Received: 18 October 2017 | Accepted 20 February 2008 Dok to:111/1365-2745-12900 RESEARCH ARTICLE Forbs, grasses, and grassland fire behaviour Peter D.Wragg® | Troy Mielke | David Tilman Deparmentof Ecology; Evolution, and Behavior, Uiverstyo! Minesota St.Paul, | Abstract Minesots 41. Ingrasslands and savannas, fire regime—frequently a major determinant of woody ns encroachment, herbaceous species composition and diversity, and nutrient cy- Peter. Waa cling~Is influenced by the quantity and characteristics of plant fuel. Laboratory Ema: wragg@umn eds ad studies reveal variation in flammability among herbaceous species, but field ex: Present ress jeriments are needed to assess whether herbaceous species composition mean: Peter. Wrage, Department of Forest e — Resources University of Mimesot, 115, ingfully affects ecosystem-scale fire behaviour. eee Hal 2500 Cevelaa ve. N-St-Padk | 2, In our North American tallgass prairie study system, grasses’ thinner leaves and mes longer leaf retention appeared to create a finer, more aerated, more connected Funding nrormetion fuel bed than forbs. We tested the hypothesis that grasses promote fire spread Nefina ese Foeneton oat area, fre intensity, and associated facets of fire behaviour more strongly than an ‘ia University of Minnesota US National Science Foundation Lang. Term Ecological ‘equivalent mass of forbs, Resear Program, Garter Number: 3. We characterized spring fires over multiple years in 315 annually ignited plots (DEB. 0620659 and DEB-1734162 spanning profound gradients of plant biomass, cover, and grass.forb ratio that re- sulted from species richness and composition treatments, in a 20-year grassland biodiversity experiment. 4. Grasses increased fire spread and associated facets of fire behaviour, compared. with an equivalent biomass or cover of forbs. Grass dominance increased fire spread area—or equivalently increased fire frequency at any given point. For fire to spread through 50% of the 9 mx 9:m plot area required approximately twice as high an abundance of forbs as of grasses, Grass dominance also resulted in fires that advanced faster, were more intense (higher rates of heat release per unit fireline length), caused more damage to plants, and released heat to greater heights. Fire temperature at 50 cm above-ground was about twice as high in plots with only grasses as in plots with the same biomass of forbs. 5. Synthesis. Even within herbaceous ecosystems that may appear homogenously flammable compared with less flammable woody ecosystems, fuel quality—spe- cifically, the proportional abundance of grasses—combines with fuel quantity and ignitions to determine effective fire regime at a given point. In spring burns, grass dominated plots burn more completely and generate higher temperatures, and thus better suppress woody plants and volatilize more nutrients, than forb-domi nated plots (holding all else equal). Handing ator: Hans Coratisen kKeyworDs ‘ecosystem function and services, fire intensity fire spread, flammability, forb, fuel oad, ‘functional composition, grass “Journal of Ecology. 2038;1064983-2001, ‘wileyonlinelibrary.com/journal/jec ‘©2018 The Authors. Journal of Ecology | 1983 {©2018 Beh Ecological Society1904 | surat ology WRAGGera. 1 | INTRODUCTION In grasslands and savannas, the frequency, spread and intensity of fire can be a major determinant of the abundances of woody and herbaceous species (Bond, Woodward, & Midgley, 2005: Cavender: Bares & Reich, 2012). It seems plausible that the species composition ‘of plant communities mayin turn influence the characteristis of fires (Mutch, 1970), and thus might create feedbacks that could ead to a temnate woody-dominated and herbaceous-dominated states (Bond etal, 2005; Staves, Archibald, & Levin, 2011), tant composition can affect fuel quantity and other characteristies. Higher fuel quantities promote fire spread and intensity (Byram, 1959; Cheney, Gould, & Catchpole, 1993; Pausas, Keeley, & Schwik, 2017). Independent of ‘uel quantity, the fraction biomass thats dry, fine, loosely packed, well aerated and connected is positively associated with fire spread area, fire intensity (the rate of heat release per unit irene length) ‘maximum temperature, rate of spread, lame lengthand flaming zone depth (Byram, 1959; Pausas eta, 2017; Schwilk, 2015), Accordingly, finer herbaceous fuels (grasses and forbs) cary fite more readily per unit fuel quantity than coarser woody fuels (trees and shrubs) Within these broad categories, there is evidence, mainly from lab oratory studies, for variation in flammability among woody species (De Magaihaes & Schwik, 2012; Mutch, 1970; Zhao, Cornwell, van Pomeren, van Logtestiin, & Cotnelissen, 20%) and among hetba- ceous species (Simpson et al, 2016) Field experiments though, are needed to assess whether herbaceous species composition mean Ingully affects ecosystem scale fre behaviour, independent of fuel ‘quantity Fernandes & Cruz, 2012; Prior etal, 2017; Schwilk, 2015). ‘The effects of grassland and savanna plant community com position on fire spread and intensity have been difficult to study experimentally in field conditions because local variability In the species compositions of plant communities is rarely at scale suit able for replicated exnerimental burn units. Here, we report on the Cedar Creek biodiversity experiment, created in 1994 with more than 300 Independently burned 9m%9m plots planted with 1, 2,4, 8 16 oF 32 perennial grassland species, with about 20 repli ‘ates of each diversity level having randomly chosen species com positions and an equal number of addtional plots having randomly chosen functional group composition (Tian, Dodd, et aly 1997). ‘Annual above-ground productivity tured out to be an increasing function of species richness (Tilman, Knops, et al, 1997). Thus, in total, the experimental treatments created plots with a wide range in biomass (and fuel) types and amounts that enabled us to examine how plant species composition and plant productivity impact fire behaviour under annual igitions. In this North American tallarass pit, grasses appeared to create finer, etter aerated, more con ‘ected fuel bed than forbs because the grasses have narrower and longer leaves and retain leaves longer; we tested the hypothesis that {grasses promote fre spread ares, fre intensity, and associated fac: ets of fire behaviour more strongy than an equivalent mass of forbs. We proceeded in three steps. In many landscapes, fre extent is ‘most limited by the quantity and charactersties of pant fue, but the frequency and timing of gritions and weather can also influence fire regime (Archibald, Roy, Van Wilgen, & Scholes, 2009), Thus, we frst characterized the influence of fuel vs. ignition and westher on fire behaviour, using the fact that fuel loads in each experimental plot ‘were relatively consistent across years whereas ignition pattem and weather varied (Our second step was to characterize the influence of fuel quan~ tity-measured as herbaceous biomass or cover~on fire behaviow Te quantity of herbaceous biomass available to fuel grassland and savanna surface ites depends on herbaceous productivity and rates ‘of loss to herbivory and decompesition. Herbaceous productivity and loss rates, in tur, depend on external factors including climate, soil parent material, and topography; but also on species richness _and composition. Biodiversity experiments in perennial herbaceous systems, including the one we report here, have shown that biomass increases strongly with increasing planted species richness (Tilman, ‘Naeem, ef a 1997). Blomassis our principal measure of plant abun- dance and fuel quantity, but we also used cover as an equivalent ‘measure during the early years of ths experiment, when cover rarely reached 100%, (Our third and focal step was to test whether herbaceous species composition influenced fire behaviour independent af productivity We focus on grasses vs. forbs (ie, non-grass herbaceous plants, all broad:-eaved aicotyledons inthis experiment) because most grass species have narrower (fine) leaves which should dry and ignite more readily, and longer leaves that may increase fuel connectiv- ity and ignitability (Pausas etal, 2017), compared with leaves andl stems of many forb species, Moreover, grasses tend to retain thelr leaves over winter and form a well-aerated fuel bed for spring fires, compared with forbs, which-at least in our study system-tend to {rop mote of ther eaves in winter, resulting in a iter layer on the soll surface with litle aeration that may reduce fre spread and in- tensity (Schwil, 2015). However, ie spread and intensity tend to be correlated with flame lena, flaming zone depth, and maximum ‘temperature, other facets of fire behaviour can vary more or less Independently depending on the system, including total heat release (Pausas et al, 2017; Schl, 2015), fre residence time (Prior etal, 2017), and reaction intensity (the rate of heat release per unit areal. ‘These other facets have been proposed to influence soi heating and ‘mortality of underground plant structures (Gagnon etal, 2015), but ‘we do not have expectations for how these cifer between grassand Forb Fuels: esidence time and reaction intensity depend on how fast the fire moves and on how deep the flaming zone is (Byram, 1959), and grass effects on spread rate and flaming zane depth could offset ‘each other to an unpredictable extent. Forts are abundant in many grasslands and savannas (Peterson, Reich, & Wrage, 2007), and co-dominate or dominate in some cal- careous grasslands (Willems & van Nieuwstadt, 1996), old fields (Tilman, 1987), tallrass praiie (Fublendorf & Engle, 2004), and ‘montane meadows (Harte & Shaw, 1995). The possibilty that forbs ‘and grasses may differentially impact fte behaviour has received Ite attention but could be of importance. Positive Feedback cy- cles can occur when grasses dlsplace forest by increasing fire fre- ‘quency and intensity, which in turn promotes fre‘tolerant grassesWRAGG era soar ot egy | 1995 ‘over fireintolerant trees (Brooks eta, 2004; D/Antonio& Vitousek, 1992). This feedoack may maintain grassland or savanna as alter- nate stable states in areas climatically suitable for forest (Bond etal, 2005; Staver et al, 2011). Forbs are rarely mentioned inthis context but If grasses create more intense fires they may suppress tree establishment more strongly than forbs, with implications for biome distributions. In addition, different effects of grasses vs. forbs (on fre may influence grassland species composition (lair & Prat, 2013: Gagnon, Harms, Pat, Passmore, & Myers, 2012; Thaxton & Platt, 2006). Finally if grasses result in greater ire spread and higher fire temperatures this could lead to greater volatilization of nitro- en, and potentially of phosphorus and suiphur which volatilize only in hotter fires; this could in turn reduce ecosystem pools of these utrents We tested the hypothesis that grasses promote fire spread, in- tensity and correlated fire measures more strongly than forbs for 2 given abundance. We assessed the influence of herbaceous fuel ‘quantity and quality by measuring fire behaviour in a 20-year stass- land biodiversity experiment, where species richness and com- Paosition treatments created highly replicated gradients in plant abundance and the ratio of grasses to forbs. We measured three complementary suites of fire behaviour variables in replicate plots. (0) Fite spread area the area burned following one ignition) and fire spread distance. The larger the fire spread area, the more frequently any given point within a parce burns (Le. the higher the ‘point fire frequency? as we call it), holding ignition frequency constant. (2) Fie intensity, re temperature at two heights, rate of spread, lame length, flaming zone depth, fire residence time and reaction inten- sity. (3) Fre severity, which we assessed using fire damage to plants. We ask how these facets of fie behaviour are influenced by (D) Fuel (be. the properties ofa plot, principally fuel quantity and character- istics) vs. other factors such as weather or ignition pattern (2) the ‘quantity of herbaceous fuel, ie. total herbaceous biomass or cover, and (3) the relative abundance of grasses vs. forbs. Finally, we tested ether plots with higher fre spread or intensity accumulated sol Natlowerrates. 