A meta-analysis of the response of soil moisture to experimental warming

In this meta-analysis, we reviewed more than 200 published papers on experimental warming studies recovered from the Web of Science (1980–2013). We selected data according to the following criteria: (1) the studies reported changes in soil moisture in both warming and control groups; (2) the measurements were conducted at least over an entire year; (3) the means, standard deviations of soil moisture and sample sizes were reported or could be calculated. In cases in which no standard errors were reported, we assigned standard deviations that were 1/10 of means (Luo et al 2006). (4) Relevant experimental information was reported, including the mean annual precipitation (MAP), mean annual temperature (MAT), warming methods, increased soil temperature, soil properties, and ecosystem types.

Overall, this study included 41 study sites with 139 paired experimental records for investigation of the response of soil moisture to warming (supplemental table S1, supplemental figure S2, supplemental information S3 available at stacks.iop.org/ERL/8/044027/mmedia). The measurements under different ecosystem managements (i.e., clipping, grazing, etc) or plant communities were considered as independent treatment data. Therefore, the same environmental conditions were maintained for all selected experiments (including rainfall conditions), except for temperature. Moreover, we did another literature research and collected 76 paired records at 31 sites to examine the changes in net primary production (NPP), above-ground net primary production (ANPP), below-ground net primary production (BNPP), leaf area index (LAI) or ecosystem evapotranspiration (ET) in response to warming. Only a few experimental records were collected from the same articles as for soil moisture.

The response ratio (RR, the ratio of the mean value of a concerned variable in warming treatment to that of a control treatment) is used here as an index of the magnitude of experimental warming effects (Hedges et al 1999). An RR value larger than 0, indicates a positive impact of experimental warming on soil moisture with increased soil moisture. The response ratio (RR) was calculated as:

$$\begin{eqnarray} &&\mathrm{RR}=\ln (\overline{{X}_{\mathrm{t}}}/\overline{{X}_{\mathrm{c}}})=\ln (\overline{{X}_{\mathrm{t}}})-\ln (\overline{{X}_{\mathrm{c}}}) \end{eqnarray} \tag{ 1 }$$

where $\overline{{X}_{\mathrm{t}}}$ and $\overline{{X}_{\mathrm{c}}}$ are the mean values of the warming and control groups, respectively.

The variance (ν) was estimated by:

$$\begin{eqnarray} &&\nu =\frac{{S}_{\mathrm{t}}^{2}}{{n}_{\mathrm{t}}{X}_{\mathrm{t}}^{2}}+\frac{{S}_{\mathrm{c}}^{2}}{{n}_{\mathrm{c}}{X}_{\mathrm{c}}^{2}} \end{eqnarray} \tag{ 2 }$$

where n_t and n_c are the sample sizes for the warming and control treatments, respectively, and S_t and S_c are the standard deviations for the warming and control groups, respectively.

The weighted response ratio (RR₊₊) of each ecosystem type is calculated from the RR of individual pair comparisons between warming and control groups, and the standard errors (S) and weighting factor (w_ij) are calculated by:

$$\begin{eqnarray} {\mathrm{RR}}_{++}&=&\frac{\sum _{i=1}^{m}\sum _{j=1}^{k i}{w}_{i j}{\mathrm{RR}}_{i j}}{\sum _{i=1}^{m}\sum _{j=1}^{k i}{w}_{i j}} \end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray} S({\mathrm{RR}}_{++})&=&\sqrt{\frac{1}{\sum _{i=1}^{m}\sum _{j=1}^{k i}{w}_{i j}}} \end{eqnarray} \tag{ 4 }$$

$$\begin{eqnarray} {w}_{i j}&=&\frac{1}{\nu }. \end{eqnarray} \tag{ 5 }$$

If the 95% confidence interval (95% CI = RR₊₊ ± 1.96S(RR₊₊)) value of RR₊₊ for a variable does not cover zero, the responses of soil moisture to experimental warming differ significantly between the warming and control treatments at a given ecosystem type (Lu et al 2011a, 2011b, Lu et al 2013).

