant leaf N and P in relation to CRAN.R-project.org/package=gbm. (2019). temperature and latitude. Proc. Natl Acad. Sci. USA 101, 11001–11006 (2004). 48. Bivand, R. & Piras, G. Comparing implementations of estimation methods for 22. Yuan, Z. Y., Chen, H. Y. H. & Reich, P. B. Global-scale latitudinal patterns of spatial econometrics. J. Stat. Softw. 63,1–36 (2015). plant fine-root nitrogen and phosphorus. Nat. Commun. 2, 344 (2011). 23. Vitousek, P. M. Litterfall, nutrient cycling, and nutrient limitation in tropical Acknowledgements forests. Ecology 65, 285–298 (1984). We thank all of the authors whose data were included in our study and Marijke Heenan 24. Hou, E. et al. Effects of climate on soil phosphorus cycle and availability in and Bruce Jaffee for improving the language. This work was supported by the National natural terrestrial ecosystems. Glob. Change Biol. 24, 3344–3356 (2018). Natural Science Foundation of China (31870464; 41973076; 41471443; 41401326), the 25. Chapin, F. S., Vitousek, P. M. & Vancleve, K. The nature of nutrient limitation US Department of Energy (DE-SC00114085), National Science Foundation Grant DEB in plant-communities. Am. Nat. 127,48–58 (1986). 1655499, US Department of Energy, Terrestrial Ecosystem Sciences Grant DE- 26. Sterner, R. W. & Elser, J. J. Ecological Stoichiometry: The Biology of Elements SC0006982, and the subcontracts 4000158404 and 4000161830 from Oak Ridge National from Molecules to the Biosphere. (Princeton University Press, 2002). Laboratory (ORNL) to the Northern Arizona University. 27. Čapek, P. et al. A plant–microbe interaction framework explaining nutrient effects on primary production. Nat. Ecol. Evol. 2, 1588–1596 (2018). Author contributions 28. Viereck, L. A., Dyrness, C., Cleve, K. V. & Foote, M. J. Vegetation, soils, and E.H., Y.L. and D.W. designed the study. E.H. collected the data. E.H. and Y.L. analyzed forest productivity in selected forest types in interior Alaska. Can. J. Res 13, the data. E.H., Y.L., Y.K., C.C., X.Lu, L.J., X.Luo and D.W. contributed significantly to the 703–720 (1983). writing of the manuscript. 29. Ringeval, B., Nowak, B., Nesme, T., Delmas, M. & Pellerin, S. Contribution of anthropogenic phosphorus to agricultural soil fertility and food production. Glob. Biogeochem. Cycles 28, 743–756 (2014). Competing interests 30. Lu, C. & Tian, H. Global nitrogen and phosphorus fertilizer use for agriculture The authors declare no conflict of interest. production in the past half century: shifted hot spots and nutrient imbalance. Earth Syst. Sci. Data 9, 181–192 (2017). Additional information 31. Roy, E. D. et al. Soil phosphorus sorption capacity after three decades of Supplementary information is available for this paper at https://doi.org/10.1038/s41467- intensive fertilization in Mato Grosso, Brazil. Agr. Ecosyst. Environ. 249, 020-14492-w. 206–214 (2017). 32. Sterne, J. A. et al. Recommendations for examining and interpreting funnel Correspondence and requests for materials should be addressed to E.H. or D.W. plot asymmetry in meta-analyses of randomised controlled trials. BMJ 343, d4002 (2011). Peer review information Nature Communications thanks the anonymous reviewer(s) for 33. Lambers, H., Raven, J. A., Shaver, G. R. & Smith, S. E. Plant nutrient- their contribution to the peer review of this work. Peer reviewer reports are available. acquisition strategies change with soil age. Trends Ecol. Evol. 23,95–103 (2008). Reprints and permission information is available at http://www.nature.com/reprints 34. George, T. S. et al. Organic phosphorus in the terrestrial environment: a perspective on the state of the art and future priorities. Plant Soil 427, 191–208 Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in (2018). published maps and institutional affiliations. 35. Hou, E. et al. Soil pH predominantly controls the forms of organic phosphorus in topsoils under natural broadleaved forests along a 2500 km latitudinal gradient. Geoderma 315,65–74 (2018). Open Access This article is licensed under a Creative Commons 36. Tiessen, H., Cuevas, E. & Chacon, P. The role of soil organic matter in Attribution 4.0 International License, which permits use, sharing, sustaining soil fertility. Nature 371, 783–785 (1994). 37. Harpole, W. S. et al. Nutrient co-limitation of primary producer communities. adaptation, distribution and reproduction in any medium or format, as long as you give Ecol. Lett. 14, 852 (2011). appropriate credit to the original author(s) and the source, provide a link to the Creative 38. Penuelas, J. et al. Human-induced nitrogen-phosphorus imbalances alter Commons license, and indicate if changes were made. The images or other third party natural and managed ecosystems across the globe. Nat. Commun. 4, 2934 material in this article are included in the article’s Creative Commons license, unless (2013). indicated otherwise in a credit line to the material. If material is not included in the 39. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very high article’s Creative Commons license and your intended use is not permitted by statutory resolution interpolated climate surfaces for global land areas. Int. J. Climatol. regulation or exceeds the permitted use, you will need to obtain permission directly from 25, 1965–1978 (2005). the copyright holder. To view a copy of this license, visit http://creativecommons.org/ 40. Trabucco, A. & Zomer, R. Global aridity index (global-aridity) and global licenses/by/4.0/. potential evapo-transpiration (global-PET) geospatial database. https:// cgiarcsi.community/data/global-aridity-and-pet-database/. CGIAR Consortium for Spatial Information. (2009). © The Author(s) 2020 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 9 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nature Communications Springer Journals

Global meta-analysis shows pervasive phosphorus limitation of aboveground plant production in natural terrestrial ecosystems

Download PDF
9 pages

Loading next page...