2 | MATERIALS AND METHODS 2.4 | Experimental design ‘We addressed our questions in the Big Biodiversity experiment ex: periment number £120} at Cedar Creek Ecosystem Science Reserve, Minnesota, USA (Tilman, Dodd, etal, 1997). This experiment was established in 1994 using dominant and common perennial species of tallrass prairie oak savanna, which dominated this egion before European settlement. ‘A grid was established of 342 plots, each 99 m, separated by ‘mown aisles atleast 4 m wide. OF these plats, 154 core plots were randomly placed across this full grid and were assigned random Graws of 1, 24,8, oF 16 species from a pool of 18 species (Tilman etal, 2001) and maintained by weeding through 2014. The spe- cies poo! had four species each of C4 grasses, C3 grasses, legumes, and nor-legume forbs, as wel as two woody species (Quercus sop Taman etal, 2001). Additional plots were planted with 1,2, 4, 8, 16 ‘0 32 species from an expanded species pool containing an ad- tional four species of each of the four herbaceous groups (Tilman, Knops. etal, 1997): forthe years in which they were also sampled (until 2000), we analysed these plots combined with the core plots {ota total of up to 315 plots, We refer to non-srasses as forbs be- cause woody plants were removed in 2010 before collecting most of the data reported here and were very rare during the earlier fre ‘measurements (1996-1999, when their mean cover was 0.04% and their maximum caver was 239. ‘The experiment was estabished in an old field where organic ‘matter had been depleted by cultivation. Moreover, the year before the experiment was established, the top 6-8 cm of soll was removed by bulldozer to reduce the seedbank and the soil was ploughed (ounara & Tian, 2008) 22 | Fire conditions Fire scar records indicate that fire was approximately annul before European settlement (Tian et al, 2000). Most fires in the past 200 years in a nearby oak savanna occurred when trees were dor- ‘man, outside ofthe late summer lightning peak, suggesting that ig- nitions by people were important even before European settlement (Wott, 2004) ‘The experiment was burned every spring, as soon as possible after snow-melt and while vegetation was stil dormant; burn dates ‘ranged from 25 March to 28 April in the years that fre behaviour was measured, 1997-2000 and 2010-2014, Until 2006, tive experiment was burned as one block by frst es- {ablishing biaclines along the downwind edges and then igniting hheadfires in the upwind rove of plots and allowing fire to jump from plot to plot, Buring the whole experiment took about 2. From 2007 onwards, each plat was ignited independently along its entire upwind edge using a driptorch: rows of plots were ignited sequen tially, starting with the dover- wind row and working up-wing, which took about 4 hr. The spatial positions of pots did not sinificantiy Influence the resuits. There was no evident spatial pattern in the fire behaviour measures when mapped nor plotted against their x ‘andy co-ordinates. Nor was there significant spatial autocorrelation Jn the residuals from our multiple regressions of fre behaviour re- sponse measures against grass and forb biomass (Mantel tess for correlation between differences in residuals and distances in space between plots: p>.15, mostly much large). Moreover, the species richness and composition treatments were randomly allocated to plots, so any correlation among plots in fre behaviour would not lead to spurious associations with treatments. "Mean air temperatures, recorded hourly and averaged over the uration ofeach bur, ranged from 8.5 to 18.4°C in 1997-1999, and (0.4-24.3°C in 2010-2014, Mean relative humidity ranged from 21% 032% in 2010-2014; mean wind speeds ranged from 2.7 t04.3 n/s in 1997-1999, and 1.5-39 m/sin 2010-2014, We mostly present av rages for these two sets of years, but we also present yearspecific1906 | seul eoloey WRAGGera. data for 2000, 2011 and 2013: in these burs, respective mean air ‘temperatures were 147, 18.0 and 243°C, and mean wind speeds were 19, 2.9 and 2.8 m/s. Mean relative humidity in 2011 was 32%, ‘and in 2013 was about 34%, Weather was measured about 1 km {rom the experiment (Cedar Creek dataset F080), except for humid ity in 2013 (measured 8 kmaway) 2.3 | Fire behaviour ‘We characterized fire behaviour in two broad ways (1) the areal and linear extent of fire spread; and (2) the characteristics ofthe fire's behaviour within the area that it burned. Fist, we visually estimated fie spread area as the percent of each plot that burned following one ignition in 2010, 2011, 2013 and 2014. Each year, plots that had not burned completely follow ing one ignition were re-ignited: this modesty increased the pro- portion burned in some plots, though many remained incompletely burned, Fire spread area was estimated following any re-ignitions leach year 1997-1999 and 2010-2014. (.. in 2020, 2011, 2013 and 2014 fire spread area was estimated both following one ignition and again after any re-ignitions) We supplemented areal fire spread with a measure of linear fire spread. Each year 2010-2014, we inferred ‘rom post-burn photographs whether fire had spread a distance of, atleast 7 m following one ignition, from each plot's upwind ignited edge toa line 7 m downwind, Video of fire spread through subsets of plots (23 plots in 2010, 13 in 2014, and 69 in 2013) confirmed photographic inferences. Second, we measured fie temperature as our main metic of fire behaviour within burned areas. Fire temperature was estimated sing metal tags—pyrometers—with Omega Lag paints (Omega Engineering, Stamford, CT, USA) of varying melting points, placed in each plot before burning (Wally, Menges, & Weekley, 2006). In 2011, 13 paints with melting points spanning 79-788°C were spot ‘ed on copper plant tags (National Band and Metal, Newport, KY, USA, covered by 2 second tag. and placed at ground level and on stakes 50 em above-groundat three cornersofa5 x 5 msquare cen {red in each ofthe 154 core plots, After the bur, we estimated the proportion ofa circle with 20 cm radius around each pyrometer that cated fire, We used identical tags in a subset of 100 plots in 2010 ‘and 90 plots in 2014, We verified the paints’ rated melting points by placing sample tagsin a calibrated muffle fumace for 1 min at in- creasing temperatures and seoved whether paints had melted using a dissecting microscope and reference images. Tags were assigned ‘the melting temperature ofthe highest meted paint; when no paints ere melted, they were assigned 20°C to approximate ambient t= perature. We calculated the median fre temperature at each height foreach pt in each year. We also measured flame height and angle, flaming zone depth {the distance from the flame front to the back of the flame zone), and the time requited forthe flame front to advance Sm, all in the central 5 x Smof each plot (hich was marked with stakes of known height that served as dimensional references in digital video record ings). From these measurements, we calculated flame length and forward ate of spread. To calculate fireine intensity rate of enerey release per unit length of fire front, Byram, 1959), we multiplied the forward rate of spread by the fuel load (approximated as biomass per unit area late the previous summer) and the approximate heat of ‘combustion 20 MJ/kg (Williams, Gl, & Moore, 1998). We assumed the heat of combustion to be constant across species compositions because itis very sinilar across herbaceous species (Byram, 1959: kKidnie, 2009). We calculated reaction intensity per unit area by di viding fireine intensity by the flaming zone depth. 24 | Fire severity ‘Asa bolic index of fire severity, we measured fire damage to simi- larly sized individuals of two plant species in a subset of plots that spanned the species richness gradient Fist, we planted and marked 12 seeds of Quercus macrocarpa the dominant tree in bur oak sa- vvanna.tthissite at systematilocations in each of 32 plotsafter the spring 2010 burn and assessed how the spring 2011 burn damaged the 372 resulting seedings. Second, we marked 99 systematically, selected flowering plants ofthe for Lats aspera in 37 plots in fall 2013 and assessed how the spring 2014 burn damaged them. 2.5 | Plant abundance and fuel load (Climate s continental and most growth is during summer from May until August. Cold winters, largely below freezing inhibit decompo~ sition and herbivory so we measured plant cover and biomass nfate summer the previous year to represent the fuel for spring fires. Each late July trom 1996 to 1999, percent cover of each species was es- timated using four 1 m? quadrats in each plot. These estimates indi- ‘cate absolute (not relative) abundance of plants: plants, bare ground and liter together summed to 100%, Simultaneously, dry plant bio- ‘mass, excluding litter from previous years, was measured by clipping strips (Tan, Reich, & Knops, 2006}; this biomass was not sorted to species, and included Quercus spp. tree seedings in addition to herbs, but woody cover was negligible as described above. Woody plants were removed from the experiment in 2010. Each late July ‘rom 2010 to 2014, dry herbaceous biomass was measured using, clip strips and sorted to species and previous years Fite. Litter from previous years was excluded from analyses; it was minimal and in- cluding it did not change results (not shown). To assess how well summer (late July) biomass represented fuel loads the following spring, we clipped biomass (not separat- ing previous-season biomass from older iter) immediately before bbuming in spring (March) 2085 in strips adjacent to those clipped the previous summer in 20 plots ranging from erass-dominated to forb-dominated. Spring herbaceous biomass, that is, actual fuel load, ‘was closely related to herbaceous biomass the previous summer (correlation coetfcient r~.87, p<.001). Moreever, a multiple re- _session of spring biomass vs. blomass the previous summer and its interaction with the proportion grass the previous summer, withthe Intercept set at zero, showed that forb-dominated plots didnot lose significantly more biomass than grass-dominated plots (interactionWRAGG era soul ottcey | 1987 p= 431), This regression estimated that in exclusively forb plots spring biomass {eel load] was 67% of previous summer biomass ‘whereas in exclusively grass plots spring biomass was 79% of previ- us summer biomass (Figure 1). 2.6 | Soil nitrogen Soll N samples were collected in summer 1994 and 2015. In each plot in each year, nine cores to-20 em depth were aggregated by pt, sieved to remove roots, and analysed for total N using methods de- tailed by Foxnara and Tilman (2008), 27 | Analyses We assessed the effects of grasses vs. forbs on measures offre be haviour an severity using multiple regressions with both plant types as independent variables. To perform two-tailed tests for whether the two plant types differedin their effects on fire, we used 10, 000 bootstrap samples each drawn randomly with replacement from the ata rows) to estimate the sampling distibution of the difference between ther coefficients. Fire spread area, a percentage, was logit transformed when used as a dependent variable to linearize its rela- ‘tionship with independent variables (Warton & Hui, 2010) The logit transformation, loglo/1-p) where p isa proportion, is not defined for (064) or p= 1 (100%), so before transforming we added 0.4% ta values below 50% and subtracted 0:1% from values above 50%. Because plots without fuel have zero fie spread and intensty, we fit these regressions through the origin by omitting intercepts. For log-transformed dependent variables, and for logistic regres- sions, omitting an intercept fits the regression through 0 onthe logit scale, or 0.5 on the probability scale, when the independent var- ables are 0; to instead fit the regression through approximately 0.0n the probability sale, we used the logit of 01% as an offset (Geman {Hill 2007). (The logit transormation is undefined for O; our con- clusions were not sensitive to using 0.1% vs. other arbitraly small values) For the fire temperature dependent variables, we fit ines through the ambient temperature when the independent variables ‘were 0 by subtracting the ambient temperature from the dependent variables before iting regressions and adding the ambient tempera- ‘ure to fitted values for plotting. We estimated variance components and mixed-effects regres- sion models using reduced maximum likelihood with the 2 package Ime4 Bates, Maeclr,Bolker, & Walker, 2014, We performed boot- strapping using the x package boot (Canty & Ripley, 2015) and calcu- lated other statistics using standard x functions. 