Spatial heterogeneity across five rangelands managed with pyric‐herbivory Authors: Devan A. McGranahan, David M. Engle, Samuel D. Fuhlendorf, Steven J. Winter, James R. Miller, & Diane M. Debinski This is the peer reviewed version of the following article: see full citation below, which has been published in final form at https://doi.org/10.1111/j.1365-2664.2012.02168.x. This article may be used for non-commercial purposes in accordance with Wiley Terms and Conditions for Self- Archiving. Devan A. McGranahan, David M. Engle, Samuel D. Fuhlendorf, Steven J. Winter, James R. Miller, & Diane M. Debinski. "Spatial Heterogeneity Across Five Rangelands Managed with Pyric- Herbivory" Journal of Applied Ecology Vol. 49 Iss. 4 (2012) p. 903 - 910, DOI: 10.1111/ j.1365-2664.2012.02168.x Made available through Montana State University’s ScholarWorks scholarworks.montana.edu Spatial heterogeneity across five rangelands managed with pyric-herbivory 1 Devan Allen McGranahan (Corresponding author) Environmental Studies Department, The 2 University of the South, mcgranah@alumni.grinnell.edu 3 Correspondence: 735 University Avenue, The University of the South, Sewanee, TN 37375. e-4 mail: mcgranah@alumni.grinnell.edu; phone: 515.708.5148; FAX: 931.598.1145 5 David M. Engle, Department of Natural Resource Ecology and Management, Oklahoma State 6 University, david.engle@okstate.edu 7 Samuel D. Fuhlendorf, Department of Natural Resource Ecology and Management, Oklahoma 8 State University, sam.fuhlendorf@okstate.edu 9 Stephen J. Winter, United States Fish and Wildlife Service, Winona, MN 10 James R. Miller, Department of Natural Resources and Environmental Sciences, University of 11 Illinois, jrmillr@illinois.edu 12 Diane M. Debinski, Department of Ecology Evolution and Organismal Biology, Iowa State 13 University, debinski@iastate.edu 14 Running title: Spatial heterogeneity and pyric-herbivory 15 Total word count: 6,202 16 Summary: 363 17 Main text:3,773 18 Acknowledgements:126 19 References:1,339 20 Legends:278 21 22 One (1) table and three (3) figures. Number of references: 5323 2 24 Summary 25 1. Many rangelands evolved under an interactive disturbance regime in which grazers 26 respond to the spatial pattern of fire and create a patchy, heterogeneous landscape. 27 Spatially heterogeneous fire and grazing create heterogeneity in vegetation structure at 28 the landscape level (patch contrast) and increase rangeland biodiversity. We analysed five 29 experiments comparing spatially heterogeneous fire treatments to spatially homogeneous 30 fire treatments on grazed rangeland along a precipitation gradient in the North American 31 Great Plains. 32 2. We predicted that, across the precipitation gradient, management for heterogeneity 33 increases both patch contrast and variance in the composition of plant functional groups. 34 Furthermore, we predicted that patch contrast is positively correlated with variance in 35 plant functional group composition. Because fire spread is important to the fire–grazing 36 interaction, we discuss factors that reduce fire spread and reduce patch contrast despite 37 management for heterogeneity. 38 3. We compared patch contrast across pastures managed for heterogeneity and pastures 39 managed for homogeneity with a linear mixed-effect (LME) regression model. We used 40 the LME model to partition variation in vegetation structure to each sampled scale so that 41 a higher proportion of variation at the patch scale among pastures managed for 42 heterogeneity indicates patch contrast. To examine the relationship between vegetation 43 structure and plant community composition, we used constrained ordination to measure 44 variation in functional group composition along the vegetation structure gradient. We 45 3 used the meta-analytical statistic, Cohen’s d, to compare effect sizes for patch contrast 46 and plant functional group composition. 47 4. Management for heterogeneity increased patch contrast and increased the range of plant 48 functional group composition at three of the five experimental locations. 49 5. Plant functional group composition varied in proportion to the amount of spatial 50 heterogeneity in vegetation structure on pastures managed for heterogeneity. 51 6. Synthesis and applications. Pyric-herbivory management for heterogeneity created patch 52 contrast in vegetation across a broad range of precipitation and plant community types, 53 provided that fire was the primary driver of grazer site selection. Management for 54 heterogeneity did not universally create patch contrast. Stocking rate and invasive plant 55 species are key regulators of heterogeneity, as they determine the influence of fire on the 56 spatial pattern of fuel, vegetation structure and herbivore patch selection, and therefore 57 also require careful management. 