Positive legacies of severe droughts in the Inner Mongolia grassland

Abstract

Global change–induced extreme droughts are increasing in grasslands worldwide, and drought legacies may greatly affect the responses of grassland ecosystems to these changes. However, it remains poorly understood whether and how severe droughts have a positive legacy effect on grassland productivity. By combining a 4-year precipitation manipulation experiment with a 40-year observational study in a semiarid grassland, we showed that extreme droughts could create strong positive legacies on community productivity and that such legacies could last for multiple years. The mechanism behind this was the coupled effect of the drought-induced increase in annuals and the favorable precipitation pattern that facilitated the flourishing of annuals in subsequent years. This study provides experimental and observational evidence for positive drought legacies and reveals their underlying mechanisms. Our findings suggest that positive drought legacies should be incorporated into Earth system models to better predict the impact of extreme droughts on grassland ecosystems.

Community structure change induced by severe droughts has positive legacies on grassland productivity.

INTRODUCTION

Precipitation is a primary climatic factor that determines the aboveground net primary productivity (ANPP) in grassland ecosystems (1–3). Over a broad spatial scale, ANPP is usually strongly related to annual precipitation, but the temporal precipitation-ANPP relationship at a given site is often weak (4–6). For example, annual precipitation explained 94% of the ANPP variation on 9500 sites in the central grassland of North America (1) and 76% on 21 sites in the temperate grassland on the Inner Mongolian Plateau (7); however, annual precipitation, on average, accounted for only 13% of the temporal ANPP variation based on 16-site, long-term observations in grasslands (4), and no significant relationship was detected at certain sites (4, 8). It has been suggested that the legacy effect (i.e., delayed impact of antecedent precipitation on current-year ANPP) is a major mechanism leading to the weak temporal precipitation-ANPP relationship (9–11) because the legacy effect can explain a notable portion of the unexplained variance in ANPP (4), and in some cases, precipitation legacies determine the grassland responses to rainfall regimes (12, 13).

In general, drought years tend to have negative legacies on ANPP (4, 5, 14, 15), and drought-induced loss of meristematic tissues, such as stems, tillers, and buds, or plant mortality is the dominant mechanism for negative drought legacies (14, 16). Several studies have reported that drought years may have positive legacies on grassland ANPP (17–20), but the underlying mechanisms remain unclear. It has been suggested that high soil nitrogen (N) availability, a result of continued N mineralization but reduced N uptake during the drought period, is a mechanism for positive drought legacies (4, 16, 17, 21–23). However, without the recovery of the reduced abundance of meristems, how does the increased available N quickly translate into positive legacies? A recent study indicated that N enrichment could not mitigate the negative impacts of drought-induced meristem loss (14).

Severe drought years reduce the meristem density of perennial species but provide an opportunity for the potential flourishing of annuals. Severe droughts have been frequently observed in the Inner Mongolia grasslands and can cause a shift of vegetation structure from a perennial-dominated community to an annual-dominated community (24). This can be ascribed to the special features of annuals, such as rapid germination, fast growth, and short life span. Furthermore, the flourishing of annuals requires favorable environmental conditions, such as suitable temperature and sufficient precipitation. For example, on average, annuals accounted for less than 5% of community ANPP in the Xilingol grassland (1980–2019); however, the annual species Chenopodium aristatum thrived, leading to a greater than 30% contribution in 2018 owing to less precipitation occurring in the early growing stage (May and June) and more occurring in the middle stage (July and August). As a result, the landscape in this region was greatly altered (fig. S1).

