Volume 215, Issue 3 p. 1186-1196
Full paper
Free Access

Circular linkages between soil biodiversity, fertility and plant productivity are limited to topsoil at the continental scale

Manuel Delgado-Baquerizo

Corresponding Author

Manuel Delgado-Baquerizo

Hawkesbury Institute for the Environment, Western Sydney University, Penrith, NSW, 2751 Australia

Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, CO, 80309 USA

Author for correspondence:

Manuel Delgado-Baquerizo

Tel: +1 303 7355185

Email: [email protected]

Search for more papers by this author
Jeff R. Powell

Jeff R. Powell

Hawkesbury Institute for the Environment, Western Sydney University, Penrith, NSW, 2751 Australia

Search for more papers by this author
Kelly Hamonts

Kelly Hamonts

Hawkesbury Institute for the Environment, Western Sydney University, Penrith, NSW, 2751 Australia

Search for more papers by this author
Frank Reith

Frank Reith

The Sprigg Geobiology Centre, School of Biological Sciences, The University of Adelaide, Adelaide, SA, 5005 Australia

Land and Water, Environmental Contaminant Mitigation and Technologies, PMB2, Glen Osmond, SA, 5064 Australia

Search for more papers by this author
Pauline Mele

Pauline Mele

The Department of Economic Development, Jobs, Transport and Resources, AgriBio Centre, Bundoora, Vic., 3083 Australia

School of Applied Systems Biology, La Trobe University, Bundoora, Vic., 3083 Australia

Search for more papers by this author
Mark V. Brown

Mark V. Brown

School of Biotechnology and Biomolecular Sciences, UNSW, Sydney, NSW, 2052 Australia

Search for more papers by this author
Paul G. Dennis

Paul G. Dennis

School of Agriculture and Food Sciences, The University of Queensland, Brisbane, St Lucia, Qld,, 4072 Australia

Search for more papers by this author
Belinda C. Ferrari

Belinda C. Ferrari

School of Biotechnology and Biomolecular Sciences, UNSW, Sydney, NSW, 2052 Australia

Search for more papers by this author
Anna Fitzgerald

Anna Fitzgerald

Bioplatforms Australia Ltd, Sydney, NSW, 2109 Australia

Search for more papers by this author
Andrew Young

Andrew Young

Oceans and Atmosphere, CSIRO, Hobart, Tas., 7004 Australia

Search for more papers by this author
Brajesh K. Singh

Brajesh K. Singh

Hawkesbury Institute for the Environment, Western Sydney University, Penrith, NSW, 2751 Australia

Global Centre for Land-Based Innovation, Western Sydney University, Penrith South DC, NSW, 2751 Australia

Search for more papers by this author
Andrew Bissett

Andrew Bissett

Oceans and Atmosphere, CSIRO, Hobart, Tas., 7000 Australia

Search for more papers by this author
First published: 13 June 2017
Citations: 85

Summary

  • The current theoretical framework suggests that tripartite positive feedback relationships between soil biodiversity, fertility and plant productivity are universal. However, empirical evidence for these relationships at the continental scale and across different soil depths is lacking.
  • We investigate the continental-scale relationships between the diversity of microbial and invertebrate-based soil food webs, fertility and above-ground plant productivity at 289 sites and two soil depths, that is 0–10 and 20–30 cm, across Australia.
  • Soil biodiversity, fertility and plant productivity are strongly positively related in surface soils. Conversely, in the deeper soil layer, the relationships between soil biodiversity, fertility and plant productivity weaken considerably, probably as a result of a reduction in biodiversity and fertility with depth. Further modeling suggested that strong positive associations among soil biodiversity–fertility and fertility–plant productivity are limited to the upper soil layer (0–10 cm), after accounting for key factors, such as distance from the equator, altitude, climate and physicochemical soil properties.
  • These findings highlight the importance of surface soil biodiversity for soil fertility, and suggest that any loss of surface soil could potentially break the links between soil biodiversity–fertility and/or fertility–plant productivity, which can negatively impact nutrient cycling and food production, upon which future generations depend.

