Soil disturbance and invasion magnify CO2 effects on grassland productivity, reducing diversity

Climate change, disturbance, and plant invasion threaten grassland ecosystems, but their combined and interactive effects are poorly understood. Here, we examine how the combination of disturbance and plant invasion influences the sensitivity of mixed‐grass prairie to elevated carbon dioxide (eCO2) and warming. We established subplots of intact prairie and disturbed/invaded prairie within a free‐air CO2 enrichment (to 600 ppmv) by infrared warming (+1.5°C day, 3°C night) experiment and followed plant and soil responses for 5 years. Elevated CO2 initially led to moderate increases in biomass and plant diversity in both intact and disturbed/invaded prairie, but these effects shifted due to strong eCO2 responses of the invasive forb Centaurea diffusa. In the final 3 years, biomass responses to eCO2 in disturbed/invaded prairie were 10 times as large as those in intact prairie (+186% vs. +18%), resulting in reduced rather than increased plant diversity (−17% vs. +10%). At the same time, warming interacted with disturbance/invasion and year, reducing the rate of topsoil carbon recovery following disturbance. The strength of these interactions demonstrates the need to incorporate disturbance into predictions of climate change effects. In contrast to expectations from studies in intact ecosystems, eCO2 may threaten plant diversity in ecosystems subject to soil disturbance and invasion.


