Engineering Fire: the drivers of low fire intensity over gopher tortoise mounds in a Florida sandhill

\papertype Original Article While large grazers are well known to alter fire regimes, small herbivore effects on fire have received compar- atively little attention. The gopher tortoise (Gopherus polyphemus) is a small herbivore that acts as an ecosystem engineer in upland, fire-dependent ecosystems of the southeastern U.S. They dig burrows that locally decrease fire intensity, and these burrows provide a refuge for many animals during fires. Importantly, their burrowing and foraging activities have the potential to modify fire regimes via several mechanisms. Gopher tortoises may actively decrease fire intensity near their mounds, either by reducing plant biomass and/or by altering the flammability of the adjacent plant community. Alternatively, tortoises may preferentially burrow at microsites with relatively nonflammable vegetation and low fuel loads. To test these hypotheses, we leveraged data from intensive monitoring of gopher tortoises at Archbold Biological Station in south-central Florida, USA. We selected 30 existing burrows varying in activity status (active, inactive, abandoned) as well as nearby, non-mound control points. We characterized plant biomass and community composition within 15 m of mounds and non-mound points and quantified 11 fire-related traits for 23 common plant species. Mounds of both active and inactive tortoise burrows had lower plant and litter cover than abandoned mounds and the surrounding vegetation matrix, but these differences in vegetation and litter did not extend beyond the mound itself nor persist following burrow abandonment. Tortoise effects on community-level flammability were minor and unlikely to modify fire intensity. Overall, the highly localized soil disturbance associated with burrowing is likely the primary means by which gopher tortoises may alter sandhill fire regimes. Critically, our study highlights how small animals can shape fire behavior via direct reduction of fuel loads. 1 1 Title 1 Engineering Fire: the drivers of low fire intensity over gopher tortoise mounds in a Florida sandhill 2

3 While large grazers are well known to alter fire regimes, small herbivore effects on fire have 4 received comparatively little attention. The gopher tortoise (Gopherus polyphemus) is a small herbivore 5 that acts as an ecosystem engineer in upland, fire-dependent ecosystems of the southeastern U.S. They 6 dig burrows that locally decrease fire intensity, and these burrows provide a refuge for many animals 7 during fires. Importantly, their burrowing and foraging activities have the potential to modify fire 8 regimes via several mechanisms. Gopher tortoises may actively decrease fire intensity near their 9 mounds, either by reducing plant biomass and/or by altering the flammability of the adjacent plant 10 community. Alternatively, tortoises may preferentially burrow at microsites with relatively nonflammable 11 vegetation and low fuel loads. To test these hypotheses, we leveraged data from intensive monitoring of 12 gopher tortoises at Archbold Biological Station in south-central Florida, USA. We selected 30 existing 13 burrows varying in activity status (active, inactive, abandoned) as well as nearby, non-mound control 14 points. We characterized plant biomass and community composition within 15 m of mounds and non-15 mound points and quantified 11 fire-related traits for 23 common plant species. Mounds of both active 16 and inactive tortoise burrows had lower plant and litter cover than abandoned mounds and the 17 surrounding vegetation matrix, but these differences in vegetation and litter did not extend beyond the 18 mound itself nor persist following burrow abandonment. Tortoise effects on community-level 19 flammability were minor and unlikely to modify fire intensity. Overall, the highly localized soil 20 disturbance associated with burrowing is likely the primary means by which gopher tortoises may alter 21 sandhill fire regimes. Critically, our study highlights how small animals can shape fire behavior via direct 22 reduction of fuel loads. 23

Gopherus polyphemus, ecosystem engineer, flammability, functional traits, Florida sandhill 24 2

25 Herbivores have clear influences on fire behavior through the consumption of plant biomass. 26 Herbivores may limit fire spread and intensity by consuming fuels before fires can (Archibald et al., 2005; 27 Bruegger et al., 2016; Davies et al., 2015; Karp et al., 2024), or may constrain fire intensity by selecting 28 for less flammable plants as they graze, e.g., by creating “grazing lawns“ of grazing-tolerant but relatively 29 less flammable grasses(Archibald et al., 2005; Hempson et al., 2019). Yet, research on herbivore-fire 30 interactions has focused nearly exclusively on large herbivores while the effects of small herbivores 31 remain largely unexplored, despite their potential to also decrease fire intensity. 32 One small herbivore that may alter fire behavior is the gopher tortoise (Gopherus polyphemus), a 33 keystone species of the pyrogenic southeastern USA, a global biodiversity hotspot (Noss et al., 2015). 34 Gopher tortoises dig burrows (mean length 4.6m; Hansen, 1963) around which low fire intensities have 35 been documented during prescribed burns in a Florida sandhill (Kaczor & Hartnett, 1990). These burrows 36 have outsized benefits for the biodiversity of their surrounding ecosystems; over 360 species rely on 37 gopher tortoise burrows for habitat and refuge from fires (Knapp et al., 2018). Thus, understanding how 38 gopher tortoise burrows influence fire intensity could help clarify the unique properties that make 39 burrows a suitable fire refuge for so many species. However, neither the mechanism nor the spatial scale 40 by which burrows decrease fire intensity is understood. Here, we evaluate three possibilities. 