Introduction

A universal circular mechanism, including positive feedback cycles between soil biodiversity, fertility and plant productivity, is implied in current ecological theories (Wardle et al., 2004; van der Heijden et al., 2008; Hol et al., 2015; Nielsen et al., 2015; Delgado-Baquerizo et al., 2016; Soliveres et al., 2016). However, empirical evidence for such a mechanism is lacking. Some of the major gaps in our knowledge are outlined below. First, previous work has focused on isolated parts of this circular association, providing specific evidence for the relationship between soil biodiversity–fertility, biodiversity–plant productivity or plant productivity–fertility (Brady & Weil, 2002; Wagg et al., 2014; Nielsen et al., 2015; Delgado-Baquerizo et al., 2016; Soliveres et al., 2016). However, we lack initiatives aiming to disentangle the magnitude and direction of these relationships when assessed in combination. Second, most biodiversity–ecosystem functioning studies have focused on one or several components of soil fertility or biodiversity. However, the links between the diversity of multiple soil trophic levels and components of soil fertility and plant production need to be considered simultaneously to achieve an integrated understanding of the role of biodiversity for ecosystem functioning (Wagg et al., 2014; Soliveres et al., 2016). Finally, previous studies have assessed the relationships between soil biodiversity, fertility and plant productivity under very specific ecological contexts, for example within a particular biome on Earth or within a particular soil depth (Wagg et al., 2014; Hol et al., 2015; Nielsen et al., 2015; Delgado-Baquerizo et al., 2016; Soliveres et al., 2016). This, in turn, limits our understanding of the circumstances in which the tripartite relationships between soil biodiversity, fertility and plant productivity occur, and, most importantly, on whether this circular mechanism is universal.

The top soil is key for food production but is highly vulnerable to degradation and erosion (Jobbágy & Jackson, 2000, 2001; Banwart, 2011). The soil surface is often covered by litter from above-ground communities and commonly contains the highest concentrations of soil organic matter (Jobbágy & Jackson, 2001; Banwart, 2011). Moreover, soil biodiversity and fertility have been found to decrease with depth as the amount and diversity of nutrients, but also the concentration of oxygen, become limiting (Hooper et al., 2000; Jobbágy & Jackson, 2001). In addition to variations with soil depth, it has also been shown that in subsurface layers, microbial community composition is driven less by these edaphic properties, compared to surface soils, and more by geogenic factors and/or stochastic processes (Powell et al., 2015; Reith et al., 2015). The essential co-occurrence of resource inputs from above-ground communities with high soil biodiversity may, therefore, support the strongest linkages between soil biodiversity, fertility and plant productivity at a topsoil level. However, reductions in soil biodiversity and fertility with soil depth can potentially break the link between soil biodiversity, fertility and plant productivity in the subsoil. In this regard, if tripartite linkages between soil biodiversity, fertility and plant productivity are limited to the topsoil, any loss of the top layer of soil due to environmental change will significantly constrain our ability to provide food for our growing population. Thus, assessing if these tripartite linkages are maintained across deep soil layers or if they are limited to the top few centimeters below the soil surface is a fundamental requirement for developing appropriate monitoring, management and conservation policies, and for predicting how ecosystem functioning will respond under changing environments. Such guidelines are urgently required, because soils are highly vulnerable to global environmental drivers, such as climate change, land-use intensification and desertification (Banwart, 2011; Maestre et al., 2015), which are acting to reduce plant productivity, biodiversity and fertility (Gans et al., 2005; Wall et al., 2010; Maestre et al., 2015; Nielsen et al., 2015). In this study, we tested the hypothesis that the positive associations between soil biodiversity, fertility and plant productivity are demonstrable at large spatial scales, but limited to the first top few centimeters of soil. To test this hypothesis, we used standardized methods for sampling and analyses of soil fertility and diversity of microbial and invertebrate-based soil food webs at 289 sites and two soils depths, 0–10 and 20–30 cm, across the Australian continent.