| INTRODUC TI ON
Elevated CO 2 , global warming and disturbances associated with land use threaten biological diversity and ecosystem services globally (Bellard et al., 2012;Dale, 1997;Hoekstra et al., 2005). To date, these processes have been studied largely in isolation, with studies of climate change (here used broadly to include elevated CO 2 [eCO 2 ]), focusing on relatively undisturbed ecosystems, and disturbance studies considering effects under present-day climate (Perring et al., 2016). However, given that climate and disturbance influence many of the same processes, it is critical to understand their combined and interactive effects (Oliver & Morecroft, 2014;Ostberg et al., 2015;Perring et al., 2016).
For example, soil food webs more effectively resist drought, and therefore retain soil carbon (C) and nitrogen (N), within extensively managed pastures than within intensively managed wheat fields (De Vries et al., 2012). While such interactions are likely to be important, predicting them is complex, requiring data-rich models or multi-factor experiments (Oliver & Morecroft, 2014;Perring et al., 2016).
In grasslands, land use can involve several disturbances that may alter effects of climate change. A few studies have examined particular climate-disturbance combinations. Grazing was found to counteract negative effects of warming on plant species richness on the Tibetan Plateau (Klein et al., 2004), and effects of eCO 2 on legume abundance in New Zealand (Newton et al., 2014).
Grassland fires, which can be intentionally or unintentionally associated with land use, are expected to interact with warming and precipitation change, with significant effects on ecosystem services (Fernandes et al., 2013;Thonicke et al., 2001;Westerling et al., 2006). In California annual grassland, fire led to more positive effects of eCO 2 and nitrogen (N) addition on productivity (Henry et al., 2006).
Soil disturbance is also common in grasslands, resulting from cultivation, energy development, or other intensive forms of land use. Soil disturbances often damage root systems and alter soil structure, with correspondingly large effects on soil properties, resource availability, and plant communities (James et al., 2003;Wilson & Tilman, 1993). These effects of disturbance are likely to influence how plants and soils respond to CO 2 and warming.
For example, greater nutrient availability can increase plant productivity responses to eCO 2 (Reich & Hobbie, 2013;Terrer et al., 2019), suggesting the potential for similar patterns with disturbance. Soil resource availability can also interact with eCO 2 and warming to influence plant communities and C cycling, effects that can be mediated by changes in productivity, soil respiration, and the sensitivity of individual species (Pastore et al., 2021;Reich, 2009).
Despite the prevalence of anthropogenic soil disturbance, and its potential for interacting with eCO 2 or warming, there have been few empirical studies of such interactions. Plowing did not interact with warming to influence grassland cover or species richness in Japanese montane grasslands (Suzuki, 2014). In separate studies of intact and disturbed calcareous grassland in Switzerland, eCO 2 had broadly similar effects on plant productivity and species evenness (Leadley et al., 1999;Niklaus et al., 2001).
Soil disturbance is also frequently associated with plant invasion (Davis et al., 2000;Hobbs & Huenneke, 1992), which itself can interact with climate change (Dukes, 2000;Liu et al., 2017;Sorte et al., 2013). Disturbance may reduce biotic resistance to invasion from native species (D'Antonio & Thomsen, 2004). Furthermore, many invasive plants have traits that allow them to respond positively to changes that involve increases in resource availability, including not only disturbance, but also N deposition, eCO 2 , and increases in precipitation (Bradley et al., 2010;Dukes & Mooney, 1999;Seabloom et al., 2015). Across studies, invasive species tend to respond more positively to eCO 2 and warming than do native species (Liu et al., 2017). Consequently, the combination of soil disturbance and invasion has the potential to alter the way grasslands respond to eCO 2 and warming.
Here, we describe an experimental study of how eCO 2 and warming interact with disturbance to influence plant communities and ecosystem properties in mixed-grass prairie. The semiarid mixed-grass prairie is the largest remaining grassland in North America, sustaining both biological diversity and livestock production (Samson et al., 2004). It is threatened by climate change as well as disturbance, particularly energy development and crop production, both of which involve soil disturbance and associated plant invasion (Heisler-White et al., 2009;Mueller et al., 2016;Ott et al., 2020;Samson et al., 2004).
We have previously documented effects of eCO 2 and warming on intact mixed-grass prairie, including increased plant productivity, reduced nutrient availability and forage quality, altered soil C, and reduced dominance by common grass species (Augustine et al., 2018;Carrillo et al., 2018;Dijkstra et al., 2010;Morgan et al., 2011;Mueller et al., 2016;Zelikova et al., 2014). We have also found that both eCO 2 and warming can promote plant invasion Reeves et al., 2015). In the present study, we compare effects of eCO 2 and warming on intact mixed-grass prairie to effects on prairie treated with soil disturbance and seeded with five native and five invasive forb species.
Our hypotheses were: (1) eCO 2 and warming would increase plant productivity (Mueller et al., 2016) more in disturbed/invaded than in intact prairie due to greater resource availability in combination with species adapted to grow rapidly in disturbed environments; (2) eCO 2 would affect plant diversity differently in disturbed/invaded than in intact prairie, potentially increasing diversity (Zelikova et al., 2014) (a) more in disturbed/invaded prairie, if disturbance reduced dominance by native grass species, or (b) less in disturbed/invaded prairie, if planted native or invasive forbs became dominant; (3) disturbance/invasion would reduce soil C, and eCO 2 and warming effects on soil C would follow from their effects on plant productivity, increasing C more in disturbed/invaded than in intact prairie.