58 59 Keywords: Biodiversity conservation, Fire–grazing interaction, Grazing management, 60 Heterogeneity, Patch contrast, Pyric-herbivory, Working landscapes 61 62 4 63 Introduction 64 Many rangelands worldwide are working landscapes managed to meet economic goals as 65 well as biological goals (Polasky et al. 2005; Ellis & Ramankutty 2008). When economic 66 objectives take precedence, rangeland biodiversity is imperilled, such as when rangeland is 67 converted to cropland or overgrazed by livestock (Samson & Knopf 1994; Fuhlendorf & Engle 68 2001; O’Connor et al. 2010). Moreover, conventional rangeland management promotes 69 spatially-uniform, moderate grazing and the homogeneous removal of biomass by grazers at the 70 pasture scale (Holechek, Pieper, & Herbel 2003) even though uniform moderate grazing 71 degrades habitat quality and contributes to the decline of rangeland biodiversity (Fuhlendorf & 72 Engle 2001; Derner et al. 2009). 73 Many rangelands evolved under patchy disturbance regimes that vary in frequency and 74 intensity across multiple spatial scales (Fuhlendorf & Smeins 1999), therefore, reconciling 75 conservation and agricultural production in rangeland probably depends upon heterogeneity-76 based management analogous to historical patterns of disturbance (Fuhlendorf & Engle 2001). 77 Heterogeneity is an important driver of biodiversity and an essential component of conservation 78 in ecosystems worldwide (Ostfeld et al. 1997). Although heterogeneity consists of many 79 ecosystem attributes, we apply the concept of patch contrast, which describes the degree of 80 difference between patches of otherwise similar properties (Kotliar & Wiens 1990). Patch 81 contrast is a useful concept for rangeland heterogeneity because many rangelands evolved under 82 a shifting mosaic of fire and grazing, in which grazing is concentrated on the most recently-83 burned portions of the landscape in response to the high-quality forage that grows after fire and 84 5 focal grazing (Archibald & Bond 2004; Allred, Fuhlendorf, Engle, et al. 2011). Patch contrast is 85 created as grazers and vegetation respond to the pattern of fire in the landscape (Adler, Raff, & 86 Lauenroth 2001). This fire–grazing interaction – or pyric-herbivory – is an ecological 87 disturbance that differs from the effects of fire and grazing alone (Fuhlendorf et al. 2009). 88 When applied in a management context as patch burn–grazing, pyric-herbivory supports 89 rangeland biodiversity by increasing the diversity of habitat types, ranging from low stature 90 grazing lawns in recently-burned patches to tall, mature plants in patches unburned for several 91 years (Fuhlendorf & Engle 2004; Winter et al. 2012). Such differences in vegetation structure are 92 driven by the pattern of grazing as well as by differential plant responses to the fire–grazing 93 interaction among patches: the relative abundance of plant functional groups varies across 94 patches according to the length of time since a patch was burned (Fuhlendorf et al. 2006; Winter 95 et al. 2012). Again, patch contrast is a useful term to describe heterogeneity among patches 96 because habitat diversity reflects the degree of difference in vegetation structure among 97 rangeland patches (Fuhlendorf et al. 2006; Coppedge et al. 2008). 98 Heterogeneity clearly benefits biodiversity on rangeland, but universal efficacy of the 99 fire-grazing interaction is less clear. We use vegetation structure and plant functional group 100 composition data from five experiments that compare management for heterogeneity (pyric-101 herbivory) with management for homogeneity (grazing with homogeneous fire regimes).The five 102 experimental locations span several gradients, including precipitation and plant community type 103 and land-use history. Given that evidence supporting an operative fire-grazing interaction has 104 been demonstrated in a breadth of ecosystems worldwide (Allred, Fuhlendorf, Engle, et al. 105 2011), we did not expect the strength of the fire–grazing interaction to vary across the ecological 106 gradient (plant community types and precipitation). However, because invasive species and 107 6 intense grazing both influence fuel load and continuity, which in turn affect fire spread (Davies 108 et al. 2010; McGranahan et al. 2012), we had reason to believe invasive species and intense 109 grazing might reduce the strength of the fire–grazing interaction. 