Long term prevention of disturbance induces the collapse of a dominant species without altering ecosystem function

Limitation of disturbances, such as grazing and fire, is a key tool for nature reserve management and ecological restoration. While the role of these disturbances in shaping ecosystem structure and functioning has been intensively studied, less is known about the consequences of long-term prevention of grazing and fire. Based on a 31-year study, we show that relative biomass of the dominant grass, Leymus chinensis, of grasslands in northern China declined dramatically, but only after 21 years of exclusion of fire and grazing. However, aboveground net primary productivity (ANPP) did not decline accordingly due to compensatory responses of several subdominant grass species. The decline in dominance of L. chinensis was not related to gradually changing climate during the same period, whereas experimentally imposed litter removal (simulating fire), mowing (simulating grazing), fire and moderate grazing enhanced dominance of L. chinensis significantly. Thus, our findings show that disturbances can be critical to maintain the dominance of key grass species in semiarid grassland, but that the collapse of a dominant species does not necessarily result in significant change in ANPP if there are species in the community capable of compensating for loss of a dominant.

litter decreased species richness and evenness 14 . However, our understanding is limited concerning the long-term (e.g., > 30 years) impacts of removal of disturbances on plant community dynamics and ecosystem function for ecosystems in which disturbances are integral to their structure and function. Long term studies on dynamics of these "protected" ecosystems are needed, especially within the backdrop of global change.
Disturbances generally impact ecosystems by altering biotic structure 15 , and it is the changes in community structure and associated biogeochemical processes that drive subsequent ecosystem dynamics over the long-term 16 . In some cases, ecosystem recovery from disturbance can be rapid 8 , whereas in other cases it can be prolonged, particularly if species that contribute disproportionately to ecosystem function (i.e., dominant species 17 ) are negatively impacted [18][19][20][21] . Grazing and fire are disturbances in grassland that are often modified by humans, with these alterations ranging from increased frequency to complete exclusion 22 . The long-term prevention of grazing and fire can have varying impacts as well, depending on the degree to which the disturbance(s) being excluded impacts dominant species 4 . The expectation is that if the exclusion of grazing and fire does not favor dominant species and/or strongly modifies biogeochemical processes, then the ecosystem impacts of the disturbance prevention can be large and potentially nonlinear, particularly if disturbance is excluded over the long-term and in tandem with other global change drivers, such as climate change 22 . Thus, it remains unclear whether ecosystem structure and function changes gradually (linearly) or abruptly (nonlinearly) after long-term prevention of disturbance and whether shifts in abundance of dominant species drive these dynamics.
Here, we examine the impacts of long-term exclusion of two key disturbance regimes -fire and grazing-in a semi-arid grassland in Inner Mongolia. Leymus Chinensis, a C 3 rhizomatous grass, is the dominant plant species in these grasslands. Historically, Inner Mongolia grasslands experienced an intermediate disturbance frequency 23 , but these grasslands have more recently been degraded because of heavy grazing and alterations in fire regimes (primarily fire exclusion). Our study addresses the following questions: (1) Whether dominance of L. chinensis, plant community structure, and ecosystem functioning (aboveground net primary productivity, ANPP) changes gradually or abruptly during long-term (30 + years) exclusion of two key disturbances -fire and grazing? and (2) Does the shift of dominant species result in significant change in ecosystem function? Our expectation was that long-term exclusion of disturbances would negatively impact dominance by L. chinensis over time, and that this decline in dominance would alter plant community structure and negatively impact ANPP. The goal of our research was to document these dynamics and assess potential mechanisms underlying the expected decline in L. chinensis dominance.

Results
The long-term ( 1A). However, biomass of L. chinensis and PB in the grazed area remained relatively stable compare to fenced area. Despite the large change in biomass of L. chinensis in fenced area, aboveground net primary productivity (ANPP) did not exhibit a significant decline over the study period, although was variable over time (Fig. 1B). ANPP in the fenced area decreased with the decrease of L. chinensis from 1997-2000, while increased with the increase of PB from 2000-2003. ANPP in the grazed area was much lower than that in fenced area. In the period from 1980-2000, the relative abundance (biomass) of L. chinensis in the fenced area was on average 50.25 ± 2.32 (Mean ± SE, Fig. 1C), varying from 33% to 76%. However, it declined sharply from 1997, reaching a minimum in 2003, and then fluctuated between 0.4%-12.4% (6.18 ± 0.75). The relative biomass of L. chinensis in the grazed area was lower than that in fenced area from 1980-2000, however, higher from 2001-2010, with no clear trend of increase or decrease.
