Consequences of Changing Precipitation Patterns for Ecosystem Functioning in Grasslands: A Review

Abstract

Grassland ecosystems worldwide provide agricultural goods and important ecosystem services. Productivity and other ecosystem processes in grasslands are, in most cases, strongly linked to the ecosystems’ water status, a factor that is predicted to experience major alterations with global climate change. Future predictions include changes in the amount, distribution, frequency, and intensity of precipitation, which, particularly in grasslands, may have important consequences for ecosystem state and functioning. This review analyses the effects of experimental precipitation manipulation on plant productivity, species diversity, soil/ecosystem respiration, and soil nitrogen in grassland-type ecosystems over a wide range of climate types, synthesising the results from 72 studies.

We found that sensitivity of ecosystem processes to changes in precipitation amounts increased with aridity. In addition, ecosystem processes were more responsive to precipitation addition than to precipitation reduction. However, we did observe high resilience of grassland ecosystems to both changing precipitation amounts and variability, which may be explained by the fact that the applied manipulation scenarios often lie within the range of the natural inter-annual precipitation variability experienced by ecosystems, and by evolutionary adaptation of grassland ecosystems to these natural inter-annual differences. Long-term effects of altered precipitation regimes on ecosystem processes, i.e. by changes in species composition and soil properties, are rarely covered within the time frame of most studies and thus cannot be ruled out as a possible consequence of a gradually changing climate.

Increasing the comparability between individual precipitation manipulation studies is needed to facilitate the evaluation of ecosystem responses to altered precipitation regimes. We recommend future precipitation manipulation studies to aim at capturing possible long-term effects with comparable designs and standardised data compilation.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

1.1 Climate Change Influence on Precipitation Regimes

Anthropogenic fossil fuel emissions continue to impinge on global climate change, resulting in ecosystems worldwide being subjected to altered temperature and precipitation regimes. Basic theory and empirical evidence suggest that state, functioning and service provision of ecosystems worldwide are increasingly influenced by this development. A major goal in current research is to achieve a comprehensive understanding of the possible consequences of climatic changes for ecosystem processes and subsequently develop adequate mitigation strategies. However, in view of the complex tangle of co-dependent ecosystem processes (e.g. productivity, biodiversity, mineralisation, soil carbon and nutrient cycling and storage), all potentially exhibiting differential responses to environmental change, the scientific community is still far from achieving this aim.

One of the major factors affected by climate change is the water status of terrestrial ecosystems all over the globe. With the ongoing temperature increase, and a 7 % higher water holding capacity of the atmosphere per 1 °C warming (Wentz et al. 2007; Trenberth 2011), overall evaporation and atmospheric water vapour concentrations will increase. The latter will promote cloud formation in humid regions (Meehl et al. 2007; Trenberth 2011), leading to increases in precipitation (Fig. 1). In contrast, arid regions are expected to experience a decrease in precipitation (Fig. 1), as here adequate surface moisture, a determinant of increased evaporation, is lacking (Meehl et al. 2007; Trenberth 2011). Indeed, long-term observations for the period between 1900 and 2005 have already demonstrated significant increases in precipitation in eastern North and South America, northern Europe, and northern and central Asia, and significant decreases in precipitation in the Sahel, southern Africa, the Mediterranean, and southern Asia (Trenberth et al. 2007). In addition to the expected changes in precipitation amount, the type of precipitation might change in temperate regions, with much of the precipitation usually falling as snow increasingly falling as winter rain, thus reducing water storage in snow packs, and concomitantly reducing water availability in summer (Trenberth et al. 2007).

Fig. 1

Multi-model mean changes in precipitation (mm day⁻¹). Regions are stippled where at least 80 % of the climate models agree on the sign of the mean change. Changes are annual means for the SRES A1B scenario for the period 2088–2099 relative to 1988–1999. Taken from Meehl et al. (2007, their Fig. 10.12)

While alterations in the amount of precipitation may have large implications for ecosystem functioning in the affected regions, precipitation distribution, frequency and intensity are regarded as nonetheless important (Easterling et al. 2000). Over the twentieth century, estimates suggest that atmospheric water vapour concentrations have increased by 5 %, which has generally increased the intensity of precipitation events (Trenberth et al. 2007). Scenarios predict an almost universal increase in precipitation intensity (Fig. 2a), although particularly at middle and high latitude regions where mean precipitation (Fig. 1) is also projected to increase (Meehl et al. 2007). However, both increases and decreases in consecutive dry days between precipitation events can be found (Fig. 2b), with regions located in the subtropics and lower mid-latitudes exhibiting an increased run of dry days, thus having a concomitant greater risk of drought (Meehl et al. 2007).

Fig. 2

Changes in spatial patterns of simulated precipitation intensity and dry days between two 20-year means (2080–2099 minus 1980–1999) for the A1B scenario. Stippling denotes areas where at least five of the nine models concur in determining that the change is statistically significant. Changes are given in units of standard deviation, following Frich et al. (2002). Taken from Meehl et al. (2007, their Fig. 10.18)

In this review we will concentrate on understanding the effects of altered precipitation regimes on one of the most important terrestrial ecosystem types worldwide: grasslands. With precipitation being one of the most important determinants of the majority of ecosystem processes, the observed and predicted changes in amount, intensity and frequency of precipitation are particularly threatening for future stability and functioning of these ecosystems.

1.2 Grassland Ecosystems

Grasslands are disturbance-dependent terrestrial ecosystems dominated by graminoid and herbaceous vegetation in climates with a distinct seasonality of productivity (Smith 1973), which are maintained by fire, grazing, drought, and/or freezing temperatures (Anderson 1982). These factors provide selective pressure for a short ruderal life cycle involving early reproduction with a high number of seeds, a high turnover of aboveground biomass, high belowground carbon investment and the location of perennating organs near the soil surface (Sala et al. 1996), which promotes the dominance of graminoids and forbs. However, grasslands encompass not only non-woody systems but also savannas, woodlands, shrublands, and tundra (White et al. 2000). Estimates of the extent of the earth’s land area in grasslands (excluding Greenland and Antarctica), depending on land cover characterisation, range from ~42 to 56 million km², or ~31 to 43 % (Whittaker and Likens 1975; Atjay et al. 1979; Olson et al. 1983). Figure 3 gives the global distribution of grasslands, with the largest areas found in central and southern Asia, southern South America, Africa and central North America (Sala et al. 1996).

Fig. 3

Grasslands and climate zones. From White et al. (2000) and UNEP (1992)

Grassland ecosystems are of high economic importance for provisioning of agricultural goods as, together with the livestock they sustain, they constitute one of the earth’s major food resources (Singh et al. 1983). Apart from their vital role in food production, grasslands goods and services include wildlife and biodiversity conservation, resource storage, prevention of soil degradation as well as supporting tourism and recreational activities, and offering aesthetic and spiritual gratification (White et al. 2000). The provision of these ecosystem services depends on the maintenance of grassland ecosystem state (Miller et al. 2011). As the extent of drought, fire and grazing determines the state transition of grasslands into deserts or shrublands and forest ecosystems (Sala et al. 1996), state shifts may easily occur with climate and land use change, afforestation, eutrophication, or the invasion of neophytes (Faber-Langendoen and Josse 2010), with restoration of previous conditions being difficult, costly, or effectively unfeasible (Miller et al. 2011). In relation to changing temperature and precipitation patterns, this problem has increased the interest of grassland ecosystem research on biotic and abiotic attributes conferring ecosystem resilience with changing environmental conditions and thus subsequently reducing system susceptibility to state shifts.