110 In this study, we test the following hypotheses using comparable data from five 111 experiments: 1. Patch contrast is greater in rangeland managed for heterogeneity when compared 112 to rangeland managed for homogeneity; 2. Heterogeneity-based management increases variance 113 in the composition of plant functional groups; and 3. Patch contrast is positively correlated with 114 variance in plant functional group composition. We found that patch contrast was associated with 115 variance in plant functional group composition and that management for heterogeneity created 116 variation in vegetation structure. However, management for heterogeneity did not universally 117 create patch contrast across our five study locations. Stocking rate and invasive plant species 118 appear to regulate patch contrast more than primary productivity despite the precipitation 119 gradient and differences in plant communities across our study locations. 120 121 7 122 Methods 123 Study locations 124 To compare the effect of spatially heterogeneous and spatially homogeneous fire regimes 125 on grazed rangeland, we combined vegetation structure and plant functional group composition 126 data from five experimental locations in central North America that span circa 650 km from 127 mixed prairie in the southwest to eastern tallgrass prairie in the northeast (Table 1). The five 128 locations include: Hal and Fern Cooper Wildlife Management Area, Woodward County, 129 Oklahoma; Marvin Klemme Range Research Station, Washita County, Oklahoma; Oklahoma 130 State University Range Research Station, Paine County, Oklahoma; Tallgrass Prairie Preserve, 131 Osage County, Oklahoma; and the Grand River Grasslands, Ringgold County, Iowa. While each 132 experiment was established independently, similarity of experimental design, treatment structure, 133 and data collected provide the opportunity to test for a connection between heterogeneity-based 134 management and actual heterogeneity in vegetation across a broad geographic area. 135 Data 136 We used vegetation structure and plant functional group composition data from each of 137 the five locations. Data were similar across all locations. Appendix S1 in Supporting Information 138 includes detailed accounts of the types of data and their specific collection methodologies. At 139 each location, cattle (Bos taurus) were stocked continuously during the grazing season on all 140 pastures and cattle were allowed unrestricted access to grazing and water within each pasture, 141 without interior fencing. Across all five locations, vegetation structure was quantified with visual 142 obstruction measurements, which combine vegetation height and vegetation density (Harrell & 143 8 Fuhlendorf 2002). Visual obstruction methods used in this study include visual obstruction 144 reading (Robel et al. 1970) and angle of obstruction (Kopp et al. 1998). 145 Plant functional group data were collected once each year at each location. Canopy cover 146 estimations follow the Daubenmire (1959) cover class index at all but the Cooper location, where 147 canopy cover was estimated to the nearest five per cent. While sampling periods varied slightly 148 across locations (see Appendix S1), the timing of the sampling periods was consistent from year 149 to year within each location. Sampling at each location followed a nested hierarchical design in 150 which pastures were divided into patches and patches were divided into transects. Sampling 151 points were randomly located along transects to measure visual obstruction and plant functional 152 group canopy cover (sampling points were located within avian point count areas rather than 153 along transects at the Tallgrass Prairie Preserve). 154 Data analysis 155 Spatial heterogeneity in vegetation structure.—To compare spatial heterogeneity in vegetation 156 structure (patch contrast) across heterogeneously-managed and homogeneously-managed 157 rangeland, we used a linear mixed-effect (LME) regression model to determine the proportion of 158 variance in vegetation structure attributable to each sampled spatial extent, and compared the 159 average proportion of variance in the patch term across treatments within each location (Winter 160 et al. 2012). We created an LME regression model with an intercept-only fixed-effect term (+1) 161 and a random-effect term that included the spatial extents that were sampled in common to each 162 location – sampling point, patch, and pasture – and a year factor to account for repeated 163 measures using the lmer function in the lme4 package for the R statistical environment (Bates & 164 Maechler 2010; R Development Core Team 2011). Because of the hierarchical and annually-165 repeated design common to all five experiments, the random-effect term for each location was 166 9 fully crossed to account for statistical interactions between sampled spatial extents and time. 167 Variance estimates were returned for each factor in the random-effect term plus an additional 168 residual error factor (Baayen, Davidson, & Bates 2008). We calculated the proportion of 169 variance contributed by each factor by applying the sum of the variance estimations as a divisor 170 to each factor’s original variance estimate. The LME model was applied to each pasture within 171 each location. 