3 | RESULTS 3.1 | Fire behaviour controlled principally by fuel Most of the variance in fre spread area and fire temperature ‘was accounted for by the abundances of the (perennial) plants ibich changed slowly across years, or other similarly stable plot characteristics, according to variance components analyses using random effects regressions (with plot and year as non-nested ran- dom effects and an intercept as a fixed effect). A random effect representing persistent plot effects (Le. variation among plots that Js consistent across all years in an analysis) aecounted for most of the variance in fire spread area in 1997-1999 (79% of variance) and in fire spread area following a sinsle ignition in 2010-2014 (59% of Variance), as well sn fre temperature in 2010-2014 (62% of var- ‘ance at ground level, 61% at 50 em above-ground). Only 2%-6% of the varlance in these measures was attTbuted to a random effect ‘representing variation among years averaging across plots) due to either weather before and during burns or mean fue properties. The ‘emaining 199-358 of vaiance in fre spread area and temperature arose froma combination of measurement error and variation in plot effects between years, whieh could be due to minute-to-ninute variation in burn weather, or year-specife variation among plots in ignition pattern or fuel properties. 3.2 | Fuel quantity, driven by plant species richness, affects fire behaviour Herbaceous biomass increased linearly with increasing log planted species richness treatments (Figure te, Fss3~ 1565, p <.001), a5 previously reported for this experiment (Reich etal, 2012: Tilman, Knops, eta, 1997, 200%). cordingly. re spread area and fie tem- perature increased similarly and significantly (p < O01) with herba- ‘ceousbiomass, cove, and planted species richness repressons with biomass, cover or species richness as the sole independent variable, ‘an estimated intercept, and responses transformed as described in ceptions to Figures and 2). Of these tee independent variables, ‘we focus on biomass as our main independent variable because its closer to the mechanism diving fie spread than is species richness, andbecause we have more data.on biomass than on cover. Firealore- sponded siilorty to effective species ricnesse%, where His Shannon diversity estimated from species relative biomass (not shown), Fire spread through a larger percentage of the area of pots that thad more plant biomass late the preceding summer, after igniting ‘each plot once along its upwind edge (Figure 2a. Pots with more than about 300 g/m of summer biomass consistently burned com- pletely whereas, in plots with less biomass, fire spread area varied from near zer0 to near complete, both averaged across multiple yeas (2010-2014, Figure 2a) and in one representative year (2011, Figure S2a). Increased fire spread area is equivalent to increased ef- fective fire frequency at each point, oF fire return interva: a given point within a plot would be expected to burn every Li{proportion burned) years. Thus, given annual ignitions fire spread area of 20% is equivalent toa point fre frequency of every 5 years: fire spread ‘area of 100% is equivalent to an annual point ire freauency. Fire spread area showed a similar saturating increase with plant cover (Figure 24) as with biomass (Figure 2c, averaging over the years (1997-1999) that both abundance measures were estimated, Consistent with an approximately linear relation between cover and biomass observed during this period (Pearson's correlation1968 | sual Eobogy WRAGGera. @ 2010-2014 ma, I zg a4 Fire spread area (%) 2011 Fire temperature (°C) 338 8 8 8 2010-2013 eg 8 8 Herb biomass (g/m*) 8 3 Planted species richness coefficient ¢=.79) In 1997-1999, only 3-5 years after the expert ‘ment was established from seed in 1994, numerous pts still ad ow biomass (50 i), Most of these low-biomass plots had low fire spread and formed a lower tall of a sigmoid curve consistent FIGURE 4 Among 154 grassland plots seeded to establish a species richness gradient in 1994 and maintained by weeding rnon-planted species and annual burning. plots with higher planted Species richness had fire spread through a larger percentage of the plt area following one ignition (a higher fre temperature ‘at ground level (band higher herbaceous plant biomass (c. Fire spread area (a) was averaged for each plot over 2010, 2011, 2013 and 2014; late summer biomass (c was averaged over available preceding years (2010, 2012, 2013) Fire temperatures are medians of three measurements per plot. Curves are linear fits against log planted species richness; ire spread area was logittransformed tort the curve, Planted species richness treatments jttered for clarity [Colour figure canbe viewed at wileyoninelbrary.coml with fe spread probability increasing at a threshold plant abun dance (inflection point) of about 200 g/* (Figure 2c) or 40% cover (Fleure 24, black curve), though this relationship is noisy without distinguishing grasses from forbs (done below, In 1997-1999, fire spread area was estimated after reigniting incompletely burned plots; fre spread atea may have been modesty overestimated at low cover compared with 2010-2014 when we report fire spread area after igniting each plot once. However, this effect was likely small re‘igition increase fire spread by only 5.4% ofthe plot area, on average, in 2010, 2011, 2013 and 2014 when spread was est ‘mated both before and after any e-gnitions Fire spread through a larger faction of plots’ areas as planted species richness increased from 1 to 16 Figure 1a), atid for bio mass and cover. Averaged over 4 years, fre spread following one: rition in monocultures ranged from less than one-fifth of plot area ‘to.complete,butin 16-species plots was consistently near complete. "Mean ire spread approached completeness even in an average four species plot but became more consistently complete with further increases in species richness. However, fire spread area saturated at 100%, fre temperature increased approximately linearly with increasing total plant biomass (Floure 3a, Figure Sa) and cover (Figure 3b) late the preceding summer, and with log planted species richness (Figure 1b) Plotlevel ‘median fire temperatures were about 300°C at ground level in plots with 400 g/m? of biomass the previous year, near the upper limit of above-ground productivity inthis experiment, in 2014 (Figure 3a). Similarty in 2000 median temperatures of about 300°C were mea suted 10.cm above the ground in plots with 400 g/m? of biomass (Figure $3a) or 2 mean plant cover of about 80% (Figure $30). Every 16-species plot had a higher median fire temperature than any ‘monoculture (Figure 1). Next, we examine how composition of plant fuel influenced fire behaviour independent of total plant abundance. 3.3 | Grass and forb relative abundance affects fire spread Fire spread area following one ignition increased more strongly with grass biomass than with forb blamass in the 154 core pots, in multiple regressions forced through the origin (bootstrap test ofWRAGG era uma ot cology | 1909 FIGURE 2 Insossandbiodvesy eerinent os ted o 2010-2014 Cates ron elie raeston with gen eran ‘oo ‘bundenceaindeperdent vores show et Me spend rea = ladanlpeteuttyercaryugttestn’?moyerpraiced © ap < escent eerste bent na 3 8 sai eases OS pn PS 5 ao 3 Sanat on freeones. 0 The Boek 3 z Stycast reco atlas 7m each ot ova 2 a % in () ae jttered for cart. Here, p-values are bootstrap tests of| 5 8 the equalty of gras and non-arass biomass effects. Black curves g 04, ° areftesvabes emincartcpesionsgansttotibioasser peat Gay apaie 27 aOR pun gcenioes ° Sete aror werent aoa oe a so oe sencesia wrcsler of yrs at evan cra Herb biomass (gi?) tilyened seaprseB Fre spend arn wen aeed ost yorsforcadotisephisin20l0 20 wnentiaenecange 2010-2014 bition Gol ondsispoh B97 Dvnentiecmatee — «S104 ‘been more than one ignition (c, d). All burns were in spring; late & sur cs cer ecu ie eoaig yeas foe senedtcone grecntevewtstlserctionycon) Ey | & Sony culty of the cote of he ram biomass and feb bomoss inspedent vale: pO for 2010-2004 means Fgue2 BO? ‘p= 036 in 2011, Figure $20). The figures plot predicted values from Foo ps.001 ‘these multiple regressions for plots containing only grasses (blue/ Oo é 6. 3th aa 40) 0 ea ey Wer wo Tt Gemeente ot f fire spread was attained a alower total biomass if that biomass was Herbibiomase (Gfm;) entirely sass thn ews etl forbs For example Fre was , _ preced io spcal tush ON aptsarefor apeinatly—_gag 1997-1999 120 g/m? of excusively grass biomass vs. 210m? of exclusively RRO ecard a eer padineeenneneclaienotera | in Figure 2a, and more flexible curves (not shown), indicate that fire = protec’ S0gin?ofgrssionas suite mice posesand fobs ere rede beltemedte Sw Thestioeerinteefetsefpaswsfabatuniancentie Ey | sored are appeared ren stronger rom 19971999 nen oe cies’ relative abundances were assessed using their percent cover in- o 100-200-300 400 500 stead of their biomass and more plots (315] were sampled: a multiple Herb biomass (g/m?) reresion pected fre spread of 50h twit OK ras-onty cover, oF 80% forb-only cover (Figue 24), Re-iting incompletely ‘0 ee bared pts before mewuing hespeadnean i997 199% un Tit have cmtrbted to the tsk measured arma of | © an | grasses: 2010-2014, both absolut and proportional increasesin. 8 Frouimicticgivommlacicninea, 2% lower in grass donated pot Spearman's coven cotfident Bg betwcon proportion gaat and ole or lathe ncemeinfe jah Ate MabsketoennsONeI- as gH maaan | Another measure of a plot's propensity to carry fire is whether it a sev enn ne one ss Moe BB Spend ate, with wich nas gy coated (= 085, 2010~ Hetb cover (%) 2014 means), the probability of carrying fire at least 7 m increased1990 | Jaumalo Ecology @ ype 8 as 838 oe 1504 100 504 Fire temperature (°C) 0 100 200 300 400 500 ® __50.cm above-ground, 2011 1 p= 012 et 300 250] 200] Grass fractior 150 4 1007 © 50-5 Fire temperature (°C) 0 100 200 300 400 500 Herb biomass (g/m?) FIGURE 3. In grassland biodiversity experiment plots, median fire temperature of three pyrometers in each of 154 plots at ground level a) did not increase more strongly wit grass biomass than ‘with forb biomass, whereas temperature of similar pyrometersina subset of 135 plots at 50 cm above-ground (b did, Lines ae fitted ‘values for plots with only grass biomass (bIue/dark grey or only {orb biomass (yellow/light gre), from multiple regressions forced through the origin with grass and for biomass as independent variables. Here. p-values are bootstrap tests ofthe equality of 11258 and forb biomass effects, The relationship between measured ‘temperatures and unadjusted grass an forb biomass (shown by the points and dotted lines, statistics not shown) conflates| fire temperature with whether fire spread around pyrometers. Tisolate the effects of bured biomass an fre temperature, Independent of fie spread, we fit regressions to erass and foro biomass independent variables that had been multiplied by the ‘mean propartion of area burned in a20 em radius around the pyrometers ina plot solid ines and p-values [Colour figure can be ‘viewed at wileyontineliorary.com] more rapidly with grass biomass than with forb biomass, both av raged over 2010-2014 (p <.004, Figure 2b) and in 2011 (p= 001, Figure S2b), The muitiple regression fitted to 2010-2014 means WRAGGera. indicates that a 50% probabiity of carrying fire required about 70 / im? of exclusively grass biomass, or about 270 g/m of exclusively forb biomass (Figure 2). 