Materials and Methods

Study sites

We used a subset of sample locations (289) across Australia (Fig. 1) from the Biome of Australia Soil Environments (BASE) project (Bissett et al., 2016). This subset of plots had available information on soil biodiversity at multiple trophic levels: bacteria (16S rRNA), fungi (internal transcribed spacer, ITS) and eukaryotes (18S rRNA) (Supporting Information Table S1). Samples and field information were collected between 2011 and 2014. Soil samples were collected according to the methods described at the BASE data-portal (Bissett et al., 2016). In brief, at each plot (25 × 25 m quadrat), composite soil samples (nine discrete soil samples) at two depth ranges (0.1 and 0.3 m) were collected. By systematically collecting soil samples from two discontinuous depths we aimed to alleviate the dependency of surface and shallow subsurface samples and to allow comparison of soil samples at the continental scale. A total of 578 soil samples (289 sites × two depths) were included in this study. Each sample was separated into two subsamples. The first subsample, used for molecular analyses, was frozen and transported to the Adelaide node of the Australian Genome Research Facility (AGRF) laboratories for DNA extraction. The second subsample was air-dried and transported to CSBP laboratories, Perth, Australia, for chemical and physical analyses. At the time of sampling all other contextual data were collected, including sample location (coordinates taken at the center point of the sampling quadrat) and elevation above sea level.

Details are in the caption following the image
Location of study sites in Australia (= 289).

Molecular analysis

Soil DNA was extracted in triplicate, according to the methods employed by the Earth Microbiome Project (http://www.earthmicrobiome.org/protocols-and-standards/dna-extraction-protocol/), at the AGRF. We used an Illumina MiSEQ for sequencing as described by Bissett et al. (2016). Briefly, amplicons targeting the bacterial 16S rRNA gene (27F – 519R) (Lane, 1991), fungal ITS region (ITS1F – ITS4) (White et al., 1990) and eukaryotic 18S rRNA gene (Euk_1391f – EukBr, http://press.igsb.anl.gov/earthmicrobiome/protocols-and-standards/18s/) were prepared and sequenced for each sample at the Australian Genome Research Facility (Melbourne, Australia) and the Ramaciotti Centre for Genomics (Sydney, Australia). 16S and ITS amplicons were sequenced using 300 bp, paired-end sequencing, while 18S amplicon reads were generated using 150 bp paired-end sequencing.

Bioinformatic analysis

Bioinformatic analyses were performed as described by Bissett et al. (2016). Briefly, for rRNA genes, sequence read quality was visually assessed using FastQC (Andrews, 2010), sequences were trimmed to remove poor quality bases (5′ end of R1 by 10 bp and 5′ end of R2 by 70 bp) and merged using Flash (Magoč & Salzberg, 2011). Efficacy of merging was checked manually after merging. Sequences < 400 bp (16S) or < 100 bp (18S), or containing N or homopolymer runs of > 8 bp were removed (Mothur v.1.34.1; Schloss et al., 2009). The remaining sequences were passed to the open reference (operational taxonomic unit, OTU) picking and assigning workflow. ITS sequences followed a similar protocol, except the ITS1 region was extracted from the Illumina R1 read (Bengtsson-Palme et al., 2013) before OTU picking. In all cases, OTUs were clustered at 97% sequence similarity using Uparse (Edgar, 2013) and OTU abundance tables constructed by mapping reads to these OTUs (usearch_global, 97%). OTU abundance tables were rarefied at 9051 (16S rRNA gene), 5000 (fungal ITS region) and 5000 (eukaryotic 18S rRNA gene) sequences/sample to ensure even sampling depth. The Shannon diversity index was calculated on rarefied OTU tables using Primer v.7 (Clarke & Gorley, 2015) for ten main groups of soil microbes and fauna: bacteria (16S rRNA gene), fungi (fungal ITS region) and stramenopiles, amoebozoa, annelida, arthopoda, excavata, rhizaria, rotifera and nematoda (eukaryotic 18S rRNA gene). We selected this metric for our study because it provides a robust and informative estimation of taxonomic diversity for microbial communities (Haegeman et al., 2013).

Analysis of soil properties

Soils were extracted in deionized water for 1 h to achieve a soil : solution ratio of 1 : 5. Soil pH was then determined using a combination pH electrode. Texture was determined as explained by Bissett et al. (2016).

Climate

Mean annual temperature (MAT) and aridity index (AI; mean annual precipitation/potential evapotranspiration) were derived from the Worldclim database (http://www.worldclim.org) (Hijmans et al., 2005; Zomer et al., 2008). Climate gaps in the dataset were completed using local and regional databases. For clarity, we used aridity (maximum AI value in the dataset – AI) instead of AI (Delgado-Baquerizo et al., 2013). We used aridity instead of MAP, because aridity includes both mean annual precipitation and potential evapotranspiration, and is therefore a more accurate metric of the water availability at each site. Spatial and climate data collection were performed in ESRI ArcMap (ESRI 2011; Environmental Systems Research Institute, Redlands, CA, USA).