172 We tested for a difference in mean proportion variance in vegetation structure to compare 173 pastures managed for heterogeneity and homogeneity within each location using a Student’s t 174 test in the R stats package. A significantly greater proportion of variance in the patch term for 175 pastures managed for heterogeneity within a location indicates that heterogeneity-based 176 management created patch contrast in vegetation structure within these pastures. 177 Spatial heterogeneity in plant functional group composition.—To test the hypothesis that 178 management for heterogeneity increases variance in plant functional group composition, we first 179 calculated the range of plant functional group composition in constrained ordination space. We 180 specified vegetation structure as the constrained axis in a redundancy analysis (RDA) of plant 181 functional group data for each location, and calculated the range of values, or site scores, along 182 the RDA constrained axis for each pasture. Redundancy analysis is a constrained ordination that 183 calculates variation in multivariate data with respect to a priori constraints (Ter Braak 1986; 184 Oksanen et al. 2011). This method allowed us to compare variation in plant functional group 185 composition with specific reference to the vegetation structure gradient, specified as RDA axis 1 186 (RDA1). We used the rda function in the vegan package for the R statistical environment 187 (Oksanen et al. 2011). 188 10 We scaled RDA1 output to allow the comparison of ordination results across all 189 locations. The overall range of possible variation in each ordination varied by location because a 190 separate ordination was performed for each location, and each ordination was based on the 191 specific plant functional groups measured at each location (see Appendix S1). Thus, prior to 192 further analysis, we combined RDA1 site scores into a single dataset and scaled the data to create 193 a standardized distribution that allows comparison across locations. 194 The range of site scores for a given pasture along RDA1 represents the variation in plant 195 functional group composition, as pastures with a greater range of functional group composition 196 span a larger range of site scores along RDA1. We tested for a difference in the mean range of 197 RDA1 scores to compare pastures managed for heterogeneity and homogeneity within each 198 location using a Student’s t test in the R stats package. Again, a significantly greater range for 199 pastures managed for heterogeneity within a location indicates that heterogeneity-based 200 management created variance in plant functional group composition within these pastures. 201 Calculating effect sizes.—We used a meta-analytical statistic to compare the effect of 202 heterogeneity-based management on patch contrast and plant functional group composition 203 across all five locations. Effect size statistics use a single value to quantify the difference 204 between two replicated groups by comparing the mean and variance of each group (Harrison 205 2011). Effect size has been used elsewhere to compare the effect of ecological management 206 across studies testing common hypotheses (Côté & Sutherland 1997). Here, the greater the effect 207 size for a location, the more pronounced the difference between response variables among 208 pastures managed for heterogeneity compared to pastures managed for homogeneity. We 209 calculated the meta-analysis statistic Cohen’s d (Cohen 1977) for each response variable, 210 11 proportion variance and range of RDA1 scores, to determine effect size with the following 211 formula: 212 d = ( µhet – µhom) / √ ( σmean), 213 In which µhet and µhom represent the mean value of the response variables in pastures 214 managed for heterogeneity and homogeneity, respectively, and σmean represents the mean 215 standard deviation of each response variable. Using the R statistical environment, we estimated 216 95% confidence intervals with a two-part iterative re-sampling algorithm. First, a sampling 217 distribution for each Cohen’s d was generated by 1000 simulations of each treatment groups’ 218 mean and standard deviation. Second, the calculated Cohen’s d was compared to the generated 219 sample distribution with 9999 iterations at alpha = 0.05 to generate the 95% confidence interval. 