Here, we hypothesize that severe drought years can lead to an increased abundance of annuals, and an increase in annuals coupled with a favorable precipitation pattern can create positive drought legacies (Fig. 1). The Inner Mongolia grassland was used as an example, and its plant communities are generally dominated by perennials (normal community structure, state I). Severe droughts lead to an increase in annuals, as indicated by the increased ratio of annuals to perennials (A/P ratio) (state II). Such changes may have strong or weak positive legacies on ANPP, depending largely on the subsequent-year precipitation distribution pattern. Considering that the seed germination of annuals usually occurs in early summer in this system (25), if less precipitation is received in the early growing stage and more in the middle stage, a positive legacy effect should be evident because of the thriving of annuals (state III). Moreover, if such a favorable pattern lasts for several years, the positive legacy effect is expected to last accordingly. Conversely, if the precipitation pattern returns to normal, the legacy effect may weaken, and the community structure may return to normal. We tested these hypotheses using a precipitation gradient experiment in combination with a long-term observational study (1980–2019) in the Inner Mongolia grassland. The precipitation gradient experiment had nine levels of growing season precipitation (100, 150, 200, 275, 300, 350, 400, 450, and 500 mm) and was conducted in six rainout shelters (35 m × 11 m) from 2014 to 2017. The lowest and highest levels represented extremely dry and wet years, respectively. To examine the legacy effects of different levels of previous precipitation, the treatments were stopped, and all plots received ambient precipitation in 2018 and 2019. This study focused on three questions: (i) Do drought or wet treatments cause alterations in the A/P ratio? (ii) Do alterations in the A/P ratio have a lagged effect on ANPP in subsequent years, and if so, how? (iii) Does the precipitation distribution pattern among the early (May and June), middle (July and August), and late (September and October) stages of the growing season in the subsequent year play a role in affecting the legacy effect?

Fig. 1. Conceptual framework showing how drought-induced community structure change and subsequent-year precipitation distribution pattern interactively drive the positive legacy effect.

Grassland communities (e.g., in Inner Mongolia grassland) are normally dominated by perennials, while annuals are the minority (state I). Climate change (e.g., severe droughts) may lead to community structure changes, as indicated by the increase in the A/P ratio (state II). Such changes may have a strong or weak positive legacy effect, to a great extent depending on the precipitation distribution in subsequent years. If the precipitation pattern is favorable to the germination and growth of annuals, the positive legacy effect should be evident due to the flourishing of the annuals (state III). Moreover, such legacies can last for several years as long as the precipitation patterns are favorable to the growth of annuals in the subsequent years. In contrast, if the precipitation distribution pattern is favorable to the growth of perennials (e.g., returning to normal), the positive legacy effect becomes weak, and the community structure returns to normal.

RESULTS

We first compared the community ANPPs in the final treatment year (2017) and the two posttreatment years (2018 and 2019) in the precipitation gradient experiment. We found that after 4 years of successive precipitation treatment, the ANPPs of the two severe drought treatments (100 and 150 mm) significantly declined in 2017 (fig. S2); however, these two treatments achieved the highest ANPPs in 2018 (Fig. 2A). In contrast, there was no significant difference in the ANPP among treatments in 2019 (Fig. 2A). As a result, the two severe drought treatments produced significantly higher positive legacy values than the other treatments in 2018 but not in 2019 (Fig. 2B).

Fig. 2. Community ANPPs and legacy values.

Community ANPP in 2017, 2018, and 2019 (A) and the legacy values in 2018 and 2019 (B) from the precipitation gradient experiment. The labels on the abscissa axis represent the precipitation treatments in 2017, and all plots received ambient precipitation in 2018 and 2019. Significant differences (P < 0.05) among precipitation treatments in the same year are indicated by different capital letters for 2017, lowercase letters for 2018, and Greek letters for 2019.

Why did severe drought treatments create positive legacies in ANPP in 2018? To address this question, we examined the performance of the two plant functional groups (perennials and annuals) presented in the community in 2017 and 2018. The ANPPs of perennials under the two severe drought treatments were significantly lower than those in the other treatments in 2017 but not in 2018 (Fig. 3A). In contrast, the ANPPs of annuals under the two severe drought treatments significantly increased in 2017, and their ANPPs increased much more strongly in 2018 than in 2017 (Fig. 3B). As a result, the A/P ratios had a similar increasing trend with decreasing precipitation in both years but were stronger in 2018 (Fig. 3C and fig. S3). These results evidently indicate that severe drought can lead to an increase in the A/P ratio. To confirm the role of the previous year’s A/P ratio increase in driving the positive legacy effect on ANPP, we quantified the relationship between the A/P ratios under different treatments in 2017 and their corresponding legacy values in 2018 and found that they were closely related (Fig. 4A). The A/P ratio in 2017 explained 91% of the variance in the legacy value in 2018.

Fig. 3. ANPPs of plant functional groups and the A/P ratios.

ANPP of perennials (A), annuals (B), and the A/P ratio (C) in 2017, 2018, and 2019 from the precipitation gradient experiment. The labels on the abscissa axis represent the precipitation treatments in 2017, and all plots received ambient precipitation in 2018 and 2019. Significant differences (P < 0.05) among precipitation treatments in the same year are indicated by different capital letters for 2017, lowercase letters for 2018, and Greek letters for 2019.