41 First, we hypothesize that gopher tortoises could actively reduce plant and litter cover on and 42 adjacent to mounds, thereby reducing the amount of fuel, which would be expected to reduce fire 43 intensity (hypothesis 1, hereafter H1). Gopher tortoises are a generalist grazer in their upland habitats 44 (Diemer, 1986) and might consume enough biomass on and near their burrows to interrupt the fuel 45 layer. Exclosure experiments have demonstrated the gopher tortoise’s ability to reduce both plant cover 46 (Lloyd et al., 2023) and abundances of preferred species (Richardson & Stiling, 2019). Gopher tortoises 47 3 could also reduce fuels locally by burying vegetation as they excavate their burrows. As tortoises dig 48 their burrows, soil accumulates just outside the burrow, creating a large mound of sediment at the 49 burrow opening, potentially reducing fuel loads. Similar decreases in vegetation cover due to burying 50 have been documented in other burrowing species (Hobbs & Mooney, 1985; Huntly & Reichman, 1994; 51 Kurek et al., 2014). 52 Second, we hypothesize that gopher tortoise behavior actively promotes a less flammable plant 53 community (H2). The disturbances in the landscape left by gopher tortoises and other fossorial animals 54 favor the establishment of early successional “fugitive species” (Hobbs & Mooney, 1985; Kurek et al., 55 2014), which selectively recruit on overturned, nutrient-poor soils (Kaczor & Hartnett, 1990). These early 56 successional plants possess functional traits, including decreased leaf surface area to volume ratio and 57 increased tissue density, which decrease flammability (Dimitrakopoulos & Papaioannou, 2001; Mason et 58 al., 2016). Alternatively, a tradeoff between flammability and herbivore tolerance could lead gopher 59 tortoises to select for palatable, but less flammable plants on and near mounds. Inferring the effects of 60 changes in plant composition on fire behavior requires assessing multiple metrics of plant flammability 61 for individual species (Zylstra et al., 2016). 62 Third, we hypothesize that gopher tortoises preferentially site their burrows in areas that are 63 already nonflammable (H3), either because of reduced fuels or reduced community flammability. The 64 extent to which gopher tortoises bias their burrow sites with respect to litter (Mushinsky & Esman, 1994) 65 or cover (Aresco & Guyer, 1999; Jones & Dorr, 2004; Lau & Dodd, 2015) varies across the gopher 66 tortoise’s range. Gopher tortoise burrow site selection in sandhill could potentially be biased towards the 67 ubiquitous bare patches with little vegetation and litter cover to facilitate thermoregulation or burrow 68 construction. Bare patches remain present even in long unburned sites (Abrahamson et al., 1984), 69 4 suggesting that there could be properties inherent to these sites that maintain their low fuel load or 70 promote a nonflammable plant community. 71 Here, we consider these three hypothesized mechanisms of reduced fire intensity over gopher 72 tortoise mounds in a sandhill in peninsular Florida. We characterized the vegetation and litter 73 surrounding gopher tortoise burrows of varying activity regimes as well as naturally occurring areas 74 devoid of vegetation and litter (hereafter bare patches). We further quantified the flammability of 75 common species of shrubs, grasses, forbs and palmettos and used these species-level observations to 76 evaluate community-level flammability around tortoise mounds. To evaluate our three hypotheses, we 77 compared response variables on mounds at active, inactive, and abandoned burrows, as well as at bare 78 patches and on transects extending into the areas surrounding mounds to evaluate the spatial extent of 79 potential tortoise impacts. If gopher tortoises reduce fuel loads (H1), we expect lower fuel on active 80 mounds compared to inactive and abandoned mounds. If gopher tortoises alter plant community 81 composition and flammability-related traits (H2), we would expect to find differences in plant traits on 82 active mounds compared to inactive, abandoned, and bare patches. Finally, if gopher tortoises select 83 intrinsically nonflammable areas (H3), we expect to find similar plant cover, litter cover, and plant traits 84 on mounds of all classifications compared to bare patches, as well as lower fuel loads close to mounds. 85

86 Study System 87 This study was conducted at Archbold Biological Station (hereafter ‘Archbold’) on the southern 88 tip of Florida’s Lake Wales Ridge (27.19°N, 81.34°W) in two management units (2A and 2C) characterized 89 as southern ridge sandhill vegetation (Figure 1). The sandhill community at this site is dominated by 90 resprouting shrubs, particularly oaks (e.g. Quercus geminata, Q. myrtifolia, Q. laevis), hickory (Carya 91 floridana) and palmetto (Sabal etonia and Serenoa repens) with sparse overstory pine trees (Pinus 92 5 elliottii) and groundcover of grasses (particularly wiregrass, Aristida beyrichiana), sedges, legumes, and 93 other herbaceous plants (Abrahamson et al., 1984a). The study area is situated at approximately the 94 highest elevational point of Archbold (67m above sea level) and has well-drained, sandy Astatula and 95 Lake soils (Abrahamson et al., 1984). Sandhill units are scheduled to be burned every 2-5 years; at the 96 time of data collection in the summer of 2022, units 2A and 2C, had been burned in 2019 and 2017 97 respectively. 