1.3 Drivers of Ecosystem Processes in Grasslands

There are a range of factors controlling ecosystem processes in grassland ecosystems, of which precipitation and temperature are thought to exhibit the strongest ties to grassland functioning (Sala et al. 1996). Accordingly, a wide range of studies describe the relationship of these factors with important ecosystem traits such as productivity, biodiversity, soil and ecosystem carbon cycling, and soil nutrient dynamics. Precipitation and consequently soil moisture have been shown to modulate the carbon cycle of grassland ecosystems, with arid systems generally exhibiting significantly lower plant productivity (e.g. Huxman et al. 2004a), biodiversity (Sala et al. 1996), soil and ecosystem carbon fluxes (e.g. Merbold et al. 2009), and soil nutrient cycling (Aranibar et al. 2004), as compared to semi-arid or mesic systems. However, negative effects of precipitation on the carbon and nitrogen cycles in grasslands have been observed, particularly with large precipitation pulses interrupting long dry periods (e.g. Kim et al. 2012; Unger et al. 2012). These precipitation pulses can lead to large carbon and nitrogen losses through high soil respiration rates (Kim et al. 2012) and leaching of nitrogen, particularly nitrate, below the rooting zone (Austin et al. 2004). In contrast to arid and semi-arid grasslands, ecosystem processes in mesic grasslands, where moisture is not limiting, are controlled by temperature, with increasing temperatures generally enhancing productivity (e.g. Flanagan and Adkinson 2011) and soil microbial activity (e.g. Davidson et al. 2006), thus resulting in higher carbon and nitrogen cycling (Aranibar et al. 2004).

Apart from the climatic controls on ecosystem functioning in grasslands, natural disturbance factors in the form of fire and grazing or land use factors like fertilisation, mowing, and tillage are known to affect ecosystem functioning. While disturbance is necessary to maintain ecosystem state in grasslands, particularly in the more humid regions, where natural succession favours tree growth (Anderson 1982), an increase in disturbance with land use intensification or even transformation of grasslands into croplands has a negative effect on ecosystem functioning and stability leading to diversity loss, increasing ecosystem vulnerability to climate change or irreversible ecosystem collapse (MacDougall et al. 2013).

1.4 Experimental Designs to Study Precipitation Impacts at the Ecosystem Scale

Knowledge of the impacts of alterations in precipitation on ecosystem processes can be gained from multi-year observational records. For example, over a 20-year period (1986–2005), inter-annual variation in aboveground productivity in a Mediterranean grassland was found to correlate with annual precipitation (Vázquez-de-Aldana et al. 2008). However, a 22-year measurements series (1982–2003) from an Inner Mongolia grassland revealed a correlation of aboveground productivity with previous-year precipitation (Ma et al. 2010).

More recently, several studies have interpreted multi-year eddy flux data to explain productivity–precipitation relationships. For example, on an annual basis, ecosystem CO₂ exchange correlated with annual precipitation in a Mediterranean grassland in Portugal (Jongen et al. 2011), whereas the inter-annual variation in ecosystem CO₂ exchange in a Mediterranean grassland in California depended primarily on the timing of precipitation, rather than total annual precipitation (Ma et al. 2007).

However, analysis of impacts of precipitation on ecosystem state using both of the above-mentioned approaches, observational records or eddy flux measurements, may be confounded by other co-varying factors. Precipitation manipulation experiments enable replication, control for confounding factors, and allow for multiple scenarios to be studied simultaneously. They can therefore contribute to our understanding of impacts of precipitation on ecosystem state. Water manipulation experiments at the pot and mesocosm scale are widely conducted (e.g. Li et al. 2011; Nagy et al. 2013; O’Brien et al. 2013). However, they do not necessarily reflect natural conditions, often are of short duration, and are limited in the number of parameters studied (i.e. through destructive sampling). Extending the outcome of these studies to the field or ecosystem scale or incorporating their results into modelling approaches is therefore in most cases not appropriate. To achieve a thorough picture of potential effects of changes in precipitation on ecosystem functioning, field studies with experimentally manipulated precipitation in natural ecosystems with global representation are needed.

Although not always a requirement (e.g. in those studies with only simulation of addition scenarios), precipitation manipulation experiments often involve the use of rainout shelters, which enables control of the amount and/or timing of precipitation received by the vegetation. To date three different designs have been employed:

• 1. ‘Closed shelters’ (Fig. 4a), covered by a complete roof, which excludes almost all precipitation from the experimental plots. Consequently, the researcher has full control over the precipitation regime, generally through fixed sprinklers. Although shelter sides and ends remain open to maximise air movement and minimise temperature and relative humidity artefacts, closed shelters have the disadvantage of altering the microclimate, in particular solar radiation (Fay et al. 2000). Comparison to unsheltered control plots to evaluate the impacts of the shelter is thus desirable (Owens 2003). However, closed shelters have the advantage of being easily constructed and dismantled and comparatively inexpensive, allowing for a necessary number of replications. There are a range of shelter designs with different sizes, supports, and roofing materials, the choice of which depends largely on site-specific demands and research objectives (e.g. short-term versus long-term investigations). Nevertheless, shelters should be sufficiently large to allow for the exclusion of edge effects. The RaMPS (Rainfall Manipulation Plot Study) installed at the Konza Prairie Biological station in northeast Kansas, USA, is an example of this experimental set-up, with timing and quantity of precipitation being experimentally manipulated since 1997.

  Fig. 4

  

  Experimental designs to study the effects of precipitation manipulation: (a) closed shelters after Jongen et al. (2013b); (b) throughfall shelters, after Yahdjian and Sala (2002), picture from K. Tielbörger; and (c) movable shelters, after Báez et al. (2013), picture from S. Collins

• 2. ‘Throughfall shelters’ (Fig. 4b), originally based on the design of Yahdjian and Sala (2002), have angled roofs composed of bands of transparent acrylic, blocking a certain amount of precipitation. This design gives the possibility to collect the intercepted rain and subsequently use it to irrigate other experimental plots (e.g. Holub et al. 2013), making throughfall shelters useful for manipulation experiments where a combination of addition/reduction scenarios is investigated. However, the throughfall shelter design does not allow for a full control of the amount and timing of precipitation received by the experimental plots, with vegetation being subjected to intra- and inter-annual variations in the amount of precipitation. In comparison to closed shelters, alterations in microclimate are reduced, with wind convection, temperature, and solar radiation less affected (Gherardi and Sala 2013).

• 3. ‘Movable shelters’ (Fig. 4c) have a design with roofs or curtains that slide diagonally along rollers to cover the experimental plots, activated by a rain sensor (e.g. Báez et al. 2013). These shelters have the advantage that the effect of the shelter on the microclimate is small. Nevertheless, care must be taken to ensure that the parked shelter does not create shade on a part of the experimental plot (Owens 2003). The disadvantages of movable shelters are higher costs for acquisition and maintenance.

1.5 Problems Associated with Precipitation Manipulation Experiments

In order to adapt current models on future ecosystem behaviour with climate change, manipulation experiments should cover the global range of ecosystems under representative current and future climate scenarios (Beier et al. 2012). In a comprehensive review, Beier et al. (2012) describe that, whereas precipitation manipulation experiments in grasslands and forest ecosystems are widely conducted, other important ecosystem types, such as arable lands and tundra, are underrepresented. In addition, experiments representing the Southern hemisphere, in particular Africa, and experiments conducted in high rainfall zones (>1.500 mm) were scarce. Additionally, the chosen scenarios are often relatively conservative and related to historical or current conditions, while in the future many ecosystems are likely to be exposed to climates exceeding historical and current climatic variations (Beier et al. 2012; IPCC 2012). In many studies, the applied precipitation regimes lie within the range of natural year-to-year climatic variability, explaining the often-found resilience of many ecosystem processes. Therefore, a major issue with precipitation manipulation experiments is to choose the right climate change scenarios for the ecosystems under study. Further issues identified by Beier et al. (2012) are the (1) relatively short duration of most studies, which therefore do not reflect long-term integrated effects on ecosystem performance that might be expected from a future climate scenario, (2) the lack of appropriate reference conditions (i.e. controls being subjected to large inter-annual variation), (3) the unequal distribution of precipitation, caused by some irrigation set-ups or the presence of slopes and soil type gradients, (4) the lack of grazing in fenced off areas altering plant communities and productivity measures, and (5) the often limited plot size constraining sampling strategy and exacerbating problems associated with edge effects and disturbance caused during sampling and measurements. However, many of these issues are hard to avoid and have to be accounted for when incorporating data into modelling approaches.