Fig. 4. The relationship between the A/P ratio and ANPP legacy value.

A positive linear relationship was found in 2018 (A) but not in 2019 (B) in the precipitation gradient experiment.

To further explore the mechanism of the A/P ratio increase under drought treatments in the precipitation gradient experiment, we evaluated the density and individual plant biomass of perennials and annuals in 2017 and 2018, respectively (Fig. 5). Compared with the long-term mean precipitation treatment (275 mm, control), the densities of the perennials under the two extreme drought treatments (100 and 150 mm) in 2017 decreased by 62.6 and 38.4%, respectively. In contrast, their individual biomass was not significantly affected under the 150-mm treatment and slightly increased under the 100-mm treatment in the same year (Fig. 5, A and C). The densities of annuals under the 100- and 150-mm treatments in 2017 increased by 268.5 and 162.7%, respectively, but their individual biomass decreased by 15.5 and 32.3% (Fig. 5, B and D). In 2018, the densities of the perennials under the 100- and 150-mm treatments remained significantly lower than that of the control treatment (275 mm), but their individual biomass increased significantly (Fig. 5, A and C). For the annuals, density and individual biomass increased markedly in 2018 (Fig. 5, B and D). These results suggest that years of severe drought greatly decreased the density of perennials and increased the density of annuals.

Fig. 5. Densities and individual biomass of plant functional groups.

Densities (A and B) and individual biomass (C and D) of perennials and annuals in 2017 and 2018 from the precipitation gradient experiment. Labels in the abscissa axis were precipitation treatments in 2017, and all plots received ambient precipitation in 2018. Significant differences between treatments in 2017 are indicated by different lowercase letters, while those in 2018 are indicated by different capital letters. * and *** indicate significant difference between the pair of columns for the same treatment at P = 0.05 and P = 0.001, respectively.

However, why were no positive drought legacies observed in 2019 (Fig. 2B)? In this year, neither the perennials nor the annuals in the two severe drought treatments showed significant differences in ANPP compared to the other treatments (Fig. 3, A and B). As a result, the A/P ratio under the previous drought treatment was not different from that under the other treatments (Fig. 3C), and there was no relationship between the A/P ratios in 2018 and the legacy values in 2019 (Fig. 4B). These results suggest that the increased A/P ratio in 2018 did not translate into positive legacies in ANPP in 2019. To understand the underlying mechanism for this alteration, we analyzed the precipitation distribution pattern among the early (May and June), middle (July and August), and late (September and October) stages of the growing season in 2018 and 2019 in comparison with the 40-year average pattern. We found that the precipitation fractions in the late stage in 2018 and 2019 were not significantly different from the long-term average, and the major difference occurred in the early and middle stages (Fig. 6). Specifically, a significant decline in the early stage and a noticeable increase in the middle stage characterized the precipitation pattern in 2018 but not in 2019 (Fig. 6). These results suggest that drought-induced increases in annuals are likely to couple with an “early less, middle more” precipitation pattern (i.e., less precipitation in the early stage and more precipitation in the middle stage of the growing season) in the subsequent year to create a positive legacy effect.

Fig. 6. The precipitation distribution patterns in 2018 and 2019 compared to the 40-year average pattern.

The growing season precipitation was separated into three stages: early (May and June), middle (July and August), and late (September and October) stages. * indicates a significant difference between the pair of columns at the same stage (P < 0.05), and “ns” indicates no significant difference between the pair of columns (P > 0.05).

We further tested this mechanism with the results of a long-term observational study. Compared with the 40-year average value of the A/P ratios, a significant A/P ratio increase (P < 0.05) that created positive legacy effect on the ANPPs in subsequent years occurred in 14 of the 40 years (Fig. 7A and fig. S4). Such a positive legacy effect on ANPP could last for 2 or 3 years as the early less, middle more precipitation pattern reoccurred. For example, a significant A/P ratio increase was first observed in 1983, and positive legacies of ANPP were observed in the following 3 years (1984, 1985, and 1986). Similarly, an increased A/P ratio occurred in 1989, and positive legacies were observed in 1990 and 1991. Furthermore, compared with the long-term (1980–2019) average precipitation distribution pattern, these 14 years with positive ANPP legacies were characterized by receiving less precipitation in the early stage and more in the middle stage, whereas there was no difference in the late stage (fig. S5). These results further provided observational support for our hypotheses.