Above-ground plant productivity

We used the normalized difference vegetation index (NDVI) as our proxy of plant productivity (Pettorelli et al., 2005; Delgado-Baquerizo et al., 2016). These data were obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard NASA's Terra satellites (http://neo.sci.gsfc.nasa.gov/). This productivity index is largely used in the current literature (Pettorelli et al., 2005; Delgado-Baquerizo et al., 2016) and provides a global measure of the ‘greenness’ of vegetation across Earth's landscapes for a given composite period, and thus acts as a proxy of photosynthetic activity and large-scale vegetation distribution (Pettorelli et al., 2005; Delgado-Baquerizo et al., 2016). We calculated the monthly average value for this variable between the periods of 2011–2014 (0.1° resolution), when all soil sampling was conducted.

Assessing soil fertility

The BASE project contains information on multiple surrogates of soil fertility including soil organic carbon (a common proxy of organic matter), macronutrients (i.e. ammonium, nitrate, phosphorus, potassium and magnesium) and micronutrients (i.e. copper, sodium, iron, manganese, zinc and aluminum). A detailed explanation on the rationale of the selected variables is available in Methods S1. The details and protocols for soil analyses are given in Bissett et al. (2016). In brief, soil organic carbon was determined using the method of Walkley & Black (1934). Ammonium and nitrate were determined from 2 M KCl extracts. Available phosphorus was measured using the Colwell method (Walkley & Black, 1934). Exchangeable aluminum, magnesium, potassium and sodium were determined using a 1 : 5 soil : water extraction. This test was used in combination with the NH4Cl2/BaCl2 extractable exchangeable cations test, where the value for water-soluble exchangeable cations is subtracted from the value for NH4Cl2/BaCl2-extractable exchangeable cations (Rayment & Higginson, 1992). Finally, diethylene-triamine-pentaacetic acid (DTPA)-extractable trace elements (copper, iron, manganese and zinc) were determined by atomic absorption spectroscopy following extraction with DPTA for 2 h (Rayment & Higginson, 1992).

Soil fertility and diversity indices

To obtain a quantitative index of soil fertility for each sample, we first normalized (log-transformed when needed) and standardized each of our 12 surrogates of fertility using the Z-score transformation ((actual value − mean)/SD) as described by Maestre et al. (2012), Wagg et al. (2014) and Delgado-Baquerizo et al. (2016). Following this, the standardized surrogates of fertility were averaged to obtain a soil fertility index (Maestre et al., 2012). Using the same approach, we obtained a quantitative index of soil biodiversity including the diversity (Shannon) of bacteria, fungi and multiple eukaryote communities (stramenopiles, amoebozoa, annelida, arthopoda, excavata, rhizaria, rotifera and nematoda) as described by Wagg et al. (2014) and Allan et al. (2014). This approach allowed us to evaluate links between soil biodiversity, fertility and plant productivity in an integrative manner, rather than focusing on a particular component of the soil community or surrogate of fertility, which may not accurately represent the highly interactive nature of consumer–resource interactions during nutrient cycling (Wardle et al., 2004; Bardgett & van der Putten, 2014; Wagg et al., 2014; Nielsen et al., 2015). As such, the effect from soil biodiversity on fertility and plant productivity will be a combined result of the close interaction among highly diverse groups of soil fauna and microbes. For example, groups of soil fauna, such as arthropods (e.g. ants and termites) and annelids (e.g. earthworms), play an essential role in helping to incorporate dead organic material into the soil (Nielsen et al., 2015), and bringing it into contact with microbial degraders. Additionally, soil microbial communities degrade recalcitrant components of organic matter into more labile molecules (Schimel et al., 2005), releasing soil nutrients, and supporting essential ecosystem services.

Statistical analysis

We first tested for differences in soil biodiversity and fertility between soil layers using one-way ANOVA with soil layer as a fixed factor, and conducted linear regression analyses to evaluate the relationship between soil biodiversity, fertility and plant productivity for the upper (0–10 cm) and lower (20–30 cm) layers. We repeated these analyses within characteristic biomes: tropical, temperate and arid ecosystems (Köppen classification), as well as croplands. In addition, we evaluated the relationship between the diversity of individual groups of organisms and soil fertility.

