In the face of ongoing and projected climatic changes, precipitation manipulation experiments (PMEs) have produced a wealth of data about the effects of precipitation changes on soils. In response, researchers have undertaken a number of synthetic efforts. Several meta-analyses have been conducted, each revealing new aspects of soil responses to precipitation changes. Here, we conducted a comparative analysis of the findings of 16 meta-analyses focused on the effects of precipitation changes on 42 soil response variables, covering a wide range of soil processes. We examine responses of individual variables as well as more integrative responses of carbon and nitrogen cycles. We find strong agreement among meta-analyses that belowground carbon and nitrogen cycling accelerate under increased precipitation and slow under decreased precipitation, while bacterial and fungal communities are relatively resistant to decreased precipitation. Much attention has been paid to fluxes and pools in carbon, nitrogen, and phosphorus cycles, such as gas emissions, soil carbon, soil phosphorus, extractable nitrogen ions, and biomass. The rates of processes underlying these variables (e.g., mineralization, fixation, and (de)nitrification) are less frequently covered in meta-analytic studies, with the major exception of respiration rates. Shifting scientific attention to these less broadly evaluated processes would deepen the current understanding of the effects of precipitation changes on soil and provide new insights. By jointly evaluating meta-analyses focused on a wide range of variables, we provide here a holistic view of soil responses to changes in precipitation.
Soil is an important component of terrestrial ecosystems through which carbon, nitrogen, phosphorus, and other elements cycle. Biological processes in soils, such as those driven by plant roots, microbes, and enzymes, regulate nutrient cycling, with direct impacts on aboveground plant and animal communities (Bardgett et al., 2008). Rates of biological activity in soils are largely determined by physical parameters, one of the most influential being soil moisture (Stark and Firestone, 1995; Brockett et al., 2012; Schimel, 2018). Historical observations have shown that annual precipitation has either increased or decreased significantly in many regions, and the intensity and frequency of precipitation extremes (heavy rainfalls and droughts) have likewise increased in many regions (Frei et al., 2006; Lenderink and van Meijgaard, 2008). These changes in precipitation patterns are projected to continue in the future, possibly at a faster rate (Bao et al., 2017).
The activity of plant roots, microorganisms, and enzymes is maximized at optimal soil water content, which is unique to each group of organisms, soil type, and ecosystem (Bouwman, 1998; Schimel, 2018). Water in soil functions as (1) a resource to promote metabolism of microbes and plants, (2) a solvent of nutrients, and (3) a transport medium to provide pathways to solutes and microorganisms (Schimel, 2018; Tecon and Or, 2017). In a water-limited environment, reduced belowground activities are common (Borken et al., 2006; Sardans and Peñuelas, 2005). The negative responses of soil processes to decreased precipitation are attributed to reduced metabolism of the organisms (Salazar-Villegas et al., 2016; Schimel et al., 2007), limited substrate availability or diffusivity (Manzoni et al., 2016), restricted mobility of the organisms (Manzoni et al., 2016), or a combination of these (Schimel, 2018). Increased precipitation, on the other hand, generally promotes processes by shifting the soil moisture level closer to the optimum (Zhang et al., 2013; Zhou et al., 2013). However, excess water in soil often results in lower biological activity due to the limitation of oxygen flow (Bouwman, 1998; Reinsch et al., 2017), while anaerobic processes such as methane production are greatly promoted (Le Mer and Roger, 2001).
Natural variation in precipitation provides opportunities to observe responses of belowground activities (e.g., Goldstein et al., 2000; Granier et al., 2007), but targeted studies of belowground responses are difficult. Controlled precipitation manipulation experiments offer the opportunity to specifically study ecosystem responses to changes in precipitation compared to naturally occurring fluctuations and have become common in recent decades (Beier et al., 2012; Borken et al., 2006; Knapp et al., 2017). Precipitation manipulation experiments (PMEs) involve constructing an experimental structure in the field, such as rainout shelters, curtains, and/or sprinklers, to simulate alternative precipitation patterns (Beier et al., 2012). These setups enable direct comparisons between a manipulated precipitation treatment and a control (ambient precipitation) in the same study system, while keeping other environmental conditions nearly identical. PMEs have been established across ecosystem types and characteristics (biome, ecosystem, soil type, and land type) and often use different methodological approaches (e.g., in terms of the magnitude and duration of the precipitation manipulation, size of the experiment, method of rain exclusion, and/or variables measured; Vicca et al., 2014).