98 Gopher tortoises dig burrows in sandy, well-drained soils throughout their range from Louisiana 99 to South Carolina (Diemer, 1986). After constructing a burrow, a tortoise spends approximately 90% of its 100 time sheltered inside (Eubanks et al., 2003), only leaving to bask, visit potential mates (Johnson et al., 101 2009), and forage, typically within 30-50m of the burrow (B. Rothermel, unpublished data). While a 102 burrow is in use, the inhabiting gopher tortoise maintains the burrow’s integrity through re-excavation 103 (Kinlaw & Grasmueck, 2012). As social interactions or forage availability change, gopher tortoises 104 abandon their burrows to reoccupy vacant burrows or dig new ones (Aresco & Guyer, 1999). Tortoises 105 are known to use as many as 10 different burrows per year (Castellón et al., 2018); at one site, 20% of 106 vacant burrows were reoccupied within a year (Aresco & Guyer, 1999). Some burrows are abandoned 107 permanently, particularly if they become shaded by woody plants, after which they collapse leaving 108 behind only the mound of excavated soil (Aresco & Guyer, 1999; Mushinsky & Esman, 1994). 109 Archbold is the site of a long-running gopher tortoise demographic study that began in 1967 110 (Layne, 1989). Population density within recently burned sandhill units during 2012-2017 was 111 approximately 1.4 adult tortoises per ha (Howell et al., 2020). Since 2012, burrow surveys have been 112 performed every 1-3 years within the core study area by searching along tightly spaced transects and 113 mapping all burrows using GPS (Howell et al., 2020). Burrows are classified as active (entrance clear of 114 6 debris, with footprints or other tortoise signs present), inactive (entrance clean, but no fresh tortoise 115 sign) and abandoned (unusable, collapsed within 1, of entrance). 116 Study Design 117 Based on records from past burrow surveys and field visits, we identified 11 active, 11 inactive, 118 and 8 abandoned burrows in fire-maintained sandhill. We also randomly selected 20 additional ‘bare 119 patch’ sites. To do so, we initially chose 25 areas that were farther than 30m from all previously known 120 gopher tortoise burrows. After surveying each area for any unmarked burrows, the first encountered 121 bare patch with no visible vegetation in an area roughly the size of a gopher tortoise mound (mean = 6.7 122 m2) was chosen. A “focal point” was established over mound or bare patch centers. Focal points were 123 distributed across both management units with ≥30 m of separation from other burrows measured from 124 GIS layers in Esri Field Maps (Figure 1). 125 Two types of vegetation surveys were centered on focal points. Four 15-m transects were 126 established radiating from the center of the focal point. These were oriented along the major and minor 127 axes of the ellipse roughly describing the mound or bare patch. In the case of an active mound, the 128 minor axis roughly corresponded to the direction of the burrow opening. First, we used a line-intercept 129 approach to estimate shrub cover and composition along the four transects, recording the species of 130 each shrub that intersected the transect and where the shrub’s foliage began and ended along the 131 transect. Second, to estimate litter and plant cover, we visually estimated the percentage of shrub, 132 palmetto, grass, forb and litter cover within a 1-m quadrat at each focal point and at 1, 3, 5, 10, and 15m 133 from the center of the site along each transect (21 quadrats per focal point). A composite vegetation 134 cover percentage was also visually collected in each quadrat (hereafter plant cover), after which the 135 percentage of litter in unoccupied space was recorded (hereafter litter cover). 136 Plant Flammability 137 7 Although low-intensity fires burn frequently in the Florida Sandhill (Abrahamson et al., 1984; 138 Ashton et al., 2008), and are recognized to have profound effects on plant life history strategies 139 (Abrahamson et al., 2021; Menges & Kohfeldt, 1995), a systematic quantification of common sandhill 140 plants’ fire related traits has yet to be completed. Flammability was quantified as a combination of 141 ignitability (how easily fuels ignite); combustibility (the intensity at which the fuels burn); and 142 sustainability (how long the fuels maintain flame) (Anderson, 1970). Each of these dimensions can be 143 measured directly, by burning the plant, or indirectly by measuring physical and chemical properties. For 144 example, high biomass (bulk) density makes fuel more combustible and ignitable by connecting fuel 145 more cohesively (Simpson et al., 2016). Alternatively, a fuel’s moisture content limits the amount of heat 146 it can absorb, making fuels less combustible and ignitable (Pyne, 1996). We quantified the flammability 147 of 23 common sandhill species by measuring 11 relevant plant traits (Supporting Information). 148 Statistical Analyses 149 All statistical analyses were performed in R version 4.3.2 (R Core Team 2024). All parametric tests 150 were conducted using a type 1 error rate (alpha) of 0.01. Differences in total litter cover and the cover of 151 all living shrubs, palmettos, grasses, and forbs (hereafter plant cover) in quadrats around tortoise 152 mounds/bare patches were evaluated using beta generalized linear mixed effects models (GLMM), with 153 burrow ID as a random effect. Litter and plant cover were compared using three models. First, type of 154 focal point (active, inactive, abandoned, or bare patch) was a fixed effect in the GLMM and evaluated 155 using an analysis of deviance test. Second, observations directly over focal points were compared to all 156 quadrats (hereafter the surrounding matrix) through the GLMM’s asymptotic Z tests. These tests were 157 conducted separately for each mound status. Third, distance from the mound edge was included as a 158 continuous fixed effect and its significance was similarly evaluated with an asymptotic Z test for each 159 mound status. 