 
/lp/springer-journals/global-meta-analysis-shows-pervasive-phosphorus-limitation-of-BVEa6qENQm

References (55)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2020
eISSN
2041-1723
DOI
10.1038/s41467-020-14492-w
Publisher site
See Article on Publisher Site

Abstract

ARTICLE https://doi.org/10.1038/s41467-020-14492-w OPEN Global meta-analysis shows pervasive phosphorus limitation of aboveground plant production in natural terrestrial ecosystems 1,2,3 3 1,2 4 1,2 3 Enqing Hou *, Yiqi Luo , Yuanwen Kuang , Chengrong Chen , Xiankai Lu , Lifen Jiang , 1,2 1,2 Xianzhen Luo & Dazhi Wen * Phosphorus (P) limitation of aboveground plant production is usually assumed to occur in tropical regions but rarely elsewhere. Here we report that such P limitation is more wide- spread and much stronger than previously estimated. In our global meta-analysis, almost half (46.2%) of 652 P-addition field experiments reveal a significant P limitation on aboveground plant production. Globally, P additions increase aboveground plant production by 34.9% in natural terrestrial ecosystems, which is 7.0–15.9% higher than previously suggested. In croplands, by contrast, P additions increase aboveground plant production by only 13.9%, probably because of historical fertilizations. The magnitude of P limitation also differs among climate zones and regions, and is driven by climate, ecosystem properties, and fertilization regimes. In addition to confirming that P limitation is widespread in tropical regions, our study demonstrates that P limitation often occurs in other regions. This suggests that previous studies have underestimated the importance of altered P supply on aboveground plant production in natural terrestrial ecosystems. Key Laboratory of Vegetation Restoration and Management of Degraded Ecosystems, South China Botanical Garden, Chinese Academy of Sciences, 2 3 Guangzhou 510650, China. Center of Plant Ecology, Core Botanical Gardens, Chinese Academy of Sciences, Guangzhou 510650, China. Center for Ecosystem Science and Society, Northern Arizona University, Flagstaff, AZ 86011, USA. Australian Rivers Institute, School of Environment and Science, Griffith University, Nathan, QLD 4111, Australia. *email: [email protected]; [email protected] NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w utrient limitation of aboveground plant production has from tropical to arctic regions, spanning over 131 in latitude 1–5 o o been widely acknowledged . In terrestrial ecosystems, (54.8 S–76.5 N), and occurring on all continents except Antarc- Nnitrogen (N) has been considered as the most important tica, where no data were available (Figs. 1–3). Phosphorus lim- 3,6 limiting nutrient of aboveground plant production ; phosphorus itation of aboveground plant production occurred on all studied (P) has also been viewed as important, but mainly in lowland continents (Figs. 1–3), although the proportion of P limitation 4,7 tropical regions where soils are generally strongly weathered . instances differed among ecosystem types (Supplementary This prevalent view, however, has been challenged by an Table 3). Globally, 301 of the 652 experiments (46.2% of all increasing number of significant P limitation cases in areas other experiments; 45.0% of experiments in natural terrestrial ecosys- 1,8,9 than the lowland tropical regions (e.g., tundra regions) . For tems; and 48.6% of experiments in croplands) revealed significant example, significant P limitation on aboveground plant produc- P limitation of aboveground plant production (Fig. 1). These tion has also been found in some temperate areas with strongly findings provide convincing evidence that P limitation of above- 10–12 weathered soils . Despite the important role of P in above- ground plant production in terrestrial ecosystems is a worldwide ground plant production, we still lack a clear understanding of phenomenon. where, to what degree, and under what conditions P limits The conventional notion that P limits aboveground plant 5,8,13 aboveground plant production over the global land surface . production mainly in tropical regions is based on the following As a consequence, none of the tens of models in the fifth phase of patterns: relative to temperate regions, tropical regions generally 2 21,22 the Coupled Model Intercomparison Project (CMIP5) archive have older and more weathered soils , higher plant N:P ratios , 22,23 represents terrestrial P biogeochemistry, which causes substantial higher plant P use efficiencies , and lower plant and soil P 2,24 uncertainty in estimates of strength of the terrestrial carbon (C) concentrations . All of these latitudinal patterns, however, can 13,14 sink through the 21st century . only indicate the relative magnitude rather than the actual Here we report the distribution, magnitude, and drivers of P magnitude of nutrient limitation across regions. The actual limitation of aboveground plant production in terrestrial ecosys- magnitude of nutrient limitation is determined most reliably by tems worldwide. To accomplish this, we use a global database of experiments in which the response of aboveground plant produc- 3,25 652 P-addition field experiments compiled from 285 papers tion to nutrient addition is quantified . Our meta-analysis of P- published between 1955 and 2017 (Supplementary Figs. 1a and 2). fertilization field experiments shows that P significantly limits The database includes P-addition experiments in all major types of aboveground plant production across tropical, subtropical, tempe- terrestrial ecosystems, including both natural terrestrial ecosys- rate, and (sub)artic regions, although both the magnitude of P tems (436 experiments in forests, grasslands, tundras, or wetlands) limitation and the percentage of P limitation instances were greater and croplands (216 experiments) (Supplementary Tables 1 and 2). in tropical and subtropical regions than in temperate and (sub)artic The number of P-addition experiments in natural terrestrial regions (Fig. 3 and Supplementary Table 3). ecosystems in this study is 3.8–8.8 times greater than the number The worldwide occurrence of P limitation of terrestrial 1,8,9,15 in the previous meta-analyses (N= 50–117) . In addition, aboveground plant production may be explained by the 1,26 41.7% of the experiments in our database were published after biochemical machinery shared by autotrophs , the great 2007 and few of these were included in previous syntheses dedi- variability of plant characteristics and environmental conditions cated to N–P interactions (Supplementary Fig. 2). The collected within regions , and the multiple pathways resulting in P experiments are located on all continents except Antarctica limitation of aboveground plant production . Researchers have (Supplementary Fig. 1a) and have wide ranges of mean annual proposed that the P and N demands of core biochemical −1 precipitation (MAP, 80–5302 mm yr ) and mean annual tem- machinery (mainly concerning rRNA and proteins) shared by all perature (MAT, −12.1 to 27.5 C) (Supplementary Table 1). photoautotrophs may cause plant growth to be limited by P and 1,26 Compared to previous datasets, this up-to-date dataset better N to a similar degree . Although ecosystem properties such as captures Earth’s diverse terrestrial habitats and thereby provides a soil P availability are key drivers of P limitation of aboveground 8,27 much clearer understanding of the role of P supply on above- plant production in terrestrial ecosystems , ecosystem proper- ground plant production. ties vary greatly across sites . For example, the soil P supply rate −2 −1 To explore the global distribution of P limitation, we first ranges from ~1 to > 10000 g m yr in both tropical and estimate a threshold value of P limitation, i.e., a critical P effect temperate regions . P limitation in tropical regions is often size that best corresponds to a critical z-score at P = 0.05, based attributed to the occlusion of P in soil by chemical or physical on the statistical