A number of meta-analyses have assembled and synthesized large and diverse PME datasets (Blankinship et al., 2011; Canarini et al., 2017; Wu et al., 2011). The first to examine soil responses to precipitation changes was conducted by Wu et al. (2011), compiling 85 manipulation studies and presenting the changes in aboveground and belowground carbon dynamics. Since then, several additional meta-analyses have considered belowground responses to precipitation changes. As of April 2019, according to our search criteria (details below), a total of 16 meta-analyses in this area were published. These meta-analyses focused on different but complementary soil properties (e.g., soil C in Zhou et al., 2016, or N in Yue et al., 2019). A combined analysis of these meta-analyses would provide a holistic view of the potential effects of projected precipitation changes on soil processes.
In this paper, we conduct a comparative analysis of 16 meta-analyses that have examined soil responses to manipulated (increased and decreased) precipitation in situ, encompassing 42 response variables including greenhouse gas exchanges, carbon and nitrogen dynamics, phosphorus content, microbial community, and enzyme activities. By collating the results of the published meta-analyses, we aimed to (1) provide a more holistic view of the effects of precipitation changes on soil composition and functioning, (2) discuss the potential underlying mechanisms of each response, and (3) identify knowledge gaps and propose future research directions. This study covers an unusually wide range of soil processes and examines the responses of individual variables as well as nutrient cycles.
We collected peer-reviewed meta-analyses focused on the effects of decreased and/or increased precipitation on soil variables. We collected
meta-analyses that included only field studies where the magnitude of precipitation was manipulated. Some meta-analyses included both field and
laboratory or greenhouse experiments, but we only analyzed field data in our comparisons. We used Google Scholar and Web of Science with the search terms
“meta-analysis” AND “soil” AND (“respiration” OR “
List of meta-analyses used in this study.
List of soil variables and their definitions as analyzed in the meta-analyses. The numbers indicate the meta-analysis number corresponding to Table 1, examining the effects of decreased precipitation (DP) and increased precipitation (IP) on each soil variable.
From each meta-analysis, we obtained the mean effect size of each soil variable. In this review, effect sizes are the natural log of response ratios
(
Our purpose in conducting a comparison of existing meta-analyses was to visualize (in)consistencies among meta-analyses and identify variables that have received more (or less) attention. We did not account for overlapping empirical data between meta-analyses and thus do not provide a unified dataset for new analyses. Instead, we present the sample sizes and publication year of each meta-analysis to help interpret the results.
Meta-analyses on autotrophic (
To understand the effects of precipitation on
Some responses vary by biome. For example, the effect of DP on total C is negative in temperate forests and positive in tropical forests and grassland (Yuan et al., 2017; Zhou et al., 2016). Total C reflects a balance of plant inputs and microbial outputs, so differences in responses among systems may reflect differences in the strength of PME effects on plants vs. microbes across those systems. Responses of this metric also depend on the size of the initial pool relative to fluxes and so may be differentially dampened across systems.
Responses of
As with
Overall, responses of
Microbial activity in soils is strongly controlled by the actions of enzymes (Ren et al., 2017). Many of these enzymes, which are produced and released by microbes, depolymerize complex carbon compounds (Ren et al., 2017). While enzyme activity is relatively unresponsive to IP (Fig. 2), DP increases hydrolytic enzyme activity (breakdown of labile carbon) and inhibits oxidative activity (depolymerization of recalcitrant carbon; Fig. 2). This indicates that under dry conditions, the relative contributions of substrates from labile carbon sources increase, while the respective relative contributions from recalcitrant sources decrease.