3.4 | Grass and forb relative abundance affects fire intensity "Multiple regression showed that fire temperature at ground level increased similarly with previous-season grass biomass and forb biomass in 2011 (p = 916 for test of equality of grass and forb coe ficients, Figure 3a), after adjusting biomass for the proportion of the area within 20.cm of pyrometers that burned. (Measured fire temperature would have depended on whether the fuel around Pyrometer burned, as well as how hot it burned: multiplying plot biomass by the mean proportion burned around pyrometers in that plot accounted for whether the fuel burned. to reveal effects on fre temperature independent of effects on fre spread) In contrast, at 50 cm above-ground fire temperature depended strongly on spe ies composition even after adjusting biomass for the proportion burned around the pyrometers (p=.012, Figure 2. Fora biomass oF 300 p/n the fitted temperature was twice as high (ebout 160°C) in plots with only grasses compared with plots with only forbs. Aspects of fire behaviour related to spread rate also increased more strongly with grass than forb biomass and were correlated with fre temperature. In 2013, multiple regression indicated that flame length (Figure 4a) and forward rate of spread (Figure 4b) both Increased more strongly with grass biomass than with forb biomass {0 =.005 and p =,002, respectively). Ata biomass of 300 w/in’, plots with only grass were estimated to have flames 2.8 m long advancing, 0.8 m/s, about three times greater than plots with only forbs. Both flame length and rate of spread were more strongly correlated with fire temperature 50 cm above-ground (Pearson's r= 86 and 64, re spectively than with fre temperature at ground level (r=.70 and r= 85, respectively, across 15 plots with suitable measurements in 2010 or 2014. The difference in lame length between grass- and forb-dominated plots arose from differences in flame height, not angle (not shown) Freine intensity, the rate of heat release per Unit time per unit length of fire front, increased more strongly with sarass biomass than with forb biomass (Figure 4c, p <.001) Using a multiple regression, plots with 300 g/m? of grass-only biomass had Fitted values of approximately 4,500 KW, more than double that of forb-only plots with similar biomass. Feline intensity was cor related similarly with fie temperature at ground level (-=.75) or 50 cm above-ground (r= 76). In contrast, reaction intensity, the rate of heat release per unit time per unit area tended to Inerease slightly more stronely with forb than with grass biomass, though this difference was not sig nificant (p= 157, not shown), This was because the flaming zone ‘extended farther behind the fire front in grass-dominated than foro dominated plots Figure 4d, p< 001):for a biomass of 300 g/m the fitted flame zone depth was about 3.5 m for plots with only grass ‘and about 0.8 m for plots with only forbs. Thus, the greater Inte sity per length of fireline in grass dominated plots was spread over aWRAGG era Jounal taaoey | 1991 FIGURE 4 In 63 grassland biociversity experiment pots, lame length (2) rate of forward spread (b), irene intensity (rate of energy release per length offre font (and flaming zone depth (dlall increased more strongly with grass biomass than with forb biomass. Lines are fitted values for plots with only grass biomass (blue/dark grey) or anly foro biomass (yelow/ight ary) rom le regressions forced through the origin with grass and forb biomass as independent variables. p-values are bootstrap tests of the equalty of grass and forb biomass effects [Colour figure can be Viewed at wleyonlinelibrary.com larger area, compared with forb-domrinated plots of similar biomass. Also in contrast to the other measures, reaction intensity was more tightly correlated with ground level temperature ¢=.65) than with temperature 50 em above-ground (r= 40}. Fire spread area and fire intensity measures were not redundant: all six measures of ire intensity in Figures 3 and 4 were low to mod crate in incompletely burned pots (ie spread area <95%) but were highiy variable in completely burned plots (Figure $4) 3.5 | Grass and forb relative abundance affects fire severity Two plant species, each with pre-marked individuals of consist- ent size, showed a stronger increase in fire damage with increas ing grass biomass than with increasing forb biomass in multiple regressions (both p <.004, Figure 5), Seedling Q. macrocarpa trees in 2011 and adult. aspera forbs in 2014 were fitted as reaching hal of their maximum damage scores in plots with about 200 g/m? of grass biomass, vs. about 300-350 g/m? of forb biomass. Pre- fire plant size varied relatively tle within species (height M + SD: Quercus = 655 £0.22, Liars = 78.2 + 15.2 em), and did not si nificantly influence damage score when added asa predictor to mul tiple regressions (not shown). Fire damage scores were correlated with ground level fre temperature (Spearman's rank correlation rho =0.79-0.89) and fire spread area (tho = 0.67-0.82); these fire behaviour measures predicted fre damage just as wellas plot herba ceous biomass did (nat shown). 3.6 | Fire, plants, and soil nitrogen accumulation Total sol N from 0 to 20cm depth increased over 21 years more stronaly in plots with higher mean fire spread (p <.001, R =.10) or higher mean fire temperature (p<.001, R? =.29), in separate simple linear regressions with percent soilN in 2015 minus percent soil Nin 1994 as the response vatlable and the fire behaviour variables aver aged over all available years (spanning 1997-2014), 4 | DISCUSSION Species richness and composition treatments established profound sradients in plant biomass and cover, and in functional composition, 2013 Flame length (m) p=.005 0 100 200 300 400 500 600 eo) 2013 Spread rate (m/s) 2 p=.002 0 100 200 300 400 500 600, ® 2013 o pen Flaming zone depth (m) 10 200 300 «00 500 600 Herb biomass (g/m?) that were independent enough to reveal that grasses increased fire spread and assoctated facets of fire behaviour, compared with ‘an equivalent abundance of forbs: arass dominance increased fire1952 | saul Eeobogy WRAGGera. © Quercus macrocarpa, 2011 e 34 o 5 3 — 24 5 ® & 8 g a | Q 2" é Ss oO 3 no 0 100 200 300 400 500 Liatris aspera, 2014 e oo Severity score (medi: T1—1—1T—T 100 200 300 400 500 Herb biomass (g/m2) FIGURE 5 Severity offre effects on plants increased more strongly with grass biomass than with fob biomass. measured the preceding ate summer. Severity offre damage to smlarly= Sized marked individuals was scaree: unburned, 1 scarched (darkened), 2 severely damaged [black or stem fallen, 3 consumed: points are medians of 12 1-year old Quercus seedings in each of 22 plats (aor of two to five reproductive Ltrs plants in each ‘of 37 plots b). Curves are Fitted values for plots with only grass biomass (blue/dark grey or ony forts biomass (yellow light grey), {rom mixed effects multiple regressions forced through the origin with grass and forb biomass a independent variables, logit ‘transformed severity score ofeach plant as response, and plot as, a random intercept. p-values are bootstrap tests of the equality of grass and forb biomass effects [Colour figure can be viewed at vileyoolinelibrary.com] spread areaor equivalently increased point fire frequency, or decreased point fire return interval~and resulted in fires that ad vanced faster, were more intense (higher rates of heat release per ‘nit Freline length, caused more damage to plants, and released heat to greater heights, These results are consistent with grasses Forming finer, more aerated, etter connected fuel beds than forbs in this system. This highly replicated vegetation-sale field experi ‘ment complements plant:scale lab flammatlty studies (Fernandes & Cruz, 2012; Schwdk, 2015; Simpson etal, 2016) in providing evi dence that herbaceous plants vary meaningfully in ther influence, per unit abundance, on ecosystem fire behaviour. This experiment ‘also complements a continent scale study linking tel sampling with remotely sensed fire measurements (Prior et al, 2017) that found that grasses increased fire spread compared with more densely packed liter from other plants, 41 | Fuel controls fire behaviour We infer from associations between plant abundance and fire be- haviour thot plant abundance influenced fie behaviour. because fuel ‘quantity and characteristics are known from first principles to infl- ence fire behaviour (Byram, 1959) and no other persistent pt char acteristics (eg, spatial variables) could explain why plots difered profoundly in fie behaviour in ways that were consistent across years, There may have been itl variation in fre spread area and fire temperature betwreen years (2%-6% of total variation) because most burns were conducted under low-humiity and milly windy conditions conducive to fire, after fuels had dried, We infer that tem- porally consistent plant abundances were stronger controls of fie spread than inconsistent plot ignition patterns or weather a the mo ‘ment of burning, because 2) year x plot interactions combined with measurement error contributed only modestly (19%-35%) to varia tion in fre spread area after a single ignition, and 6} re-igniting in completely burned plots increased the percent area burned by only 5.4% on average 4.2 | Herbaceous plant abundance promotes fire spread Our results ae broadly consistent with measurements of fireine in tensity in relation to total fue load from Alcan, South American, ‘and Australian savannas (Govender, Trollope, & Van Wigen, 2006; Trollope, Trollope, Potseter, & Zambatis, 1996; Willams etal, 1998 with measured flame heights in South African savannas (Trollope etal, 1996); and, for our grass-dominated plots, with rates of for ‘ward spread in grass-dominated Aftican savannas (Trollope etal, 1996) and Australian grasslands (Cheney etal, 1993). Nonetheless, ‘there are some factors to consider when interpreting ou ire behav- jour measures. Biomass and cover late the previous summer over- represented fuel loads for spring fires because some growth and decomposition occurred between late summer abundance sampling and spring burning ina sample of 30 plots spring fuel biomass aver _aged 73% of previous-summer peak biomass, Our estimates of rates of heat release are upper bounds because they assume that uel was entirely consumed (Stronach & McNaughton, 1989), However, any overestimation i kely modest because fuel consumption typically appeared fairly complete (P. D. Wragg pers. abs), consistent withWRAGG era seat otteey | 1999 fuel consumption 299% in early spring fires in tallrass prairie else- ‘where (Bragg, 1982}; further, anther method of estimating grass- land firetine intensity using an established relationship with flame length (Alexander & Cruz, 2012) yielded median values 19% larger than those presented. Fires might have been more intense had they ‘eccurredin auturnn before fuel was compressed by winter snowpack and before most forbs dropped their leaves. because these factors Tikety reduced aeration and convective and conductive heat transfer similarly to how cutting grass before burning it reduced fire spread rate by 19% (Cheney et a, 1993). Conversely, severe winter cold cured (killed) herbaceous fuels, enabling them to dry more rapidly (favouring fre in spring than they would have in autumn. ire inten- sity may have been higher in more extensive burns (Byram, 1959), and fie spread would be faster uphil and slower downhill than in this lat experiment ‘The planted species richness treatments influenced fre spread area and temperature indirectly through ther effects on biomass. Planted species richness increased biomass, on average, as reported Previously from this experiment (Reich etal, 2012; Taman, Knops, etal, 1997, 2006) and explained by mechanisms including comple- mentary resource use and positive feedbacks with sil fertility (Reich etal, 2012) a5 well as more diverse mixtures being more fikelytoin- ‘lude the most productive species, which were CA grasses (Tilman, Kops, etal, 1997). Biomass, in tun, influenced fire, Planted species richness may also have weakly increase fie spread area and tem- perature over and aboveits effects on biomass, for example, through Increased probability of including highly flammable species, but the variation in fire behaviour uniquely explained by species richness {and not total biomass) was small ot shown). The wide variation in fire spread within low-diversity plots implies that even lowrdiversity mbtures can achieve maximum fie spreadif they include highly pro- ductive or flammable species. 