2 Methods

2.1 Data Compilation

In this review on the effects of changing precipitation patterns on ecosystem processes, data collection was restricted to studies in which precipitation was experimentally manipulated in the field, as they represent the best way to explore cause–effect relationships between water availability and ecosystem functioning. We incorporated studies conducted in grassland ecosystems in the broad sense, i.e. grasslands, savannas, woodlands, shrublands, and tundras (White et al. 2000). In addition, we only incorporated those studies with a precipitation manipulation for a period of at least two consecutive months. For each selected study, we collected information on ecosystem type, latitude, longitude, mean annual temperature (MAT), mean annual precipitation (MAP), data on experimental duration and set-up, and magnitude of manipulation. In the case of multifactor experiments (e.g. precipitation in combination with warming, nutrients, or CO₂), we only used data of the precipitation treatments, in relation to the controls, with the other factors kept at ambient levels. In relation to ecosystem type we used the following criteria: ecosystems with MAP < 300 mm were denoted as arid, MAP between 300 and 600 mm as semi-arid, and MAP > 600 mm as mesic. Studies in which the amount of precipitation was manipulated (addition or reduction) are assembled in Table 1, while studies aiming at altering the variability of precipitation, without changing the absolute amount of rainfall received by the vegetation, are assembled in Table 2.

Table 1 Summary of precipitation manipulation studies with rainfall addition or reduction—effects on productivity, biodiversity, respiration, and soil nitrogen

Table 2 Summary of precipitation manipulation studies with changes in the distribution of precipitation

2.2 Data Analysis

Collected data on the magnitude of change with precipitation manipulation was grouped into three categories of response variables: (1) productivity, including aboveground net primary productivity (ANPP), belowground net primary productivity (BNPP), absolute growth rate (AGR), gross ecosystem photosynthesis (GEP), vegetation cover, and additional growth parameters (e.g. shoot length, leaf length, number of leaves), (2) diversity, including data on species richness (i.e. number of species), species composition, and diversity indices such as the Shannon-Wiener and Simpson’s indices, and (3) soil respiration (R _S), ecosystem respiration (R _ECO) and responses of soil nitrogen (N-availability, N-mineralisation or concentration of nitrate, ammonium, or total N). For multi-year studies, we calculated the average response for inclusion in the analysis. The aridity index (AI) was calculated according to Köppen (1923), with AI = MAP/(MAT + 33). To normalise the responses to the magnitude of the treatment imposed, we calculated the sensitivity index as the % change in response variable divided by % change in precipitation following the manipulation. The sensitivity index yields positive values if the response is unidirectional with the treatment imposed (i.e. productivity increase with water addition, or productivity decrease with water reduction). Finally, for the Partial Least Squares Regression (PLSR) analysis we differentiated the data according to ecosystem (arid, semi-arid, and mesic), with an additional climate classification: cold (MAT < 5 °C), temperate (5 °C < MAT < 15 °C), and warm (MAT > 15 °C), thus giving a set of nine different biomes. PLSR (Wold et al. 1983, 2001) was applied to model percentage change in aboveground productivity (response variable, Y data) using MAT, MAP and percent manipulation as explaining variables (X data). By visually inspecting the scores and loading plots, the main factors determining the percent change in productivity can be assessed depending on ecosystem types, and relationships between potential explaining variables can be evaluated. X and Y data were mean centred and weighted by 1/(standard deviation). NIPALS algorithm was used. PLSR calculations were performed using the software package The Unscrambler X 10.3 (CAMO Software AS, Oslo, Norway).

3 The Impact of Changes in the Amount of Precipitation

3.1 Precipitation Addition

A total of 61 publications were included in the analysis for effects of changing precipitation quantity on ecosystem processes, 44 of which describe results of rainfall addition experiments, with a total number of 34 ecosystems (7 arid, 13 semi-arid, and 14 mesic systems) and 45 addition scenarios studied (Table 1). Addition experiments usually do not require shelters, and are often conducted in combination with precipitation exclusion, using the intercepted rainfall of a throughfall manipulation experiment (e.g. Talmon et al. 2011; Hoeppner and Dukes 2012; Byrne et al. 2013; Flanagan et al. 2013). The applied addition scenarios increased precipitation between 20 and 115 %, with the exception of McCulley et al. (2007), adding 300 % of rainfall. Some of these addition studies were conducted in ecosystems where climate change scenarios predict an increase in precipitation, such as in northern Europe (Press et al. 1998; Robinson et al. 1998; Keuper et al. 2012; Weijers et al. 2013), central Asia (Niu et al. 2008; Liu et al. 2009; Yan et al. 2010, 2011a, 2011b), or cold tundra regions in northern Russia (Keuper et al. 2012) and North America (McGraw 1985). However, several other addition studies, conducted in warm and temperate arid and semi-arid systems (e.g. Harpole et al. 2007; Evans et al. 2011; Talmon et al. 2011; Potts et al. 2012; Báez et al. 2013; Xu et al. 2013), investigated the effect of an increase in precipitation, although this is in contrast to future climate change scenarios, with these studies generally done by mere opportunity (e.g. using intercepted precipitation from throughfall shelters) or aiming at studying the effects of inter-annual variation in precipitation. The main parameters studied in precipitation addition experiments in the field were aboveground productivity (AP), totaling 40 observations, and soil/ecosystem respiration (R _S/ECO), with 16 observations; less attention was given to effects on biodiversity (9 observations) and soil nitrogen properties (6 observations).

Figure 5 shows the variation of percentage response in the above-mentioned parameters to precipitation addition scenarios. None of the addition studies reported a negative response to any of the studied parameters. Both AP and R _S/ECO showed the largest response to precipitation addition in arid regions, with average increases of 54 % and 46 %, respectively. Parameter increases in semi-arid (40 % for AP, and 31 % for R _S/ECO) and mesic regions (20 % for AP and 21 % for R _S/ECO) were lower. A 0-response was shown for diversity (Fig. 5d) in semi-arid and mesic ecosystems, a result which has to be interpreted carefully due to the low observation count. Similarly, 5 out of 6 studies reporting on soil N properties in semi-arid and mesic ecosystems gave a 0-response (Fig. 5h). In arid ecosystems, none of the studies reported on diversity and soil N properties (Fig. 5d, h). The highest percentage responses to precipitation addition were found in an arid shrubland in central China, with maximum increases of 266 % in AP and 59 % in R _S (Song et al. 2012), and a semi-arid Mediterranean grassland in California, with maximum increases of 275 %, 30 %, and 60 % in AP, R _ECO, and soil N, respectively (Potts et al. 2012).

Fig. 5

Boxplot of variation in the range of percentage change in (a, b) aboveground productivity (AP), (c, d) diversity, (e, f) soil/ecosystem respiration (R _S/ECO), and (g, h) soil N properties observed in arid, semi-arid, and mesic ecosystems (see Table 1) in response to precipitation addition (a, c, e, g) or reduction (b, d, f, h). Number (n) refers to sample size. Boxplots visualise the first, second, and third quartile, and the mean values (dotted lines), with whiskers indicating the maximum and minimum. Outliers are shown as dots

Relating the percentage of manipulation to the percentage of productivity response (Fig. 6) or respiration response (Fig. 7) for all addition studies revealed a higher responsiveness of arid and semi-arid ecosystems to precipitation addition, as compared to mesic ecosystems. A 50 % increase in the amount of precipitation would result, on average, in a productivity increase of 37 %, 27 %, and 23 % in arid, semi-arid, and mesic ecosystems, respectively (Fig. 6). In addition, respiration would increase by 29 %, 35 %, and 13 % in arid, semi-arid, and mesic ecosystems, respectively, with a 50 % water addition (Fig. 7). However, the respiration results have to be regarded with care, due to the low number of replicate studies. Further, site-specific differences in other well-known drivers of soil respiration, such as soil nutrient and carbon availability, can be expected to affect these results.