We then used structural equation modeling (SEM; Grace, 2006) to obtain a system-level mechanistic understanding of the direct and indirect relationships between distance from the equator (absolute latitude), altitude, aridity, MAT, percentage of clay, soil pH, plant productivity, soil biodiversity and fertility. This approach further allowed us to evaluate whether the links between soil biodiversity, fertility and above-ground plant productivity were maintained after accounting for multiple factors simultaneously. We conducted independent SEMs for each soil layer to provide insights on the tripartite linkages between soil biodiversity, fertility and plant productivity for the upper and lower layers, and to further evaluate the relative importance of key drivers of primary productivity, soil biodiversity and soil fertility (i.e. geolocation, climate and soil properties) at different depths. The use of SEM is particularly useful in large-scale correlative studies, as it allows us to partition causal influences among multiple variables, and to separate the direct and indirect effects of the predictors included in the model (Grace, 2006). The first step in SEM requires establishment of an a priori model based on known effects and relationships among the drivers of soil fertility (Fig. S1). Some data manipulation was required before modeling to improve the normality and linearity of our data. Altitude and percentage of clay were log-transformed to improve normality. Similarly, distance from the equator and aridity were x2-transformed and MAT was square-root-transformed. With a good model fit (see further below in this section ), we were free to interpret the path coefficients of the model and their associated P-values. A path coefficient is analogous to the partial correlation coefficient, and describes the strength and sign of the relationship between two variables (Grace, 2006). Since some of the variables introduced were not normally distributed, the probability that a path coefficient differs from zero was tested using bootstrap resampling (Delgado-Baquerizo et al., 2013). Bootstrapping is preferred to the classical maximum-likelihood estimation in these cases, because in bootstrapping probability assessments are not based on the assumption that data follow a particular theoretical distribution. Thus, data are randomly sampled with replacement to arrive at estimates of standard errors that are empirically associated with the distribution of the data found in the samples (Grace, 2006). When these data manipulations were completed, we parameterized our model using our dataset and tested its overall goodness of fit. There is no single universally accepted test of overall goodness of fit for SEM, applicable in all situations regardless of sample size or data distribution. We used the chi-square test (χ2; the model has a good fit when 0 ≤ χ2/df ≤ 2 and 0.05 <  1.00) (Schermelleh-Engel et al., 2003) and the root mean square error of approximation (RMSEA; the model has a good fit when 0 ≤ RMSEA ≤ 0.05 and 0.10 <  1.00) (Schermelleh-Engel et al., 2003). Additionally, and because some variables were not normal, we confirmed the fit of the model using the Bollen–Stine bootstrap test (the model has a good fit when 0.10 < bootstrap  1.00) (Schermelleh-Engel et al., 2003). Our a priori model attained an acceptable fit by all criteria, and thus no post hoc alterations were made. All the SEM analyses were conducted using Amos v.20.0 (IBM, Armonk, NY, USA).

Data accessibility

The primary data used in this study are available in Table S1. The raw sequence data have been deposited in the Sequence Read Archive (SRA) from the National Center for Biotechnology Information (USA) (https://www.ncbi.nlm.nih.gov/sra). The SRA accession number for each sample is available in Table S1.

Results

Both soil biodiversity and fertility were significantly higher in the upper compared to the lower soil layer (Figs 2a, 3), for bacteria, fungi and microbial eukaryotes, with strong correlations observed between soil layers and biodiversity (Spearman's ρ = 0.605; < 0.001) as well as fertility (Spearman's ρ = 0.791; < 0.001). Most importantly, we showed that the stronger tripartite linkages among soil biodiversity, fertility and above-ground plant productivity were in the upper, compared to the lower, soil layer (Fig. 2b–d). Note that soil biodiversity (calculated as Shannon diversity indices) was positively and strongly related to the alpha diversity calculated from species richness (Fig. S2). In fact, similar results were obtained when using soil biodiversity calculated from species richness indices (Fig. S3), indicating that our results are robust to the choice of soil biodiversity metric. Moreover, the stronger correlations between soil biodiversity, fertility and plant productivity at the upper layer were apparent when exploring the relationship between the diversity of single soil communities (Amoebozoa, Annelida, Arthopoda, Excavata, Rhizaria, Rotifera, Stramenopiles, Nematoda and Fungi) and both the fertility index and individual surrogates of fertility (Table S2; Fig. 4).