results provided in the 285 papers that com- mechanisms, the chronic loss of dissolved inorganic and organic prised our database (see “Methods” section; Supplementary P by leaching, and/or the exhaustion of soil primary minerals Fig. 3). We then map the global distribution of significant and during long-term soil development . However, there are also non-significant P limitation cases. We quantify the magnitude of other pathways that can cause P limitation in different regions P limitation at the global scale as well as in various groups of and at different timescales (from years to millions of years), ecosystems using a meta-analysis approach typically used in including the formation of soil layers (e.g., iron pans) that ecological studies, i.e., the natural logarithm transformed physically prevent/inhibit access by roots to potentially available response ratio (Ln(RR)) of aboveground plant production to P P, transactional limitations in which the input of P by weathering additions weighted by the inverse variance (details in “Methods” is less than the input of other resources, low-P parent material, 16–19 section) . Finally, we explore the effects of climate, ecosystem sinks that reduce P levels, and anthropogenic increases in the properties, and fertilization regimes and their relative importance supply of other resources and especially N and atmospheric CO . in predicting the P effect size using a boosted regression tree Permafrost, for example, can isolate plants from deeper portions 20 4,28 method . In general, we show a more widespread and much of the soil profile in cold regions . Low-P parent materials stronger P limitation of aboveground plant production in natural explain P limitation in some temperate regions . These and other 1,8,9,15 terrestrial ecosystems than previously suggested . pathways (e.g. the precipitation of P with Ca in arid soils) can cause P limitation in many temperate and (sub)arctic regions. Results and discussion Globally distributed P limitation. Our synthesis revealed that P The magnitude of P limitation. In natural terrestrial ecosystems, limitation of aboveground plant production is globally distributed, P additions increased aboveground plant production over controls 2 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE P limitation in natural terrestrial ecosystems Significant (196) Non-significant (240) P limitation in croplands Significant (105) Non-significant (111) Fig. 1 Locations of the 652 experiments in which the effect of P addition on aboveground plant production was assessed. a Natural terrestrial ecosystems. b Croplands. Experiments were determined to have significant P limitation based on the Ln(Response Ratio). If the Ln(Response Ratio) was higher than a threshold value (0.23 for natural terrestrial ecosystems and 0.09 for croplands), it was considered a significant case (Z test, P < 0.05) of P limitation. Determination of the threshold values is described in the “Methods” section and is supported by the Supplementary Fig. 3. Numbers in brackets are the number of experiments in the indicated group. Source data are provided as a Source Data file. by an average of 34.9%, with a 95% confidence interval of to 41.1% (N = 44) and 43.0% (N = 46) of the 107 terrestrial P- 30.0%–40.1% (N = 436; Table 1). The estimates were robust given addition experiments in the study by Elser et al. were performed our large sample size, as suggested by our sensitivity tests (not in Europe and North America, respectively (Supplementary sensitive to outliers; Supplementary Fig. 4), publication bias tests Fig. 1b). The P effect size in North America (36.9%) in the (no significant publication bias; Supplementary Fig. 5a) and current study was close to the global average, but the P effect size temporal change test (a significant but minor change in effect size in Europe (21.7%) was much smaller than the global average and with publication year; Supplementary Fig. 6a). Our average the averages in Australia (50.6%), Asia (40.4%), and South (34.9%) was about two times greater than the average reported in America (37.7%) (Fig. 3). Therefore, the global magnitude of P a recent global meta-analysis (17.7%) that used the same meta- limitation is much larger than that based on data mainly from analysis method but with a much smaller sample size (N = 60). To Europe and North America. The experiments in Augusto et al. compare our meta-analysis with three other previous meta- were spread quite evenly over the global land surface (Supple- 1,8,9 analyses , we also calculated the magnitude of P limitation in mentary Fig. 1c). Their relatively low estimates might be partly the natural terrestrial ecosystems by weighting the Ln(RR) uni- explained by their use of a slightly different way (relative to our formly or by weighting the RR with the inverse variance (see study) to remove pseudo-replications, i.e., the latest measurement details in “Methods” section). The estimated averages (36.3% and for forests but the earliest measurement for other ecosystems. 47.6%, respectively) were again higher than those reported in the Moreover, all of the previous syntheses had a much smaller 8 1 9 previous meta-analyses (23.4% –29.3% and 31.7% , respectively, sample size (Table 1) and therefore a relatively poorer N= 50–117; Table 1). Moreover, we found that the P effect size representation of global natural terrestrial ecosystems than our increased with quantity of P added and with the experimental study (Supplementary Figs. 1, 4). duration (Fig. 3). Therefore, the magnitude of P limitation in the The P effect size was much smaller in croplands than in natural natural terrestrial ecosystems was even larger after the Ln(RR) was terrestrial ecosystems (Table 1) and was even smaller after it was weighted by the quantity of P added (40.5%) or experimental adjusted with the trim-and-fill method (4.1%; Supplementary duration (48.4%) (Table 1). Fig. 5b). The pattern holds true for most continents and climate The estimate for natural terrestrial ecosystems was lower in zones (Fig. 3), under all fertilization regimes (Fig. 3), on all types Elser et al. than in our study (Table 1). This perhaps because up of soils (Supplementary Fig. 7), and for all meta-analytic methods NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w Table 1 The magnitude of P limitation in natural terrestrial ecosystems is larger than previously estimated. Method This study Previous studies Effect size [Lower CI, Number of Effect size [Lower CI, Number of Upper CI] (%) experiments Upper CI] (%) experiments Natural terrestrial ecosystems Ln(RR) weighted by the inverse variance 34.9 [30.0, 40.1] 436 17.7 [11.1, 24.8] 60 Ln(RR) weighted uniformly 36.3 [31.0, 41.9] 436 29.3 [14.9, 45.4] 107 23.4 [16.8, 30.4] 117 RR weighted by the inverse variance 47.6 [34.5, 60.8] 436 31.7 [21.6, 41.8] 50 Ln(RR) weighted by P addition amount 40.5 [27.5, 54.8] 436 Ln(RR) weighted by experimental duration 48.4 [33.5, 64.9] 436 Croplands Ln(RR) weighted by the inverse variance 13.9 [11.1, 16.8] 216 Ln(RR) weighted uniformly 16.3 [13.0, 19.7] 216 RR weighted by the inverse variance 15.4 [7.0, 23.7] 216 Ln(RR) weighted by P addition amount 24.7 [11.2, 39.8] 216 Ln(RR) weighted by experimental duration 25.9 [12.0, 41.6] 216 The magnitude of P limitation in the natural terrestrial ecosystems and also in the croplands was calculated using the five methods listed in the table. CI indicates confidence interval. Ln(RR) indicates ln transformed response ratio. (Table 1). The pattern can be at least partly explained by the fertilization regimes, soil properties, and plant properties each higher availability of soil P in croplands than in natural terrestrial explained some percentage (9.1%–40.0%) of the total explained ecosystems (Supplementary Fig. 8). Before P-addition experi- variation in P effect size in both the natural terrestrial ecosystems 2 2 ments were performed, some of the croplands likely received P (R = 0.59) and the croplands (R = 