160 8 Species flammability was analyzed using three approaches. First, we used K-means to find 161 clusters of plant species with similar flammability traits. The quality of these clusters was assessed with a 162 comparison of the average cluster sum of squared error (SSE) against 250 randomly permutated datasets 163 (Peeples, 2011). Second, we tested for correlations between each pairwise combination of flammability 164 traits using Pearson’s product moment correlation (𝜌) tests. Third, we used principal components 165 analysis (PCA) to examine which subset of the 11 flammability traits capture the most variance among 166 species. Of our 11 traits, bulk density, heat of combustion, hours to ignition, and burn duration had the 167 highest loadings along the first three principal components (see Results and Table 2) and were used to 168 summarize the combustibility, ignitability and sustainability of each species. 169 Finally, we analyzed community-weighted trait means (CWM) for a subset of flammability traits. 170 Estimates of bulk density, heat of combustion, hours to ignition, and burn duration were combined with 171 relative species abundances along transects to compute CWM for each quadrat; only community 172 composition of shrubs and palmettos were used for CWM calculations due to their dominance (averages 173 of 58 and 24 percent quadrat coverage respectively, compared to 6 for forbs and 12 for grass). Linear 174 mixed effect models were used to examine variation in CWMs of flammability traits in the same model 175 design used for plant cover/litter above. A multivariate analysis of variance was used to evaluate 176 differences in multivariate means of flammability traits among mound types and asymptotic T tests were 177 used to test the significance of model coefficients. 178

179 Gopher Tortoise Effects on Fuel Load 180 Burrowing activity was generally associated with reduced fuel loads. When considering plots 181 directly over mounds, both active and inactive mounds had lower mean cover of plants (χ2 = 90.84, df = 182 3, p <0.001) and litter (χ2 = 35.29, df = 3, p <0.001) compared to abandoned mounds, but those of active 183 9 and inactive mounds did not differ significantly from each other (Figure 2, Supporting Information). Bare 184 patches had significantly lower plant cover than both active and inactive mounds, but not lower litter. 185 However, fuel reductions were largely limited to the mound itself. Active and inactive mounds 186 had significantly less cover of plants (between 71 and 74%) and litter (between 35 and 46%) relative to 187 off-mound plots (Figure 3, Table 1). Fuel load returned to baseline levels following a burrow’s collapse 188 judging by the lack of difference in plant and litter cover on versus off abandoned mounds. There was 189 less plant and litter cover at bare patches compared to the surrounding matrix. Furthermore, there was 190 no effect of distance from mounds of any type on either plant or litter cover (Supporting Information). In 191 contrast, as distance from bare patches increased, there were significant increases in cover of both 192 plants (Z = 4.65, n = 383, p < 0.001) and litter (Z = 3.53, n = 383, p < 0.001). 193 Plant Species Flammability 194 Our examination of species-level flammability metrics (Supporting Information) revealed 195 significant pairwise correlations between maximum temperature and bulk density (𝜌 = 0.786, 95% CI: 196 (0.552, 0.905), t = 5.82, df = 21, p < 0.001), mass loss rate and percent burned (0.667, (0.352, 0.847), t = 197 4.11, df = 21, p < 0.001), moisture at ignition and hours to dry (0.677, (0.367, 0.851), t = 4.21, df = 21, p < 198 0.001), and heat of combustion and bulk density (0.554, (0.184, 0.787), t = 3.05, p = 0.006) (Supporting 199 Information). Cluster analysis did not reveal any clear groupings of plants based on their flammability 200 traits, as SSE for the data was not less than the SSE of permutations for any number of clusters 201 (Supporting Information). 202 Four principal components explained 73.7% of the variance in plant flammability traits among 203 species (Table 2). PC1 was associated with combustibility (with positive loadings on bulk density, percent 204 burned, maximum burn temperature, and heat of combustion; 29.2% of variance explained), reflecting 205 hot, complete burns. PC2 was associated with low ignitability (with positive loadings on minimum 206 10 temperature for ignition, hours to dry and hours to ignition; 21.4% explained). PC3 was associated with 207 low sustainability (with negative loadings on percent burned and burn duration; 12.7% explained) and 208 flame extinguishing quickly following ignition. PC4 was associated with high moisture content (with a 209 negative loading on dry matter content and a positive loading on hours to ignition; 10.4% explained). 210 The first two principal components revealed strong variation in flammability among functional 211 groups (Figure 4). Grasses and palmettos generally had the higher values of both combustibility (high 212 PC1 values) and ignitability (low PC2 values), although each was represented by only two species. 213 Neither palmetto species was as ignitable but both were as combustible as the two grasses. Shrubs 214 showed a wide range of ignitability and combustibility. However, shrub PC1 and PC2 values were strongly 215 negatively correlated (𝜌 = -0.81, 95% CI: (-0.95, -0.37), t = -3.92, df = 8, p = 0.0044), indicating that 216 combustibility and ignitability are positively correlated. Forbs were generally not combustible but had 217 moderate to high ignitability. 218 Gopher Tortoise Effects on Community Flammability 219 Community flammability in on-mound plots did not differ among burrow types (Wilk’s Lambda = 220 0.906, df = 3, p = 0.119). There was evidence that burrowing activity modified community flammability 221 when compared to the surrounding matrix. Active mounds had a significantly higher bulk density 222 compared to the surrounding matrix (1762.0 g per m3 more, 95% CI: (674.9, 2847.0), df = 74.4, t = 3.2, p 223 = 0.002) and abandoned mounds had a significantly lower burn duration (1.97 sec less, 95% CI: (-3.28, -224 0.665), df = 55, t = -2.98, p = 0.004). No other flammability traits differed significantly among burrow 225 types. There was no strong evidence of gradual changes in flammability traits from mounds either, as no 226 traits varied significantly with distance from mounds. 227