Fig. 7. Positive legacy evidence from long-term observations.

(A) A/P ratios in terms of ANPP in the long-term observational study (1980–2019). Points with red circles indicate the value years of A/P ratios that are significantly greater than the long-term mean (P < 0.05). Gray columns indicate the years with positive ANPP legacies and high previous-year A/P ratios. The height of each column indicates the strength of the corresponding legacy effect. The legacy values are denoted above the corresponding columns. (B) Relationship between legacy value and the ratio of middle- to early-stage precipitation amounts in the growing season in the years with significantly greater A/P ratios. The growing season precipitation was separated into three stages: early (May and June), middle (July and August), and late (September and October) stages. (C) Relationship between the legacy value and A/P ratio for the years with positive ANPP legacies and high previous-year A/P ratios.

To better understand the role of an early less, middle more precipitation pattern in creating positive ANPP legacies, we evaluated the impacts of precipitation amounts in the growing season and the early (May and June) and middle (July and August) stages on the annuals, perennials, and A/P ratios in the long-term observational study. The ANPP of the perennials followed an exponential rise to a maximum pattern in response to the precipitation of the growing season and the early and middle stages (Fig. 8, A to C). Concomitantly, the ANPP of the annuals and the A/P ratios increased linearly with the precipitation across the growing season and in the middle stage but were not related to the early-stage precipitation (Fig. 8, D, F, G, and I). These results suggest that the early-stage precipitation increased the ANPP of the perennials but not that of the annuals. In contrast, the middle-stage precipitation promoted the annuals better than the perennials, as the ANPP of annuals increased linearly with the middle-stage precipitation, whereas the ANPP increase of perennials followed a decreasing rate. Thus, an early less, middle more precipitation pattern may depress the growth of perennials and promote the growth of annuals.

Fig. 8. ANPPs of plant functional groups and A/P ratio in relation to precipitation in different stages.

ANPPs of perennials (A to C) and annuals (D to F), and A/P ratio (G to I) in relation to growing season (May to October), early-stage (May and June), and middle-stage (July and August) precipitation in the long-term observational study.

To determine the relative role of the previous-year A/P ratio and current-year precipitation pattern in driving the positive legacy effect, we quantified their contributions to legacy values in the long-term observational study. Considering that the years with a positive legacy effect were characterized by less precipitation in the early stage and more precipitation in the middle stage, we used the ratio of middle- to early-stage precipitation (M/E ratio) as a parameter for the precipitation distribution pattern. We found that the legacy values were significantly and positively related to the A/P and M/E ratios for the 14 years with positive legacies and high previous-year A/P ratios (Fig. 7, B and C), with the previous-year A/P ratio accounting for 46.9% of the variance in the legacy value and the current-year M/E ratio accounting for 10.7% (fig. S6).

DISCUSSION

This study provides experimental evidence that severe droughts can produce strong positive legacies in community ANPP. In contrast, no significant legacy effect was observed in the plots that experienced increased precipitation. These results differed from those of previous studies suggesting that droughts usually create negative legacies, whereas wetness creates positive legacies (4, 15). Furthermore, we demonstrated that 4-year extreme droughts greatly reduced soil moisture and increased soil temperature (figs. S7 and S8), which decreased the community ANPP, but increased the A/P ratio substantially in the last two treatment years (2016 and 2017) (figs. S9 and S10). Our results further suggest that the drought-induced thriving of annuals is a major mechanism driving positive drought legacies in this grassland ecosystem (Figs. 4A and 7C). The increase in the A/P ratio under severe drought treatment was mainly due to the increased abundance of annuals rather than the decreased abundance of perennials (Fig. 3, A and B). This was especially apparent for the extreme drought treatment (100 mm) in the precipitation gradient experiment; annuals contributed 37.5% to community ANPP in 2018, whereas its contribution was, on average, less than 5% across the 40 years (1980–2019). Moreover, we demonstrated that the flourishing of annuals in 2018 could be ascribed to the increases in plant density and size (Fig. 5), as the density of annuals under the extreme drought treatment (100 mm) increased by 132.9%, and their individual biomass increased by 272.7% relative to those of the long-term mean treatment (275 mm) in 2018.