0.79) (Fig. 4 and Supple- fertilizers that increased soil P availability, although no pre- mentary Figs. 9−11. That the P effect size is regulated by multiple experiment fertilization was recorded in the source literature (see factors rather than by any single factor once again explains why P “Methods” section for the selection criteria). In croplands in limitation is widespread on the globe. For example, P limitation France, for example, an average of 82% of soil P was estimated to can occur in the tropical and subtropical natural ecosystems due have originated from former fertilization applications . Our to the high temperatures and precipitation that drive plant P results, therefore, suggest that P limitation in croplands has been demand and to the low soil extractable P concentration that limits 30,31 24 largely alleviated by historical fertilizations and that a soil P supply (Supplementary Fig. 10a, c, d). In contrast, the reduced amount of P fertilizer is needed to increase crop occurrence of P limitation in the temperate and (sub)arctic nat- production in the future. It follows that to accurately predict the ural ecosystems may be attributed to their generally high soil future fertilizer effect on crop production, models require organic matter contents and pH values (Supplementary Fig. 10e, fertilization history. Our results concerning differences between h). High soil organic matter content can reduce soil P availability croplands and natural terrestrial ecosystems may also be related by occluding P in the organic forms and by enhancing microbial 25,27,34 to the lower soil organic matter contents and shorter experi- immobilization of P in the soil . High soil pH can reduce mental durations in croplands (Fig. 3 and Supplementary soil P sorption capacity and thus increase the use efficiency of P Table 1). fertilizer by plants (i.e., increase the P effect size). Moreover, A positive asymmetric distribution of the P effect size was both high soil organic matter content and moderate soil pH can observed in the croplands (Supplementary Fig. 5b). This does not enhance the availability of nutrients such as N, potassium, and 35,36 necessarily indicate a publication bias (i.e., the tendency of journals calcium in soils , which may exaggerate the response of plant 32 9,31,37 to favor the publication of statistically significant results) . growth to P addition . Fertilization experiments in croplands typically have multiple As our study did for P, the study of LeBauer and Treseder nutrient (e.g., P, N, and K) treatments, and P addition is only one examined plant responses to N additions alone; they found a of the multiple treatments. Therefore, there is no apparent global distribution of N limitation on terrestrial primary 1 37 tendency of journals to favor publication of statistically significant production. Elser et al. and Harpole et al. , however, reported P effects. Instead, the asymmetric distribution may mainly result a prevalent co-limitation of terrestrial primary production by N from a true heterogeneity . The growth of plants in some regions and P. Although the results seem conflicting (i.e., globally is known to be strongly limited by P supply (e.g., Australia and distributed P limitation, N limitation, or N and P co-limitation), 4,7 27,37 lowland tropical areas with strongly weathered soils) .However, they can be reconciled by the multiple limitation hypothesis . we are unaware of any large, negative P effect on plant growth at A prevalent co-limitation of terrestrial primary production by N the community level, although there are rare reports of P toxicity and P suggests a generally balanced N and P limitation in global 1,37 symptoms in some plants that have adapted to low soil P terrestrial ecosystems , while a globally distributed N limitation 33 3,6 availability and of P-driven limitation of plant growth by N via indicate widespread N limitation in terrestrial ecosystems . soil microbes in some N-limited ecosystems .Therefore,the Given that both are reasonable, a worldwide occurrence of P detection of many more positive P effects than negative P effects is limitation in terrestrial ecosystems is expected, as observed in the reasonable (Supplementary Fig. 5). present study; the absence of a worldwide occurrence of P limitation would either imply an imbalance of N and P limitation, which would counter the finding of prevalent N and P co- Predictors of the magnitude of P limitation. Although our limitation (if widespread N limitation is true), or imply a less results show that P limitation is a global phenomenon, the widespread N limitation, which would counter the finding of magnitude of P limitation did vary greatly among experiments, globally distributed N limitation (if prevalent N and P co- with the Ln(RR) ranging from −0.48 to 2.44 (Fig. 2). Climate, limitation is true). Furthermore, the widespread P limitation 4 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE 2.5 ab Forest (52.5%) Grassland (36.8%) 2.0 Tundra (43.8%) Wetland (52.9%) 1.5 1.0 0.5 0.0 –0.5 020 40 60 80 0 20 40 60 80 100 1.5 cd Cropland (48.6%) 1.0 0.5 0.0 –0.5 0 20 40 60 80 0 20 40 60 80 100 Absolute latitude (°) Accumulated proportion (%) Fig. 2 Consistent occurrence of significant P limitation in all types of ecosystems. a Significant P limitation occurred in the natural terrestrial ecosystems at almost all latitudes. The magnitude of P limitation decreased with latitude in wetlands (meta-regression, R = 0.35, P < 0.05, N = 85) but not in any other type of natural terrestrial ecosystem (meta-regression, P > 0.05). b Significant P limitation occurred in all types of natural terrestrial ecosystems. Values in the brackets indicate the percentage of significant P limitation cases in the total sample size in each type of natural terrestrial ecosystem. ° ° c Significant P limitation occurred in croplands at all explored latitudes (absolute latitude between 0.1 and 56.5 ), and the magnitude of P limitation decreased with absolute latitude (meta-regression, R = 0.16, P < 0.05, N = 216). d Significant P limitation was found in 48.6% of the P-fertilization experiments in the croplands. As in Fig. 1, statistical significance of P limitation was assessed based on the Ln(Response Ratio). Dashed lines in all plots (0.23 in (a) and (b), 0.09 in (c) and (d)) indicate where the magnitude of P limitation approximates a 0.05 significance level (Z test). In (a) and (c), the size of each point is proportional to the weight used for meta-regression analysis. Source data are provided as a Source Data file. identified in this study and the widespread N limitation reported plants in natural terrestrial ecosystems (Fig. 3 and Supplementary in LeBauer and Treseder together imply that P limitation and N Fig. 10). This may lead to an underestimation of the global limitation are largely independent of each other . This possibility average of P limitation in natural terrestrial ecosystems. More- is supported by another synthesis study, which reported that the over, additional uncertainties can be introduced by the statistical effects of N supply and P supply on aboveground plant analyses (Table 1), missing variances of aboveground plant production are additive in most terrestrial ecosystems . Taken production (Supplementary Table 4), and the various measures of together, the evidence suggests the worldwide occurrence of both aboveground plant production used in different experiments P limitation and N limitation in terrestrial ecosystems , which (Supplementary Table 5). Missing measurements of ecosystem supports the multiple limitation hypothesis and challenges properties such as soil extractable P concentration and pH 27,37 Liebig’s Law of the Minimum . (Supplementary Table 1) can lead to an underestimation of their Although our dataset is much larger than those in previous relative importance in predicting the magnitude of P limitation. syntheses, there is still a large uncertainty in our estimate of the Third and last, given the