228 11 In this study, we leveraged a well-studied population of a keystone species, the gopher tortoise, 229 to explore how this small herbivore may alter fire intensity. We found evidence that gopher tortoises 230 directly reduced vegetation and litter (H1), but little evidence that gopher tortoises actively changed 231 local plant composition and flammability traits (H2) or selected sparsely-vegetated areas as burrow sites 232 (H3). Our study demonstrates that small herbivores can alter fire behavior through fuel limitation. In 233 addition, our study provides a new dataset of flammability traits for many common plant species of the 234 Florida sandhill and provides insights into how functional groups interact with fire. 235 Gopher Tortoises Reduce Fuel at Mounds 236 We found clear evidence of fuel limitation over gopher tortoise mounds in Florida sandhill and 237 that this is attributable to gopher tortoise burrowing activity. On the mounds of active and inactive 238 burrows, both plant and litter cover were greatly reduced compared to the surrounding vegetation 239 matrix. These effects were highly localized and ended abruptly at the edge of the mounds themselves. 240 The abrupt cutoff of tortoise effects suggests that fuel limitation is likely the result of burial rather than 241 trampling or herbivory. Gopher tortoises regularly forage at distances beyond 30m of their burrows, 242 indicating that any clear herbivory effects should extend beyond the mounds. 243 Rather, soil deposition during repeated excavation likely buries litter, herbaceous vegetation and 244 dormant seeds as well as imposes constraints on colonizing species. The soils that gopher tortoises and 245 other fossorial taxa displace onto their mounds are nutrient poor in relation to the surrounding matrix, 246 since organic matter that accumulates at the soil surface get covered by nutrient-poor subsurface soils 247 (Kaczor & Hartnett, 1990; Kurek et al., 2014). These nutrient constraints have been hypothesized to limit 248 vegetative growth on mounds and favor recruitment from seed (Kaczor & Hartnett, 1990). Even so, 249 continuous excavation on active mounds could bury seeds deep in soil as shown on the mounds of the 250 pocket gopher (Thomomys bottae), requiring time and additional disturbance to resurface the seeds in 251 12 order for the recolonization of the mound to occur (Hobbs & Mooney, 1985). The seed storage facilitated 252 by burial could make mounds an important place for seed-based recruitment in the southern ridge 253 sandhill. 254 Here we show evidence that recolonization does take place following burrow abandonment. 255 Plant and litter cover over abandoned mounds were greater compared to active and inactive mounds 256 but similar to the surrounding matrix. Further studies are needed to determine the rate at which 257 recolonization occurs and whether the new plant community is shaped by the seeds stored through 258 tortoise burial or by early successional species. The recovery of the fuel layer on abandoned mounds 259 also suggests that there were no inherent properties of the mound location that prevent fuel 260 accumulation (contrasting with H3) and further supports that tortoises have substantial impacts on fuel 261 for fires in the immediate vicinity of their mounds (supporting H1). 262 Plants Vary Widely in Flammability Traits 263 We found substantial differences in plant ignitability and combustibility across the different 264 functional groups in principal component (PC) space. These differences could be due in part to the 265 composition of their biomass, as entire shoots with leaves attached were used for collecting flammability 266 measurements. In other systems, shoot and leaf flammability are shown to be decoupled (Alam et al., 267 2020). To explain the flammability differences among functional groups, future studies could focus on 268 the properties of leaves and shoots separately to disentangle how traits that are known to influence 269 flammability, such as the lignin composition in shoots or volatile compounds and surface area to volume 270 ratio of leaves, contribute to integrated flammability estimates at the whole plant level (Mason et al., 271 2016). 272 While most functional groups tended to occupy distinct quadrants of PC space, the shrubs 273 included in this study had a positively correlated combustibility and ignitability but varied substantially in 274 13 their overall flammability. The variation in shrub flammability could be explained by how different life 275 history strategies interact with fire. The reseeders Ximenia americana and Palafoxia feayi were the least 276 flammable shrubs in this study. Since reseeders have heavy investments in their aboveground 277 reproductive organs (Pausas & Verdu, 2016), resistance to fire can help preserve these investments and 278 make postfire reproduction possible. In particular, P . feayi has been shown to engage in high 279 reproductive output shortly after fire (Ostertag & Menges, 1994). Moreover, P . feayi has slow postfire 280 growth rates compared to other sandhill plants (Maguire & Menges, 2011), incentivizing them to resist 281 fire rather than compete with plants that recover faster. On the other hand, the clonal resprouters Lyonia 282 ferruginea and Quercus laevis were the most flammable shrubs in this study. Since resprouting shrubs 283 have the competitive advantage after fires and do not invest heavily in the biomass that will be 284 destroyed, promoting fire spread could increase the range and intensity of fires, thereby maximizing the 285 postfire landscape where they outcompete other species. 