Effect sizes for soil enzyme and physical variables with respect to decreased (red) and increased (blue) precipitation. Filled points represent a significant effect size (95 % CI not overlapping 0), and open points represent a nonsignificant effect size. Variable names correspond to Table 2. No. is meta-analysis number as listed in Tables 1 and 2. The sample size is indicated by
The summary diagrams (Fig. 1c and d) illustrate how DP generally slows the belowground carbon cycle, while IP promotes it. Nearly all steps of the carbon cycle – carbon stock, substrates, microbial activity, and respiration – are altered by both types of precipitation changes. However, enzyme activity tends to be relatively unresponsive, particularly to IP, and the observations of biomass and carbon variables vary both in direction and significance among meta-analyses. These variables also tend to vary across biomes, ecosystems, and soil types.
We found only one meta-analysis that addressed the effects of precipitation on soil
The results of Yan et al. (2018) were significant across a wide range of ecosystem types, treatment durations, and magnitudes of precipitation
manipulation. The effects of DP were greater in farmlands than in other land types, in shorter-term (
Several soil nitrogen variables, including root nitrogen (N),
Mineralization rate, defined as N supply by Homyak et al. (2017), does not change under DP despite the increase in substrate (i.e., DON; Fig. 3). However, the product of mineralization and
Extracellular enzyme activity, here shown as both total proteolytic activity (pro-enzyme) and three particular N acquisition enzyme activities
(
In contrast to DP, soil nitrogen cycling is accelerated by IP (Fig. 3c). Although no mineralization indicator was included in the meta-analyses, ample
evidence shows that nitrogen mineralization is likely to increase with IP (Hu et al., 2014; Sierra, 1997; Pilbeam et al., 1993; Mazzarino et al.,
1998). Along with greater
Soil nitrogen undergoes a wide range of chemical and biological transformations, some of which are difficult to quantify. Despite the large number of
empirical studies included in meta-analyses, some nitrogen variables, such as rates of mineralization (for IP), nitrification, denitrification, and
We found four meta-analyses that examined how precipitation changes affect the soil phosphorus (P) cycle (He and Dijkstra, 2014; Yan et al., 2018;
Yuan et al., 2017; Yue et al., 2018). The results differ among meta-analyses; for instance, according to these meta-analyses, IP can have a negative,
positive, or nonsignificant effect on total P (Fig. 4). Yuan et al. (2017) assembled the largest dataset and found that IP decreases total P, while
DP increases total P. As phosphorus is commonly a limiting nutrient for vegetation, plant P uptake and concentration are frequently studied, but
studies of soil phosphorus stocks are rarer (He and Dijkstra, 2014; Yue et al., 2018). The timescale of precipitation experiments can be as short as
one growing season (or less), and the effect of such short-term precipitation manipulations on slow processes such as chemical weathering is
negligible. However, phosphorus cycling through faster processes such as decomposition of organic matter, plant uptake, and consumption by microbes
can respond (Van Meeteren et al., 2007). Plant P uptake tracks in the same direction as changes in precipitation (He and Dijkstra, 2014). The effects
on total P are strongly linked to soil type (Yuan et al., 2017). Although Yuan et al. (2017) found significant effects of DP and IP on total P, the
effects were small (
Effect sizes for soil phosphorus variables responding to decreased (red) and increased (blue) precipitation. Filled points represent a significant effect size (95 % CI not overlapping 0), and open points represent a nonsignificant effect size. Variable names correspond to Table 2. No. is meta-analysis number as listed in Tables 1 and 2. The sample size is indicated by
Microbial biomass (MB) in soil either decreases or does not respond to DP (Fig. 5a), and these responses depend on the amount of precipitation removed
(Zhou et al., 2016; Ren et al., 2017, 2018), the length of droughts (Ren et al., 2018), vegetation type (Zhou et al., 2016; Ren et al., 2017, 2018),
and mean annual precipitation (MAP; Ren et al., 2017). MB is affected by DP only when precipitation is reduced by more than
Effect sizes for
In contrast, IP stimulates microbial growth and thus increases MB unless the proportion added is very high (
In contrast to the responsiveness of MB to altered precipitation, the composition of bacterial and fungal communities is rather unresponsive
(Fig. 5b). Although Blankinship et al. (2011) and Yan et al. (2018) estimated significant effects on the abundance of fungi (both positive and
negative effects of IP) and F : B ratio (negative effect of DP;