long span of time of the datasets global magnitude of P limitation in terrestrial ecosystems. There (Supplementary Fig. 2), the nature of nutrient limitation has are three sources of this uncertainty. First, while ecosystems in likely changed over much of the land covered in this analysis, due, Australia, East Asia, and West Asia are better represented in our for example, to the changes in atmospheric N deposition and to dataset than in the previous ones, ecosystems in North Asia and changes in fertilization practices in the croplands . In spite of the tropics are still largely underrepresented (Supplementary these uncertainties, our large dataset of P addition experiments Fig. 1a). More experiments from North Asia may lower the global provides a much clearer pattern of the global distribution of P averages of P limitation, while more experiments from the tropics limitation and a more robust estimate of the global magnitude of would likely increase the global averages. Mature mixed forests P limitation in terrestrial ecosystems than previous datasets. were also underrepresented (Supplementary Table 3), and the Our findings have important implications for understanding inclusion of an increased number of forests may lower the global the role of P supply in controlling aboveground plant production average of the natural terrestrial ecosystems (Supplementary in terrestrial ecosystems. The results show a more widespread and Fig. 7). Second, most experiments were performed for ≤ 10 yrs much stronger limitation of aboveground plant production by P −1 and with a cumulative P addition ≤ 500 kg ha , which may not in natural terrestrial ecosystems than previously thought. The be long enough or high enough to fully stimulate the growth of results confirm the necessity of incorporating P limitation in NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 5 Ln (Response ratio) Ln (Response ratio) Vegetation Soil weathered P addition amount Experimental Fertilizer type –1 (kg ha ) duration (year) ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w Forest Australia Grass Asia Tundra South America Wetland North America Europe Strongly Africa Intermediately Slightly ≤1000 m >1000 m >500 200–500 50–200 Tropic ≤50 Subtropic Temperate >10 (Sub)arctic 5–10 1–5 Humid ≤1 Sub-humid Dry subhumid SSP Semi-arid TSP Arid Others 020 40 60 0 20 40 60 80 100 Effect size (%) Effect size (%) Fig. 3 P limitation was significant in all regions and major types of ecosystems and under various fertilization regimes. Exceptions are the non- significant P limitation in two groups of experiments in the croplands that had a small sample size (N = 8). Natural terrestrial ecosystems are shown in green (total N= 436), and croplands are shown in yellow (total N = 216). Values represent effect sizes ± 95% confidence intervals. The size of each point is proportional to the sample size (sample sizes are listed in Supplementary Table 3). The dashed lines indicate the no-fertilization effect. SSP is single superphosphate, and TSP is triple superphosphate. Source data are provided as a Source Data file. Natural terrestrial ecosystems Croplands 25 25 20 18.5 20 17.8 16.5 14.6 15.1 15 15 12.7 12.0 12.0 11.9 11.5 10.2 9.2 10 10 8.6 8.9 6.2 5.0 5 5 3.8 2.7 2.0 1.0 0 0 Fig. 4 Relative influence of climate, fertilization regimes, and ecosystem properties on the magnitude of P limitation. a Natural terrestrial ecosystems. b Croplands. The number above each bar indicates the percentage of the total explained variation accounted for by the variable. Fertilization regimes are in pink, climate factors are in blue, soil properties are in gray, and vegetation properties are in green. Source data are provided as a Source Data file. 6 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications Mean annual temperature P addition amount Mean annual precipitation Soil available P Soil organic C Experimental duration Vegetation type Soil pH Fertilizer type N fixation vegetation Soil available P Mean annual temperature P addition amount Crop species Soil organic C Soil pH Mean annual precipitation Fertilizer type Experimental duration N fixation crop Aridity Climate zone Altitude Continent Relative influence (%) NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE 13,14 earth system models . The results also show a much smaller P treatments (concentrations of available P, organic C, and total N; pH in water; and particle size), and parent material type. For each experiment in forest ecosystems, fertilizer effect in croplands than in natural terrestrial ecosystems, forest composition (i.e., pure or mixed forest) and the average forest age during the which suggests that P limitation in croplands has been generally experiment were also recorded. alleviated by historical fertilizations. Finally, the co-regulation of plant response to altered P supply by climate, ecosystem Data preparation. In cases where the referenced studies did not report the latitude properties, and fertilization regimes highlights the importance or longitude of the P-addition experiment (52% of the studies did not report both of taking a systems approach to study how nutrient supply affects latitude and longitude), the approximate latitude and longitude were derived by geocoding the name of the location in Google Earth 7.0 (the free version). In cases aboveground plant production. where the referenced studies did not report MAT (76%), MAP (54%), or altitude (65%), the values were derived from WorldClim using site geographic location (i.e. latitude and longitude). The aridity index (AI) of each site was obtained from Methods 40 CGIAR-CSI using data from WorldClim ; the AI value decreases as aridity Data collection. With the aim of constructing a comprehensive database of the increases. experimentally determined effects of P additions on aboveground plant production Soil type was classified according to the U.S. Department of Agriculture soil in global terrestrial ecosystems, we collected as many experiments that fulfilled our classification system . Soils were grouped based their degree of weathering 42,43 criteria (described below) as possible. Relevant studies were identified by searching according to previous studies : Andisols, Histosols, Entisols, and Inceptisols ISI Web of Knowledge, Google Scholar, and China Knowledge Resource Integrated were considered to be slightly weathered soils; Aridsols, Vertisols, Mollisols, and Database using combinations of keywords such as “phosph* addition”, “phosph* Alfisols were considered to be intermediately weathered soils; and Spodosols, fertili*”, “phosph* enrich*”, “aboveground biomass”, “primary product*”, “crop Ultisols, and Oxisols were considered to be strongly weathered soils. Parent yield”, and “grain yield”. Our survey also included studies summarized in pre- material types were grouped into four geological classes according to a previous viously published syntheses and the subsequent relevant studies citing those study : acid, intermediate, mafic, and calcareous rocks. syntheses. A PRISMA flow diagram (Supplementary Fig. 12) shows the procedure For comparison of P effect sizes among regions, experiments in the database we used for the selection of studies. were grouped in four different ways. First, experiments were grouped according to To be included in our database, published experiments were required to satisfy their continental locations: Australia, Asia, Africa, Europe, North America, and the following criteria: (1) the P-addition experiment was conducted in the field and South America. Second, experiments were grouped based on absolute latitude into o o o included P-addition and control treatments within the same ecosystem under the four latitude belts or regions: tropic (23.4 S–23.4 N), subtropic (23.4–35 Sor same environmental conditions, and also included measures of aboveground plant o o o o o N), temperate (35–50 Sor N), and (sub)arctic (>50 Sor N). Third, production in both P-addition and control treatments; (2) no