43 | Why do grass and forb relative abundances affect fire behaviour? Thete are three groups of explanations for differences between {1085 and forb effects on fire. Fist, differences in heat released per Unit of fuel burned, and the completeness with which fuel burned, ate unikely to explain why grasses increased fire intensity more strongly than did forbs per unit biomass. We assumed a constant heat yield per unit biomass because the heat of combustion of var- ‘us grassand forb species hada coefficient of variation in only 5.3% in a Canadian tallgrass prairie (Kidnie, 2009), and 1.2% in a South American grassland (Britton, Dodd, & Weichert, 1976), We also as- sumed complete combustion if (coarse) forbs were less completely consumed by fire than (finer) grasses then intensity would be over- estimated for forbs more than for srasses, making our inference that frasses more strongly increase fireline intensity conservative. Second, biomass (or cover the previous summer may have over- estimated spring fuel loads more for forbs than for grasses; this ely contributed modestly to the more positive effect of grass than farb abundance on fire behaviour. Plant abundance was measured in late July, when cover and biomass were near their maxima, and these abundances are unlikely to be biased because grasses and fotbs did not differ consistently in their phenology (Whittington, Tiiman, Wrage, & Powers, 2015). However, forbs may have had higher decomposition or herbivory. Multsite analyses have found that forbs decompose faster that gramincids (Correll et al, 2008: Reich, Wright, & Lusk, 2007), and Cé grasses—paticularty in tribe ‘Andropegonese, which includes three of our four species~tend to have relatively low decomposition and herbivory rates due to high tissue CIN and lignin ratios Bond, Midgley, & Woodward, 2003: Cornwell etal, 2008; Hobbie, 1992: Taylor, Parkinson, & Parsons, 1989), Conversely, there was strong overlap between prairie grasses and forbs in decomposition-elated traits at this site (Crane etal, 2002), Seedlings of some of the grass species from this experiment tending tobe less palatable to crickets than some ofthe forbs (Burt Smith, Gime, & Tilman, 2003), implying that insect herbivory during ‘autumn could have targeted forbs over grasses; herbivores larger than mice were excluded. However, the net effect was that biomass was reduced from one summer ta the following spring only 12% ‘more forb-dominated than in grass-dominated plots, which can be only a small part of why grasses increased fre spread and intensity about twice as strongly as forbs per unit abundance. ‘The third, apparently most influential, set of factors that can ‘explain why fires in grass-dominated plots tended to spread more ‘rapidly and intensely than fires in forb-dominated plots with sim- ilar plant abundance involves differences in leaf retention, as well 2s leaf and stem morphology. These factors apparently contributed to grasses forming a more aerated, fine, connected fuel bed (P. D. ‘Wragg & T. Mielke pers. obs), though we cannot separate traits! roles because grasses differ from forbs in many ways. Grasses tended to retain thelr senesced leaves aver winter and form a well _2erated fuel bed in the spring, whereas the for leaves n this exper ‘ment tended to drop during winter into a compact, poorly aerated Iitter layer (Figure 6), After lat fll the forb stems formed fuel that was coarser (thicker in dlameter, and for several species somewhat ‘wood? than grass stems. The forb species (all dicotyledons) had shorter and broader leaves than the grass species, perhaps also con- tributing to coarseness, though lea thickness, surface area-yolume ‘aio, specificleafatea, and leaf dry matter content—associated with flammability elsewhere (Lavore & Garnier, 2002; Murray, Hardstaff, & Philips, 2013)—did not differ consistently between grasses and forbs in this experiment (unpubl data; Cadotte, Cavender-Bares, ‘Tiiman, & Oakley, 2009), Grass-dominated fuels may also be better ‘connected than forb-dominated ones at least above the less flam- mabe compact Iter layer, which would also promote fire spread ‘tea and intensity (Brooks et al, 2004: Pausas etal, 2017). The leafless upright forb stems were disconnected, whereas the grasses {formed a connected fuel bed. Spatial arrangement of stems can also infivence fuel connectivity, but grasses and forbs didnot appear to lifer consistently in this (P.D. Wragg pers. bs). The more upward dstsbution of heatin gass- than forb-dominated pots, indicated by higher temperature at 50 cm above-ground (but not at ground level) and longer flame lengths for given total biomass, may have arisen1994 | sual Eeobogy WAG Grass, medium biomass Forb, medium biomass fe) © © Before burn After burn. FIGURE 6 Fire spread more completely, and was more intense, in grass- than forb-dominated plots of similar biomass. Photos of fre spread la-k] were taken of the same set of vee plots at different times: immediately before burs in April 2011 (wide views, ac and April 2015 (close views, df), and immeciately after urns in April 2011 (wide views, g-i and close views, jk. Averaging across annual spring burn in vars with data (1997-1999 and 2010-2014) fire spread through 98% ofthe grass-dominated plot mean late summer biomass across these years 114 g/m’, of which 89% grass-principaly the C, buncharass Schizachyrium scoparum: left column, ad 8. compared with 45% fire spread in a plot dominated by the legume forb Lespedeza capitata (mean biomass 132 g/m of which 92% fort: middle column, be, h, k)and just 14% ire spread in a plot dominated by the Asteraceae forb Lites aspera (mean biomass 143 g/m”. 97% forb:right column, cf). Compared with grasses that formed well aerated and connected fuel beds of fine leaves and stems (forbs formed fuel beds of poorly aerated fallen leaves and disconnected coarse stems () or coarse and weakly connected leaves and stems (f) that burned incompletely (k]. Other plots illustrated that when fire did spread, lames were longer and the flaming zone was deeper in plots ‘dominated by grasses (Mt: mean biomass 188 g/m?, 66% grass, principally Panicum vrgatum) than in plats dominated by forbs (N- mean biomass 180 g/m, 91% forb, princially the legume Dalea purpurea; and P: mean biomass 143 g/m? of which 97% for, principally L. aspera, same plot asc fA hisher biomass plot Mlustrates a more intense fire (L: mean biomass 366 g/mn?, 56% grass and 44% forb), Photos \-0) were taken in April 2010 [Colour figure can be viewed at wileyonlineibray.com)WRAGG era soar ot egy | 1995 through the same processes that lead to faster spread, because these ‘acets of fite tend to be correlated (Byram, 195; Schwik, 2015); there was ne consistent difference in ive height between grass and ‘orb species inthis experiment (Cadotte et al, 2009), though greater leaf drop by forbs may have resulted in a lower fuel height distribu- tion for forbs atthe time of burning. Our results are directly relevant to prescribed burs conducted during the spring. Grasses and forbs Tkety afer less in thei effects on lightning fires during late summer, before forbs have dropped their leaves nto a compact iter bed, and likely differ to an intermediate degree (more than in late summer, less than in spring) in their effects on prescribed fall burns, when forbs have dropped some leaves. However, fire spread and severity and several facets of fre ine tensity intensity, flame length, flame zone depth, temperature at 5D.em above-ground) were comrelated and greater in grass- than forb-dominated plots, other measuresfie temperature at ground level, residence time, and rate of heat release per unit area—were independent ofthe spread-associated variables and not associated ‘with grass vs. forb abundance, because increased fieline intensity in .r2s5-dominated plots was offset bya deeper flaming zone and more ‘upward heat distribution, Both the correlation of the fire spread- associated variables and the independence of the other variables are consistent with other studies (Pausas etal, 2017; Schull, 2015). Below-ground damage and mortality of resprouters such as those In this study, a8 well as heat scaification of legume seeds, may be more closely associated with measures such as fie residence time, ‘which may better predict sol heating, than with spread-associated measures (Bradstock & Auld, 1995; Gagnon eta, 2015}. In that case, mortality and scarification may be independent of grassfor ratio, as fire residence time is, Too few marked Liatrs and Quercus invid- uals (254 failed to resprout following our burns to analyse correlates cof mort 44 | Alower fire spread percolation threshold for grass-dominated plots The sigmoidal increase in proportional fire spread area with increas- ing plant cover (Figure 2c) s consistent with fire spread as a perco- lation process, and grasses resulted in fire spread at a much lower apparent percolation threshold than forbs. Percolation processes exhibit thresholds in occupancy in this case, the fraction of the landscape that has flamable fuel that ignites readily under the con- tions of the buat which the probability offre spreading from ‘one patch to at least one other patch rapidly increases from zero to one (Cox & Durtett, 1988}. The apparent threshold in our data was about 40% total plant cover respective of composition, or 30% for exclusively grass plots (Figure 2), suggesting that the threshold is lower for grasses than for forbs, Indeed, the fitted line for forb-only Plots suggested a much higher threshold of about 80% cover. The lower cover threshold for percolation in erass-dominated than frb- dominated plots may be due to a higher probability of fie spread- ing between grasses than forbs because grasses are finer fue see above) ur results are consistent with invasive grasses radically inereas- Ing fire spread area in arid shrublands in the Mediterranean (Grigulis tal, 2008) and the North American Great Basin Balch, Bradley, DAAntonio, & Gémez-Dans, 2013). Invasive grasses are understood {tohave made the landscape more connected in terms of fire spread. (Our results suggest this could be due to either an increase in total fuel above a percolation threshold or a lower percolation threshold with mare flammable ful. Steep declines in bum area as tree cover increases above 40% Archibald etal, 200%, presumed to correspond to flammable 5728s cover falling below 60%, have been interpreted as consistent with a theoretical percolation threshold of 59% (Abades, Gaxiola, & Marquet, 2014; Cox & Durrett, 1988; Schertzer, Staver, & Levin, 2014). The apparent threshold in our grass-dominated plots (-30%, Figure 2c) may have been lower than the theoretical 59% for at least three reasons. Fist, fite burning with wind Is biased perco- lation, which may have a lower threshold for spread (Ohtsuki & Keyes, 1986), Second, the theoretical threshold is for well mixed landscapes; clumping of flammable patches allows fire to spread at Tower cover (O'Neil, Gardner, Turner, & Romme, 1992). Third, our estimates of percent cover may have been at 3 finer seale than the scale offre spread. Models in which fire can jump over unoccupied patches predict lower percolation thresholds and are equivalent to coarser resolution models (Q Nell, Mle, Turner, & Gardner, 986). Cover was estimated with a grain size as fine as 1 em#, Except in the sparsest plots, much of the estimated bare ground was composed ‘of numerous small bare patches interspersed between plant matter, ‘key jumped readily by fre (P. D. Wragg pers obs). Therefore, our ‘measures of plant cover would correspond to higher cover values at ‘ coarser resolution that accounts forthe ability of fire to jump over small gap, resulting in a threshold estimate closer tothe theoretical value. ‘We focus on howe grasses vs forbs influence fire behaviour hold ing abundance constant, but C4 grasses can also promote fire by in- ‘creasing total plant abundance and fuel load. In this experiment, the presence of CA grasses in a plot strongly increased herbaceous pro- ductivity, apparently because ther lower tissue N allowed them to roduce more biomass per unit area than C3 plants (Tiiman, Knops, cet al, 1997); similarly, C4 pants can have higher water use efficiency (Ehieringer & Monson, 1993). C4 grasses’ ow N and high lignin con- tents can also increase fue loads by inhibiting decomposition and herbivory (Ehleringer & Monson, 1993; Masubelele, Bond, & Stock, 2007), Thus in addition to having lower fre percolation thresholds duc to grasses’ high flammability per unit abundance, grasslands dominated by C4 grasses may also be move likely to exceed their percolation threshold—and thus to burn—because CA grasses can increase fuelload 45 | Forbs, tree-grass interactions, and biome shifts ‘The hypothesis that feedbacks between vegetation and fire can ead to grassland, savanna or forest as alternate stable states hinges on grasses promoting and being promoted by fie, and trees suppressing1996 | Jeumalo Ecology WRAGGera. and being suppressed by fire Abades etal, 2014; Bond et a, 2005; Grimm, 1983; Schertzer etal, 2014; Staver & Levin, 2012; Staver etal, 2011), Forbs are missing from these treatments, implying that ‘they are assumed to be (I) negligible in either abundance or per abundance impact, or (2) equivalent to grasses. Forbs may indeed bbe negligible in abundance in many tropical savannas and other ecosystems. However, our results suggest that forbs have substan {tal per-abundance impacts