The thriving of annuals under drought treatments could be explained by three mechanisms. First, severe droughts greatly reduced the meristems of perennials, as indicated by the significant declines in their density in the 100- and 150-mm treatments in 2017 (Fig. 5B). The meristem loss of perennials greatly eliminated their shading effect on the soil, resulting in an increased soil temperature (fig. S8), thus facilitating the seed germination of annuals. In addition, drought treatments decreased the density but not the individual biomass of the perennials (Fig. 5, A and C) and reduced their uptake of soil nutrients, subsequently increasing the nutrient supply to the annuals (fig. S11) and facilitating their thriving (26). Second, the annuals in this system were usually C₄ plants such as C. glaucum and Salsola collina. These species have higher water use efficiency and growth rate than perennial C₃ species (9). Taking these advantages into account, the annuals under drought treatments could quickly take up water after each watering treatment and grow faster than the plants under high precipitation levels. Consequently, their abundance increased under severe drought treatments in 2017. The increased abundance of annuals greatly increased their transient seed banks (seeds from the previous year), providing a basis for the flourishing of annuals in 2018 because the germination and growth of annuals substantially depend on the size of transient seed banks (27, 28). Last, the annuals in our system have advantages in seed dispersal, as their seeds are small but large in quantity. As indicated in Fig. 5, the density of annuals increased in all plots in 2018, likely due to seed dispersal between plots. Thus, the annuals could recover from drought more quickly than perennials (29) and have the potential to create positive ANPP legacies.

The weak legacy effect under wetness treatments can be ascribed to two factors. First, this system is characterized by high potential evapotranspiration, which is almost two to three times the annual precipitation (30). As a result, increased soil moisture after a heavy precipitation event lasted only 48 to 120 hours (31); thus, it would be difficult for moist soil to be carried over from the previous year to the current year (fig. S12) (32). Second, high-level precipitation may trigger a shift in the primary limiting resources from water to N. In this system, community ANPP is colimited by water and soil N (7). Increased precipitation can remove water limitation but exacerbate soil N stress because more biomass production requires more N. Thus, the shift in limiting resources from water to N may constrain the ANPP response to high-level precipitation and consequently debilitate the legacy effect of wetness.

Our results demonstrate that the drought-induced, increased abundance of annuals needs to be coupled with a favorable precipitation pattern for annuals to create positive legacies. In this grassland, the overwintered tillers of the perennials usually become active in early spring (late April or early May), whereas the germination of seeds of the annuals usually occurs in early summer (late June or early July). Droughts in the early growth season substantially constrain the growth of perennials (33, 34), as the ANPP of perennials was closely related to precipitation during this period (Fig. 8B). When summer is coming, the temperature becomes suitable for the germination of annuals. Sufficient precipitation during this stage facilitates their germination and rapid growth. As shown in Fig. 8F, the ANPP of annuals increased linearly with an increase in middle-stage precipitation. Moreover, the drought-induced increase in the abundance of annuals in the previous year produces a larger transient seed bank. As a result, annuals can rapidly establish and expand their populations (35). As indicated in Fig. 5, the density and individual biomass of annuals in the previous drought treatments increased sharply in 2018. Our results also highlight the role of the current-year precipitation pattern in driving positive ANPP legacies. On the basis of the regression equation (y = 15.28x − 24.41) of the ANPP legacy values and the M/E ratios across the 40 years (fig. S13), zero legacy was achieved when the M/E ratio was 1.6, while an M/E ratio higher than 1.6 might produce positive ANPP legacies. The long-term mean M/E ratio across the 40 years (1980–2019) was 1.67, which is very close to the M/E ratio (1.6) for zero legacy. These results suggest that compared with the long-term mean precipitation distribution pattern, an early less, middle more pattern with an M/E ratio higher than 1.6 can be defined as a favorable pattern for the thriving of annuals in this grassland. Under such a pattern, the increase in ANPP of annuals may override the decrease in ANPP of perennials and potentially produce positive legacies on community ANPP. The long-term observational study further demonstrated that if the favorable precipitation distribution for annuals lasted for several years, the legacy effect on the ANPP also lasted accordingly.