| Experimental design
Climate treatments included: (1) ct: ambient CO 2 and temperature, (2) cT: infrared warming to increase canopy temperature 1.5°C during the day and 3°C at night, (3) Ct: free-air CO 2 enrichment (FACE) to 600 ppmv, and (4) CT: CO 2 enrichment plus warming. The target CO 2 and temperature levels correspond with late 21st century projections for the region (CMIP3 emissions scenario B1; Kunkel et al., 2013). Global change treatments started in spring 2006 (eCO 2 ) and spring 2007 (warming) (Figure 1). Five replications of each treatment were randomly assigned to 20, 3.3-m diameter circular plots ("rings") within two blocks with slightly different soil types (Morgan et al., 2011). The warming treatment was imposed year-round.
Enrichment of CO 2 was limited to daylight hours when green plants were present and the site was not covered with snow. We controlled for potential infrastructure effects by installing dummy FACE tubing and heaters at untreated rings. To minimize dilution of treatment effects through lateral water flow, we buried a 60-cm-deep plastic barrier around each ring. Additional details on the experimental design and system performance are reported in Morgan et al. (2011) and LeCain et al. (2015).
Each experimental ring was randomly divided into intact prairie and disturbed prairie (including multiple smaller experiments) by a 30-cm deep, belowground metal barrier oriented parallel to the prevailing northwest winds ( Figure 1).
The disturbed/invaded treatment was designed to mimic key aspects of soil disturbances in mixed-grass prairie-partial plant removal, soil turnover, colonization by invasive species-at a scale small enough for this climate change experiment. Similar changes occur with anthropogenic disturbances such energy development, where road and well-pad construction can involve removal of plants, storage and subsequent reapplication of topsoil, partial revegetation, and invasion by non-native species (Ott et al., 2020). Within the disturbed prairie section of each ring, we established a 70-cm × 100-cm subplot and partially tilled it (57% of the area) to a depth of 10 cm (Figure 1, Figure S1).
Specifically, we tilled two 20-cm × 100-cm strips, separated by a 15-cm × 100-cm strip of existing prairie vegetation in the center, and bordered by two 7.5-cm × 100-cm strips of prairie vegetation on the sides. Tillage involved turning over, breaking up, and homogenizing the soil with mattocks, hoes, and rakes, and was conducted in October 2007. Effects of disturbance alone (not presented here) were monitored for 1 year.
In November 2008, the entire subplot was seeded with five invasive and five native forb species, six of which successfully established in the study: A. frigida, Centaurea diffusa, Chenopodium leptophyllum, Grindelia squarrosa, Linaria dalmatica, and Salsola tragus ( Table S1).
We focused on forbs because many of the most important invasive plants in mixed-grass prairie are forbs and because this enabled us to locate and track seedlings within the graminoid-dominated ecosystem. Competition between native species and the invasive grass, Bromus tectorum was examined in a separate study within the PHACE experiment . For species selection, we paired each invasive species with a native species by life history, habitat, and where possible phylogeny (Table S1). We selected locally common species to ensure that they were well-adapted to local environmental conditions. All seeds were collected from nearby field sites in Colorado and Wyoming. We sowed 3.43 g seed species −1 m −2 by hand, taking care to distribute seed evenly over both the tilled and vegetated sections of each disturbed/invaded subplot, after which the surface soil was raked and lightly packed to improve seedsoil contact.

| Measurements
We sampled peak aboveground biomass in mid-July. Vegetation was clipped at the ground surface, separated by species, dried at 60°C, and weighed. The harvest plan was designed to mimic effects of grazing in this typically grazed ecosystem. In the intact prairie, a 1.5-m 2 area was divided into 24 squares, half of which were harvested each year. Individual squares were therefore harvested two or three times during the 5-year study. In the disturbed/invaded subplots, re-harvest was not realistic because dominant species F I G U R E 1 Diagram of study design. Colors indicate the timing and location of treatments and sampling. Soil sampling dots indicate approximate locations of multiple smaller samples. Individual rings (whole plots) were 3.3 m in diameter. Underground barriers separated subplots from one another, and rings from the surrounding prairie. were unpalatable. In addition, some of the short-lived forbs relied on seed set occurring after the time of harvest. Consequently, separate 0.07-m 2 areas (including both tilled and vegetated strips) were harvested each year. In the final year, we harvested all of the plot area not previously harvested (0.28 m 2 ). To minimize effects of different clipping frequencies among subplots, all areas not harvested during the growing season were clipped to a height of 2 cm following full senescence. To assess the potential effects of different clipping frequencies on biomass production we examined results from a study of grazing intensity at the same site. Methods are described in Derner and Hart (2007).
Sampling for soil C and belowground biomass was conducted in the week following peak biomass clipping. We collected five soil cores in intact subplots (3 cm diameter, 15 cm depth) and four soil cores in tilled strips within disturbed/invaded subplots (2.54 cm diameter, 15 cm depth). All cores were then split into 0-to 5-and 5-to 15-cm samples and composited by depth. Roots were removed by sieving followed by hand picking of sieved soil, then washed, dried, and weighed. A subsample of the composited soil was sieved to 2 mm, inspected to remove visible particulate organic matter, and then treated to remove carbonates (1 M H 3 PO 4 ) (Sherrod et al., 2002).
The resulting dried soil was ground to powder and weighed for C analysis on a Finnigan DeltaPlus XP connected to a Carlo Erba NC-