220 To test our third prediction that patch contrast is positively correlated with variance in 221 plant functional group composition, we plotted the patch contrast effect size against the plant 222 community composition effect size and calculated a correlation coefficient using Kendall’s Τ, a 223 non-parametric test for association between two variables based on similarity of rank (Kendall 224 1938). 225 226 12 227 Results 228 Management for heterogeneity increased patch contrast at three of the five experimental 229 locations used in this study (Cooper, Stillwater, and the TGPP) (Fig. 1). At two locations, 230 Klemme and the GRG, management for heterogeneity did not increase spatial heterogeneity in 231 vegetation structure compared to management for homogeneity, and thus did not create patch 232 contrast. 233 At Klemme and the GRG, variance in vegetation structure among pastures managed for 234 heterogeneity was lower and variance in vegetation structure among pastures managed for 235 homogeneity was higher than at Cooper, Stillwater, and the TGPP. In other words patch-level 236 variation was neither as great as expected on pastures managed for heterogeneity at Klemme and 237 the GRG, nor was patch-level variation as low as expected on pastures managed for homogeneity 238 at these two locations. 239 Management for heterogeneity increased the variance in plant functional group 240 composition at two of the five locations (Cooper and the TGPP) (Fig. 2). An outlier among 241 pastures managed for homogeneity at Stillwater increased the variation around the mean such 242 that, despite generally higher variance in plant functional group composition among pastures 243 managed with heterogeneity, the difference was not significant (P = 0.08). As above, there was 244 no difference between pastures managed for heterogeneity and those managed for homogeneity 245 at Klemme and the GRG. 246 13 Calculated effect sizes for patch contrast and variance in plant functional group 247 composition were positive for both measures at all five locations, but at only three locations 248 (Cooper, Stillwater, and the TGPP) was Cohen’s d significantly non-zero based on estimated 249 95% confidence intervals (Fig. 3). This trend was consistent for both patch contrast and variance 250 in plant functional group composition. In no instance did management for heterogeneity produce 251 a negative effect size in relation to management for homogeneity. The positive association 252 between patch contrast and variance in plant functional group composition (Τ = 0.40) indicated 253 that the amount of spatial heterogeneity in vegetation structure on pastures managed for 254 heterogeneity generally varied in proportion with plant functional group composition. 255 Notably, differences in patch contrast and plant functional group composition were 256 associated with neither environmental factors along the geographic gradient, nor with differences 257 in management, including pasture size, number of patches, or fire regime (Table 1). For example, 258 pastures managed for heterogeneity at the most arid location in the mixed-grass prairie (Cooper), 259 and in two of the three mesic, tallgrass prairie locations (Stillwater and TGPP) had significant 260 patch contrast compared to pastures managed for homogeneity. Thus, whether patch contrast 261 followed management for heterogeneity was independent of climate and vegetation type. 262 Likewise, pasture area did not appear to affect whether patch contrast followed management for 263 heterogeneity, as the area of pastures at Stillwater was similar to the area of pastures at Klemme 264 and the GRG. Historical stocking rate, however, was associated with differences in patch 265 contrast: only Klemme and the GRG were stocked heavily prior to the beginning of the 266 experiments (Table 1), and management for heterogeneity at these locations did not create patch 267 contrast compared to management for homogeneity. 268 269 14 270 Discussion 271 We found that management for heterogeneity applied through patch-burn grazing 272 increased patch contrast and increased the variance in plant functional group composition at 273 three of the five locations. Overall, patch contrast increased with variance in plant functional 274 group composition. Whether management for heterogeneity created patch contrast was 275 unaffected by precipitation, vegetation type, primary productivity, pasture area, patch area or 276 number of patches per pasture (Table 1), which is congruous with previous work noting the 277 range of ecosystems in which the fire–grazing interaction has been reported (Allred, Fuhlendorf, 278 Engle, et al. 2011). At the same time, the fact that heterogeneity-based management did not 279 universally create patch contrast underscores the fundamental link between fire and grazing in 280 pyric-herbivory. 