Biomass of L. chinensis contributed more than 50% of ANPP in the fenced area from 1980-2000 (Fig. 1A), and explained 69% variation in ANPP during this period ( Fig. 2A). However, there was no significant relationship between L. chinensis biomass and ANPP during 2001-2010 ( Fig. 2A). Biomass of PB was significantly and positively related with ANPP in both periods (Fig. 2B). Furthermore, biomass of L. chinensis and PB was negatively correlated (Fig. 2C) from 1980-2010.
The large and rapid reduction in L. chinensis biomass also resulted in large change in community structure. During 1980-2000, L. chinensis was by far the most dominant member of the community, with Stipa grandis, Achnatherum sibiricum, Carex korshinskyi, Agropyron cristatum, Caragana microphylia ranked (by relative biomass) 2-6 in the community (Fig. 3A). These 2-6 th ranked species contributed 27% to ANPP, whereas the remaining 26 species in the community contributed 23% to ANPP. However, during 2001-2010, with S. grandis becoming the top-ranked species, the increased abundance in S. grandis, A. cristatum and A. sibiricum collectively contributed 57% of ANPP (Fig. 3B). In this second period, L. chinensis was relegated to 4 th ranked species, and along with Cleistogenes squarrosa, C. korshinskyi, and Artemisia frigida (ranked from 5-7) contributed 24% of ANPP. Several other species also exhibited dramatic changes in relative abundance between the two periods. For example the forb, Artemisia Scientific RepoRts | 5:14320 | DOi: 10.1038/srep14320 pubescens decreased its rank from 7 to 27, while Chenopodium glaucum increased its rank from 29 to12. For plant functional groups (see details for species composition of each plant functional group in Bai et al. 2004), perennial rhizomatous grasses (PR, comprised of only L. chinensis) decreased significantly in abundance (P < 0.05), while perennial bunchgrasses (e.g., S. grandis, A. cristatum and A. sibiricum) and shrubs and semi-shrubs (e.g., Kochia prostrata) increased significantly ( Fig. 3C, P < 0.05). The increase of PB compensated for most of the decrease of PR.
To assess potential mechanisms underlying the dramatic change in abundance of L. chinensis, we examined relationships between biomass of L. chinensis and annual precipitation or mean daily temperature over the study period. In addition, we utilized several short-term experiments to assess the impacts of litter removal (simulating fire), mowing (simulating grazing) and grazing intensity on L. chinensis abundance. Although annual precipitation decreased and mean daily temperature increased significantly from the 1980-2000 period to the 2001-2010 period (P < 0.05), dominance of L. chinensis was not significantly related with annual precipitation (P = 0.14, r 2 = 0.07, Fig. 4A) or mean daily temperature (P = 0.08, r 2 = 0.11, Fig. 4B) from 1980-2010 and for both periods. For the experiment removing litter and mowing vegetation at the end of the growing season, dominance (relative biomass) of L. chinensis was significantly higher in 2009 and 2010 than undisturbed (control) grassland (P < 0.05, Fig. 4C,D). Similarly, relative abundance of L. chinensis was enhanced by both early season fire and end-of-season mowing (Fig. 4E), as well as light grazing by sheep (Fig. 4F). However, with high intensity grazing, relative abundance of L. chinensis decreased when compared to the ungrazed treatment.

Discussion
Limiting the frequency of disturbances is a key tool for nature reserve management and ecological restoration 8,24 . Yet, we have limited understanding of the long-term (e.g. > 30 yrs) consequences of such prevention of disturbance regimes. As expected, the dominant grass species, L. chinensis, which can strongly drive ecosystem functions, such as ANPP, declined with fire and grazing exclusion in a semi-arid Inner Mongolian grassland. However, contrary to expectations, this decline was not gradual, but rather considerably lagged and abrupt. The dominance of L. chinensis remained relatively stable in the first 21 years of cessation of disturbance in this study, but then L. chinensis declined rapidly and was relegated to subdominant status in the community within only a year or two. If our study had lasted less than 21 years, we would have concluded that prevention of disturbance had no significant effects on L. chinensis, highlighting the importance of long term studies. Thus, even two decades may not be 'long-term' enough for some processes in plant communities.
Although L. chinensis dominance declined abruptly, aboveground productivity remained surprisingly unaffected by the dramatic loss of this key grass species. It is well known that for most ecosystems the dominant species mediate most ecological processes 17,25 , control the majority of the resources (including Scientific RepoRts | 5:14320 | DOi: 10.1038/srep14320 space) and have disproportionate impacts on species interactions and ecosystem functions 17 . Consistent with this, we found that the biomass of L. chinensis explained 69% variation of ANPP in the first 21 years of the study. However, given their role in communities, changes in abundance of dominant species are expected to impact ecosystem processes and function significantly or even disastrously 17,[26][27][28] . But contrary to this expectation, we did not observe large effects on ecosystem function due to compensatory responses of a subset of the community-perennial bunchgrasses (PB, S. grandis, A. cristatum and A. sibiricum)-which were subdominant in the community during the first 21 years of disturbance exclusion. Such compensatory responses have been shown to be important in maintaining ecosystem function in a 24-year study in the same site 29 and in other ecosystems 20,30,31 . Our results point to the possibility that compensation may not result from one dominant species being replaced by a functionally similar species (in this case a perennial rhizomatous grass), but that several species may collectively result in compensation and that these species may belong to a completely different functional group -perennial bunchgrasses.