286 Gopher Tortoises Have Minor Effects on Plant Community Flammability 287 There were no widespread changes in flammability on mounds due to changes in community 288 composition (rejecting H2). There were only two significant differences in the community weighted 289 means of flammability traits on mounds when compared to the surrounding matrix: the plant 290 community had a higher fuel bulk density on active mounds and a shorter burn duration on abandoned 291 mounds. These differences may signify slight changes in community flammability on active and 292 abandoned mounds but are unlikely to have any consequential impact on fire behavior on either mound 293 types. First, active mounds are severely fuel limited; species abundances are so low that any community 294 flammability traits will be overshadowed by the effects of fuel limitation. Second, the effect size of 295 burrow activity status on fire duration (13.2 seconds on abandoned mounds vs 15.2 in matrix) is likely 296 too small to be ecologically meaningful. 297 14 The data presented here could be used to expand the scope of flammability studies to include 298 sandhill elements of the larger biodiversity hostspot of the North American Costal Plain(Noss et al., 299 2015). Models predicting consumption height (Zylstra et al., 2016) and fire severity (Schwilk & Caprio, 300 2011) show that species composition and functional traits respectively greatly improve model accuracy 301 when predicting fire behavior at landscape scales and assessing the relative importance of biotic vs 302 abiotic influences on fire intensity. Further studies can now extend trait-based predictive modelling to 303 the southern ridge sandhill and other comparable ecosystems by using the dataset of species-specific 304 functional traits and flammability we present here alongside community composition measures. 305 Taken together, our results suggest that previous observations of decreased fire intensity around 306 gopher tortoise mounds in sandhill (Kaczor & Hartnett, 1990) arise due to fuel limitation via two 307 pathways: limitation of litter, a key component to fire spread and intensity in similar ecosystems (Schwilk 308 & Caprio, 2011), and reduction in live plant cover available for fire to consume. Together, these patches 309 of sparse fuels may create localized areas of low fire intensity over gopher tortoise mounds. Our findings 310 are consistent with other findings in Florida sandhill (Mushinsky & Esman, 1994) but contrast with 311 findings from other habitat types (Aresco & Guyer, 1999). Since gopher tortoises are found in a variety of 312 ecosystems and at different population densities throughout the southeastern U.S. Coastal Plain, further 313 research is needed before extrapolating effects on surrounding vegetation and decreases in fire intensity 314 to all gopher tortoise mounds. 315

316 Gopher tortoises effects on small-scale fire behavior are likely related to decreases in fuel for 317 fires directly over tortoise mounds This finding provides a mechanism by which tortoises provide fire 318 refugia for hundreds of other small animals (Knapp et al., 2018), as well as broadly demonstrating how 319 our understanding of herbivore-fire interactions should expand to include small-bodied herbivores. 320 15 Implications for fire management could be substantial, particularly as climate change reduces the 321 number of days on which fire can be safely prescribed, and increases the frequency of burning under 322 suboptimal conditions(Kupfer et al., 2022). Further work is needed to understand how small herbivores 323 may alter fire behavior in other pyrogenic ecosystems, particularly in how they create fire refugia for 324 other animals. 325 16

326 Abrahamson, W. G., Abrahamson, C. R., & Keller, M. A. (2021). Lessons from four decades of monitoring 327 vegetation and fire: maintaining diversity and resilience in Florida’s uplands. Ecological 328 Monographs, 91(2), 1–20. https://doi.org/10.1002/ecm.1444 329 Abrahamson, W. G., Johnson, A. F., Layne, J. N., & Peroni, P. A. (1984). Vegetation of the Archbold 330 Biological Station, Florida: an Example of the Southern Lake Wales Ridge. Florida Scientist, 331 47(Autumn), 209–250. 332 Alam, M. A., Wyse, S. V., Buckley, H. L., Perry, G. L. W., Sullivan, J. J., Mason, N. W. H., Buxton, R., 333 Richardson, S. J., & Curran, T. J. (2020). Shoot flammability is decoupled from leaf flammability, but 334 controlled by leaf functional traits. Journal of Ecology, 108(2), 641–653. 335 https://doi.org/10.1111/1365-2745.13289 336 Anderson, H. E. (1970). Forest fuel ignitibility. Fire Technology, 6(4), 312–319. 337 https://doi.org/10.1007/BF02588932 338 Archibald, S., Bond, W. J., Stock, W. D., & Fairbanks, D. H. K. (2005). Shaping the landscape: Fire-grazer 339 interactions in an African savanna. Ecological Applications, 15(1), 96–109. 340 https://doi.org/10.1890/03-5210 341 Aresco, M., & Guyer, C. (1999). Burrow Abandonment by Gopher Tortoises in Slash Pine Plantations of 342 the Conecuh National Forest. Journal of Wildlife Management, 63(1), 26–35. 343 Ashton, K., Engelhardt, B., & Branciforte, B. (2008). Gopher Tortoise ( Gopherus polyphemus ) Abundance 344 and Distribution after Prescribed Fire Reintroduction to Florida Scrub and Sandhill at Archbold 345 Biological Station. 42(3), 523–529. 346 Bruegger, R. A., Varelas, L. A., Howery, L. D., Torell, L. A., Stephenson, M. B., & Bailey, D. W. (2016). 