281 Pyric-herbivory – the unique ecological disturbance created by the fire–grazing 282 interaction – depends upon fire to influence grazing behaviour such that both grazing and 283 vegetation respond to the spatial pattern of fire (Fuhlendorf et al. 2009). However, our results 284 clearly indicate that the influence of fire on the pattern of grazing and vegetation in the landscape 285 is weak unless fire and grazing function as an interacting disturbance. A universal response to 286 pyric-herbivory requires the pattern of fire in the landscape to influence vegetation structure and 287 grazing behaviour and create a contrast between patches that attract grazing (magnet patches) 288 and patches that deter grazing (deterrent patches). However, the influence of fire is weak if it 289 fails to override other environmental factors that contribute to grazer selectivity at the landscape 290 level (Adler, Raff, & Lauenroth 2001; Allred, Fuhlendorf, & Hamilton 2011). 291 15 Grazing followed the spatial pattern of fire and created patch contrast at three of our five 292 locations, but heterogeneity-based management failed to couple fire and grazing into an 293 interacting disturbance at two locations. We attribute the lack of a fire–grazing interaction at 294 Klemme and the GRG to poor fire spread in the burned patches created by a history of 295 overgrazing at each location and invasive plant species that modified the fuelbed in the GRG. 296 Severe grazing in years preceding fire reduces fire spread by reducing the fuel load and creating 297 gaps in the fuelbed (Kerby, Fuhlendorf, & Engle 2007; Davies, Svejcar, & Bates 2009; Leonard, 298 Kirkpatrick, & Marsden-Smedley 2010; Davies et al. 2010). At Klemme and the GRG, stocking 299 rates prior to experimental treatment were much greater than pre-treatment stocking rates at 300 Cooper, Stillwater and the TGPP (Table 1). Heavy grazing reduced fuel loading, which reduced 301 fire spread. As such, subsequent grazing preference was not determined by pyric-herbivory but 302 rather by environmental variability at spatial scales other than the burned patches– e.g., areas 303 close to water, shade, or patches of preferred forage species (Senft et al. 1987; Bailey et al. 304 1996). 305 Overstocking contributed to reduced fuel load in the GRG, but discontinuity in the 306 fuelbed appears to have been caused not by gaps of bare ground but by an abundance of invasive 307 tall fescue (Schedonorus phoenix (Scop.) Holub). Tall fescue creates a barrier to fire spread: 308 during the conventional prescribed burning period, live fuel moisture content in tall fescue 309 exceeds that required to sustain fire spread (McGranahan et al. 2012). In the GRG, grazing 310 reduced accumulated dead fuel and increased proportion of live tall fescue in the fuelbed, which 311 thereby reduced fire spread (McGranahan 2011). 312 Our multivariate method for determining variance in plant functional groups 313 accommodated functional group classifications for each location. This approach is both flexible 314 16 in combining data from individual experiments into a comparative analysis and allowed for 315 insight into the role specific plant functional groups play in the fire–grazing interaction. For 316 example, Cooper had the greatest shrub component in the vegetation, and patch contrast at this 317 location is likely due to the adaptation of the dominant shrub, sand sagebrush (Artemisia filifolia 318 Torr.), to quickly resprout after fire (Winter et al. 2011). At the other end of the productivity 319 gradient, management for heterogeneity failed to create patch contrast in the GRG, which had a 320 much lower abundance of native plant species (Pillsbury et al. 2011) than the other tallgrass 321 prairie locations, which were not only relatively free of invasive plant species but were 322 dominated by native plants (Fuhlendorf & Engle 2004; Fuhlendorf et al. 2006). Given that patch 323 contrast increases with variance in plant functional group composition (Fig. 3), native plant 324 species with an evolutionary history of pyric-herbivory are likely important in ensuring that 325 management for heterogeneity achieves the desired outcomes. 