on fire behaviour and that these citer, from the effects of grasses. Where forbs ae abundant, considering them separately from grasses could improve understanding of fe behaviour, tree demography, and fire-mediated biome shifts in two ‘ways, First, fora given herbaceous biomass, rass-dominated fuels produced more extensive fre that were more intense and released heat higher above-ground and were thus more ely to top-kll trees by damaging their aerial buds or other tissues (Trollope, Trollope, & Hartnett, 2002) Increased fire spread with Inereased biomass and cover, especialy of grasses, profoundly inhibited the establish ment of oak trees (Q. macrocarpa and Quercus elipsoidal) inthis experiment (Wragg, 2015} fie held trees ina "fire trap” (Bond & van igen, 1996) by repeatedly top-klling them before they could row large enough for their stems to survive fre. Second, our results suggest the percolation threshold for fire spread, may be steeper and occur at lower herbaceous cover for grass-than forb-dominated systems (Figure 2d, To the extent that percolation processes con tribute to the suddenness of biome shifts (Schertzer et al, 2044), thisimplies that shifts between grassiand or savanna and forest may bbe moce sudden and occur at lowes herb abundances fn grass- than forb-dominated systems. [A stronger positive influence of grasses than of forbs on fire spread and tree suppression could help explain why acid, sandy soils that Favour grasses over forbs tend tobe grasstand or savanna whereas more basic, day-rich soils that favour forbs over grasses tend to be forested, in some regions, Sol texture only weakly pre- cts tree cover globally (Lehmann et al, 2014; Sankaran eta, 2005), perhaps because its effects can reverse depending on context. For ‘example, high lay content may be associated with high tee cover (forest) in South America (Rugsero, Bata, Pivllo, Meirelles, 2002), yet high clay content may be associated with low tree cover in some Australian and Aftican savannas where clay suppresses ‘trees through water-logging or high grass productivity and fire in tensity (Bond, 2008; Wiliams, Duff, Bowman, & Cook, 1996) High clay content may favour trees through high water holding capacity and nuttient availability that increase tree growth rate and thus the Ticettnood that trees avoid top kil by tre (Murphy & Bowman, 2012). (Our findings raise another possibility. Ia European grasslands and perhaps other regions, sols with high lay content are often also rch in cations and relatively basic, and catiorich and basic sols pro ‘mote forb dominance ina least some regions Tilman, 1982; Tilman etal 1994; Willems & van Niewwstadt, 1996). Thus, f basic sols rich in clay promote forbs over grasses, and if this reduces fre intensity, {his could help explain why such sol are more key tobe forested than sandler, more acid soil. Similarly, N deposition favours grasses ‘over forbs in European and other grasslands (Stevens etal, 2009; io, Ni, & Wan, 2009: Xia & Wan, 2008), which could increase flam- ability and thus favour grassy biomes over forest. 46 |_ Fire-mediated interactions between plant functional groups Grasses and forbs may influence neighbouring vegetation through their profoundly different per-abundance effects on fire behav- Tour, in addition to other mechanisms such as resource consumo~ tion, How might differences in fire behaviour between grasses ‘and forbs affect different plant functional groups in perennial- dominated grasslands and savannas? Woody plants tend to be Inhibited by grass-driven increases in fie, because fire that de~ stroys above-ground tissue requires small woody plants to regrow from the base (if not Killed) instead of compounding previous- season growth, Thus, in addition ta inhibiting tree establishment (discussed above), grass-driven fire may inhibit woody subshrubs (such as Amorpha canescens inthis experiment) that are commonly considered forbs because they never grow taller than the grass layer yet can grow from woody previous-season stems. Within ‘ruc herbs, howeves sponse to fire, Speces' responses to fire are understood mostly at the level of qualitative traits, such as seeding vs. re-sprouting after fire, which classify all the herb species inthis experiment as “pet- ennial resprouting herbs" (Lavorel & Garner, 2002), Accordingly, srasses and forbs do not appear to differ consistently in their re- sponse to fie. For example, in some tallgrass praiies and mesic srasslands, forb and grass abundance responded simiary to fire frequency (Engle & Bidwel, 2001; Kirkman et a, 2014). I grasses and forbs differ in their effects on fire but not in their response to fire, that would accord with synthetic findings that traits that determine plants’ impacts on fire are largely different than, and not necessarily correlated with, tats that determine ther response to fire (Lavorel & Garnier, 2002) However, high fire frequency did promote grasses over forbs nat least one tallgass prairie, Konza Prairie (Colin, 1987; Gibson Hulbert, 1987). In particular, CA grasses may respond more pos: Invely to high fire frequency than mostly C3 forbs and C3 grasses (Forrestel, Donoghue, & Smith, 2014: Li, Zuo, & Knops, 2013}, for several reasons, First, C4 grass growth tends to be more concen: ‘uated inthe middle ofthe summer growing season than growth of C3 forbs and C2 grasses, so C4 grasses are less lily to be green and vulnerable during spring and autumn fires. Second, many CA arasses are intolerant of iter that builds up whe fireis infrequent (Knapp & Seastedt, 1986). Third, C4 grosses’ lower tissue N’con- centrations may give them a competitive advantage (Tilman, 1920) where high fire frequency depletes N. If grasses promote fire (rel ative to forbs), C4 grasses respond positively to fire (relative to C3 forbs, especially subshrubs), and C3 grasses are uncommon (as is ‘the case in many subtropical and tropical grasslands), then positive feedbacks between CA grass abundance and fire could establish C4 grass-dominated and forb-dominated patches as alternative stable states there ate not such clearcut differences in re-WRAGG era seat otteey | 1997 427 | Im) nutrient losses ions of grass vs. forb effects on fire for Nitrogen (N), phosphorus (P), and sulphur (5) were more likely to be volatilize by fire and lost in plots dominated by grasses than in plots dominated by forbs with equivalent biomass. Nitrogen was likely volatilized at a higher rate in grass-dominated plots due to both higher fire spread and higher canopy fire temperatures. First, higher fire spread in grass-dominated plots would certainly have caused higher N volatilization because fire temperatures near the ground were high enough to volatilize N (above 70°C; Knops, Bradley, & Wedin, 2002) in all but the barest plats (65 g/m, Figure 3a and Figure S33; >15% cover, Figure S3b) Second, within ateas that burned, grasses likely promoted nutrient volatilization more strongly than did forbs through higher canopy fire tem= peratures: temperatures high enough to volatilize N at SO em above ground required only 100 g/m of grass biomass but more than 200 g/m of forb biomass. Temperatures were also more Tkely high enough to volatilize P (at 281°C) and S (at 444°C) in grass-dominated plots (Haynes, Bruno, & Lide, 2014: Kauffman, Cummings, & Ward, 1994). Plot median temperatures ranged up to approximately 300°C at ground level in 2011), 400°C at 10 em above-ground {in 2000), and 250°C at SO cm above-ground (in 2011}; our highest pyrometer measurement was 610°C. These measures are comparable with results from similar pyrometers in Kansas tallgrass prairie (Gibson, Hartnett, & Merl, 1990). However, higher temperatures may have occurred—making vola- tilization of P and $ more fikelybecause pyrometer estimates of ‘maximum fire temperature are conservative compared with maxi- mum (Iverson etal, 2004; Kennard, Outealt, Jones, & O'Brien, 2005) or peak t-min mean (Wally etal, 2006) temperatures from continuously logged thermocouples, and actual flame tempera- tures can be even higher than thermocouple measures (Martin, Cushwa, & Miller, 1969). Why did total sol N increase with increasing fire spread (corre- sponding to higher point fre frequency) and increasing fire tem- perature in this experiment, in contrast with multi-decade fire ‘manipulations in which high fire frequency typically reduced total Nin surface sol (Pellegrini etal, 20177? In this experiment, plant abundance and species composition strongly determined both ef- fective fire frequency (this paper and soil N accumulation (Fornara ‘5 Tilman, 2008), Soll Naccurulates in proportion to organi matter inputs from root biomass in early-successional, infertile ol fields lke this one (Knops & Tilman, 2000; Li, Knops, Zuo, & Laungani, 20U), and soil N increased especially strongly with the abun dance of CA grasses in this experiment (Fornara & Tikman, 2008). Relatively litle N was volatilzed by fire from shoot and litter bio- ‘mass, compared with the much arger amount of N in unburned root biomass: roots constituted approximately 84% of total plant bio- ‘mass inthis experiment in 2006 (Fomnara & Tilman, 2008). Thus plant abundance and species composition confounded the effect of fire spread on soll N. This confounding may be stronger here than in experiments that directly manipulate fire frequency. It may also be stronger in early-successional, infertile old fields ike this ‘experiment (in which mean soil Nincreased from 0,045%in 1994 to 0.059% in 2015) than later in succession, Early in succession, when plant-and especialy C4 grass-abundance powerfully promotes soll N accumulation via root biomass) as well as fire frequency via above-ground biomass) ire-driven N losses may not detectably af {ect total sil N. Accordingly in another infertile od field at Cedar Creek undergoing natural succession and rapidly accumulating soil IN. manipulating fire frequency over 27 years id not alter total sol N(Lietal, 2014). In contrast, studies that found that frequent fire reduced soil Nat Cedar Creek (Kops, Ritchie, & Tilman, 2000; Nomis & Reich, 2009; Reich, Peterson, Wedin, & Wrage, 2001) and often elsewhere (Pellegrini etal, 2017)—were in unploughed ¢erasslands and savannas with sll N levels closer to steady state. In ‘unploughed prairies, if anthropogenic N deposition increases the ratio of grasses to forbs (Stevens etal, 2009; Xia & Wan, 2008: Xia etal, 2009) and thus increases fire spread and nutrient vola- Itization, this may reduce sol N availabilty (and increase surface light] and mitigate species loss due to eutrophication (MeLauchlan, Crane, Nipert, & Ocheltree, 2014), 5 | CONCLUSIONS Fire spread area, point fire frequency, fre intensity and associated facets of fire behaviour increased more strongly with grass than with forb abundance, holding total fuel quantity constant, in this long-term, replicated field experiment with annual spring initions ‘and randomized species richness and composition fand hence fuel ‘quantity and quality) treatments, This highlights that, even among herbaceous species uel quality isa major determinant of effective fire regime. Moreover, these results imply that forbs and grasses may play distinc roles in fie-mediated interactions between plants, Including interactions between herbaceous and woody vegetation that influence biome boundaries, and have distinct effects on nutt- {ent loss through volailization. ‘Widespread decreases in grassland and savanna fire intensity resulting. from changing ignition regimes, fie suppression, and land management (Archibald, Lehmann, Gémez-Dans,& Bradstock, 2013), may increase the importance of the distinctly higher flam- ‘mabilty of grasses and the impacts of this higher flammability on ecosystem dynamics. ACKNOWLEDGEMENTS We thank Joe Farglone, Jane Cowles, Mara Sagedahl, Christine Carroll, Laura Jaskéewicz and others for help with fie temperature measurements; Adam Clark for statistical advice: Kally Worm, Jim Krueger, and numerausinters for help with burning; Dan Bahauiin for help with archived data: Jane Cowles, Adam Clark, Alex Reich ‘and Carla Staver for discussions; and Jeannine Cavender-Bates ric Seabloom, Sarah Hobbie, Melissa Pastore and two anonymous re- viewers for comments. PDW. was supported by 2 Scholarship for1956 | surat ology WRAGGera. Doctoral Study Abroad from the National Research Foundation (South Africa) and by Graduate School, Doctoral Dissertation, and Crosby Fellowships from the University of Minnesota, This work was supported by grants from the US National Science Foundation Long: “Term Ecological Research Program (LTER) including DEB-0620852 and DEB4234162, The Cedar Creek Ecosystem Science Reserve and the University of Minnesota provided further support. The au: ‘thors declare no contlc of intrest. AUTHORS! CONTRIBUTIONS, P.DW, TM. and DIT, conceived the ideas and designed methodol ‘oy: BD.W. and TM. collected the data; PDLW. and DT. analysed the data; PWV. led the writing of the manuscript Al authors contrib ‘uted critically tothe drafts and gave final approval for publication. DATA ACCESSIBILITY The data are available through both the Cedar Creek Ecosystem Science Reserve data catalog (htps//cedarcreek.umn.edu/research/ data} and the LTER Network Data Portal (htin“iportaliter netedu/). Weather: https://doi.org/10.6073/pasta/actb5f9%4cb £838860adt30545409413 (Seeley, 2018); plot coordinates: https// doi.o1g/10.6073/pasta/940934edb437799Lba2ca09961072¢ (man, 20186; fire behaviour: https//dol.org/10.6073/pas talcedfaTibbest5979e84cate15(9c8353 (Wragg, 20182); fre tem peratures: htps://doLorg/10.5073/pasta/c7866d0297068Seee6e3 didb(7743529 (Wrage, 2018d); fire severity: htipsi/doi org/0. 