Fig. 6

Correlation between the percentage manipulation versus the percentage change in aboveground productivity (AP), for arid, semi-arid, and mesic ecosystems

Fig. 7

Correlation between the percentage manipulation versus the percentage change in soil/ecosystem respiration (R_S/ECO), for arid, semi-arid, and mesic ecosystems

Considering the strong moisture relationship with plant productivity generally found in terrestrial ecosystems (e.g. Nippert et al. 2006), it is remarkable that many of the applied addition scenarios did not result in significant increases in ecosystem processes. In total, 18 out of the 40 studied scenarios did not find increases in productivity responses, with approximately equal representation of arid and semi-arid ecosystems (44 % and 40 %, respectively), while 8 out of the 16 addition scenarios in mesic systems did not result in a productivity response. Notably, the four studies finding no response in arid systems were all performed in cold Tundra climates (McGraw 1985; Robinson et al. 1998; Keuper et al. 2012; Weijers et al. 2013). Respiration was generally found to be more responsive to water addition than productivity, with only 3 out of 16 studied scenarios reporting no effects.

3.2 Precipitation Reduction

Of the 61 publications that were included in the analysis for effects of changing precipitation quantity on ecosystem processes, 41 described results of rainfall reduction experiments (Table 1). In total, these studies covered 38 ecosystem types (5 arid, 12 semi-arid, and 21 mesic systems) with 52 reduction scenarios applied. Experiments were performed using any type of the shelter designs described in Sect. 1.4, with the applied scenarios reducing precipitation between 20 and 100 % of natural precipitation during the experimental period. Most studies were conducted in ecosystems with future climate change prognoses indicating decreasing amounts of precipitation (e.g. Miranda et al. 2009, 2011; Evans et al. 2011; Talmon et al. 2011; Cherwin and Knapp 2012; Potts et al. 2012; Báez et al. 2013; Byrne et al. 2013). However, some studies, especially those conducted in northern European mesic ecosystems, chose to study precipitation reduction without a clear consensus of climate model predictions (e.g. CLIMAITE project studies of Emmett et al. 2004; Peñuelas et al. 2007).

AP, totaling 48 observations, was the most studied parameter, followed by biodiversity (18 observations), R _S/ECO (17 observations), and soil N properties (11 observations). None of the featured reduction studies reported a positive response of either parameter with a decrease of precipitation (Fig. 5). Both biodiversity and soil N were found to be resilient to water reduction (Fig. 5c, g), with the exception of biodiversity in two arid ecosystems, exhibiting a negative response of 60 % and 27 % (Fig. 5c; Miranda et al. 2009), and a semi-arid grassland giving a 65 % decrease in soil N availability (Potts et al. 2012).

AP and R _S/ECO showed the highest responsiveness to precipitation reduction in arid ecosystems, with average decreases of 31 and 33 %, respectively. Parameter decreases in semi-arid (17 % for AP, 18 % for R _S/ECO, respectively) and mesic ecosystems (12 % for AP, 7 % for R _S/ECO, respectively) were considerably lower (Fig. 5). Most responsive to precipitation reduction was an arid steppe in Mongolia, with a decrease in AP of ~75 % (Shinoda et al. 2010), and a semi-arid warm grassland in California (Potts et al. 2012), where AP decreased by 70 % in response to a mere 23 % water reduction. Surprisingly large reductions in AP were reported for an Alpine grassland (mesic ecosystem), where a 2–3 month exclusion of precipitation resulted in a decrease in AP of ~67 % (Gilgen and Buchmann 2009). In addition, in an arid warm grassland ecosystem in Spain (Miranda et al. 2009), exposed to a 50 % precipitation reduction, AP was reduced by 60 %.

Relating the percentage of manipulation to the percentage of productivity response (Fig. 6) or respiration response (Fig. 7) for all reduction studies revealed a higher responsiveness of arid ecosystems to precipitation reduction, as compared to semi-arid and mesic ecosystems. On average, a 50 % precipitation reduction would lead to productivity decreases of 37 %, 13 %, and 8 % in arid, semi-arid, and mesic grassland ecosystems, respectively (Fig. 6). However, the magnitude of responses of R _S/ECO to precipitation reduction was substantially lower, with average decreases of 14 % and 4 % in semi-arid and mesic ecosystems, respectively, with a 50 % precipitation reduction (Fig. 7). The response of arid systems could not be estimated due to the lack of observations.

In total, 24 out of the 48 studied precipitation reduction scenarios found no significant treatment responses in AP (38, 45, and 69 % of the studies in arid, semi-arid, and mesic systems, respectively). With regard to R _S/ECO, 7 out of the 17 studied scenarios reported the absence of a significant treatment effect, with this non-responsiveness particularly pronounced in mesic ecosystems (5 out of 9 studies).

3.3 Sensitivity to Changes in the Amount of Precipitation

Our findings (Figs. 5, 6, and 7) indicate a hierarchy in the responsiveness of ecosystems to water manipulation, with largest responses to changing precipitation amounts in arid ecosystems, immediately followed by semi-arid ecosystems, while most mesic ecosystems were resilient to both water addition and reduction. In addition, in all three ecosystem types, responsiveness to water addition was higher as compared to water reduction.

The magnitude of the changes in a response parameter likely depends on the magnitude of the exposed addition/reduction scenario. For example, Song et al. (2012), being the only study including more than one addition scenario, reported a progressively increasing productivity response (from 7 to 266 %), with increasing precipitation addition (from 25 to 100 %). Similarly, several studies, including two reduction scenarios (e.g. Evans et al. 2011; Cherwin and Knapp 2012; Evans and Burke 2013), reported higher parameter reductions with increasing precipitation reduction. Thus, the higher responsiveness of arid ecosystems to precipitation may be due to a coincidentally higher amount of manipulation in these ecosystems. To estimate the possible impact of this effect, we calculated a sensitivity index, by weighing the ecosystem parameter response to the relative magnitude of exposed manipulation. Subsequently, we evaluated the effect of aridity index (after Köppen 1923), which accounts for both temperature and precipitation, by combining MAP and MAT, on the sensitivity index. Note that by definition a high aridity index means a humid climate while a low aridity index means an arid climate.

Figure 8 demonstrates this cause–effect relationship, with sensitivity of ecosystem parameters to changes in the amount of precipitation strongly decreasing with the aridity index. The average sensitivity index to water addition was 0.70, 0.77, and 0.18, while reduction scenarios gave values of 0.44, 0.40, and 0.14 for arid, semi-arid, and mesic ecosystems, respectively.

Fig. 8

Sensitivity index of arid, semi-arid, and mesic ecosystems with precipitation reduction or addition versus Köppen aridity index

Deviating from the sensitivity–aridity relationship are the results reported in Potts et al. (2012). In this study, a semi-arid Mediterranean-type grassland, subjected to relatively small manipulations of the precipitation amount, gave exceptionally high sensitivity indices in both addition scenarios (5.3, 0.6, and 1.1 for AP, R _ECO, and soil N, respectively) and reduction scenarios (3.0, 2.2, and 2.8 for AP, R _ECO, and soil N, respectively). However, the study was performed during a year characterised by natural drought, with severe consequences for the ambient treatment, which highlights the importance of considering climate conditions in the controls with precipitation manipulation studies. Omitting this study, the sensitivity indices for semi-arid ecosystems are smaller, i.e. 0.57 with water addition and 0.18 with reduction. In short, arid ecosystems showed the highest sensitivity to water manipulation, followed by semi-arid and mesic ecosystems. Thus, the observed hierarchy of ecosystem sensitivity to changes in the amount of precipitation with aridity is not an artefact of either the magnitude of manipulation or MAT.