The large shift in abundance of L. chinensis may have resulted from a number of factors. Here we explore two potential drivers: gradually changing climate and the long-term exclusion of disturbances. We found annual mean precipitation decreased and annual mean temperature increased significantly from 2001-2010 (P < 0.05), consistent with the decrease of L. chinensis' dominance. However, climate and L. chinensis' dominance were not significantly correlated. Furthermore, L. chinensis did not decline during the same period of gradually changing climate in an adjacent site which had been fenced since 1999 32 . If climate change induced the decrease of L. chinensis in our long-term site from 2001-2010, we would expect to observe the same decrease in adjacent sites. Furthermore, we did not observe any extremes in climate (drought, high heat) during the time of rapid decline of L. chinensis. Climate extremes, such as drought, have been shown to be important for triggering abrupt declines in dominant species in other systems 33 . Thus, taken together with the observation from 1980 to 2010, we inferred that neither gradually changing climate nor climate extremes were the primary drivers of the drastic decline of L. chinensis.
To assess the role that lack of disturbance played in the change in L. chinensis abundance over time, we examined results from several short-term studies manipulating fire and mowing, as well as the presence of litter. Litter removal and mowing increased the dominance of L. chinensis in the same site significantly, while in other sites mowing, fire and low-density grazing also increased the dominance of L. chinensis significantly. These results clearly show the importance of disturbance in maintaining the dominance of L. chinensis in Inner Mongolian semi-arid grasslands. However, another mowing experiment conducted in the same research site showed that the mowing treatment decreased the dominance of L. chinensis before 1998 23 , indicating that the same disturbance type could induce different responses if acting in different successional stages. Similarly, intensive grazing decreased the dominance of L. chinensis 34 , so the relative degree of disturbance also is an important consideration. The experimental prevention of disturbance underlies the loss of dominance of L. chinensis in the 31 year long observation. Thus, it appears that moderate disturbance is critical to maintain ecosystem structure of Inner Mongolian grasslands.
Although these short-term experiments demonstrate that disturbances are important for maintaining dominance by L. chinensis, they do not explain the abrupt change in abundance that was observed only after 21 years of exclusion of disturbances. The nonlinear change suggests that a threshold was crossed 22 , but the mechanisms underlying the abrupt change remains unclear. One candidate mechanism is the buildup of litter over the long-term negatively impacted the resprouting of L. chinensis, which reproduces primarily asexually via rhizomatous growth. Short-term litter removal increased L. chinensis abundance, and therefore it may be that it takes a long time for the buildup of litter to have negative demographic effects on this species. Similar negative effects of litter have been observed for other rhizomatous grasses in the absence of fire and grazing 35 . Another potential mechanism could be that it may take some time for the primarily sexually reproducing bunchgrasses to increase in abundance to a point in which they could negatively impact L. chinensis. Finally, changes in resource availability could also be an important mechanism affecting L. chinensis, either directly or indirectly by favoring less abundant species 36 . L. chinensis is known to have high stoichiometric homeostasis 37 and low growth rate 38 , which may promote L. chinensis under natural disturbance regimes, but accumulation of resources under lack of disturbance could allow other less competitive species to become more abundant over time. Unfortunately, data is not available to fully test these mechanisms, and thus further studies are needed to fully understand the factors that lead to the abrupt shift in L. chinensis with exclusion of disturbances over the long-term.
In conclusion, our 31-year study showed that relative biomass of L. chinensis declined sharply only after 21 years of exclusion of grazing and fire. Thus, long-term management to prevent disturbances can have large impacts on plant communities but these impacts may be considerably lagged. ANPP did not decline accordingly, suggesting collapse of dominant species does not necessarily result in significant change in ecosystem function, as a consequence of compensatory responses of a functionally different subset of the community -perennial bunchgrasses -which were subdominant in the community, rather than species that were functionally similar. Therefore, it is clear that we need to understand functional roles of less common species in communities, as these often overlooked species may be critical for maintenance of ecosystem function over the long-term with alterations in disturbance regimes and other global changes.