elemental analyzer. Pretreatment soil C was collected in 2005
for each ring, using the same methods, from three cores located around the edge of each ring.
To measure soil NO 3 − availability, we installed resin probes vertically from 2 to 7.6 cm depth, between May and October (or late

| Analysis
Statistical analyses were conducted in JMP version 15 (SAS Institute, Cary, NC, USA). We analyzed biomass, soil C, and soil N responses with linear mixed models that included random effects of block (to account for spatial covariation) and ring (to account for the split-plot design), and fixed effects of eCO 2 , warming, disturbance/invasion, and their interactions.
For dependent variables measured over time, models also included a random effect of ring × disturbance (to account for repeated measures within subplots), and fixed effects for year and its associated interactions. In analyses of biomass, soil N, and diversity indices, year was a categorical variable, reflecting the expectation that its effects would depend mostly on climatic fluctuations (Mueller et al., 2016). Analyses of soil N also included a random effect to account for the number of days probes were left in the soil each year. In analyses of soil C, year was a continuous variable, to test for differences in the rate of soil C recovery following disturbance. To account for preexisting spatial variation in soil C, these models also included pretreatment soil C as a covariate (Carrillo et al., 2018). To meet model assumptions of homoscedasticity some dependent variables were ln transformed (aboveground biomass, NO 3 − ) or squared (light penetration) prior to analysis. Significant interac- tions between year, disturbance/invasion, eCO 2 , and warming treatments were investigated with post hoc analyses within disturbance treatments (Table S2).
Shannon's diversity and evenness were calculated from biomass data in R, using the vegan package (Oksanen et al., 2015).
Diversity metrics were not quantitatively comparable across subplots because measurement areas differed in size and only disturbed/invaded prairie was seeded with additional species.
We therefore analyzed diversity metrics only within disturbance treatments (Table 2). We used bivariate linear regression to test whether species diversity and resin-available NO 3 − were correlated with aboveground biomass, and whether belowground biomass was correlated with diversity and soil C, within disturbance treatment and year.
To assess the relative importance of direct and indirect effects of eCO 2 and warming on diversity, we conducted separate confirmatory path analyses for intact and disturbed/invaded prairie. The models included effects of eCO 2 , warming, and eCO 2 × warming on both aboveground biomass and diversity, and effects of aboveground biomass on diversity. Thus effects of climate change treatments on diversity could be either direct or indirect, mediated by aboveground biomass.

| Aboveground biomass
Elevated CO 2 increased aboveground biomass, and the strength of this effect varied with disturbance and year (disturbance × eCO 2 × year; Figure 2, Table 1). In disturbed/invaded prairie, eCO 2 interacted with year (Table S2) In intact prairie, eCO 2 also increased aboveground biomass in later years of the study (eCO 2 × year: Figure 2, Table S2), but its effects were smaller than in disturbed/invaded prairie: −3%, 1%, 20%, previously (Mueller et al., 2016). Declines in biomass in intact prairie over time were closely associated with similar trends in precipitation (Mueller et al., 2016).

| Belowground biomass
We measured belowground biomass in both disturbed/invaded and intact prairie only in 2011, the year with maximum aboveground biomass of C. diffusa. At 0-5 cm depth, eCO 2 increased belowground biomass across disturbance treatments by an average of 55% ( Figure S2, Table 1). At 5-15 cm depth, belowground biomass was higher in disturbed/invaded than intact prairie, but was not significantly altered by global change treatments.
We examined the relative strength of direct and indirect predictors of diversity using path analyses. Marginal R 2 values for diversity were .12 for intact prairie and .35 for disturbed/invaded prairie.
Model results support the interpretation that there are strong indirect effects of eCO 2 on diversity in disturbed/invaded prairie, mediated by increases in aboveground biomass ( Figure 4). In disturbed/ invaded prairie, eCO 2 increased aboveground biomass (p = .04), and aboveground biomass was negatively associated with diversity (p < .0001). In intact prairie, all pathways were non-significant (p > .25).