We plotted frequency distributions of RR to display variability among individual studies. The frequency distributions were assumed to follow normal distributions and fitted by a Gaussian function (i.e., normal distribution).

$$\begin{eqnarray} &&y=a\exp \left [-\frac{(x-\mu )^{2}}{{2 \sigma }^{2}}\right ] \end{eqnarray} \tag{ 6 }$$

where x is RR, y is the frequency (i.e., number of RR values), a is a coefficient showing the expected number of RR values at x = μ, and μ and σ are the mean and variance of the frequency distributions of RR, respectively. We used a t-test to examine whether the response ratio in the warming treatment was significantly different from that in control. The percentage change of a variable was calculated by the formula: (e^RR₊₊ − 1) × 100% (Lu et al 2013).

Our results showed significant negative effects of experimental warming on soil moisture in most study sites. The overall responses of soil moisture to warming decreased significantly from 8.50% to 11.73% in forest, grassland and cropland (figure 1, table 1). The negative responses of soil moisture to warming conditions could be attributed to the enhanced plant growth and transpiration processes (Wan et al 2002, Xia et al 2010). Evaporation from the soil surface and transpiration by plants are major avenues of water loss from the soil reservoir (Chapin et al 2002). Previous studies have showed that warming can alter plant phenology, leading to changes such as earlier flowering and leafing (Wolkovich et al 2012), increased height and cover of plants (Walker et al 2006), and prolonged growing season (Sherry et al 2007), which may enhance plant growth and eventually increase losses of water from plant transpiration (Xia et al 2010).

Experimental warming stimulated plant growth from both above- and below-ground NPP, leading to increased NPP (figures 4(a)–(c); table 1). Our synthesized results are consistent with those from other meta-analyses that focused on the response of above-ground plant growth to warming (Rustad et al 2001, Lin et al 2010, Wu et al 2011). For example, the latest study indicated that plant growth (NPP, ANPP and BNPP) was enhanced under warming conditions in 16 experimental samples (Lu et al 2013).

The warming induced significant increase LAI in tundra, while shows insignificant changes in shrubland (figures 4(d) and (e)). Previous studies supported our conclusion over various regions. A meta-analysis showed that warming increased plant height and cover at 11 sites across the tundra biome (Walker et al 2006). Another study showed that warming treatment increased the effective green leaf area index (green LAI) at most sites from six European shrublands located along a north–south climatic gradient (Mänd et al 2010). An increased LAI will strengthen plant transpiration and eventually exacerbate the losses of soil moisture with warming.

The responses of soil moisture to warming may differ considerably among ecosystem types. The results obtained in this synthesis show that experimental warming significantly decreased soil moisture in forests, grasslands and croplands (p < 0.05; figures 1(b), (d), (f) and 2), while the warming induced soil moisture was insignificant in tundra and shrubland (p > 0.05; figures 1(c), (e) and figure 2). Although the weighted response ratio (RR₊₊) of tundra was negative, decreases of soil moisture were not significant (figure 1(c)), owing to changes in surface hydrology induced by warming. In general, tundra ecosystems are located at high latitude regions, where permafrost thaw is an important source of soil moisture in addition to precipitation. Experimental warming may accelerate permafrost melting, leading to increases in soil moisture (Natali et al 2011).

An insignificant decrease in soil moisture was also found in shrublands. More than half of shrubland sites showed increased soil moisture under warming treatments; however, it should be noted that the increase was very low (figure 1(e)). According to previous studies, increases in soil moisture in response to experimental warming were not significant when compared with control treatments (Beier et al 2004). Several studies have reported that warming decreased soil moisture via enhanced plant growth, as indicated by increasing shrub growth and biomass with experimental warming (Klein et al 2007, Peñuelas et al 2007, Mänd et al 2010). Due to the limited experimental data available for shrubland ecosystems, the current meta-analysis can not reflect comprehensive responses of soil moisture. Accordingly, this analysis should be repeated when more experimental data are available.