Overall, ecosystem processes were more sensitive to water addition than to water reduction, with sensitivity indices of 0.45 and 0.28, respectively (Fig. 8). This effect was particularly pronounced in arid (0.70 with addition, 0.44 with reduction) and semi-arid regions (0.77 versus 0.40, respectively), while mesic ecosystems did not differ much in their responsiveness to either water addition or reduction (0.18 with addition, 0.14 with reduction). However, the trend towards higher sensitivity with aridity was maintained (Fig. 8).

4 The Impact of Changes in Precipitation Variability

In many ecosystems, a consequence of future climate change will be increasing intra-annual precipitation variability, with heavier rainfall events but longer intervening dry periods (see Sect. 1.1; Fig. 2). In recent years, several precipitation manipulation experiments have been conducted to study these scenarios. While some studies (e.g. Fay et al. 2003; Miranda et al. 2009) include a scenario of decreased precipitation amount (see Sect. 3.2), others try to single out the sole effect of altered precipitation distribution, frequency, and intensity on ecosystem performance (Knapp et al. 2002; Laporte et al. 2002; Heisler-White et al. 2008, 2009; Jongen et al. 2013a, b, c; Koerner et al. 2014). The experimental design usually involves closed shelters, with the collection and subsequent redistribution of natural precipitation according to the implemented scenario.

To date, the effects of precipitation variability, without altering total annual precipitation inputs, are still rarely studied, with only 18 publications included in the analysis. These studies were performed in 12 different ecosystem types (5 arid, 3 semi-arid, and 4 mesic systems) and 28 variability scenarios were studied (Table 2). The applied variability scenarios differed widely, with 0.5- to 14-fold increases in the both the extent of the normal dry period and the applied precipitation intensity (on average ~2-fold). Two studies given in Table 2 (i.e. Bates et al. 2006; Miranda et al. 2011) followed a different approach by altering the seasonal distribution of precipitation between the winter and the spring/summer periods. Main records made are on AP (27 observations), whereas observations on biodiversity (16), R _S (11), and soil N properties (10) are reported upon with less frequency. Figure 9a shows the variation of responses in AP to all applied precipitation variability scenarios in arid, semi-arid, and mesic systems. Observations in arid ecosystems consistently reported that AP was non-responsive to changes in the distribution pattern of precipitation (Miranda et al. 2009; Thomey et al. 2011; Vargas et al. 2012). However, Thomey et al. (2011) did report significant productivity increases in the dominant grass species with large infrequent rainfall events. In marked contrast, Heisler-White et al. (2008; 2009) with 6 observations in two different semi-arid ecosystems found a strong positive effect of larger infrequent precipitation pulses, with an increase in AP of on average 59 %, whereas negative effects on AP (−10 to −31 %) have been reported for mesic systems (Knapp et al. 2002; Laporte et al. 2002; Fay et al. 2003; Harper et al. 2005; Heisler-White et al. 2009). However, 9 out of 12 observations in mesic ecosystems report no significant response of AP with precipitation variability (Laporte et al. 2002; Fay et al. 2011; Jongen et al. 2013b, c; Poll et al. 2013; Koerner et al. 2014).

Fig. 9

Boxplot of variation in the range of percentage change, with mean values separately indicated (dotted lines) in measures of (a) productivity (AP), (b) diversity, and (c) soil respiration (R _S) observed in arid, semi-arid, and mesic ecosystems (see Table 2) in response to precipitation variability. Number (n) refers to sample size. Boxplots visualise the first, second, and third quartile, and the mean values (dotted lines), with whiskers indicating the maximum

In a conceptual model (‘bucket model’) Knapp et al. (2008) predicted the responses of terrestrial ecosystems to more extreme intra-annual precipitation patterns. The model is based on the expected change in the amplification of the fluctuations of soil water content with respect to certain stress thresholds (e.g. field capacity and permanent wilting point) with increasing precipitation intensities and longer intervening dry periods. According to the ‘bucket model’, in mesic ecosystems, the altered fluctuations will increase the duration of soil moisture exceeding the stress thresholds, leading to more frequent and higher water stress with increasing precipitation variability, thus resulting in future negative responses in AP. However, arid ecosystems will experience the opposite effect, with a decrease in seasonal water stress with increasing precipitation variability, as the amplification of soil water dynamics would result in deeper soil water infiltration, thereby permitting soil moisture to be maintained above drought stress thresholds for longer periods (Knapp et al. 2008). The results, with a 0- or an on-average positive response of AP in arid and semi-arid ecosystems, respectively, and an on-average negative response in mesic systems, support the predictions made in the ‘bucket model’. However, the majority of the observations (19 out of 27) report no response of AP to altered precipitation variability, similar to the studies on the impacts of precipitation addition or reduction on AP (see Sects. 3.1 and 3.2).

In contrast to AP responses, the changes in biodiversity parameters with increased precipitation variability were very small (Fig. 9b). Arid and semi-arid ecosystems proved to be non-responsive, while mesic ecosystems exhibited a slight increase in biodiversity with decreasing precipitation frequency (Fig. 9b). This increase, however, was due to a single observation (Knapp et al. 2002), with none of the other studies in mesic ecosystems reporting significant biodiversity responses.

Changes in soil respiration were found to reflect the responses of AP to increased precipitation variability, with an 11 % increase (only 2 observations) and a 10 % decrease in arid and mesic ecosystems, respectively (Fig. 9c). Unfortunately, none of the studies in semi-arid ecosystems report results on R _S. Regarding soil nitrogen, general conclusions are difficult due to the small number of studies reporting on these parameters (10 observations). Nevertheless, soil nitrogen responses to decreasing precipitation frequency were extremely variable, with 7 observations finding no significant changes, while three observations report strong responses of −30 %, 60 %, and 100 % in arid, semi-arid, and mesic ecosystems, respectively (Heisler-White et al. 2009; Yahdjian and Sala 2010). As increasing AP and R _S in the arid/semi-arid regions should promote soil nitrogen mineralisation and turnover, the 0- and negative responses of soil N are unexpected. However, as not only precipitation frequency but also pulse size was altered, the increased precipitation intensity could enhance nitrogen losses from soils, which might counteract the expected positive effects of increased AP and R _S (see Sect. 4). Further, high temperatures in warm arid and semi-arid ecosystems can be expected to facilitate NH₃ volatilisation from soils (e.g. Fan et al. 2011).

What remains is the question why the majority of the observations (45 out of 63) report a non-responsiveness of ecosystem processes with increasing precipitation variability. Of major importance is the question whether an increase in precipitation variability actually causes an increase in variability of soil water content. As the fluctuations in soil moisture can easily be expressed as the coefficient of variation of daily average soil water content (CV_SWC) (e.g. Fay et al. 2011; Jongen et al. 2013b, c), we suggest the inclusion of this parameter in future studies, as non-responsiveness can be explained by the lack of a significant change in CV_SWC. In addition, the lack of responsiveness to increased precipitation variability can be explained by the resilience of grassland vegetation to short-term decreases in soil moisture during the growing period (Miranda et al. 2011; Jongen et al. 2013c). Indeed, plants have the ability to cope with irregularities of precipitation patterns through a high degree of phenotypic plasticity (Jump and Peñuelas 2005) and the possibility to employ strategies that improve water uptake and reduce water consumption (Moreno et al. 2008). Also, manipulation of precipitation variability during sensitive periods such as germination, seedling establishment, or the peak growing period can cause strong limitations on AP and dependent parameters that would not occur when studying less sensitive times, such as the end of the growing season. Finally, the length of the period that ecosystems experienced an increase in precipitation variability could be another reason for the non-responsiveness observed in many cases (Fay et al. 2011; Beier et al. 2012; Poll et al. 2013). Long-term effects of precipitation manipulation, including the loss of resilience, the possibility for adaptation, or a steady change in soil properties caused by the manipulation, might result in different ecosystem responses than those found in short-term studies (Beier et al. 2012). For example, the studies conducted in Konza Prairie reported no effects on AP and R _S in the long term (Fay et al. 2011; Koerner et al. 2014), while negative effects were found in the first 2 years of manipulation (Knapp et al. 2002; Fay et al. 2003; Harper et al. 2005).