| Soil C
Trends in shallow (0-5 cm) soil C varied significantly with disturbance and warming (disturbance × warming × year; Figure 5, Table 1). In intact prairie, soil C declined over the course of the study, a trend that aligned with decreases in precipitation. Soil C was 25% lower in disturbed/invaded than intact prairie in 2010 but recovered substantially by 2013 (disturbance × year; Table 1) ( Figure 5). Within disturbed/invaded prairie, warming reduced the rate of soil C recovery (warming × year: F 1,54 = 5.9, p = .02; Table S2).  Table 1 and Table S2 for statistical results. The eCO 2 × warming interaction previously described for intact prairie (Carrillo et al., 2018), was marginally significant in the subset of years within this study (F 1,15 = 4.2, p = .06). In 5-15 cm soil C, trends over time were more positive in disturbed/invaded than intact prairie (disturbance × year;

| DISCUSS ION
Predicting effects of climate change requires a better understanding of how it will interact with widespread anthropogenic disturbances (Oliver & Morecroft, 2014;Perring et al., 2016). In this study, we found that eCO 2 and warming had very different effects in disturbed/invaded prairie than in intact prairie ( increased diversity by 10% in intact prairie, it reduced diversity by 17% in disturbed/invaded prairie. At the same time, disturbance/ invasion interacted with warming to influence belowground processes, with disturbance/invasion reducing soil C and warming slowing subsequent soil C accrual. Together, these patterns indicate that soil disturbance and associated invasion can increase sensitivity of grasslands to climate change. The increase in aboveground biomass we observed with eCO 2 in disturbed/invaded prairie was much larger than increases observed in other grassland experiments (Hovenden et al., 2019;Terrer et al., 2019). Both disturbance and invasion contributed to this biomass response. Across 2011-2013, 69% of the aboveground biomass in eCO 2 rings was from C. diffusa (Figure 2), a problematic invasive species in western North America. This biennial forb responded particularly strongly to eCO 2 (Reeves et al., 2015), as previously observed for another invasive forb at this site, L. dalmatica . The strong eCO 2 response of C. diffusa might be due to its rapid growth rate and its C 3 photosynthetic pathway (associated with strong eCO 2 responses among the native graminoids but not In previous studies, eCO 2 has often favored invasive over native plants (Liu et al., 2017;Sorte et al., 2013). In this case, however, the effect appears to have been compounded by soil disturbance. Almost all of the C. diffusa biomass (on a per-m 2 basis) was rooted in the tilled portions of the disturbed/invaded subplots (Figure 2), suggesting that it required soil disturbance to respond strongly to eCO 2 . Disturbances that kill plants tend to increase water and N availability in grasslands (James et al., 2003;Wilson & Tilman, 1993), including within the F I G U R E 3 Effects of eCO 2 and warming on species richness, evenness, and Shannon's diversity (H′) in intact versus disturbed/invaded prairie. Note that absolute values are not comparable across disturbance treatments and years, due to species addition in disturbed/ invaded subplots and variation in sampling effort (0.75 m 2 in intact prairie, and 0.1 m 2 in disturbed/invaded prairie in 2009-2012, and 0.28 m 2 in disturbed/invaded prairie in 2013). Treatment codes indicate global change treatments: ct = control, cT = warmed, Ct = eCO 2 , CT = warmed plus eCO 2 . Error bars are ±1 SE. See Table 2 for statistical results. An alternative explanation for the stronger biomass responses in disturbed/invaded prairie than intact prairie could be the lower growing-season harvest frequency in this treatment. However, in a long-term grazing study at the same site and the same years, biomass production was only 19% greater in ungrazed than in moderately grazed (~50% biomass removal) pastures ( Figure S7). This comparison is inexact, but suggests that differences in clipping frequency could account for only a small fraction of the observed 10-fold increase in biomass we observed with disturbance and invasion.