Environmental (e.g., mean annual temperature and mean annual precipitation) and ecosystem management (e.g., clipping, grazing) may potentially influence the responses of soil moisture to experimental warming since we extracted data from diverse manipulative experiments at a global scale. Ecosystem management (e.g., clipping or grazing) may indirectly accentuate soil water evaporation by reducing shading of the soil surface, which may in turn lead to large decreases in soil moisture (Xue et al 2011). When compared with the results of a ten years warming study, decreases in soil moisture were larger in clipped than unclipped plots with warming in a grassland in Oklahoma, USA (Luo et al 2009b, Xu et al 2012). The topography (e.g., slope) also indirectly impacts soil water evaporation through different plant cover and species composition (Liancourt et al 2012). Analyses revealed that the soil drying rate was faster on the drier upper slope or with vegetation and faster overall in drier years under warming conditions (Liancourt et al 2012). Moreover, vegetation can not only increase soil drying through plant transpiration, but also by bringing water up from deeper soil layers (Horton and Hart 1998), which may further desiccate deeper soil layers.

Moreover, warming induced phenology changes will strongly influence soil moisture response. In a previously conducted 2-year field experiment, simulated warming increased soil moisture by 5–10% during the end of the growing season (Zavaleta et al 2003). Analyses found that warming accelerated the decline of canopy greenness (normalized difference vegetation index) each spring by 11–16% by inducing earlier plant senescence. Lower transpiration losses resulting from the earlier senescence provide a possible mechanism for the unexpected increases in soil moisture.

As an important variable for terrestrial ecosystem structure and function, soil moisture will undergo substantial changes in response to warming, significantly influencing biogeochemical and biogeography processes as well as climate systems (Dai et al 2004, Asharaf et al 2012). Our results showed significantly decreased trends of soil moisture in most ecosystem types (i.e., forest, grassland and cropland), which will result in profound impacts of the vegetation production. Numerous studies showed that soil drying will restrict the plant growth in water-limited ecosystem (Austin and Sala 2002, Schenk and Jackson 2002, Yuan et al 2007). A global analysis employing monthly data records from 238 micrometeorological tower sites distributed globally across 11 biomes revealed that vegetation production was 50% more sensitive to drought than ecosystem respiration (Schwalm et al 2012). Moreover, with decreasing soil moisture, plants will alter their carbon allocation strategy, promoting allocation of photosynthetic product to the root and decreasing its allocation to the stems and leaves (Schenk and Jackson 2002, Mokany et al 2006). More roots may bring water up from deeper soil layers (Horton and Hart 1998), leading to soil drying at deeper soil layers (Shaver et al 2000).

The impacts of soil drying on soil carbon showed large uncertainty (Falloon et al 2011). Laboratory studies and field observations suggest that soil respiration is low under dry conditions, reaches the maximal rate in intermediate soil moisture levels, and decreases at high soil moisture content when anaerobic conditions prevail to depress aerobic microbial activity (Luo and Zhou 2010). Therefore, in humid regions, soil drying alone appears to increase soil respiration and result in soil carbon losses (Smith et al 2005). On contrary, in arid and semi-arid regions, soil moisture changes alone appear to have acted to increase soil carbon storage. This was presumably because drying generally acts to increase soil carbon storage by reducing the respiration rate (Falloon et al 2011).

Soil moisture modulates the land surface energy balances by changing the land surface characteristics and energy components (Chapin et al 2002). Changes in soil moisture can alter land surface albedo and soil thermal capacity (Wang et al 2005). In general, both albedo and emissivity of dry soil are larger, which decreases absorbed shortwave radiation and increases emitted long wave radiation, respectively, leading to local energy reduction (Chapin et al 2002). Therefore, the present study highlights the importance of the responses of soil moisture to warming; however, further investigations simulating soil moisture coupling with the responses of other climate variables to climate change are necessary.