326 The long-term legacy effect of historical management as regulators of pyric-herbivory are 327 not known, although recent data from Klemme suggest that when stocking rate is moderated, 328 plant productivity recovers, fuel load and fuel continuity increase, and fire drives spatial pattern 329 of grazing (Limb et al. 2011). For the period examined in this study, Klemme had a diverse 330 composition of plant functional groups despite low patch contrast, which is probably due to 331 spatially-heterogeneous grazing driven by environmental factors other than fire, because the 332 influence of fire was small (Adler et al. 2001). In the GRG, however, both patch contrast and the 333 range of plant functional group composition were slight, probably due to the great abundance of 334 tall fescue on historically severely stocked pastures (McGranahan 2011). Thus, restoration of 335 pyric-herbivory at Klemme probably depends primarily on the recovery of plant productivity, but 336 17 recovery for overstocking and invasive species control may be required before pyric-herbivory 337 can be fully restored to the GRG. 338 The five rangeland locations included here used domestic cattle Bos taurus as grazers, 339 reflecting the fact that native herbivores have largely been extirpated from central North 340 American rangelands and cattle ranching is the predominant use of many rangelands worldwide. 341 Even in ecosystems where native herbivores persist, the natural fire regimes of many rangelands 342 have been substantially altered. However, domestic livestock and prescribed fire can re-create 343 the pre-historic mosaic: evidence from the North American tallgrass prairie suggests the 344 conservation value of cattle might be analogous to that of bison Bison bison, the dominant native 345 herbivore, in heterogeneous landscapes managed with fire (Towne, Hartnett, & Cochran 2005; 346 Allred, Fuhlendorf, & Hamilton 2011). Management for heterogeneity has been shown to 347 increase the diversity of invertebrates, small mammals, large ungulates and birds in several 348 ecosystems worldwide (Archibald & Bond 2004; Fuhlendorf et al. 2006; Bouwman & Hoffman 349 2007; Coppedge et al. 2008; Engle et al. 2008; Fuhlendorf et al. 2009; Doxon et al. 2011). 350 Moreover, patch burn-grazing is an agriculturally-productive management practice in working 351 rangeland grazed by cattle (Limb et al. 2011). 352 353 18 354 Conclusion 355 Our results demonstrate that management for heterogeneity using patch burn-grazing 356 does not universally create patch contrast in rangelands. Rather, patch burn-grazing creates patch 357 contrast only if fire is the primary driver of grazer site selection across the landscape. The level 358 of patch contrast appears to correspond to the level of variance in plant functional group 359 composition. Management for heterogeneity using patch burn-grazing can increase heterogeneity 360 in vegetation structure, and therefore increase rangeland biodiversity compared to management 361 for homogeneity, but only when fire behaviour influences grazing behaviour. 362 Three important themes that apply to management for heterogeneity emerged from our 363 findings. First, managers choosing to apply patch burn-grazing should stock livestock at a 364 moderate stocking rate. Each location in our study that did not show patch contrast was 365 excessively stocked before being managed with patch burn-grazing, which suggests that 366 excessive stocking reduces fire spread and decreases the influence of fire on the spatial pattern of 367 grazing. The second theme is that invasive species that reduce fire spread render fire ineffective 368 to drive spatial pattern of grazing. Finally, by moderating stocking rate on overgrazed 369 rangelands, plant productivity and fuel load will recover and fire will again influence spatial 370 pattern of grazing (Limb et al. 2011). However, the extent to which invasive species persist as a 371 barrier to effective patch burn-grazing remains unknown. 372 373 19 374 Acknowledgements 375 This research was made possible by support from the Iowa State Wildlife Grants program 376 with the U.S. Fish and Wildlife Service Wildlife and Sport Fish Restoration Program (#T-1-R-377 15), a grant from the Joint Fire Science Program (#201814G905), and a grant from National 378 Research Initiative (#2006-35320-17476) from the USDA Cooperative State Research, 379 Education and Extension Service. The authors acknowledge the support of the Iowa Agriculture 380 and Home Economics Experiment Station and the Oklahoma Agricultural Experiment Station; 381 the assistance of R. Harr, R. Limb, B. Allred, J. Kerby, and M. Kirkwood; as well as the 382 statistical advice of P. Dixon, C. Goad, and K. Pazdernick. The manuscript was improved by 383 contributions from L. Schulte-Moore and B. Wilsey. The manuscript was also improved by 384 comments from three anonymous reviewers. 385 386 20 387 References 388 Adler, P.B., Raff, D.A. & Lauenroth, W.K. (2001) The effect of grazing on the spatial 389 heterogeneity of vegetation. Oecologia, 128, 465–479. 