Data acquisition and selection

We collected data from published articles exploring at least one of the following variables in response to drought manipulation: microbial biomass carbon (MBC), soil respiration, metabolic quotient (qCO2), cumulative respiration, dissolved organic carbon (DOC) and F:B ratio (or

individual values for both fungi and bacteria). We searched for articles in Web of Science (Helama et al.), Google Scholar and Scopus, using the search terms “dry-rewetting”, “drought”, “microbial”, “soil” (and a combination of them). We also searched for articles that were cited in the publications we found. A number of criteria had to be fulfilled to be included in the meta- analysis: experiments, either in the field, in pots, or as soil laboratory incubations, had to have a control treatment with an ambient regime of precipitation (for field experiments) or with

moisture kept constant (for pot and soil laboratory incubation experiments); a drought treatment with reduced or total exclusion of precipitation in field e xperiments, no water addition during the dry period in pot and soil laboratory incubation experiments and with a rewetting phase (where soil moisture was brought to the control), or a drought treatment with consistently lower soil moisture than the control (i.e., without dry-rewetting cycles; Fig. 1); at least one of the variables investigated had to be reported in both control and drought treatment; the treatment and control started with the same soil type and plant species, and were conducted under equal spatial and temporal scales.

The variables investigated were each measured at different time points (Fig. 1).

Observations were separated into drying and rewetting phases for laboratory incubations that had one or more rewetting events. In field experiments all observations were considered as part of the drying phase, since rewetting through precipitation occurred both in control and drought

18

treatments. Because MBC and F:B ratio can each be measured with different techniques, which may be difficult to compare, we decided to use data from a single technique for each response variable. For microbial biomass the most common technique in our data set was the fumigation- extraction technique (259 observations from 31 studies; Vance et al., 1987) while F:B ratios were predominantly based on the phospholipid fatty acid (PLFA) analysis in soil (87 observations from 16 studies) (Frostegård et al., 2011).

Following these criteria, a total of 60 studies were included, providing 629 paired observations (control vs treatment) of respiration, 162 for cumulative respiration, 259 for microbial biomass, 200 for DOC and 87 for F:B ratio (including 83 specific for fungi and 77 specific for bacteria). Mean, number of replicates and standard deviation (SD) of the response variables were calculated or extracted for all treatments and controls, considering different treatments (e.g., soil type, fertilizer application) as separate experiments (Gurevitch and Hedges, 1999). Figure data were extracted using Plot digitizer software (Huwaldt, 2013) to convert data- points to numerical values.

When values of qCO2 were not reported, the ratio was estimated if respiration and MBC

were both reported (13 studies). In those cases, standard deviations for control and drought treatments were derived with Taylor expansion (Stuart and Ord, 1994):

,

where μ indicates mean and σ2 and Cov the variance and covariance of respiration (R) and microbial biomass C (MBC), respectively. For studies not reporting the covariance or raw data, the covariance term was approximated based on reported mean values and their variance (where

19

available). A cross-study covariance estimate, , was derived from all reported mean values, Ri and MBCi, reported in the dataset, and the ratio of overall covariance to the

pooled variance of Ri and MBCi was calculated. Study-specific covariance estimates, reflecting

levels of variation in each study, were then obtained by multiplying this ratio with, the pooled variance of R and MBC in each study. Most of the data obtained for qCO2 belong to the

rewetting phase, while for the drying phase there were only 19 observations. Therefore, only results from the rewetting phase are shown.

For each observation we recorded informative data about soil a nd drought characteristics as well as other experimental settings, to be included as categorical and continuous explanatory moderators in the meta-analysis. For the soil characteristics we collected information on soil type, texture and organic C (SOC). So il texture classes were grouped into three categories: coarse (sand and loamy sand soils), medium (loam, silty loam and clay loam) and fine soils (sandy clay loam, silty clay loam, clay loam, clay), based on the “Soil survey manual” of the USDA (NRCS). When SOC was expressed as soil organic matter (Olsen and Sommers), a conversion factor of 0.5 was used to transform to SOC (Pribyl, 2010). SOC values were used only when soil had a pH below 7, and inorganic C content was most likely negligible (Shi et al., 2012). SOC values were classified into low (<2%) or high (>2%) where the 2% has been used as a threshold to describe changes in multiple soil quality parameters (Chenu et al., 2000,

Greenland et al., 1975). By using this threshold, the low and high SOC groups had similar numbers of observations. For the drought characteristics we collected information on drought type (constant soil moisture reduction or dry-rewetting cycles), relative drought length (expressed as the ratio of dry days over total number of days or D:T ratio), number of dry- rewetting cycles and drought intensity (defined here as the standardized difference, i.e. the

20

maximum difference in soil moisture content between the control and drought treatments expressed as a percentage of the control; Fig. 1). Soil moisture data were mostly expressed as gravimetric, volumetric, or as a percentage of water holding capacity. Although the three measures are different, when expressed as % change relative to the control, they can be compared as they all respond linearly to water increase. When soil moisture was e xpressed as water potential, values were transformed to gravimetric data with the use of pedotransfer functions. We used two pedotransfer functions specific to sandy and clay soils, respectively (Buitenwerf et al., 2014), which have good fits for water potentials values in the same ranges found during our meta-analysis. In a few cases the water potential was found to be even lower than that (e.g. Van Gestel et al., 1993, who reported values of -1180 MPa) and we approximated these values with a soil moisture content of 0%. Drought intensity was used as a continuous explanatory variable since previous studies observed a relationship between soil water potential and respiration (Manzoni et al., 2012). Furthermore, data were divided into field and laboratory (where laboratory included soil incubation and greenhouse experiment), and whether plants were included or not. Plants were always present in field experiments, while in laboratory experiments only 6 studies included plants with only a few of the investigated variables reported. Therefore, observations extracted from pots with plants were excluded from the analysis in order to remove effects of plant presence in laboratory experiments.

21

Figure 2. 1. General illustration of selection criteria and drought characteristics of retrieved

studies. The y-axis represents soil moisture content and x-axis the experimental length (days). Points represent time of measurement for both dry-rewetting and constant drought treatments. Length of rewetted and dry phases was variable between studies.