5 The Impact of Changes in Precipitation Intensity

With future climate scenarios predicting a change in precipitation distribution (Sect. 1.1), the effects of extended dry periods between precipitation events (Fig. 2b) on ecosystem processes will be accompanied by effects of increased precipitation intensity (Fig. 2a). Studies addressing this increase in precipitation variability (Sect. 4, Table 2) often show a lack of responsiveness of grassland ecosystems. However, the presented analysis only included studies with a minimum duration of 2 months of precipitation manipulation, with parameter responses taken for each year or growing season. Thus, the effects of individual precipitation pulses on parameter responses on the short term, and ultimately on ecosystem functioning in the long term, should be considered.

For arid and semi-arid ecosystems, it has been hypothesised that discrete precipitation pulses stimulate brief but important episodes of biological activity (Huxman et al. 2004b), with pulse size and frequency differentially affecting above- and belowground biota. With the response thresholds of plants and microbes being determined by the ability of the organisms to utilise precipitation events of different infiltration depth, there is a hierarchical view of precipitation pulse patterns and their effect on ecosystem processes (Huxman et al. 2004b; Schwinning and Sala 2004). This suggests that small rainfall events will only stimulate soil microbes in the uppermost soil layer, subsequently increasing microbial respiration, while the stimulation of carbon assimilation in higher plants requires larger rainfall events, with concomitant increase in infiltration depth. Indeed, most studies show a positive effect of increased precipitation intensity on short-term plant performance (e.g. photosynthetic carbon uptake, phenology, reproductive traits) after a precipitation pulse, which, however, is seldom reflected in productivity increases (e.g. Patrick et al. 2007; Chen et al. 2009; Angert et al. 2010).

In relation to aboveground biological activity, several studies have shown that the magnitude and duration of the photosynthetic response were related to the size of the precipitation event (Chen et al. 2009; Jongen et al. 2013d) and that photosynthetic assimilation responds with a delay to precipitation pulses (Ogle and Reynolds 2004; Potts et al. 2006b). In addition, the plants’ response differs across species and functional types, and depends on phenology, morphology, and physiological status (Huxman et al. 2004b; Ogle and Reynolds 2004; Potts et al. 2006a; Patrick et al. 2007; Angert et al. 2010).

In relation to belowground activity, a rapid increase in soil CO₂ efflux following precipitation pulses has been observed in various ecosystems (Mariko et al. 2007; Chen et al. 2009; Unger et al. 2010; Fan et al. 2012; Jongen et al. 2013a). This phenomenon of increased carbon and nitrogen losses after rewetting of dry soils, commonly termed ‘Birch effect’, has become an important subject in ecological studies. Although a complete understanding of the processes underlying the Birch effect has not yet been achieved (Borken and Matzner 2009), it is commonly accepted to be a direct response of the soil microbial and fungal community to changing moisture conditions (Borken and Matzner 2009; Inglima et al. 2009; Unger et al. 2010, 2012; Kim et al. 2012). Many studies agree on the theory of a positive effect of rewetting on soil microbial performance through a stimulation of microbial growth and matter transformation (e.g. Austin et al. 2004). However, several studies favour the hypothesis of a negative effect of precipitation pulses, with the large carbon and nitrogen losses being triggered by a hypo-osmotic stress response of soil microbes and fungi (Fierer and Schimel 2003; Xu and Baldocchi 2004; Jarvis et al. 2007; Unger et al. 2010, 2012). The majority of studies show that length and severity of the dry period prior to rewetting play a key role in the soil microbial response to such sudden changes in soil water status (Kieft et al. 1987; Fierer and Schimel 2003; Xu and Baldocchi 2004; Cable et al. 2008; Unger et al. 2010), with large and transient carbon pulses corresponding to high microbial stress (Unger et al. 2010; Meisner et al. 2013). Furthermore, wetting rate and infiltration to a soil seem to affect the microbial response to changing soil water potentials (Unger et al. 2012). Thus, the magnitude of change in soil water content related to a precipitation pulse is probably a key factor determining pulse effects on microbial activity and matter turnover.

Additionally, precipitation pulses may cause higher losses of nutrients, in particular the easily soluble components such as nitrate, due to higher water infiltration and leaching below the rooting zone (Yahdjian and Sala 2010; Jongen et al. 2013a).

As both increased Birch effects and leaching are to be expected with an increase of precipitation intensities, overall negative effects on soil microbial activity, nutrient availability, and hence productivity are a likely consequence. This effect might be exacerbated by larger precipitation pulses causing temporarily anoxic soil environments, impacting negatively on root and microbial performance (water logging, e.g. Jackson and Colmer 2005). However, the potential effects of increased precipitation intensity are masked by the concomitant effects of increasingly pulsed water availability and the longer intervening dry periods and therefore difficult to disentangle. Further, the impact of effects of increased leaching and soil nitrogen and carbon losses on ecosystem functioning is likely to be significant in the long term, and thus seldom detected in short-term observations.

For example, Patrick et al. (2007) found decreased soil respiration rates with concomitantly increased leaf level photosynthesis with a large precipitation pulse in summer during a 7-day observation period, resulting in an increase in carbon sequestration. However, the reduced soil respiration rates are likely to indicate substrate limitation or stress for soil microbes after the pulse, which could, in the long term, hamper the observed increase in carbon sequestration.

Although grassland ecosystems often show a lack of responsiveness to an increase in precipitation variability (Sect. 4, Table 2), this finding was particularly noticeable in arid and mesic ecosystems. In semi-arid ecosystems, overall positive effects of increased precipitation variability on productivity have been reported (Fig. 9), a finding that does not support the hypothesis of increased precipitation intensity causing negative effects due to the increased leaching and Birch effects in these ecosystems. Probably, the duration of most studies was not long enough to account for these long-term effects. In addition, it can be argued that the positive effect of longer periods with soil moisture being above stress thresholds, as hypothesised in the ‘bucket model’ for water-limited ecosystems (Knapp et al. 2008), counteracts negative effects of increased precipitation intensity on ecosystem functioning. However, Jongen et al. (2013c), increasing precipitation variability within the same experimental site for two subsequent years, reported lower soil nitrogen and productivity in the second year of manipulation, which could be related to negative effects of increased leaching and Birch pulses in the first year. Nevertheless, the results highlight a need for more long-term observation studies, ideally across several years, to assess the impact of precipitation intensity changes on ecosystem functioning.

6 Effects of Changing Precipitation Patterns on Synchronicity of the Matter Cycles

Due to the co-dependence of plant productivity and soil microbial mineralisation, there is a close linkage of ecosystem cycling of carbon and mineral nutrients. On the one hand, the temporal availability of carbon provided by plants, mainly through litter fall, root turnover, and exudation of organic substances to the soil, is a prerequisite for soil microbial growth (e.g. Kuzyakov and Gavrichkova 2010), while on the other hand, soil organic matter turnover and the mineralisation of nutrients, particularly of nitrogen, by soil microbes are vital for plant performance (e.g. Vitousek and Howarth 1991). Temporal synchronicity between both supply and demand of carbon and mineral nutrients is therefore crucial for ecosystem functioning (Augustine and McNaughton 2004).