Belowground stoichiometric relationships of carbon, nitrogen, and phosphorus can help researchers interpret and infer nutrient movements in soil organisms and their environments. Yet, few meta-analyses have synthesized belowground stoichiometric responses to precipitation treatments; greater attention has been paid to stoichiometry of aquatic systems and plants (Cleveland and Liptzin, 2007; Redfield, 1958; Yuan and Chen, 2015). He and Dijkstra (2014) and Yan et al. (2018) found no changes in soil C : N and N : P with DP (Fig. 3), but MBC : MBN responded to both precipitation changes (Fig. 5). Increased MBC : MBN with IP indicates that wetter conditions stimulated greater metabolic activity of microbes, which accumulated more carbon in their bodies. This suggests that the soil microbial biomass C : N : P ratio, which is well-constrained globally (60 : 7 : 1; Cleveland and Liptzin, 2007), could be altered by IP to have more weight on carbon. Soil N : P ratios can be heavily dependent on plant nutrient uptake; as discussed in Sect. 3.3, DP reduces plant nitrogen uptake, which could increase soil N : P. However, this effect depends on site aridity (Sardans et al., 2012) and could be mitigated by robust mycorrhizal symbioses (Mariotte et al., 2017), which could help maintain soil N : P ratios by sustaining plant nutrient uptake under DP.
Meta-analyses have substantially advanced our understanding of the impacts of precipitation changes on soil processes and properties. Responses of
several variables have been investigated by three or more meta-analyses and with robust datasets; these include soil respiration, nitrogen stocks,
total phosphorus, and microbial biomass. However, many other variables have received less attention. For example, sample sizes for analyses of
autotrophic respiration are smaller than for those of heterotrophic respiration; substrate availability has not been analyzed while soil C, N, and P content have;
and analyses of bacterial and fungal responses to IP are sparser than those of responses to DP.
Filling these knowledge gaps could help to reveal the mechanisms underlying soil responses to precipitation changes. For example, there is robust
agreement across studies that soil and heterotrophic respiration slow under DP and accelerate under IP. However, the relative importance of different
mechanisms in the response of heterotrophic respiration is still unknown – in other words, how much of this response comes from changes in the level
of microbial activity (e.g., entering and exiting dormancy) vs. substrate availability? Similarly, what are the most important mechanisms behind
changes in
Studies of processes that have received less attention (e.g., microbial metabolic state, nitrification, denitrification, and N fixation) have the
potential to inform models and improve predictions of the effects of precipitation changes on important fluxes and pools. This benefit can be seen in
ecosystem models that explicitly represent active and dormant microbial biomass, which can outperform those representing microbial biomass as a single
pool (He et al., 2015; Salazar et al., 2019; Wang et al., 2015). A more synthetic understanding of nitrification and denitrification responses across
ecosystems could improve projections of societally relevant nitrate leaching and soil emissions of
The meta-analyses we examined had strong geographical imbalances, as has been identified elsewhere. While all but one meta-analyses collected global
empirical data, the data are concentrated in the US, Europe, and China. Almost 90 % of the existing PMEs are located at midlatitudes
(30–60
PMEs are quite diverse, adopting a variety of approaches, treatment levels, and treatment types (Beier et al., 2012; Kreyling and Beier, 2013), and so are the data derived from them. Many PMEs use long-term rainout shelters, which unavoidably modify the ambient environment in other ways (Kreyling et al., 2017). While synthesizing the results of PMEs around the globe in the context of these experimental issues could be challenging, meta-analyses provide one somewhat simplistic approach, through an exhaustive statistical summary of empirical studies (Hedges et al., 1999). Meta-analysis can obscure the substantial influence of environmental characteristics and methodological differences on effect sizes. Categorization by environmental characteristics, such as climate, geography, ecosystem, soil, and soil biota, can provide a local to regional view of soil responses that is specific to the given environmental characteristic. Categorization by methodology, such as experimental duration, intensity of treatment, measurement method, and fertilizer use, can clarify the human-derived impacts on effect sizes. These categorization efforts help to identify when and how soil responses depend on their environmental context. While an exhaustive analysis of these categories is beyond the scope of this paper, we have highlighted the cases in which these factors affected each meta-analysis result in the text above. A further breakdown of these categories by environmental characteristics and methodology can be found in the Supplement (S1). As more and more PMEs are implemented, sample sizes available for meta-analysis are increasing (Song et al., 2019). In this regard, the recent deployment of broad networks of PMEs with standardized methodology and sampling procedures (Fraser et al., 2013; Halbritter et al., 2020) could ultimately contribute to more powerful meta-analyses with more easily interpreted outcomes (Hilton et al., 2019; Knapp et al., 2012, 2017).