6073/pasta/OPtdbccact abt 7 f6eaSBcteéaavcB1 (Wragg, 2018) plantcoversttps:/doiorg/10.6073/pasta/ae57ebe9d074dS142044 22eeec37bAA? (Tan, 2048; plant biomass: https:/dol org/10. 607 Aipasta/eecOQHt59taSdeTAMb27eSt 7ecfeSd (Tilman, 20:83) soring fue! loads htips:/4do.ore/10,6073/pasta/037b3d380aIc641 .d587462aca0aba86e (Wragg, 2018b}; ol nitrogen: https org 10.6072/pasta/e2vc093034540666cde2ed040811fe (Tian, 20184, orciD Peter 0. Wrage ® hit%//orcd.org/0000-0003-2361-4286 REFERENCES ‘Abades, S. R. Gavola, A. & Marquet, P. A. (2014). Fir, percole tion thresholds and the savanna forest transition: A. neutral ‘model approach, Journal of Ecology, 102, 1986-139. nto org/0.1111/1365-2745.2521 Alwrandet, M. £6 Cruz, M. G. (2012) nterdependencles between ‘fame length and fetne intensity in preaicting crown fre intation and crown scorch height. International loumal of Widlond Fir, 23, 95-113, htips/fdolorg/10.1071/WWFLI001 ‘Archibald, 5, Lehmann, C. ER. Gomer-Dans, J. L, & Bradstock, R ‘A. (2013). Dering pytomes and global syndromes of tive te ines, Proceedings ofthe National Academy of Sciences ofthe United States of America, 110, 6442-8447. ttps:/fdolerg/101073/ pnasa2it466t10 -AhIbaI, Sx Roy, D. Px Van Wen, B. Wh & Scholes, RJ. (2007) \Winat limits tre? An examination of erivess of but area In Southern Africa. Global Change Biology, 15, 613-650. https//dot ‘org/10.1111/.1365-2486,2008.01754.« Bale, J. Bradley, B.A, D/Antonio, CM, & Gomez-Dans 3.2013). Introduced annal grass Ineeases rglona fre activity seroes the afd western USA (1980-2009). Giobal Change Boiogy, 19, 173-183. htpsufdoiorgyt0ani1fgco.z2046 Bates, D, Maechler, M., Bole, B, & Walker, 5. (2014) iting Finesr mixed-effects models using med. AvXivexpnts, 2806 5823, Bond, W. J (2006). Wat limits trees in C, graslands ana savannas? ‘Annual Review of Ecology, Evolution, and Systematics, 9, 641-659, https: fdoLorg/10146/annurevecolsys.39.10707473414 Bond, W. J. Midgley. G. F. & Woodward, F. | (2003). What contrats ‘South Aican vegetation - Climate oF tte? South Aan Journal ot Botany, 69, 79-71 https: oi org/10.1016/S0254-6299115]30362-8 Bond, W. J, & van Wilgen, 8. W. (1996 Fe and plants. London, UK: Chapman & Hal tps://dorg/10:10071978-74-009-1499-5 Bond W.J, Woodward FI & Midgley, GF (2005). The global distribution of ecosystms na worl without fre. New Phytlogt, 165, 525-538, Bradstock, R.A. & Auld. T. D. (1995) Sol temperatures during ee perimental bushfires in relation to fre intensity: Consequences for lesume germination and Tire management in south-eastern ‘Australia, Journal of Applied Ecology, 22, 76-84, https. org/10.2307/2404417 Bragg, T. B. (1982). Seasonal variations in Fuel and fuel consump tion by fires In a bluestem praise. Ecology, 63, 7-11. htlps//eol org/10.2307/1957024 Britton, C.M., Dodd, J.D Weichert, AT {1974 Energy values of plant “species and iter of an Andropogon-Paspalum grassland Jounal ot Biogeography 3 389-395. htips://do.rg/10.2307/3037982 Brooks, M. L, D'Antoni, CM. Richardson, D. M. Grace J.B, Kooy, 4. , Dtomaso, JM, Pyke, D. (2004). EFects of invasive allen Plants on fe regimes.BoScenes, 54, 677-688. hitps/olorg/10. 641/0006-3568(2004105410677-E01APO}20.C02 BurtSwith, GS, Grime, 1. P, & Timon, O. (2003), Seedling vest {ance to nerivory 36a predictor of relative abundance in 2 syn thessed prairie community. Oikos, 101, 345-353, https/éo org/10.1034/1600-0706.2008.11052. Byram, G. M. (1959. Combustion of forest fuels. In K.P. Davis (Ea). Forest fe: Control and use gp. 61-89) New York, NY MeGraw- Hl CCadotte, M. W, Cavendr- Bares J, Tan, Ci, & Oakey, TH. (2009) ‘Using phylogenetic, functional and tat diversity to understand pat {ears of plant community productivity. PLOS ONE, , 5695. Mts// {o.018/10.1371/journa pone 0005695 Canty, A & Ripley, B. 2015) boo: Bootstrap R(S Plus functlons.R pack ‘ageversion 13.45, Cavender-Bares, J, & Reich, BB. (2012) Shocks to the system: ‘Comunity assembly ofthe oak savanna n'a 40-year re reaqvency experiment. Fzoogy 93, 52-69, ntps://doLorg/10.1890/11-0502.1 Cheney, N. Gould, J, & Catchpole, W. (1993). The influence of fuel, weather and fire shape variables on Frespread in grass: lands. International Joumal of Wieland Fre, 3, 31-44. https://aol org/10.1071/WF9950033, Colfns, S. L. (1987). interaction of isturbances in tllrass pra Ties A fleld experiment. Ecology, 68, 1243-1250. mtpsvfoot org/10.230771939208, Comma, W.K, Comelssen, J. H.C, Amatangelo, K, Dotrepaal, E, vine, V. Ts Godoy, Om Westoby, M, (2008), Pant species tats are the predominant contol on fitter decomposition rates within Dlomes worldwide. Fenlgy Lette, 11, 1065-1071. https//éol oxg/t0.1181/)1461-0248.2008.01219% Co, J. To & Durtett,R. (1988), Limit theorems for the spread of epi denies ad forest ire. Stochastic Processes aetheirAppications 30, 71-191 htps/fdoLorg/10.1016/0804.4149(88)90083 xWRAGG era Journal ot ology | 1999 rane J. Tian, D, Wea, D. Releh, Py THoeker, M. & Kops (2002) Functional tras, productivity and effects on nitrogen cycing (0 99 grassland species. Functional Ecology, £6, 562-574. psd 01g/10.1046/,1365-2435,2002.00860.x DiAntonio, CM, & Vitousek, P.M. (1992) Biological invasions by ex otc grasses. the grasstr cycle. and global change. Annual Review of| Feologyand ystematies, 28, 63-87. ttps:/otovg/10-1146/annurev. €323.110192.000481 De Magalies, R. M. Q, & Schwik, D. W. (2012). Leaf traits and fit ter Hlammabitty: Evidence for non-adetive mixture effects ina temperate forest. Journal of Ecology, 100, 1153-1163. nitps/40 org/10.1211/}1365-2745.201201987% Enleringor.J.R, & Monson, RK. (1993), Evolutlonary and ecological 25 ects of photosynthetic pathway variation. Annual Review of Ecology and Systemeris, 24, 11-499, nttosdoiore/10-1146/annureN, ex24an0199.002211 Els, DP, & Platt. WJ (2013), Fuel composition influnces fre char actristes and understorey nardwoods in pine savanna, Journal of| Ecology, 101, 192-201. https/d ong/10.1115/1965-2745.12008 Engle, D. M, & Bidwell T:G. (2004). Viewpoint: The response of can tral North American pales t0 seasonal tre. Joumal of Range Management, 54, 2-10.hitpsfdot.org/10.2907/4003519 Fernandes, PM, & Cruz, M, (2012, Pantammebiity experiments ater Imitednsightintovegetattn-fredynamicsinteractions NewPhytloi, 194, 605-609 ntps/dolorg/10:1191/.1469-81372012.04065% Fornara,0.A, Tian, {2008} Pantfunetionalcompostioninfluences ‘ates of sallcarbon and nitrogen accumulation. Jourel of zoos 96, ‘384-322. ntps:/dolorg/10:1111/, 1365-2745 200701845 Forrestal, EJ, Donoghue, M, J, & Smith, M. D. (2014). Convergent Dhvlogenetic and functional responses to altered re regimes in mesic savarma grasslands of North America and South Aria. New Phytologist, 208, 1000-2013. Fublendort, 5. D. & Engle, D. M. (2004) Applleation of the Fre: erasing Intracton to restore 2 shiftmg mosaic on tallrase Pralre. Journal of Applied Ecology, 4, 604-614. https/col oxp/t0.1211/3. 0024-8901 200.00937 CGaunon PR, Hats, K. Es Patt, Wr PASSO, HA MYEES, (2012) Small-scale variation in fuel loads diferentialy affects two ‘o-dominant bunchgrasses ina species-ich pne savanna, PLoS ONE, 7.629674, tps//do.org/10.1371/Jouratpane.0029674 Gagnon, PR. Passmore, HA, Slocum, M, Myers, J A. Harms, K.E, Platt W. J, & Palng,C. ET (2015). Fuols and fires lnfluence ves tation vis above- and belowground pathways in 2 high-alversity| lant community. Journal of Ecology, 103, 1009-1019. nits org/t0.1311/1365-2745.12601 (Gelman, A. & Hil J (2007). Data analy using regression and mutieve” hlerarchleal madels. Cambridge. UK: Cambridge Universit Press. Gibson, D. 1, Hartt .C, 6 Mert, GS (1990. Fire tamperature heterogenelty in contrasting tre prone habitats: Kansas talgrass prairie and Florida Sandhil. Bulletin ofthe Torey Botanleal Cub. 117, 249-286, tp: org/10.2907/2996822 Gibson, D.J,& Hulbert, LC. (1987) Effects of ie, topography and year te-year climate variation on species composton in tall ass pall Plant Ecology 72, 175-105. Govendes,N, Trollope, WS. W. & Van Wien, B, W. (2006). The effect of ie season, re requency, rainfall nd management on fre nen sity in savanna vegetation in South Arica. Journal af Applied Eola, 143, 748-758, hps://dovorg/40.1991 1365-2664 2006 01184 Griguts, Ke Lavorel, S» Davies, 1, Dy Dossantos, Ax Lloret, Fa & ‘Montserrat, V. (2005). Landscape-scale positive feedbacks between fre and expansion ofthe large tussock grass, Ampelodesmas mau! tanica i Catalan shrublnds. Global Change Biology, 11, 1042-1083. hitpss/dotore/10.1111/.1368-2486.2005.00780.x Grimm, E.C. (1983). Chronology and dynamics of vegetation change In the prarle-woodkand region of southern Minnesota, U.S.A. New Phytologist, 98,311-380.nttos/AdoLorg/10.1111/).1469-8197:1983. sp0s44.x Harte,1, & Shaw, R (1995) shifting dominance within amootane vegeta- tion community ~ Results or cate-warming experiment. Slece 267, 876-880. Mtps/Holorg/IO1126/scence. 267.5199 876 Haynes. W.M, Bruna, TJ & Ue. D.R. (2014) CRC handbook of chem- ‘try and physics, 95th Eaton, Internet Version 2015. Boca Raton, FLSCRC Press. Hobbie SE (1992) ettecsof plant spcieson tient eying Tencsin Ecology ‘SEvoliin, 7, 996-999 tps onp/101016/0169-5879)90126N Ivetson, LR YBUSSY DA ReDDECK, Jy Hutchinson, TF LON, Rs P & Prasad, A. M, (2004), comparison of thermocouples and tem perature paints to monitor spatial and temporal characterises of landscape-sale presrited fires. Intemational Joumal of Wildland Fire, 18, 11-822. https//doiore/10.1071/ WFO3063 Keutfman J. By Cummings, OL, & Ward D.E. (1994). Relationships of fre, bomass and nutrent dynamics along a vegetation gradient In the Braziton Cerrado. Joumal of Eongy, 82, 319-531, Mpa org/10.2307/2261261 Kennard. D OutcltK, Jones, & O'Bren, J. (2005). Comparing tach- niques for estimating fame temperature of prescribed fires. Fire cology 4, 75-84 ‘ane, 8: (2009), Fue fad andre behaviour in he southern Ontario tals pre (Thesis). University of Toront, Ketrnan, P, Cains, S|, Smith, M.D, Knapp. AK Burke. DE, ‘Burs, CE, » Wragg P-D.(2014), Responses to ie ifer between South Afican and North American grassland communities. Journal of Vegetation Slenee, 25, 793-804. tps://doLorg/10-1111/)s12130 Knapp, A.K, & Seastodt, .R. (1986) Detritus accumulation limits pro- uty of talrass prac. Bioscience, 96, 62-668. nttps//d0 org/10.2307/1310087 Kops, J. M. H. Bradley K.L, & Wedd, D. A. (20021, Mechanisms of Plant species impacts on ecosystem nitrogen cyling Ecolgy Letters, 5,454-466,htips//dol.org/10.1046/1461-0248,2002.00332.x Kops, JM. H, Ritchla. ME. & Tilman D.