Details are in the caption following the image
Relationship between soil biodiversity, fertility and plant productivity across different soil depths. (a) Z-score values (mean ± SE) for soil biodiversity and fertility in the upper layer (0–10 cm) and lower layer (20–30 cm). Significance levels of each predictor: ***, < 0.001. (b–d) The linear regression between soil biodiversity, fertility and plant productivity for two different soil layers. Solid and dashed lines represent linear regressions fitted to the upper and lower layers, respectively.
Details are in the caption following the image
Mean (± SE) values for the diversity of soil microbes and fauna in the upper layer (0–10 cm) and lower layer (20–30 cm). Significance levels of each predictor: ***, < 0.001. ‘nats’, the units for Shannon diversity calculated using the natural logarithm.
Details are in the caption following the image
Relationship between the diversity of particular groups of soil microbes and fauna (a–j) with fertility in the upper layer (0–10 cm) and lower layer (20–30 cm). Solid and dashed lines represent linear regressions fitted to the upper and lower layers, respectively. ‘nats’, the units for Shannon diversity calculated using the natural logarithm.

Structural equation models explained 79% of the variation found in plant productivity. The proportion of explained variation in soil biodiversity (24% in surface vs 9% down under) and fertility (74% vs 63%) was substantially greater in the upper than the lower layers (Fig. 5a,b) (our a priori model is available in Fig. S1). We found strong positive associations, including direct and indirect effects, among soil biodiversity, fertility and above-ground plant productivity in the upper layer (0–10 cm; Fig. 5a), but not in the lower layer (20–30 cm; Fig. 5b), after accounting for multiple environmental drivers, such as distance from the equator, altitude, climate and soil properties. In particular, we found that the strong positive associations among soil biodiversity–fertility and fertility–above-ground plant productivity are limited to the upper layer of soil (0–10 cm; Fig. 5). In addition, we found direct and significant negative effects of increasing aridity and MAT on soil biodiversity, fertility and above-ground plant productivity in the upper soil layer. However, only aridity continued to explain variation in soil biodiversity in the deeper soil layer. Finally, we found positive and significant direct effects of percentage of clay on soil fertility, which was stronger in the deeper soil layer (Fig. 5a,b).

Details are in the caption following the image
Direct and indirect effects of space, climate and soil properties on soil biodiversity, fertility and plant productivity derived from structural equation modeling in the upper (a) and (b) lower soil layers. Numbers adjacent to arrows are indicative of the effect size (bootstrap P-value) of the relationship. The width of arrows is proportional to the strength of path coefficients. R2 denotes the proportion of variance explained. Significance levels are as follows: *, P < 0.05 and **, P < 0.01. MAT, mean annual temperature; RMSEA, root mean square error of approximation.

The stronger relationship between soil biodiversity and fertility, as well as fertility and above-ground plant production in the upper layer compared to the lower layer of soil was, in general, maintained within major biomes including tropical, temperate and arid ecosystems (Fig. 6). The only exception to this was croplands, where the relationship between soil biodiversity and fertility was disrupted for both soil layers and the relationship between fertility and plant productivity became negative for the lower layer of soil (Fig. 6). However, in general, we found strong reductions in soil biodiversity and fertility with depth for all ecosystem types, including croplands (Fig. 6).

Details are in the caption following the image
Relationship between soil biodiversity, fertility and plant productivity across different soil depths for arid (= 66), temperate (= 159), tropical (= 17) and agricultural (= 47) regions. (a, d, g, j) Z-score values (mean ± SE) for soil biodiversity and fertility in the upper layer (0–10 cm) and lower layer (20–30 cm). Significance levels of each predictor: °, P < 0.10; *, < 0.05; **, < 0.01; ***, < 0.001 . (c, e, f, h, i, k, l) The linear regression between soil biodiversity, fertility and plant productivity for two different soil layers. Solid and dashed lines represent linear regressions fitted to the upper and lower layers, respectively.