347 17 Targeted Grazing in Southern Arizona: Using Cattle to Reduce Fine Fuel Loads. Rangeland Ecology 348 and Management, 69(1), 43–51. https://doi.org/10.1016/j.rama.2015.10.011 349 Castellón, T. D., Rothermel, B. B., & Bauder, J. M. (2018). Gopher Tortoise Burrow Use, Home Range, 350 Seasonality, and Habitat Fidelity in Scrub and Mesic Flatwoods of Southern Florida. Herpetologica, 351 74(1), 8–21. https://doi.org/10.1655/Herpetologica-D-17-00030.1 352 Davies, K. W., Boyd, C. S., Bates, J. D., & Hulet, A. (2015). Dormant season grazing may decrease wildfire 353 probability by increasing fuel moisture and reducing fuel amount and continuity. International 354 Journal of Wildland Fire, 24(6), 849–856. https://doi.org/10.1071/WF14209 355 Diemer, J. E. (1986). The Ecology and Management of the Gopher Tortoise in the Southeastern United 356 States. Herpetologica, 42(1), 125–133. https://www.jstor.org/stable/3892243 357 Dimitrakopoulos, A. P., & Papaioannou, K. K. (2001). Flammability assessment of Mediterranean forest 358 fuels. Fire Technology, 37(2), 143–152. https://doi.org/10.1023/A:1011641601076 359 Eubanks, J. O., Michener, W. K., & Guyer, C. (2003). Patterns of movement and burrow use in a 360 population of gopher tortoises (Gopherus polyphemus). Herpetologica, 59(3), 311–321. 361 https://doi.org/10.1655/01-105.1 362 Hansen, K. (1963). THE BURROW OF THE GOPHER TORTOISE. Quarterly Journal of the Florida Academy 363 of Sciences, 26(4), 353–360. 364 Hempson, G. P., Archibald, S., Donaldson, J. E., & Lehmann, C. E. R. (2019). Alternate Grassy Ecosystem 365 States Are Determined by Palatability–Flammability Trade-Offs. Trends in Ecology and Evolution, 366 34(4), 286–290. https://doi.org/10.1016/j.tree.2019.01.007 367 Hobbs, R. J., & Mooney, H. A. (1985). Community and population dynamics of serpentine grassland 368 18 annuals in relation to gopher disturbance. Oecologia, 67(3), 342–351. 369 https://doi.org/10.1007/BF00384939 370 Howell, H. J., Rothermel, B. B., White, K. N., & Searcy, C. A. (2020). Gopher Tortoise Demographic 371 Responses to a Novel Disturbance Regime. Journal of Wildlife Management, 84(1), 56–65. 372 https://doi.org/10.1002/jwmg.21774 373 Huntly, N., & Reichman, O. J. (1994). Effects of subterranean mammalian herbivores on vegetation. 374 Journal of Mammalogy, 75(4), 852–859. https://doi.org/10.2307/1382467 375 Johnson, V. M., Guyer, C., Hermann, S. M., Eubanks, J., & Michener, W. K. (2009). Patterns of dispersion 376 and burrow use support scramble competition polygyny in gopherus polyphemus. Herpetologica, 377 65(2), 214–218. https://doi.org/10.1655/08-029R.1 378 Jones, J. C., & Dorr, B. (2004). Habitat associations of gopher tortoise burrows on industrial timberlands. 379 Wildlife Society Bulletin, 32(2), 456–464. https://doi.org/10.2193/0091-380 7648(2004)32[456:haogtb]2.0.co;2 381 Kaczor, S., & Hartnett, D. (1990). Gopher Tortoise ( Gopherus polyphemus ) Effects on Soils and 382 Vegetation in a Florida Sandhill Community. The American Midland Naturalist, 123(1), 100–111. 383 Karp, A. T., Koerner, S. E., Hempson, G. P., Abraham, J. O., Anderson, T. M., Bond, W. J., Burkepile, D. E., 384 Fillion, E. N., Goheen, J. R., Guyton, J. A., Kartzinel, T. R., Kimuyu, D. M., Mohanbabu, N., Palmer, T. 385 M., Porensky, L. M., Pringle, R. M., Ritchie, M. E., Smith, M. D., Thompson, D. I., … Staver, A. C. 386 (2024). Grazing herbivores reduce herbaceous biomass and fire activity across African savannas. 387 Ecology Letters, 27(6), 1–13. https://doi.org/10.1111/ele.14450 388 Kinlaw, A., & Grasmueck, M. (2012). Evidence for and geomorphologic consequences of a reptilian 389 ecosystem engineer: The burrowing cascade initiated by the Gopher Tortoise. Geomorphology, 390 19 157–158, 108–121. https://doi.org/10.1016/j.geomorph.2011.06.030 391 Knapp, D. D., Howze, J. M., Murphy, C. M., Dziadzio, M. C., & Smith, L. L. (2018). Prescribed fire affects 392 diurnal vertebrate use of Gopher Tortoise (Gopherus polyphemus) burrows in a Longleaf Pine 393 (Pinus palustris) forest. Herpetological Conservation and Biology, 13(3), 551–557. 394 Kupfer, J. A., Lackstrom, K., Grego, J. M., Dow, K., Terando, A. J., & Hiers, J. K. (2022). Prescribed fire in 395 longleaf pine ecosystems: fire managers’ perspectives on priorities, constraints, and future 396 prospects. Fire Ecology, 18(1). https://doi.org/10.1186/s42408-022-00151-6 397 Kurek, P., Kapusta, P., & Holeksa, J. (2014). Burrowing by badgers (Meles meles) and foxes (Vulpes 398 vulpes) changes soil conditions and vegetation in a European temperate forest. Ecological 399 Research, 29(1), 1–11. https://doi.org/10.1007/s11284-013-1094-1 400 Lau, A., & Dodd, C. K. (2015). Multiscale burrow site selection of gopher tortoises (Gopherus 401 polyphemus) in coastal sand dune habitat. Journal of Coastal Research, 31(2), 305–314. 402 https://doi.org/10.2112/jcoastres-d-12-00201.1 403 Layne, J. N. (1989). Comparison of survival rates and movements of relocated and resident gopher 404 tortoises in a south-central Florida population. In Gopher tortoise relocation symposium 405 proceedings (pp. 73–79). 406 Lloyd, R. B., Henning, J. A., & Chupp, A. D. (2023). Effects of Gopher Tortoise (Gopherus polyphemus) 407 Exclusion on Plant Assemblages in a Longleaf Pine Forest. Journal of Herpetology, 57(4), 367–372. 408 https://doi.org/10.1670/22-067 409 Maguire, A. J., & Menges, E. S. (2011). Post-fire growth strategies of resprouting Florida scrub 410 vegetation. Fire Ecology, 7(3), 12–25. https://doi.org/10.4996/fireecology.0703012 411 20 Mason, N. W. H., Frazao, C., Buxton, R. P., & Richardson, S. J. (2016). Fire form and function : evidence 412 for exaptive flammability in the New Zealand flora. Plant Ecology, 217(6), 645–659. 413 https://doi.org/10.1007/s11258-016-0618-5 414 Menges, E. S., & Kohfeldt, N. (1995). Life History Strategies of Florida Scrub Plants in Relation to Fire. 415 Bulletin of the Torrey Botanical Club, 122(4), 282–297. 