(2000), Selective herbivory ‘on a nitrogen fing legume (Lthyrs venosus ftiuences productv= ity and ecosystem nitrogen poo's in an oak savanna, Ecascence, 7 166-174, htips//do ong/0.1080/11956860.2000.11682585, Kops, J. M. H, 6 Tilman, D. (2006). Dynamics of sol nltrogen and carbon secumulaton for 61 years after agrcuitural abandonment. Feology, 81, 88-98. htps//doLorg/20:1890/0012-9658(2000}084 ‘088:DOSNAC|20.c0:2 Lavore, 5, & Garnier, E (2002) Predicting changes in community com Position and ecosystem functioning trom plant alts: Revising the Holy Gra Functional Ecology, 16, 545-586, htips:/fdo org/10.1046/).1365-2435.2002.00664 Lenmann. CER. Anderson, TM. Sankaran, MI, Higgins SL, Archie, 5 Hoffman, W.A.,.. Bond, W. (20%). Savanna vegetation-tive- imate relationships sitter among continents. Science, 243, 548 552. https://dol.org/0.2126/science 1247355, 1, Kops, J.M.H, Zu, X, & Laungan,R (2084), Carbon and nitrox _gencycling are resistant to ire in nutrient poor grassland So Science Sockety of Ameria Journal, 78, 225-831. neps//dolorg/0.2136/ sss9]2014.02.0056 1, W, 240, & Kops, J. M. H. (2039). Different fre frequency ime acts over 27 years on vegetation sucession in an infertile oleld _rassand, Rongelond Ecology & Management, 66, 267-273. ips dekorg/40.2111/REM-D-t1-002241, Martin, RE, Gist, CT. & Mer, RL (1969. Fe as ptysal factor ‘wind management In Proceedings of the Ninn Armocl Te Timbers Fre Elegy Confrence (pp, 271-288) Taloussce, Fria: Tal Tinbass Reseach ne. Masubeele, M. L Bond, We Ja & Stock, WD. (2007). How savanna _rasses decompose? South African Joumalof Botany, 72,203 ips ‘dolorg/20.4036/}.s3).200702.084200 | seus esloey WRAGGera. “Met auction, KK, Crane J. M, Nipper, JB & Ochre, T.W. (2014). Lack of eutrophication na taligrass pale ecosystem over 27 years Feolgy, 95, 1225-1285. htps/AdoLorg/10.1890/13-1068.4 ‘Mur, B. Ps & Bovanan, DM. J. 8. (2012), What controls the ast ution of tropical forest and savanna? Ecology Letters, 15, 749-758, ttps//dol.org/101211/]1461-0248.2012.01771. ‘Murray, 8.8, Hardstft, LX, & Philips, ML. (2043) Differences ta eat ‘fammabity, eat tats and fammabity at relationships between native and exotic plant species of dry sclerphyl forest PLoS ONE, 8, £79205. https /eoi.org/10.1971/Joural pone 0079205 ‘Muten, R. Ws. (2970), Wlland rires and ecosystems ~ A hypothesis Feology, 5, 1046-1051. lips: tol. org/10-2307/1933631 NNorts, M. Dy & Releh, P. 8 (2009). Modest enhancement of nitro ‘gen cancervation via retranslocation in response to graces In N Supply and leat N status Pant and So, 916, 193-208, tps. org/0.1007511104-008.9770-6 ‘OhtsulT. & Keyes, T1986) Blased porcolatlon: Forest fires with wind “Journal of Prysles A Mathematia and Genera, 1, L281 Mpa. org/0.1088/0305-4870/19/5/012 ONL RV. Cardner, RH, Turner, M. G. & Romme, W. H (1992) Epidemiology treory and. disturbance spread on landscapes Landscape Ecology, 719-26. tps: /éoLorg/10.1007/8F02573954 Oeil \ Mne, BT Turner, MG. & Gardner RH. (1789), Resource blization scales and landscape pattern. Landscape E003) 2,63-69, ttps/dol.org/101007/BF00138908, Pausas,J.G, Keeley, JE, & Schuik,D. W. (2027), ammabilty 25 an ecological and evolutionary diver Juri ofEclogy. 105, 289-297. ttps/dol.org/101311/1365,-2745.1269 Peleg, AFA. Ahstiém, A, Hobbie, S.E, Reich, PB, Nirah LP, Staver, A. Cy JBe480N, RB (2017) Fie frequency aves ‘eeadal changes in sil carbon and ritragen and ecosystem pro: ductivity, Nature, 53, 194-298, hitpe//doLorg/10.1038/nature? 4668, Peterson, D.W. Rech, PB, & Wrage, K J (2007), Plant functional group responses fo fire frequency and tee canopy cover gradlents in oa savannas and woodlands. Joumal of Vegetation cence, 18, 3-12. tpssdo\.org/10-1111/.1654-1103.2007:002510%. Prog, LD, Murphy, 8 P, Willamson, G3, Cochrane, M.A. Jolly, W. M, {& Bowman, D. M. JS 2017}. Doos Inherent flammabilty of grass and iter fuels continu to continental pattems af landscape fre actty? Journal of Bogeosraphy, 44, 1965-2699. Reich, P.., Peterson, D. W., Wedin, DA, & Wrage, K. (2008) Fire and ‘vegetation effects on productivity and nitrogen eyting across 2 fo est-prassand continuum Ecology, 82, 1705-1719. Reich, PB, Tilman, D, Isbell F, Muelr, K, Hobbie, S. E, Flynn, OF B, & Ekonhauer, N. 2012) Impocts of blediversity loss escalate ‘through time as redundancy fades. Science, 336, 589-592. Rttps// otorg/10:1126/scence.1217909 Reich P 8 Wight IJ & Lusk, C. H. (2007. Precitng leat physio. (ty from simple plant and clmate stbutes & glabal GLOPNET anaisis. Feoloycal Applications, 17, 1982-1988. hitps/dol org/0.1890/06-1803:1 Rugsioro, .G.C. Bataha, M.A. Palla, V-R, & Mekrles, ST: (2002) Sollvegetstion relationships In eerrago (Brazlan savanna) and ‘semideckduous forest, Southeastern Brazil. Pant Ecology, 160, 1-46, itossdo\.org/10:1023/0:1015817217386 Sankaran, M, Hanan, N. Py Scholes, RJ, Ratnam, J, Augustine, D. J, (Cae, B.S, .. Zambatls,N. (2005). Determinants of woody cover in Argan savannas. Nature, 438, 886-849, Nitps//00:0%8/ 10.1039 | satre04070 Schertzer, Stave A.C. fe Levin, A. (2014) Implcations ofthe spa tal dynamics of fire spread forthe bistability of savanna and forest. Joven of Mathematica logy, 70, 1-13. ‘Schuik,. W. (2013). Dimensions of plant flammability. New Phytologist 1206, 486-488, https//dol.org/104114/nph 19372 Seeley M.(2018) Hourycimatedata Meterolosiemeasurementsatcodar reek natura histor area. Environmental Data intiatve tps: 01g/10.40753/pasta/ac1bSt994cb8388602dt50545429413 Simpson, K, J Riley, . Sy Cistn, PA Belchet CM. Lehmann C E.R, Thomas, . H, & Osborne, .P. (2016), Determinants of flan mobllty in savanna gras species Journ of Ecology 104, 138-148 ttpsfdot.org/101114/1365-2745 12508, stave, A. C, Archibald, S & Levin. A. 2012), The global extent and determinants of savanna and forest as altemative biome states, Sclence, 394, 230-282. hitps/#oLorg/10:1126/sclence 1210465, Staver,A.C. & Levin, S.A. (2072). Integrating theoretical ciate ana tke effects on savanna and forest systems. The American Natura, 180, 211-224, httos,/ido.org/10.1086/6665648 Stevens, 1, Maskell... Smart, 5... Caporn,S.J.M, Dise, NB. & owing, D.J.6.(2009) \dentityinngicators of atmospheric zen deposition impacts in acd grasslands Biological Conservation, 142, 2069-2075, hitps/deLong/101016/ bacon, 2008.04.002 Stronach, N-R. Hy & McNauehion, . (1989). Grasslandtire dynamics the serengeti ecosystem, and a potential method of retrospectively estimating fire energy. Journal of Appled Ecology, 26, 1025-2033. https/dol.org/10.2307/2403709 Tayi BR. Parkinson. & Parsons, W-F (1989) Nitrogen and grin content 35 precetos of iter decay rates: A microcosm test, Elon, 70,97-104. hitps://doi.org/10.2307/1938416 Thaxton J.M, & Platt. W. 12006). Small-scale fuel variation alters tee intensity and shrub abundance in 3 pine savanna, Ecology, 87, 2931- 41887. https//deLorg/20.1890/0012-9658(2006)8711331:5FVAFI] 2oco2 Tman, D. {1982}. Resource competition and community structure Princeton, Nl: Princeton University Pres. Timan, 0. (1987). Secondary Succession and the pattern of plant dom ‘ance along experimental nitrogen graents, Ecologcal Monographs, 57, 190-244 “Timan, D. (1990). Mechanms of plant competition for nutrients: The ‘elements of a predictive theory of competition. In J.B. Grace & D. Tiiman (Eds), Perspectives on plant competition (pp. 117-14). San ‘Diego, CA: Academic Press Tilman, D. (20188). Plant aboveground biomass dats: Biodiversity Ik Effects of plant biodherstty on population and ecosystem pro cesses, Enaronmental Data Inistive. ntpe/dore/10.6073/ pasta/S4tcc0Bas9tasac7ateaTest ected Timon, D. (20480) Plant species percent cover dats: Biodlversty I Effects of plant biodiversity on population and ecosystem process, Environmental Data nieve. nttps/édolore/10.6073/0a5ta/ aeSTebeSATAS5k2048270cec37D442 “Tian, (2018) Pot coordinates forthe E120 big blocvorslty Melt Blodiversty I: Effects of plant blodersty on population and ecosystem processes. Environmental Data Intative, httes/do (018/10.6073/p25t2/980%94ed437799082c30a9 7510726 Tuman. 0. (2018). Soll nitrogen: Bleaversty I: Etfects of plant biodiversity on population and ecosystem processes. Environmental Data initiative. tps/loiorg/10.6073¢pastar <2oco7203d5A066écdcd2cH3.98121f6 Tuman.0 Dod, M.Sivertown, J Poulton, Johnston. A. & Crawly, M. (£994) The park grass experimentinsights from the most long term ecological study. In RA, Leigh & AE. Johnston (Eds), Long term experiments in agrcitaral and eclagca scenes (pp. 287-0) Wallingford, UK: CAB Internationa. Tiiman, D. Kops, J Wedin, Dy Reich Py Ritchie, M, & Siemann, . (1997), The influcnee of funetionsl diversity and composition fon ecosystem processes. Science, 277, 1300-1302. https:/ol e19/10.1126 science 277 5930.4300 Tima, D., Naeem, S KNOPH Jy Reich, Py SlemanNn, Es Wen, Dy on Lawton, J. (1997) Biodiversity and ecosystem properties. Science, 278, 1866-1867.WRAGG era Tian. 0. etch. & Knops, (2006) Blodversity and ecosystem sta bility ina decade-long grassland experiment. Notre, 441, 629-632 hitps/dotorg/101038/naturc04722 Tian. D, Rich, P. B.. Knops, J. Wein, D., Miike, T. & Lehman, (C. (200}. Diversity and productivity ia a longterm grassland experiment. Science, 298, 849-845. tps /doiorg/10.1126) selence.1060391, Tian. D. Reich, P, Philips, H. Menton, M4 Patel, A. Vos... KROps. (2000) Fie suppression and ecosystem carbon storage. Ecology. 84 2680-2685. https:/dol.org/10.1890/0012-9658|2000)08112680: Fsaecsi20co2 Trollope, W.S. W Tollape L.A, & Hartnett, 0.C.(2002). Fre behaviour ‘8 key factor inte fre ecology of Arian prasstands and savannas. In D. X. Viegas (ed, Forest Fie Resenrch & Wldlond Fite Safety. Proceedings of the IV International Conference on Forest Fre Reseorch (pp. 1-15) Rotterdam, the Netherands: Milpess. Tolle, W. S. Ws THotpe, LA Pole, As FL & Zarb N (2996). SAFARI'92 charactenzaton of blomass and fre behavior In the small expermantal Burns i the Kruger National Park Jumma of {Geophysical ReseorchAumaspheres, 101, 28891-23539. nitps/d0 org/10.1029/96)D00671, Waly, A. L, Menges, . 5, & Woolley, C. W. (2006). Comparison of thes devices for estimating fire temperatures In ecological stus es, Applied Vegetation Science, 9, 97-108, httos/AdoLore/M0A111/ }.2654-109%.2006.1000859. Watton, DL, & Hui, K-C.(2010), The atesine sasnine: The analysis of proportions in ecology Eeoigy, 9, 3-20. Whittington, H.R, Tilman, D, Wragg, PD, & Powers, J. (2015). Phenologieal responses of pre plants vary among specles and year In a three-year experimental warming stucy. Ecospere 6, 15, tes, van Nleuwestat M.-L (2996). Long-termafter eects of fertiization on above-ground phytomass and spedes diversity in caleaeous erasing, Jounal of Vegetation Science, 7, 177-184 hitpsusdoiorg/0.2307/3236317 Wiliams, RJ, Duff, G. A, Bowman, D.M. J. 5, & Cook, G. D. 1996, Variation In the composition and structure of tropleal savannas as a function of rainfall and soll texture along large-scale climatic pra entin he Northern Tertoy, Australia Jounat of Blosengranhy, 23, 747-736, nto Hdl org/M01111/}1365-26991996 1BO0036 x Wiliams, &, Gil, A. & Moore, P. (2998). Seasonal changes In fre be: haviour Ina tropical savanra in nofthern Ausra, International Journal of Wikland Fire, 8, 227-239. hips/idolorg/103074/ wes9a0227 Welt, 1.0. ¥. (2004). 200-year fire history ina remnant ook savanna In southesstezn Wisconsin. The American Mialand NoturalSt, 152, Journal ot ology | 2001 201-213. nttps:/fdoLorg/10-1674/0003-0031/2004)15210201:AYF Hiaj2oco;2 Wings, PD. (2025). Human Impacts om how sovanna plans Interact ttvough fe, resources, and microclimate (PRD thes) University ot Minnesota, t. Paul MN Wragg, P. (20182). Fire behavior Biversity It Effects of plant bio- ‘versity on population and ecosystem processes. Environmental Data Intiative. nttps/dol.org/10.607a/pasta/ceafa7fobest5979 eet 69852 ‘Wrsgg,P. 2018) Fire fuel load: Biodiversity I: Effects of plant blo- tlversity on population and ecosystem processes. Environmental Data Intative.http://dotorg/40.6075/pasta/037bSd380atcs4td s9aszacaabas6e Wragg? (2018) Freseverity-BiodiversityIk Ettectsof plant bioversity ‘onpopulationandcensystem processes ErvironmentalDatalntiatve hitps://4o.096/10.6075/pasta/09tdbceact Sabt7iSeaSactedad eo ‘Wags. P. (2018) Fire temperatures: Blodversity Ik Effects of plant diversity on population and ecosystem processes. Eviconmental (Data native, ntps//doLorg/10.6072¢pasta/c7Es6d02770685ee" exadbt7 748529 Xt, J. Nl, 5, & Wan, 5. (2008). Response of cosystam carbon exchange “So wotming ancntragen addon uring swe nyrelogialy contrast ing growing seasons in a temperate steppe. Global Change Biloy, 15, 1544-1556. https org/I0:1111/}1365-2486.2008 01807 Xta, J, & Wan, 5. (2008). Global response pattems of terestral plant ‘species to nitrogen adation. New Phytologist, 178, 428-499. ntos/t doiorg/10.1111/, 1469-8137 2008.02488.x ‘Zhao, We, Commvel, W. van Pomoren, M, van Logtestn. Re. Pe 6 CCornaissen 1. H.C. (2016) Specles minture effects on flammabity 206s plant phylogeny: The importancect liter partclesze and the spedal ole for non-Pinus Pinaceae. Ecology and Evolution, 6, 8223 8204, toad org/30:1002/ece9 2451 SUPPORTING INFORMATION Additional Supporting Information may be found online in the supporting information tab for this article, How to cit this article: Wragg PD, Mielke T, Tian D.Forbs,, arasses, and grassland fie behaviour J Ecol 2018;1061983- 2001. https: org/10.1111/1365-2745.12980