Discussion

This study provides, to our knowledge, the first empirical evidence showing that on a continental scale the positive effects of soil biodiversity–fertility and fertility–above-ground productivity are limited to topsoil (above 20 cm depth). In these topsoils we also show that tripartite, positive feedback relationships between diversity of microbial and invertebrate-based soil food webs, fertility and plant productivity are universal across soil types and climates, and that they are comparatively weaker in lower soil layers, even at the shallow depths we studied. Furthermore, our results provide an integrated view of how climate and soil factors influence soil biodiversity, fertility and plant productivity at the continental scale and across two different soil depths, which is critical to improving soil carbon and Earth system simulation models as well as for formulating sustainable ecosystem management and conservation policies.

Most importantly, we used SEM to build a system-level understanding of the drivers of soil biodiversity, fertility and plant production in terrestrial ecosystems in the two different soil layers. This ultimately allowed us to evaluate if the reported positive associations among soil biodiversity, fertility and plant productivity are maintained across soil layers after accounting for important environmental factors such as spatial parameters, climates, soil texture and pH. In agreement with recent studies, our model provides strong evidence for a negative impact of increasing aridity and MAT on soil biodiversity at multiple trophic levels, on soil fertility and on plant productivity (Delgado-Baquerizo et al., 2013, 2016; Maestre et al., 2015) in surface soils. This effect is lost in deeper soils, where only aridity continues to explain variation in soil biodiversity. Supporting the robustness of our results, we found a strong positive link between percentage of clay and soil fertility, which was stronger in the deeper soil layer. Clay is well known to be one of the main drivers of soil fertility in terrestrial ecosystems (Brady & Weil, 2002). We identified strong positive associations (including direct and indirect effects) among soil biodiversity, fertility and plant productivity in the upper layer (0–10 cm), but not in the lower layer (20–30 cm). In particular, we found a decoupling of soil fertility–plant productivity and soil biodiversity–fertility relationships for the lower layer. These results provide strong support for the hypothesis that the positive linkages among soil diversity, fertility and plant productivity are largely limited to a thin surface soil layer of a few centimeters.

The co-occurrence of resource inputs from above-ground communities (e.g. amount and diversity of litter; Hooper et al., 2000), and of high levels of soil diversity from multiple trophic levels in the top 10 cm of soil may explain the highest link between soil biodiversity and fertility for this layer. The lack of relationship between soil fertility and plant productivity below the topsoil may be related to the lower fertility found between 20 and 30 cm and the reports of greatest plant nutrient uptake in the top 10 cm of soil (Jackson et al., 1988). Interestingly, the bivariate direct positive relationship between soil biodiversity and above-ground plant productivity was maintained for both soil layers. Although the indirect positive effects between soil biodiversity and plant productivity via soil fertility were limited to the upper soil layer (Fig. 2), other mutual benefits for above-ground plant production (from soil biodiversity) and soil biodiversity (from above-ground plant production) may still occur in the lower layer of soil. For instance, higher plant production may still benefit the diversity of root/microbe consumers and decomposers, for example nematodes and fungi, at 20–30 cm depth (Nielsen et al., 2015; Table S2). Also, higher plant productivity may promote water acquisition and storage from which soil fauna could strongly benefit. Similarly, a relatively high soil biodiversity within the lower soil layer may promote plant productivity by providing defense to plants against pests and the invasion of soil-borne pathogens (van Elsas et al., 2012).

In general, the positive correlations between soil biodiversity, fertility and plant productivity for the upper layer were maintained across major biomes, that is, ‘natural’ ecosystems from arid, temperate and tropical regions. They also remain valid when exploring the relationship between soil diversity at multiple trophic levels (Amoebozoa, Annelida, Arthopoda, Excavata, Rhizaria, Rotifera, Stramenopiles, Nematoda and Fungi; Fig. 4; Table S2), and regardless of whether the fertility index or individual surrogates of fertility are used (Table S2). An interesting exception was the diversity of bacteria, which showed a negative relationship with soil fertility and plant productivity (Table S2). However, we still found a positive effect of bacteria on nitrate availability, highlighting the robustness of our data, as nitrification is largely controlled by prokaryotes (Robertson & Groffman, 2007; Table S2). This result is supported by a recent global study that reported a higher positive influence of fungal diversity (vs bacterial diversity) on multiple ecosystem functions related to nutrient cycling in nutrient-poor dryland ecosystem soils (Delgado-Baquerizo et al., 2016), a common condition observed for Australian soils (Orians & Milewski, 2007).