Our results in disturbed/invaded prairie were similar to those seen with nutrient fertilization: Large increases in total biomass with eCO 2 were associated with reduced light availability and reduced diversity both over time and among rings (Figure 3 and 4, Figures S3 and S6).  Table 1 and Table S2 for statistical results.   (Figures S4 and S5).
Soil C responses to climate change also differed in intact and disturbed/invaded prairie ( Figure 5). In intact prairie, soil C declined between 2010 and 2013, potentially due to dry conditions during the final 2 years (Mueller et al., 2016). In contrast, soil C in disturbed/ invaded prairie was 25% lower than that in intact prairie in 2010, but recovered by 2013, 6 years after disturbance. Warming slowed that recovery process in shallow (0-5 cm) soils. Both disturbance and warming effects may have been caused by accelerated decomposition (Carrillo et al., 2018;Walker et al., 2018). Warming effects on decomposition could have been stronger in disturbed/invaded prairie due to C inputs from newly established plants with relatively labile tissue, including C. diffusa. This explanation fits with previous work demonstrating greater sensitivity of decomposition to temperature in cultivated than in grassland soils (Conant et al., 2008).
In contrast to our expectations, however, eCO 2 and warming effects on soil C did not follow from their effects on plant biomass. This might be explained by smaller plant biomass responses to eCO 2 and warming belowground than aboveground ( Figure S2) or insufficient time for aboveground plant biomass to be incorporated into the soil.
It is also possible, therefore, that increases in plant biomass with eCO 2 would eventually be reflected in soil C. Carbon loss is typical for grassland soils following disturbance, and post-disturbance recovery of soil C is important for mitigating climate change (Conant et al., 2001;McLauchlan et al., 2006). Our results suggest that warming may slow the short-term recovery of soil C in disturbed environments, even in the presence of high aboveground biomass production.
Disturbed and invaded grasslands are common within the mixed-grass prairie and around the world (Early et al., 2016;Ott et al., 2020;Samson et al., 2004). This study adds to the small set of previous experiments examining how global changes can interact with disturbance or invasion (Bradley et al., 2010;Liu et al., 2017;Perring et al., 2016). Those studies have shown that grazing, clipping, fire, and soil disturbance can sometimes alter global change effects (De Vries et al., 2012;Fernandes et al., 2013;Klein et al., 2004;Newton et al., 2014;Teyssonneyre et al., 2002), and that global changes can facilitate invasion (Liu et al., 2017;Sorte et al., 2013).
In related work from parallel eCO 2 experiments in intact and disturbed, planted grassland, disturbance initially led to high resource availability and aboveground biomass, but did not alter eCO 2 effects on plant biomass and species evenness (Leadley et al., 1999;Niklaus et al., 2001). In contrast, we observed marked differences in degree and direction of global change effects with disturbance and invasion. The results demonstrate that disturbance can substantially alter eCO 2 and warming effects, particularly when it favors invasive species (Davis et al., 2000) that respond positively to these global changes (Liu et al., 2017).
Together, the results of this study highlight the need to account for soil disturbance and associated invasion when predicting climate change effects on productivity, diversity, and C cycling (Oliver & Morecroft, 2014;Perring et al., 2016). They also have implications for management of mixed-grass prairie and other semiarid grasslands. It is well known that anthropogenic soil disturbances reduce biological diversity and soil C in grasslands (Davidson & Ackerman, 1993;Hoekstra et al., 2005;Ott et al., 2020;Samson et al., 2004). Our results suggest that eCO 2 and warming will make it more difficult for grasslands to recover from disturbance. In this case, eCO 2 led to dominance by an unpalatable forb and reductions in plant diversity, while warming slowed soil C accumulation. Consequently, limiting soil disturbance and controlling plant invasion, common grassland management objectives, maybe be even more important under future atmospheric and climatic conditions.

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interest to report.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available  (2)