416 Mushinsky, H. R., & Esman, L. A. (1994). Perceptions of gopher tortoise burrows over time. Florida Field 417 Naturalist, 22(1), 1–7. http://www.fosbirds.org/FFN/PDFs/FFNv22n1p1-7Mushinsky.pdf 418 Noss, R. F., Platt, W. J., Sorrie, B. A., Weakley, A. S., Means, D. B., Costanza, J., & Peet, R. K. (2015). How 419 global biodiversity hotspots may go unrecognized: Lessons from the North American Coastal Plain. 420 Diversity and Distributions, 21(2), 236–244. https://doi.org/10.1111/ddi.12278 421 Ostertag, R., & Menges, E. S. (1994). Patterns of Reproductive Effort with Time since Last Fire in Florida 422 Scrub Plants. 5(3), 303–310. 423 Pausas, J. G., & Verdu, M. (2016). Plant Persistence Traits in Fire-Prone Ecosystems of the 424 Mediterranean Basin : A Phylogenetic Approach. Oikos, 109(1), 196–202. 425 Peeples, M. A. (2011). R Script for K-Means Cluster Analysis. http://www.mattpeeples.net/kmeans.htm 426 Pyne, S. (1996). Introduction to Wildland Fire (2nd ed., Vol. 11, Issue 1). 427 R Core Team (2024). R: A language and environment for statistical computing. R Foundation for 428 Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. 429 Richardson, J. C., & Stiling, P. (2019). Gopher tortoise herbivory increases plant species richness and 430 diversity. Plant Ecology, 220(3), 383–391. https://doi.org/10.1007/s11258-019-00921-4 431 21 Schwilk, D. W., & Caprio, A. C. (2011). Scaling from leaf traits to fire behaviour: community composition 432 predicts fire severity in a temperate forest. British Ecological Society, 99(4), 970–980. 433 https://doi.org/10.1111/j 434 Simpson, K. J., Ripley, B. S., Christin, P. A., Belcher, C. M., Lehmann, C. E. R., Thomas, G. H., & Osborne, C. 435 P. (2016). Determinants of flammability in savanna grass species. Journal of Ecology, 104(1), 138–436 148. https://doi.org/10.1111/1365-2745.12503 437 Zylstra, P., Bradstock, R. A., Bedward, M., Penman, T. D., Doherty, M. D., Weber, R. O., Gill, A. M., & 438 Cary, G. J. (2016). Biophysical mechanistic modelling quantifies the effects of plant traits on fire 439 severity: Species, not surface fuel loads, determine flame dimensions in eucalypt forests. PLoS 440 ONE, 11(8), 1–24. https://doi.org/10.1371/journal.pone.0160715 441 442 22 Figures 443 444 Figure 1. Map of Florida (left), Archbold Biological Station (center), and burn units 2A and 2C with 445 locations of gopher tortoise mounds and bare patches sampled for this study (rightmost panel). Imagery 446 was sourced from Bing Virtual Earth 2025 . 447 448 23 449 450 Figure 2 : Mean percentage of litter (A) and plant (B) cover are greater over abandoned mounds than all 451 other types and similar to the surrounding matrix. The jittered percentages of litter or plant cover in 1x1 452 m quadrats on each mound type are displayed along with their means and standard errors. For the 453 surrounding matrix, each point is an average of all off-mound 1x1 plots at each location. Means that are 454 significantly different at the α = 0.01 level have different letters. 455 456 24 457 Figure 3 : Mean percentages of plant and litter cover are significantly lower over active and inactive 458 gopher tortoise mounds, but not abandoned mounds. For active, inactive, and abandoned mounds, 459 there was no significant trend in plant or litter cover once off of the mound. The jittered percentages of 460 litter plant cover in 1x1 m quadrats at varying distances from mound and bare patch centers are 461 25 displayed along with their means. The mean percentage of off-mound plant and litter cover is shown in 462 red. Means that are significantly different at the α = 0.01 level have different letters. 463 464 26 465 Figure 4. Scores of each plant species plotted on the first two principal component axes. Vectors (red 466 arrows) of flammability traits are overlaid, with their directions specified by their loadings on each axis. 467 Vector magnitudes have been increased by 5 times their original values to aid their representation. Refer 468 to supporting information for full species names. 469 470 27 Tables 471 Table 1: Comparisons on mounds vs the surrounding plant matrix using an asymptotic Z test from a beta 472 GLMM. All mound types except for abandoned have significantly less plant and litter cover over mounds. 473 All tests were performed at the 0.01 level. 474 475 476 Plant Cover Type Sample size on Estimate on with 95% CI Sample size off Estimate off with 95% CI Statistic p-value bare 20 0.183 (0.125, 0.259) 383 0.718 (0.697, 0.739) Z = -10.313 P < 0.001 active 11 0.499 (0.370, 0.629) 193 0.769 (0.742, 0.795) Z = -4.294 P < 0.001 inactive 11 0.414 (0.298, 0.540) 202 0.752 (0.724, 0.779) Z = -5.447 P < 0.001 abandoned 8 0.704 (0.555, 0.819) 145 0.793 (0.742, 0.836) Z = -1.503 P = 0.133 Litter Cover Type Sample size on Estimate on with 95% CI Sample size off Estimate off with 95% CI Statistic p-value bare 20 0.243 (0.162, 0.349) 383 0.740 (0.711, 0.767) Z = -7.976 P < 0.001 active 11 0.279 (0.161, 0.438) 193 0.771 (0.722, 0.813) Z = -5.920 P < 0.001 inactive 11 0.297 (0.176, 0.457) 202 0.731 (0.692 0.766) Z = -5.089 P < 0.001 abandoned 8 0.552 (0.344, 0.743) 145 0.747 (0.670, 0.812) Z = -2.100 P = 0.036 28 Table 2. Loadings for the first 4 principal components of a principal components analysis on the 477 flammability traits of plants. The percentage of total variability explained by each component is 478 displayed next to each component. 479 Measure Component 1 (29.19%) Component 2 (21.40%) Component 3 (12.73%) Component 4 (10.41%) Bulk Density (g per m ) 0.44 0.29 0.145 0.087 Minimum Temperature (°C) 0.162 0.47 0.015 -0.124 Maximum Temperature (°C) 0.443 0.156 0.195 -0.027 % Burned (% of mass) 0.366 -0.113 -0.537 0.212 Loss Rate (g per sec) 0.321 -0.243 -0.036 0.133 Dry Matter Content (% of mass) 0.326 0.092 -0.034 -0.586 Hours to Dry -0.208 0.523 -0.097 0.033 Heat of Combustion (Cal per g) 0.392 -0.011 0.12 -0.032 Moisture at Ignition (% of mass) -0.188 0.385 -0.351 -0.359 Hours to Ignition -0.013 0.403 0.24 0.611 Duration per Gram (sec) 0.096 0.083 -0.668 0.257 480