The lack of soil moisture limitation in mesic ecosystems generally assures a tight coupling between microbial nutrient supply and plant nutrient demand (e.g. Vitousek et al. 1998; McCulley et al. 2009; Bobbink et al. 2010), with synchronicity between microbial and plant processes being mediated by intra-annual temperature variation, thereby minimising the loss of available nitrogen through leaching and gaseous emissions. However, in water-limited arid and semi-arid ecosystems, several studies observed an asynchronicity between nitrogen supply by microbes and nitrogen demand by plants (Jackson et al. 1988; Augustine and McNaughton 2004), due to the large fluctuations in soil moisture, with differences in the hierarchy, intricately linked to differences in thresholds, of the responses of plants and microbes to changes in soil water (Schwinning and Sala 2004; Collins et al. 2008). Such decoupling between peaks of mineral nutrient supply and plant growth can lead to substantial losses of mineral nutrients from the system, and result in a shift from a closed internal nitrogen cycle to an ‘open’ cycle, with the excess nitrogen being leached and/or emitted from the ecosystem (de Schrijver et al. 2008). For example, Yahdjian et al. (2006) suggested that net mineralisation of nitrogen during long dry periods is less affected than plant and microbial nitrogen immobilisation, resulting in an accumulation of nitrate in the soil that is subsequently lost by leaching during first precipitation events, thereby leading to nitrogen limitation and decreased productivity.

With altered precipitation amount, frequency, and intensity, the response thresholds of plants and soil microbes might experience a larger deviation, amplifying the asynchronicity in nitrogen processes, and resulting in extended periods of decoupling between nitrogen and carbon cycles throughout the year.

Figure 10 shows the observed percentage responses of R _S/ECO versus AP in arid, semi-arid, and mesic ecosystems, considering all reduction, addition, and variability scenarios. Only those studies with simultaneous measurements of R _S/ECO and AP responses were included. Regardless of the ecosystem type, negative responses in AP, as found in precipitation reduction scenarios, were accompanied by negative responses of the same magnitude in R _S/ECO, indicating that a decrease in precipitation does not result in asynchronicity of plant and microbe performances (Fig. 10). This finding is unexpected, as moisture limitation has previously been found to enhance the possibility of asynchronicity (e.g. Evans and Burke 2013). However, although Evans and Burke (2013) found significant increases in soil inorganic nitrogen pools with a simulated long-term drought in a semi-arid grassland in California, the decoupling of AP and soil respiration due to different drought sensitivities of ecosystem processes was small (−42 % and −33 %, respectively).

Fig. 10

Percentage change in AP versus percentage change of R _S/ECO for precipitation addition (circles), reduction (squares), and variability (triangles) studies in arid (black), semi-arid (grey), and mesic (white) ecosystems. The dashed line indicates an exponential function (y = 59.7(1−e ^(−0.01x)), r ² = 0.73, p < 0.0001), whereas the 1:1 relationship is indicated by a solid line

In contrast to the observed synchronicity between AP and R _S/ECO in studies with precipitation reduction, the different ecosystem compartments (AP and R _S/ECO) tended to show larger residuals from the 1:1 line with precipitation addition. In those precipitation addition studies with positive responses in AP and R _S/ECO, the AP response was often much stronger than the R _S/ECO response. This effect became more pronounced with increasingly positive responses in AP and was observed in 2 arid (Thomey et al. 2011; Song et al. 2012), 2 semi-arid (Sternberg 2011; Talmon et al. 2011; Potts et al. 2012), and 1 mesic ecosystem (Zhou et al. 2006, 2012; Sherry et al. 2008). Thus, if precipitation manipulation increased plant growth, this was not always reflected in the same order of magnitude in soil or ecosystem respiration, which is indicative of a decoupling between carbon production and nitrogen mineralisation, potentially leading to a lagged nutrient deficiency of plants. As this effect was most pronounced in arid and semi-arid ecosystems, where asynchronicity of the matter cycles is reflected in nitrogen accumulation during longer dry periods, a higher response of plant productivity as compared to soil microbial activity might be expected, as additional water enables plants to better assimilate the readily available nitrogen pool. While this explains how, in the short term, precipitation addition accelerates plant growth in soils without nutrient limitation, the lagged effect of smaller nutrient supply by microbes with larger nutrient fixation by plants is not considered. However, the studies showing the most extreme deviations from the 1:1 line in Fig. 10 can be regarded as exceptions, with Potts et al. (2012) comparing precipitation addition to a control subjected to severe natural drought, and Song et al. (2012) studying an extreme addition scenario in the most arid ecosystem reported upon. Under these extreme conditions, a higher response of AP as compared to the response of R _S/ECO with precipitation addition might be expected, as (1) nutrient accumulation during drought periods is expected to be higher, and (2) high soil water potential changes might cause greater stress to soil microbes (Sect. 5) than to plants.

Reports on both AP and R _S/ECO in studies manipulating precipitation variability, without altering total precipitation inputs (Table 2), did not find pronounced differences between soil and plant treatment responses (Fig. 10). Dijkstra et al. (2012) showed that nitrogen release by soil microbes was enhanced as compared to plant nitrogen uptake with large and infrequent precipitation pulses, indicating that changes in precipitation event sizes could exacerbate losses of nitrogen in a semi-arid system. However, such short-term effects were not reproduced in any of the longer-term manipulation studies, most of them finding a near 1:1 response in AP and R _S/ECO (Fig. 10).

Thus, in general, we found that in most ecosystems reported upon, both AP and R _S/ECO did respond synchronously to precipitation manipulation scenarios. This is supported by the general lack of significant treatment effects on soil nitrogen availability or mineralisation with precipitation manipulation (Tables 1 and 2), with the exception of only four studies (Heisler-White et al. 2009; Yahdjian and Sala 2010; Potts et al. 2012; Evans and Burke 2013). Therefore, increased asynchronicity of the matter cycles will not likely be a threat with changing precipitation patterns as predicted with future climate scenarios.

7 Synthesis

Experimentally manipulating precipitation patterns is an indispensable tool to describe and model future climate change impacts on ecosystem processes (Reichstein et al. 2013; Reyer et al. 2013; Vicca et al. 2013). During the last two decades, a growing body of studies has emerged reporting on manipulative experiments in a variety of ecosystems, with some of these studies including the effects of changing precipitation patterns. In this review, the findings of these individual studies were synthesised in order to make inferences from the combined results and to identify and refine strategies for future research. An earlier meta-analysis of precipitation manipulation studies, carried out by Wu et al. (2011), synthesised 39 studies, conducted in 34 ecosystems with different vegetation types, their analysis focusing on plant growth parameters and ecosystem carbon balance. They found that supplemental precipitation stimulated plant productivity and ecosystem carbon fluxes, whereas reduced precipitation suppressed these parameters (Wu et al. 2011). Our review analyses the effects of in situ precipitation manipulation on plant productivity, species diversity, soil/ecosystem respiration, and soil nitrogen in grassland type ecosystems over a wide range of climate types (MAT range: −13.9 °C to 22.9 °C; MAP range: 115–1,741 mm), synthesising the results from 72 studies. Our analysis showed a hierarchy in the responsiveness of grassland ecosystems to changing precipitation quantity, with largest responses found in arid ecosystems, immediately followed by semi-arid ecosystems, while the majority of mesic ecosystems did not respond to either water addition or reduction. Furthermore, our analysis indicated that, independent of ecosystem type, ecosystem processes were more responsive to water addition than to water reduction, which agrees with Wu et al (2011), reporting higher sensitivity to increased precipitation than to decreased precipitation. In a review on precipitation reduction experiments, Vicca et al. (2013) reported no evidence of differential effects of experimental drought over sites with different MAP, which is in contrast to our results for grassland type ecosystems, where responsiveness to manipulation was higher in arid regions than in mesic regions. This, however, is plausible, as productivity losses through water reduction in shallow-rooted grassland ecosystems are potentially stronger in arid regions, which might not be the case with deeper-rooted forest ecosystems (Reichstein et al. 2013). In agreement, Knapp and Smith (2001) concluded that systems dominated by herbaceous vegetation may be more responsive to future precipitation regimes, as their productivity shows generally larger inter-annual variability than forests. Hsu et al. (2012) reported higher sensitivity of aboveground productivity to changes in mean annual precipitation in semi-arid ecosystems as compared to arid ecosystems, concluding that at the driest sites, sensitivities may be lower due to low relative growth rates, density limitations, and high evaporation rates. This is in contrast to our results, with grassland ecosystems showing higher sensitivity with increasing aridity. However, the conclusion of Hsu et al. (2012) is based on two semi-arid sites with exceptionally high sensitivity.