Our results have important implications for soil ecosystem functioning under global change scenarios. They suggest that any loss of surface soil could break the link between soil biodiversity, fertility and plant productivity, all essential components of ecosystem functioning. Indeed, it is predicted that ongoing processes such as climate change, land-use intensification and desertification will lead to significant surface soil loss in the near to medium term (Banwart, 2011). For example, agricultural intensification and increases in aridity linked to climate change (Huang et al., 2016) can potentially reduce permanent plant cover and promote physical processes such as wind-blown sand and abrasion of exposed rock surfaces, increasing the already high rates at which soils are eroded. In ecosystems where soils (e.g. croplands) are subject to strong anthropogenic pressures or are vulnerable to erosion, the strong linkage between biodiversity and soil fertility could be weakened. In support of this argument, we found no relationship between soil biodiversity and fertility in cropland samples (Fig. 6k). Thus, our results suggest that erosion of soils may weaken the strong linkage between soil biodiversity, fertility and plant productivity. Note that, even if erosion impacts only the linkage between soil biodiversity–fertility and/or fertility–plant production and/or plant production–soil biodiversity, these impacts may still have cascading effects on entire food webs with consequences for plant productivity and other ecosystem functions. Given that the highest soil biodiversity is found in the upper soil layer (Fig. 2), after losing the upper layer of soil through erosion, it may take centuries or millennia for the new topsoil to start acting as the former one given microbial dispersal/colonization dynamics (especially for fungi and soil fauna; Powell et al., 2015) and evolutionary rates.

Our results further reinforce the recently proposed concept of the soil critical zone, which is currently measured in meters, but our work suggests it may be limited to the more vulnerable top few centimeters of soil only (Pointing & Belnap, 2012). Furthermore, our results suggest that the relationship between soil biodiversity and fertility, and its positive effects for plant production, are an intrinsic property of this thin, but critical and vulnerable, layer of Earth's skin. Interestingly, the influence of environmental factors (including spatial distance, climate and soil properties) on our response variables was also significantly greater for the upper layer. Thus, the ability of climate (i.e. MAT) and spatial factors (i.e. distance from the equator) to predict soil biodiversity and fertility is reduced under conditions of weakened diversity–fertility relationships. This has consequences for understanding how global change, particularly aridity, will impact soil fertility in the coming decades. Interestingly, the negative effects of aridity on soil biodiversity were maintained across both soil layers.

Altogether, our findings indicate that the often positive theoretical linkages of soil biodiversity at multiple trophic levels, fertility and plant production are ubiquitous in terrestrial ecosystems at large scales, but limited only to a thin layer of soil close to the surface. Below the surface layer, the relationships between soil biodiversity, fertility and plant productivity weaken considerably. These findings provide a novel framework to predict impacts of soil loss and degradation. By revealing the relationships between soil biodiversity, fertility and plant productivity, which underpin critical ecosystem functions and services on which humans depend, we can begin to formulate appropriate soil and plant productivity management and conservation policies respond to likely changes in ecosystem functioning under future environments.

Acknowledgements

We would like to acknowledge the contribution of the BASE project partners, doi: 10.4227/71/561c9bc670099, an initiative supported by Bioplatforms Australia with funds provided by the Australian Commonwealth Government through the National Collaborative Research Infrastructure Strategy. We thank the BASE project and its contributors (https://downloads.bioplatforms.com/base/acknowledgements) for sequence and edaphic data used in this work. The data used in this study are available from the BASE data portal (https://downloads.bioplatforms.com/base/). M.D-B. and B.K.S. acknowledge support from the Australian Research Council (projects DP13010484; DP170104634). M.D-B. also acknowledges support from the Marie Sklodowska-Curie Actions of the Horizon 2020 Framework Programme H2020-MSCA-IF-2016 under REA grant agreement no. 702057.

    Author contributions

    M.D-B. designed this study and analyzed data from the BASE project in consultation with A.B., J.R.P. and B.K.S. The field surveys and soil samplings were conducted by K.H., F.R., P.M., M.V.B., P.G.D., B.C.F., A.F., A.Y. and A.B. under the coordination of A.B. M.D-B. conducted statistical modeling. A.B. conducted bioinformatics analyses. The manuscript was written by M.D-B. with contributions from all co-authors.