In contrast, more precipitation in the early stage of the growing season facilitates the growth of perennials (Fig. 8B), as these species usually turn green much earlier than the seed germination of annuals. Consequently, the growth of annuals was constrained because early established individuals of perennials have competitive advantages in acquiring soil nutrients and capturing light. As a result, the legacy effect of the increased abundance of annuals was dampened. Therefore, under the scenario in which the precipitation distribution pattern returned to normal, the positive legacy effect weakened, and the community structure returned to normal (Fig. 2B). These results support our hypothesis and are consistent with the prevailing Mongolian folklore that “Early rainfalls facilitate the growth of perennials; middle rainfalls facilitate the growth of annuals.”

In addition, our results demonstrated that although the growing season precipitation generally promoted the growth of perennials and annuals (Fig. 8, A and D), consistent with previous findings in this grassland (25), precipitation in the early versus middle growing season stages had differed impacts on perennials and annuals. The early-stage precipitation could promote the perennials but had little impact on the annuals (Fig. 8, B and E); in contrast, the middle-stage precipitation could substantially promote the growth of annuals (Fig. 8F). This was particularly apparent for the predrought treatments in 2018 in which thriving annuals created positive legacies in ANPP. Similarly, a previous study in a Swiss grassland found that precipitation timing-mediated recovery of grasses from drought produced positive ANPP legacies (36). These results suggest that precipitation pattern or timing is important in driving positive drought legacies on ANPP. It should be noted that our system is a cold, semiarid grassland, and the generality of these findings needs to be further corroborated in other grasslands.

In conclusion, this study provides experimental and observational evidence that severe droughts can produce positive legacies in community ANPP, consistent with the findings of previous studies (17–20, 36). In this perennial and annual coexisting grassland community, years of drought reduced the density of perennials, increased the availability of soil nitrogen, and promoted the growth of annuals. Under a favorable precipitation pattern for the growth of annuals in subsequent years, the thriving of annuals greatly increased postdrought ANPP and resulted in positive ANPP legacies. Given that positive ANPP legacies occurred in more than one-third of the past 40 years in this grassland ecosystem and were observed in other grasslands (17–20) and that global climate models predict more frequent drought extremes in grasslands worldwide (37, 38), our findings may have general implications for understanding the impacts of severe droughts on ecosystem functioning and services in grasslands. We identified that the underlying mechanism for the positive drought legacies was the coupled effect of drought-induced increases in annuals and the favorable precipitation pattern for their thriving in subsequent years. Considering that an increased abundance of annuals has occurred extensively in grazing grasslands worldwide (39), the drought-induced positive legacies in these systems should be more evident. Thus, models forecasting ecosystem feedbacks to climate change should consider positive drought legacies in grasslands.

MATERIALS AND METHODS

Study site

This study was conducted at the Inner Mongolia Grassland Ecosystem Research Station (IMGERS), Chinese Academy of Sciences (116°42′E, 43°38′N). This area has a medium temperate monsoon climate with uneven precipitation and marked temperature changes throughout the year. The mean annual temperature and precipitation were 1.1°C and 335.9 mm (1982–2012), respectively. The growing season usually lasts from early May to late October and receives more than 80% of the annual precipitation. The long-term average fractions of growing season precipitation in the early (May and June), middle (July and August), and late (September and October) stages of the growing season were approximately 30, 50, and 20%, respectively. The vegetation is dominated by perennial grasses, including Leymus chinensis, Stipa grandis, and Agropyron cristatum (8). In the relatively dry years, some annuals, such as C. glaucum, S. collina, and Axyris amaranthoides, increased or flourished following summer rainfall. The soil type was chestnut in the Chinese soil taxonomy system and Calcic-Orthic Aridisol in the U.S. soil taxonomy system (40).