Our analysis indicated that a large part of the studies reported a resilience of grassland ecosystems to changes in precipitation patterns. However, Wu et al. (2011) do not mention resilience in their findings. The resilience to changing precipitation amounts was most evident in mesic ecosystems, followed by semi-arid and arid ecosystems, for both addition and reduction scenarios. Furthermore, resilience was more evident for diversity, soil nitrogen, and productivity as compared to respiration processes. This resilience might be because the applied manipulation scenarios often lie within the range of the natural inter-annual precipitation variability experienced by ecosystems, and a long-term evolutionary adaptation of ecosystem components to these natural differences (Sardans and Peñuelas 2013).

To test the observed hierarchy in responsiveness of ecosystems towards changing precipitation amounts with increasing aridity, we performed a Partial Least Squares Regression analysis (PLSR) of the productivity responses with changing precipitation quantity. The x-score plot of the factors MAP, MAT, and manipulation amount revealed a separation between grassland responses towards manipulation in different biomes (Fig. 11). PLSR showed that x-scores at factor 1 (amount of manipulation) explained 29 % of the y-variance (response variable), while x-scores at factor 2 (MAP and MAT) explained only 2 % of the y-variance. Thus, although high loadings of MAT and MAP on factor 2 are observed, the variance in AP response is mainly explained by the amount of manipulation or other unknown factors, which is indicated in similar directions of the manipulation and response vectors. MAP and MAT influences on ecosystem response variance cannot be separated as they showed similar loading vectors; however, neither parameter contributed significantly to ecosystem responses towards manipulation.

Fig. 11

Scores plot of the first two factors of a Partial Least Squares Regression (PLSR) analysis to model percentage change in aboveground productivity in grassland ecosystems (response variable, y-data) using MAT, MAP, and precipitation addition and reduction scenarios (percent manipulation) as explaining variables (x-data). Samples originating from nine different biomes are sorted by MAP (arid, semi-arid, mesic) and MAT (cold, temperate, warm). Arrows indicate mean x-and y-loadings of the predictor variables and the response. Explained variance in x and y, respectively, is given in brackets in the axes labels. n = 87

The poor fit of the PLSR model, with only 38 % and 25 % of x-variance being explained by factors 1 and 2, respectively, can be an indication for data distribution being prejudiced by the resilience of many systems (0-responses). Thus, resilience, as indeed observed in most studies, overall seemed to have more impact than the hierarchy of different climate types in those studies that did observe ecosystem responses with precipitation manipulation. However, it has to be considered that the PLSR model is not only driven by MAT, MAP, and percentage manipulation, but also influenced by site-specific factors (e.g. soil properties, species composition, herbivory, management, climatic and site history) and means of experimental conduct (e.g. length and timing of experimental manipulation, seasonality, inter-annual variation in precipitation, shelter effects), which hampers generalisation and direct comparability of results in between sites.

In relation to changes in precipitation variability, resilience was prominent in arid and mesic grassland ecosystems, while semi-arid ecosystems showed an increase in productivity with increasing variability, although this result is based on a single study by Heisler-White et al. (2009). The hierarchy with a 0- or an on-average positive response of AP in arid and semi-arid ecosystems, respectively, and an on-average negative response in mesic systems in studies reporting effects towards precipitation variability experiments (Fig. 9), supports the predictions made in the ‘bucket model’ by Knapp et al. (2008). However, it must be considered that this conclusion is only based on a small amount of observations and, similar to the findings for changing precipitation amounts, is overshadowed by the resilience found in most studies. In addition, the hypothesised negative effects of increased leaching and gaseous carbon and nitrogen losses with higher precipitation pulse intensity in arid and semi-arid ecosystems could not be confirmed. This could be due to two reasons: (1) either the precipitation variability studies conducted are not long enough to account for negative long-term effects of soil nitrogen and carbon depletion or (2) the positive effects of soil moisture being above stress thresholds for longer periods counteract negative effects of increased precipitation intensity on ecosystem functioning.

For those studies that reported upon both AP and R _S/ECO, synchronous responses of both parameters towards precipitation manipulation scenarios were found, with the exception of some addition studies, with higher responses of productivity as compared to R _S/ECO. Therefore, increased asynchronicity of the matter cycles will not likely be a threat with the changing precipitation patterns that are predicted with future climate scenarios.

In a recent publication, Vicca et al. (2012) highlighted the necessity of a common metric to increase comparability of precipitation manipulation experiments in different ecosystems, this metric combining an index of both stress duration and stress intensity, thereby reflecting the actual treatments as experienced by plants. Indeed, it has to be considered that altering precipitation amounts might not always result in an equivalent change in available soil water for plants and microorganisms, as processes that distribute precipitation in the soil are complex, with interception, infiltration, run-off, seepage below the rooting zone, soil evaporation, plant water use, and hydraulic redistribution strongly differing between soil and ecosystem types (Loik et al. 2004) and differing between season and quantity of applied precipitation (Parton et al. 2012). Thus, these factors, by determining the available water for plants, might play a crucial role in determining whether ecosystems display either resilience or responsiveness to changing precipitation patterns. In addition, Sala et al. (2012) found that the responsiveness of ecosystem processes to changing water availability is influenced by a lagged effect of the previous year’s situation, pointing to the need to incorporate information about climate legacies when interpreting results from precipitation manipulation experiments. Furthermore, most of the studies included in our review were conducted over short time periods. Shifts in species composition that can occur in the long term are thus rarely observed in precipitation manipulation studies (Weltzin et al. 2003). Considering climate change as a directional process, ecosystem responses will depend on both the magnitude of change and the time frame being considered (Sala et al. 2012).

The problem when comparing the results of a large number of individual studies to address large-scale ecological questions has recently been raised by Fraser et al. (2013), recommending the use of coordinated distributed experiments (CDEs), a possible solution to increase the level of in-between study comparability. CDEs provide a collaborative, coordinated, and hypothesis-driven approach for standardised experimental conduct and data analysis on an international level, thus controlling for site and study effects on both spatial and temporal scales, allowing to address important large-scale ecological issues, that would otherwise be difficult to resolve (Fraser et al. 2013).

We strongly support both the necessity to introduce a common metric to improve inter-study comparability, as proposed by Vicca et al. (2012), and the use of ecological CDEs, proposed by Fraser et al. (2013). In this review, we tried to improve inter-study comparability by relating the relative amount of ecosystem response with the relative amount of manipulation (sensitivity index, Sect. 2.2) and attempting to assess whether differences in ecosystem responses were driven by MAT and MAP. In addition, particularly for those manipulation experiments investigating the effects of precipitation variability, we suggest the use of the coefficient of variation of daily average soil water content (CV_SWC) to be included in future studies.

Although this review of precipitation manipulation experiments finds a general resilience of grassland ecosystems towards a range of manipulation scenarios, the question how future precipitation changes will affect ecosystem processes in global grasslands is far from answered. Thus, coordinated precipitation manipulation experiments with long-term field observations and increased comparability are desirable to capture and compare possible long-term effects (e.g. through changes in species composition and soil properties) on ecosystem state and functioning.