fertilization was experiments were grouped according to altitude into low-altitude experiments recorded in the control treatment either before the start of the experiment or (< 1000 m a.s.l.) and high-altitude experiments (≥1000 m a.s.l.). Finally, during the experiment; (3) the P treatment received a P fertilizer that did not experiments were divided based on site aridity level into five groups: arid (AI ≤ contain N so as to avoid the effect of N; as a result experiments with application of 0.20), semi-arid (0.20 < AI ≤ 0.50), dry subhumid (0.50 < AI ≤ 0.65), sub-humid ammonium phosphate, manure or other fertilizers were excluded. (0.65 < AI ≤ 1.0), and humid (1.0 < AI). The complete dataset is available at To be considered an experiment in our analysis, a reported experiment had to Figshare . be temporally and spatially distinct and had to have internally consistent controls. Multiple experiments could be reported by one publication; for instance, the Phosphorus limitation threshold. One major objective of our study was to map application of the same experimental treatments was considered to represent the global distribution of experiments in which P significantly limited aboveground multiple experiments if the treatments were applied at several sites with different plant production. To do this, we had to define a threshold value that separated vegetation types. When multiple measures were reported over time at a single experiments that did or did not find significant P limitation. We estimated the experimental site, we used the latest measure. When multiple levels of P fertilizer threshold value separately for the natural terrestrial ecosystems and the croplands, treatments were reported, we used the measure with the highest amount of P using a method described in a recent study . In general, we first collected the addition. Choosing the latest measure and the highest P addition amount increased reported statistical responses of aboveground plant production to P additions from the likelihood that P additions fulfill plant demand and overcome the sorption of P 25,27 the source references. We then investigated the distribution of the Ln(RR) values. fertilizer by soils and soil microbial competition for P fertilizer . When multiple Finally, we identified the threshold value of the Ln(RR) that optimizes the dis- forms of P fertilizers were tested, we chose the treatment of single superphosphate tinction between statistically significant positive P effects and statistically non- or triple superphosphate if available. significant P effects. Of the 128 experiments in the natural terrestrial ecosystems We included only experiments that reported the response of community-level that reported a significant P limitation, 84% had an Ln(RR) value ≥ 0.23 (Supple- aboveground plant production to P additions. Single-species responses were not mentary Fig. 3a). Similarly, of the 162 experiments in the natural terrestrial eco- included unless drawn from a mono-dominant community. If several species from systems that reported a non-significant P effect, 85% had an Ln(RR) < 0.23 a community were individually assayed, an average across all species was used. (Supplementary Fig. 3b). When the two groups were combined, the maximum Experiments in forest or savanna ecosystems that only reported the response of percentage (84%) of correct classification (i.e., a significant positive effect was understory or herbaceous response to P additions were not included. Experiments classified as a significant case and a non-significant effect was classified as a non- with only stand biomass responses were excluded unless the stand biomass data significant case) was obtained with an Ln(RR) value of 0.23 (Supplementary could be used to calculate aboveground plant biomass production. Fig. 3c). Therefore, 0.23 was used as the threshold Ln(RR) value to distinguish Ecosystems were classified as forest, grassland, tundra, wetland, or cropland; significant from non-significant P limitation in natural terrestrial ecosystems. This natural forests, plantations, shrublands, and savannas were all classified as forest. In threshold value is close to the one used in a previous study (Ln(RR) of 0.20) .A forest ecosystems, beside aboveground plant biomass production (N = 33), we also similar approach was applied to the P-addition experiments in the croplands, such accepted proxy variables that are known to be correlated with aboveground plant that 0.09 was used as the threshold Ln(RR) value for croplands (Supplementary biomass production, such as litterfall production (1) and the rate of increase in tree Fig. 3d–f). diameter (34), stem volume (25), basal area (25), or height (16) (Supplementary Table 5). We showed that the weighted Ln(RR) did not differ significantly among the various variables used (Supplementary Table 5). In croplands, beside Meta-analysis. We quantified the magnitude of P limitation at the global scale and aboveground plant biomass production (N = 85), we also accepted marketable in various groups of ecosystems by weighting the Ln(RR) with the inverse variance yield (131), because we found that marketable yield responded similarly to P 16–19 and a random-effect model . To do this, we extracted means, standard additions as aboveground plant production in the croplands based on studies that deviations (SDs), and sample sizes (n) from the published studies. If standard error reported both measures (Supplementary Fig. 13). In tundras, beside aboveground (SE) rather than SD was reported, SD was calculated: biomass production (N = 10), we also included leaf mass per tiller (4), tiller pffiffiffi SD ¼ SE n ð1Þ biomass (1), and plot level NDVI (1) (Supplementary Table 5). In wetlands, beside aboveground biomass production (N = 72), we also included height increase (5), If neither SD nor SE was reported, we approximated the missing SD by multiplying leaf area index (3), the production of whole plants (3), and chamber based gross the reported mean by the average coefficient of variance of our complete dataset. If primary production (2) (Supplementary Table 5). sample size was not reported, we assigned sample sizes as the median sample size of In total, we collected data from 652 P addition experiments reported in 285 our complete dataset. We approximated the SDs and the sample sizes separately for published papers, including 436 experiments from natural terrestrial ecosystems the natural terrestrial ecosystems and the croplands and also separately for the (including forests, grasslands, tundras, and wetlands) and 216 experiments from control and the P-addition treatments (see details in Supplementary Table 4). croplands (see experimental locations in Supplementary Fig. 1a). Beside The Ln(RR) of an experiment was calculated as follows: aboveground plant production measures, our database also included site characteristics and fertilization regimes, which were used to explain the variation in LnðÞ RR ¼ Ln ¼ LnðÞ X  LnðÞ X ð2Þ Ln(RR). Site characteristics included site location (latitude and longitude), climate t c variables (MAT and MAP), topographical conditions (altitude and slope), plant characteristics (vegetation type, and symbiotic N fixation), soil type, soil where X . and X . are mean aboveground plant production in the P treatment and t c physiochemical properties before the experiments began or from the control control, respectively. NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w The weighted mean response ratio (Ln(RR )) of a group of ecosys was follows: fertilization regimes in predicting the Ln(RR) in the natural terrestrial ecosystems and in the croplands. Before BRT analyses, variable selections were made to avoid w ´ LnðRR Þ i¼1 i i high correlations among predictors. Specifically, (1) Soil organic C concentration Ln RR ¼ ð3Þ i¼1 i was included as an indicator of soil organic matter content, while soil total N concentration was not included in the BRT models due to its high correlation with where m is the