We identified some technical challenges during this comparative study, including data collection and the definition of samples. Data collection is perhaps the most time-consuming process of searching literature and contacting researchers. Most meta-analyses extract effect size, SD, and sample size from publication when possible, commonly with the use of digitizing software (Canarini et al., 2017; Liu et al., 2016; Ren et al., 2017; Xiao et al., 2018; Yan et al., 2018; and others). While digitizing software is helpful, the accuracy of the digitized values depends on the resolution of the figures. In some cases, digitizing is not feasible when points are too large or error bars are too close to the points. Thus, we emphasize the importance of comprehensively presenting and publishing data, both in original studies and meta-analyses, to minimize errors associated with digitizing. Secondly, we found that the definition of a sample used in meta-analyses differs by studies. Specifically, some meta-analyses treat observations over multiple years from the same experiment as distinct individual samples, which could potentially violate the assumption of sample independence. We recommend, therefore, that a meta-analysis accounts for within-study dependency (Canarini et al., 2017) or selects a single year or season to include in the analysis.
This assessment of meta-analyses provides a broad perspective on how precipitation changes affect soils and belowground processes. Belowground carbon and nitrogen cycles speed up with increased precipitation and slow down with decreased precipitation, while bacterial and fungal communities are relatively insensitive to decreased precipitation. While the responses of the fluxes and pools of each cycle – gas emissions, soil carbon, nitrogen ions, and biomass – have been studied extensively, responses of the associated process rates remain less studied or unexamined by meta-analyses. There are also gaps in the study of soil elements such as phosphorus and nitrogen ions, as well as of stoichiometric relationships and bacterial and fungal biomass under increased precipitation. We suggest that additional scientific attention to these gaps is warranted and would help to deepen and consolidate current knowledge of soil responses to precipitation changes.
The data collected from meta-analyses and used in this paper are available through the Purdue University Research Repository
(
The supplement related to this article is available online at:
AOA, AS, YO, MRU, IR, and JSD designed the research. AOA, AS, YO, SR, MRU, JL, and IR conducted the comparative analysis and contributed to writing the original draft. AOA prepared the manuscript with contributions from all co-authors.
The authors declare that they have no conflict of interest.
Ideas for this paper were developed during a distributed graduate seminar organized by the Drought-Net Research Coordination Network (RCN) in spring 2017. Drought-Net was supported by the NSF (DEB-1354732; PIs (principal investigators) Melinda Smith, Osvaldo Sala, and Richard Phillips). We thank Melinda Smith, Osvaldo Sala, and Alan Knapp for valuable feedback to this research. This is publication 2002 of the Purdue Climate Change Research Center.
This research has been supported by the Department of Forestry and Natural Resources at Purdue University and Takenaka Scholarship Foundation in Tokyo, Japan (AOA); the 529 Colciencias-Fulbright grant; the Icelandic Research Fund 2016 (grant no. 163336-052); the POA funds from the Instituto de Investigación de Recursos Biológicos Alexander von Humboldt, Bogotá, Colombia (AS); the Natural Environment Research Council award number NE/R016429/1 as part of the UK-SCAPE program delivering National Capability (SR); and the University Grants Commission, India, under Raman Fellowship Program (IR).
This paper was edited by Kees Jan van Groenigen and reviewed by Feike Dijkstra and Nameer Baker.