Inner Mongolia grassland precipitation experiment

The Inner Mongolia grassland precipitation experiment (IMGPE) was conducted at an experimental site (500 m × 500 m) fenced by IMGERS in 1999, close to the long-term observation site. In 2012, six movable-roof rainout shelters (replicates), each 35 m by 11 m in size, were established. From 2014 to 2017, nine levels of growing season precipitation were implemented in each rainout shelter by watering the plots at 100, 150, 200, 275, 300, 350, 400, 450, and 500 mm. Rainwater was collected from rainout shelter roofs. Such a precipitation gradient covered the lowest (116.3 mm in 1974) and the highest (443.6 mm in 2005) records in this region. The 275-mm treatment was the long-term (1982–2012) average growing season precipitation at IMGERS. According to the long-term (1982–2012) average fractions of precipitation amount and frequency, the imposed fractions of precipitation amount at the early, middle, and late stages were 30, 50, and 20%, respectively, and the numbers of corresponding precipitation events were 8, 10, and 5 times, respectively (fig. S14). The plots for treatments were 4 m by 4 m in size and hydrologically isolated by galvanized plates with a belowground depth of 1 m. Precipitation manipulation lasted for each growing season (May 1–October 31) from 2014 to 2017. Precipitation treatments were stopped in 2018, and all plots received ambient precipitation in 2018 and 2019 (fig. S15). On the basis of this experimental platform, aboveground biomass (a reasonable proxy for ANPP) was measured in each plot by clipping all plants within a 1-m × 0.5-m quadrat every year in late August (corresponding to the peak biomass period). The plants were sorted into species, dried, and weighed. The species in each plot were classified into two functional groups, perennials and annuals. In each plot, we randomly collected three soil cores (2 cm diameter, 30 cm depth) at the end of the growing season (25 October 2017). Fresh soil was analyzed to determine inorganic nitrogen (NH₄⁺-N and NO₃⁻-N) concentrations using a FIAstar 5000 Analyzer (Foss Tecator, Höganäs, Sweden). Soil moisture (0 to 20 cm) was measured using a portable soil moisture instrument (Diviner 2000, Sentek Pty Ltd., Balmain, Australia) at an interval of approximately 3 days. Soil temperature at a depth of 10 cm was recorded using iButton digital temperature loggers (DS1922L, Maxim Integrated Products, Inc., CA, USA) with a time step of 2 hours.

Long-term observations of ANPP during 1980–2019

The long-term observation data of community ANPP (1980–2019) were obtained from a long-term field investigation by IMGERS in a permanent experimental plot (500 m × 500 m), established in 1979. This plot represents the mature community of steppe ecosystems (40). Within this plot, an east-west transect (200 m × 100 m) was established in 1980. It was then divided into five equal-sized replicate subplots, each 40 m by 100 m in size. In late August of each year, all aboveground plants were sampled from each subplot using a 1-m × 1-m quadrat. The mean aboveground biomass of the five subplots was used as the ANPP for the corresponding year. Similar to the IMGPE experiment, the species in each subplot were classified into two functional groups: perennials and annuals. In addition, long-term meteorological data (1980–2019) were obtained from the IMGERS weather station.

Statistical analysis

The legacy value was quantified as the difference between the observed ANPP (ANPP_observed) and the expected ANPP (ANPP_expected), while the latter was estimated by the overall relationship between ANPP and the growing season precipitation in the IMGPE over 4 years (2014–2017). This relationship was best fit with a saturating function [y = 272.18x/(178.77 + x); fig. S16 and table S1] in terms of the Akaike information criterion (AIC). The legacy values in the IMGPE in 2018 and 2019 and the long-term observations were calculated as follows (4, 14)

$$\text{Legacy value}={ANPP}_{observed}-{ANPP}_{expected}$$

We used the middle- to early-stage precipitation ratio as a parameter to indicate the precipitation distribution pattern for each year. In the long-term observational study, we examined the relationship between legacy values and the ratios of middle- to early-stage precipitation amounts. AIC was used to determine the best function of the relationships (41). One-way analysis of variance (ANOVA) followed by a least significant difference (LSD) multiple comparison test was used to examine the differences in ANPP among treatments and precipitation patterns under different conditions. A t test after normality and homogeneity tests was used to examine the differences between precipitation amounts in 2018 and 2019 and the mean of the 40-year precipitation amounts at the early, middle, and late growing stages. Differences in plant density and individual biomass of annuals and perennials between 2017 and 2018 under the same treatment were also examined using a t test. Biomass-based A/P ratios were used to evaluate changes in plant community structure. In the long-term observational study, years with a significant increase in the A/P ratio (P < 0.05) were identified, and their legacy values in subsequent years were examined. Variance partitioning analysis was used to illustrate the fractions of variance in the ANPP legacy value, explained by the A/P and M/E ratios. All statistical analyses were performed using R version 3.6.0 (R Core Team, 2019).

Supplementary Materials

This PDF file includes:

Figs. S1 to S16

Table S1