number of experiments in the group (e.g., a region), and w . is the soil organic C concentration (Natural terrestrial ecosystems: r = 0.85, P < 0.001, weighting factor of the ith experiment in the group. The w . was calculated as N = 92; Croplands: r = 0.95, P < 0.001, N = 54). (2) Soil available P concentration follows: was included as an indicator of soil P availability, while soil total P concentration was not included. Ecosystem properties such as soil particle size and parent w ¼ ð4Þ v material type were not included in our BRT analyses due to the very large pro- portions of missing data (Supplementary Table 1), including which can bias the where v . is the variance of study (i) in the group. The v . was calculated as follows: i i estimate of their relative importance. Parameter values used for the BRT analyses generally followed the v ¼ v þ T ð5Þ i i recommendation of a previous study , i.e. bag fraction as 0.75, the number of where v is the within-study variance of study (i), and T is the between-studies cross validation as 10, and tree complexity as 2. Learning rate was set at a small variance. The v was calculated as follows: value (i.e., 0.005) to include a large number (>1000) of regression trees in the 2 2 models. Because Ln(RR) is a continuous numerical variable, a Gaussian S S t c v ¼ þ ð6Þ distribution of errors was used. The relative importance of each predictor 2 2 n X n X t t c represented a percentage of the total variation explained by the models. The BRT analyses were performed with the “gbm” package version 2.1.5 plus the custom where n and n are the sample sizes for the P treatment and the control groups, t c respectively, and S and S are the standard deviations for the P treatment and the code of another study in R version 3.3.1. For evaluation of the spatial structure of t c control groups, respectively, of study (i). The calculation of T can be seen in the BRT residuals, the global Moran’s I statistic was applied to determine the Borenstein et al. . significance using the “spdep” package version 0.7.7 . The standard error of the Ln(RR ) was calculated as: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Reporting summary. Further information on research design is available in s Ln RR ¼ P ð7Þ the Nature Research Reporting Summary linked to this article. þ m i¼1 i The 95% confidence interval (CI) for the Ln(RR ) was calculated as follows: + Data availability All data used in this study are available at Figshare (https://doi.org/10.6084/m9. 95%CI ¼ Ln RR ±1:96s Ln RR ð8Þ þ þ figshare.8969963). The source data underlying all Figures except Supplementary Figs. 5, If the 95% CI did not overlap with zero, the overall P addition effect in the group of 10, and 11 are also available on the above web page. Supplementary Figs. 5, 10, and 11 are experiments was considered significant. The percentage change in aboveground directly created using R functions, as described in Methods. plant production induced by P addition (i.e., the effect size) in a group of ecosystems was measured as follows: Code availability Effect sizeðÞ % ¼ðexp Ln RR 1Þ100% ð9Þ þ The code used in this study is available at Figshare (https://doi.org/10.6084/m9. figshare.8969963). The meta-analyses were performed using “meta” package in R version 3.3.1 . 1,8,9 To compare our analyses with the previous meta-analyses , we also calculated the global magnitude of P limitation using two other methods: Received: 24 July 2019; Accepted: 14 January 2020; 1,8 (1) Ln(RR) weighted uniformly , where effect size only depends on the means of control and P treatment groups. (2) RR weighted by the inverse variance , where the RR rather than the Ln(RR) is weighted by the inverse variance. We also calculated the global magnitude of P limitation by considering the effects of the quantity of P added and experimental duration: References (3) Ln(RR) weighted by the quantity of P added. The weight of each experiment 1. Elser, J. J. et al. Global analysis of nitrogen and phosphorus limitation of was calculated as follows: primary producers in freshwater, marine and terrestrial ecosystems. Ecol. Lett. 10, 1135–1142 (2007). w ¼ P ð10Þ P 2. Aerts, R. & Chapin, F. S. The mineral nutrition of wild plants revisited: a re- i¼1 i evaluation of processes and patterns. Adv. Ecol. Res 30,1–67 (1999). −1 where P is the cumulative quantity of P added during the experiment (kg ha ), 3. Vitousek, P. M. & Howarth, R. W. Nitrogen limitation on land and in the sea - and M is the number of experiments in the natural terrestrial ecosystems (436) how can it occur? Biogeochemistry 13,87–115 (1991). or the croplands (216). 4. Vitousek, P. M., Porder, S., Houlton, B. Z. & Chadwick, O. A. Terrestrial (4) Ln(RR) weighted by experimental duration. The weight of each experiment phosphorus limitation: mechanisms, implications, and nitrogen-phosphorus was calculated as follows: interactions. Ecol. Appl. 20,5–15 (2010). 5. Turner, B. L., Brenes-Arguedas, T. & Condit, R. Pervasive phosphorus w ¼ P ð11Þ limitation of tree species but not communities in tropical forests. Nature 555, i¼1 367–370 (2018). 6. LeBauer, D. S. & Treseder, K. K. Nitrogen limitation of net primary where D is the duration of the experiment (yr). productivity in terrestrial ecosystems is globally distributed. Ecology 89,371–379 We checked the quality of our meta-analysis according to the checklist of 19 (2008). Koricheva and Gurevitch . Our meta-analysis fulfilled all of the quality criteria for 7. Vitousek, P. M. & Farrington, H. Nutrient limitation and soil development: a meta-analysis in plant ecology (Supplementary Table 6). We performed two experimental test of a biogeochemical theory. Biogeochemistry 37,63–75 sensitivity tests (Supplementary Fig. 4). One was the leave-one-out meta-analysis (1997). using the “metainf” function. The other was the cumulative meta-analysis using a 8. Augusto, L., Achat, D. L., Jonard, M., Vidal, D. & Ringeval, B. Soil parent random-effect model in the “metacum” function based on inverse-variance material - a major driver of plant nutrient limitations in terrestrial ecosystems. weighted Ln(RR). The cumulative meta-analysis was repeated 1000 times with Glob. Change Biol. 23, 3808–3824 (2017). random orders of experiments. We also ran a cumulative meta-analysis using the 9. Li, Y., Niu, S. & Yu, G. Aggravated phosphorus limitation on biomass “metacum” function based on the uniformly weighted Ln(RR). We created funnel plots to detect possible publication bias using the “funnel” function. The possible production under increasing nitrogen loading: a meta‐analysis. Glob. Change publication bias was statistically tested using the “metabias” function. If a Biol. 22, 934–943 (2016). publication bias was suggested, we further adjusted the P effect size by the trim- 10. Laliberté, E. et al. Experimental assessment of nutrient limitation along a 2‐ and-fill method using the “trimfill” function. Finally, we explored the relationships million‐year dune chronosequence in the south‐western Australia biodiversity between the Ln(RR) and the absolute value of latitude and publication year using hotspot. J. Ecol. 100, 631–642 (2012). the “metareg” function. All the above functions were from the “meta” package in R 11. Peltzer, D. A. et al. Understanding ecosystem retrogression. Ecol. Monogr. 80, version 3.3.1 . 509–529 (2010). 12. Wardle, D. A., Walker, L. R. & Bardgett, R. D. Ecosystem properties and forest decline in contrasting long-term chronosequences. Science 305,509–513 Boosted regression tree analysis. Boosted regression tree (BRT) analyses were (2004). conducted to quantify the relative importance of climate, ecosystem properties, and 8 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14492-w ARTICLE 13. Reed, S. C., Yang, X. & Thornton, P. E. Incorporating phosphorus cycling into 41. Soil Survey Staff. Soil Taxonomy: A Basic System of Soil Classification for global modeling efforts: a worthwhile, tractable endeavor. N. Phytol. 