Material and Methods
We utilized four studies to address our research questions: 1) a long term field monitoring program, 2) a litter removal and mowing experiment, 3) a mowing and fire experiment, and 4) a grazing experiment. All of these studies were located at the Inner Mongolia Grassland Ecosystem Research Station (IMGERS), which is located in the Xilin River Basin, Inner Mongolia Autonomous Region, China (43°38′ N, 116°42′ E). The study area selected for the four studies is dominated by the perennial, rhizomatous C 3 grass, L. chinensis, an important, widely distributed rhizomatous C 3 grass species that dominates in the eastern part of the Eurasian steppe (40°-62° N, 87°-130° E), a 420,000 km 2 area. Interspersed in this matrix of grass is diverse suite of uncommon grass and forb species that comprise the following functional groups: perennial bunchgrasses (PB), perennial forbs (PF), shrubs and semi-shrubs (SS), and annuals and biennials (AB). The annual mean temperature in the study area is 0.3 °C with mean monthly temperatures ranging from − 21.6 °C in January to 19.0 °C in July. The annual mean precipitation is 346.1 mm with 60-80% falling during the growing season from May to August, and approximately 10% falling as snow.
Long term field monitoring. The long-term monitoring took place at a 500 × 500 m site, which was fenced to exclude grazers in 1979. Prior to that, the study area was subject to light grazing. The grassland outside the fenced area continued to be grazed with the rate of 1.5-3.0 sheep ha −1 . There was no fire during our study. As we have no replicate for fenced and grazed area, an east-west transect of 200 * 100 m was established with five equal-sized replicate blocks (40 * 100 m each) within each area to ensure the sampling area is big enough to represent the spatial heterogeneity. From 1980 to 2010, end of season aboveground biomass was sampled by clipping all plants at ground level within a 1 × 1 m quadrat that was randomly located within each block, which decreased the limitation of no replicate for fenced and grazed area. All living vascular plants were sorted to species, oven-dried at 60 °C and weighed. Because the standing crop of these steppe communities reached the annual peak at the end of August, our estimated community biomass approximated the aboveground net primary productivity (ANPP) of these ecosystems. As we sampled ANPP in the end of August for the long-term monitoring study, annual precipitation (AP) and annual mean daily temperature (MDT) collected at the study site were based on data from September 1 of the previous year to August 31 of the current year. All meteorological data were obtained from the weather stations of IMGERS. Full methods and details on experimental treatments can be found in earlier studies 29,39 . The litter removal and mowing experiment. The litter removal and mowing experiment was initiated in 2007 and consists of the following treatments: control, litter removal, and mowing, each replicated six times (3 treatments × 6 replicates = 18 plots total). The plots were 3 × 3 m, with 1 m buffer separating each plot. The experimental treatments were randomly assigned to each plot. To simulate grazing, a mowing treatment was carried out by clipping all the standing plants at 5 cm above ground on 1 September each year. To simulate the effects of decreased litter after grazing and fire, the litter removal treatment was implemented on 30 August each year by removing all litter at the ground level. Aboveground biomass was sampled within 1 × 1 m quadrats within each plot on 30 August in 2009 and 2010 using the same methods as described above.
The fire and mowing experiment. The fire and mowing experiment was established in 2005 and included three treatments: control, fire and mowing. The experimental design was a randomized block Scientific RepoRts | 5:14320 | DOi: 10.1038/srep14320 design, with 6 blocks (each separated by a 2-m walkway), with each block containing three 10 × 10 m plots, each separated by a 1-m buffer. The treatments were randomly assigned to the three plots within each block. All blocks were located at the same topographic position and on similar soils. The fire treatments were applied in early May of 2006 and 2007. The fire removed almost all the aboveground plant materials and surface litter. The mowing treatment was implemented in October 2006, with all the plants cut at 5 cm above ground. On 15 August 2007, aboveground biomass was sampled by cutting all plants at ground level within a 1 × 0.5 m quadrat located within each plot. All living vascular plants were sorted to species, oven-dried at 60 °C and weighed. The grazing experiment. In 2004, we established a grazing experiment consisting of seven stocking rates 40 . For this study, we utilized only four of the stocking rate treatments: 0, 1.5, 3.0 and 4.5 sheep ha −1 (hereafter referred to as control, SR1.5, SR3.0 and SR4.5, respectively). The area of each fenced plot was 2 ha except for treatment SR1.5, in which was 4 ha. Starting in 2005, sheep were transferred to plots in mid-June and were maintained there until mid-September of each year. We sampled the plants within a 1 × 1 m quadrat with 6 replicates in each plot in August 2007.
Statistical analysis. Linear regression analysis was used to assess the relationships between biomass of L. chinensis, biomass of PB and ANPP, as well as relationships between relative biomass (calculated as the ratio of biomass of L. chinensis to ANPP) and biomass of L. chinensis and AP and MDT. ANOVA was used to test the difference of annual precipitation, mean daily temperature, relative biomass of functional groups, and relative biomass of L. chinensis among various treatments for each experiment separately. All analyses were performed using SAS (version 9.0, SAS Inst., Cary, NC, USA).