208, Making and Interpreting Soil Surveys, 2nd edition. (US Department of 324–329 (2015). Agriculture, Soil Conservation Service, 1999). 14. Wieder, W. R., Cleveland, C. C., Smith, W. K. & Todd-Brown, K. Future 42. Yang, X. & Post, W. M. Phosphorus transformations as a function of productivity and carbon storage limited by terrestrial nutrient availability. pedogenesis: a synthesis of soil phosphorus data using Hedley fractionation Nat. Geosci. 8, 441–U435 (2015). method. Biogeosciences 8, 2907–2916 (2011). 15. Yue, K. et al. Influence of multiple global change drivers on terrestrial carbon 43. Smeck, N. E. Phosphorus dynamics in soils and landscapes. Geoderma 36, storage: additive effects are common. Ecol. Lett. 20, 663–672 (2017). 185–199 (1985). 16. Luo, Y., Hui, D. & Zhang, D. Elevated CO stimulates net accumulations of 44. Hou, E. et al. A global dataset of phosphorus addition experiments in carbon and nitrogen in land ecosystems: a meta-analysis. Ecology 87,53–63 terrestrial ecosystems. Figshare https://doi.org/10.6084/m9.figshare.8969963 (2006). (2019). 17. Gurevitch, J. & Hedges, L. V. Statistical issues in ecological meta-analyses. 45. Borenstein, M., Hedges, L. V., Higgins, J. P. T. & Rothstein, H. R. A basic Ecology 80, 1142–1149 (1999). introduction to fixed-effect and random-effects models for meta-analysis. Res. 18. Gurevitch, J., Koricheva, J., Nakagawa, S. & Stewart, G. Meta-analysis and the Synth. Methods 1,97–111 (2010). science of research synthesis. Nature 555, 175–182 (2018). 46. R Core Team. R: A Language and Environment for Statistical Computing. 19. Koricheva, J. & Gurevitch, J. Uses and misuses of meta‐analysis in plant http://www.R-project.org/. (R Foundation for Statistical Computing, Vienna, ecology. J. Ecol. 102, 828–844 (2014). Austria, 2017). 20. Elith, J., Leathwick, J. R. & Hastie, T. A working guide to boosted regression 47. Greenwell, B., Boehmke, B., Cunningham, J. & GBM Developers. gbm: trees. J. Anim. Ecol. 77, 802–813 (2008). Generalized Boosted Regression Models. R package version 2.1.5. https:// 21. Reich, P. B. & Oleksyn, J. Global patterns of plant leaf N and P in relation to CRAN.R-project.org/package=gbm. (2019). temperature and latitude. Proc. Natl Acad. Sci. USA 101, 11001–11006 (2004). 48. Bivand, R. & Piras, G. Comparing implementations of estimation methods for 22. Yuan, Z. Y., Chen, H. Y. H. & Reich, P. B. Global-scale latitudinal patterns of spatial econometrics. J. Stat. Softw. 63,1–36 (2015). plant fine-root nitrogen and phosphorus. Nat. Commun. 2, 344 (2011). 23. Vitousek, P. M. Litterfall, nutrient cycling, and nutrient limitation in tropical Acknowledgements forests. Ecology 65, 285–298 (1984). We thank all of the authors whose data were included in our study and Marijke Heenan 24. Hou, E. et al. Effects of climate on soil phosphorus cycle and availability in and Bruce Jaffee for improving the language. This work was supported by the National natural terrestrial ecosystems. Glob. Change Biol. 24, 3344–3356 (2018). Natural Science Foundation of China (31870464; 41973076; 41471443; 41401326), the 25. Chapin, F. S., Vitousek, P. M. & Vancleve, K. The nature of nutrient limitation US Department of Energy (DE-SC00114085), National Science Foundation Grant DEB in plant-communities. Am. Nat. 127,48–58 (1986). 1655499, US Department of Energy, Terrestrial Ecosystem Sciences Grant DE- 26. Sterner, R. W. & Elser, J. J. Ecological Stoichiometry: The Biology of Elements SC0006982, and the subcontracts 4000158404 and 4000161830 from Oak Ridge National from Molecules to the Biosphere. (Princeton University Press, 2002). Laboratory (ORNL) to the Northern Arizona University. 27. Čapek, P. et al. A plant–microbe interaction framework explaining nutrient effects on primary production. Nat. Ecol. Evol. 2, 1588–1596 (2018). Author contributions 28. Viereck, L. A., Dyrness, C., Cleve, K. V. & Foote, M. J. Vegetation, soils, and E.H., Y.L. and D.W. designed the study. E.H. collected the data. E.H. and Y.L. analyzed forest productivity in selected forest types in interior Alaska. Can. J. Res 13, the data. E.H., Y.L., Y.K., C.C., X.Lu, L.J., X.Luo and D.W. contributed significantly to the 703–720 (1983). writing of the manuscript. 29. Ringeval, B., Nowak, B., Nesme, T., Delmas, M. & Pellerin, S. Contribution of anthropogenic phosphorus to agricultural soil fertility and food production. Glob. Biogeochem. Cycles 28, 743–756 (2014). Competing interests 30. Lu, C. & Tian, H. Global nitrogen and phosphorus fertilizer use for agriculture The authors declare no conflict of interest. production in the past half century: shifted hot spots and nutrient imbalance. Earth Syst. Sci. Data 9, 181–192 (2017). Additional information 31. Roy, E. D. et al. Soil phosphorus sorption capacity after three decades of Supplementary information is available for this paper at https://doi.org/10.1038/s41467- intensive fertilization in Mato Grosso, Brazil. Agr. Ecosyst. Environ. 249, 020-14492-w. 206–214 (2017). 32. Sterne, J. A. et al. Recommendations for examining and interpreting funnel Correspondence and requests for materials should be addressed to E.H. or D.W. plot asymmetry in meta-analyses of randomised controlled trials. BMJ 343, d4002 (2011). Peer review information Nature Communications thanks the anonymous reviewer(s) for 33. Lambers, H., Raven, J. A., Shaver, G. R. & Smith, S. E. Plant nutrient- their contribution to the peer review of this work. Peer reviewer reports are available. acquisition strategies change with soil age. Trends Ecol. Evol. 23,95–103 (2008). Reprints and permission information is available at http://www.nature.com/reprints 34. George, T. S. et al. Organic phosphorus in the terrestrial environment: a perspective on the state of the art and future priorities. Plant Soil 427, 191–208 Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in (2018). published maps and institutional affiliations. 35. Hou, E. et al. Soil pH predominantly controls the forms of organic phosphorus in topsoils under natural broadleaved forests along a 2500 km latitudinal gradient. Geoderma 315,65–74 (2018). Open Access This article is licensed under a Creative Commons 36. Tiessen, H., Cuevas, E. & Chacon, P. The role of soil organic matter in Attribution 4.0 International License, which permits use, sharing, sustaining soil fertility. Nature 371, 783–785 (1994). 37. Harpole, W. S. et al. Nutrient co-limitation of primary producer communities. adaptation, distribution and reproduction in any medium or format, as long as you give Ecol. Lett. 14, 852 (2011). appropriate credit to the original author(s) and the source, provide a link to the Creative 38. Penuelas, J. et al. Human-induced nitrogen-phosphorus imbalances alter Commons license, and indicate if changes were made. The images or other third party natural and managed ecosystems across the globe. Nat. Commun. 4, 2934 material in this article are included in the article’s Creative Commons license, unless (2013). indicated otherwise in a credit line to the material. If material is not included in the 39. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very high article’s Creative Commons license and your intended use is not permitted by statutory resolution interpolated climate surfaces for global land areas. Int. J. Climatol. regulation or exceeds the permitted use, you will need to obtain permission directly from 25, 1965–1978 (2005). the copyright holder. To view a copy of this license, visit http://creativecommons.org/ 40. Trabucco, A. & Zomer, R. Global aridity index (global-aridity) and global licenses/by/4.0/. potential evapo-transpiration (global-PET) geospatial database. https:// cgiarcsi.community/data/global-aridity-and-pet-database/. CGIAR Consortium for Spatial Information. (2009). © The Author(s) 2020 NATURE COMMUNICATIONS | (2020) 11:637 | https://doi.org/10.1038/s41467-020-14492-w | www.nature.com/naturecommunications 9

Journal

Nature Communications – Springer Journals

Published: Jan 31, 2020

There are no references for this article.