390 Allred, B.W., Fuhlendorf, S.D., Engle, D.M. & Elmore, R.D. (2011) Ungulate preference for 391 burned patches reveals strength of fire-grazing interaction. Ecology and Evolution, 1–13. 392 Allred, B.W., Fuhlendorf, S.D. & Hamilton, R.G. (2011) The role of herbivores in Great Plains 393 conservation: comparative ecology of bison and cattle. Ecosphere, 2, art26. 394 Archibald, S. & Bond, W.J. (2004) Grazer movements: spatial and temporal responses to burning 395 in a tall-grass African savanna. International Journal of Wildland Fire, 13, 377–385. 396 Baayen, R.H., Davidson, D.J. & Bates, D.M. (2008) Mixed-effects modeling with crossed 397 random effects for subjects and items. Journal of Memory and Language, 59, 390–412. 398 Bailey, D.W., Gross, J.E., Laca, E.A., Rittenhouse, L.R., Coughenour, M.B., Swift, D.M. & 399 Sims, P.L. (1996) Mechanisms that result in large herbivore grazing distribution patterns. 400 Journal of Range Management, 49, 386–400. 401 Bates, D. & Maechler, M. (2010) lme4: Linear mixed-effect models using S4 classes. Available 402 at http://CRAN.R-project.org/package=lme4 (Accessed 22 May 2011). 403 Bouwman, H. & Hoffman, R. (2007) The effects of fire on grassland bird communities of 404 Barberspan, North West Province, South Africa. Ostrich: Journal of African Ornithology, 405 78, 591–608. 406 Cohen, J. (1977) Statistical power analysis for the behavioral sciences. Academic Press, New 407 York, NY. 408 Coppedge, B., Fuhlendorf, S.D., Harrell, W.C. & Engle, D.M. (2008) Avian community 409 response to vegetation and structural features in grasslands managed with fire and 410 grazing. Biological Conservation, 141, 1196–1203. 411 Côté, I.M. & Sutherland, W.J. (1997) The effectiveness of removing predators to protect bird 412 populations. Conservation Biology, 11, 395–405. 413 Daubenmire, R. (1959) A canopy-coverage method of vegetational analysis. Northwest Science, 414 33, 43–64. 415 Davies, K.W., Bates, J.D., Svejcar, T.J. & Boyd, C.S. (2010) Effects of long-term livestock 416 grazing on fuel characteristics in rangelands: An example from the sagebrush steppe. 417 Rangeland Ecology and Management, 63, 662–669. 418 21 Davies, K.W., Svejcar, T.J. & Bates, J.D. (2009) Interaction of historical and nonhistorical 419 disturbances maintains native plant communities. Ecological Applications, 19, 1536–420 1545. 421 Derner, J.D., Lauenroth, W.K., Stapp, P. & Augustine, D.J. (2009) Livestock as ecosystem 422 engineers for grassland bird habitat in the western Great Plains of North America. 423 Rangeland Ecology and Management, 62, 111–118. 424 Doxon, E.D., Davis, C.A., Fuhlendorf, S.D. & Winter, S.L. (2011) Aboveground 425 Macroinvertebrate Diversity and Abundance in Sand Sagebrush Prairie Managed With 426 the Use of Pyric Herbivory. Rangeland Ecology and Management, 64, 394–403. 427 Ellis, E.C. & Ramankutty, N. (2008) Putting people in the map: anthropogenic biomes of the 428 world. Frontiers in Ecology and the Environment, 6, 439–447. 429 Engle, D.M., Fuhlendorf, S.D., Roper, A. & Leslie Jr, D.M. (2008) Invertebrate community 430 response to a shifting mosaic of habitat. Rangeland Ecology and Management, 61, 55–62. 431 Fuhlendorf, S.D. & Engle, D.M. (2001) Restoring heterogeneity on rangelands: Ecosystem 432 management based on evolutionary grazing patterns. BioScience, 51, 625–632. 433 Fuhlendorf, S.D. & Engle, D.M. (2004) Application of the fire–grazing interaction to restore a 434 shifting mosaic on tallgrass prairie. Journal of Applied Ecology, 41, 604–614. 435 Fuhlendorf, S.D., Engle, D.M., Kerby, J. & Hamilton, R. (2009) Pyric herbivory: Rewilding 436 landscapes through the recoupling of fire and grazing. Conservation Biology, 23, 588–437 598. 438 Fuhlendorf, S.D., Harrell, W.C., Engle, D.M., Hamilton, R.G., Davis, C.A. & Leslie Jr, D.M. 439 (2006) Should heterogeneity be the basis for conservation? Grassland bird response to 440 fire and grazing. Ecological Applications, 16, 1706–1716. 441 Fuhlendorf, S.D. & Smeins, F.E. (1999) Scaling effects of grazing in a semi-arid grassland. 442 Journal of Vegetation Science, 10, 731–738. 443 Gillen, R.L., Eckroat, J.A. & McCollum, F.T. (2000) Vegetation response to stocking rate in 444 southern mixedgrass prairie. Journal of Range Management, 53, 471–478. 445 Gillen, R.L., Rollins, D. & Stritzke, J.F. (1987) Atrazine, spring burning, and nitrogen for 446 improvement of tallgrass prairie. Journal of Range Management, 40, 444–447. 447 Gillen, R.L. & Sims, P.L. (2004) Stocking rate, precipitation, and herbage production on sand 448 sagebrush-grassland. Journal of Range Management, 57, 148–152. 449 Hamilton, R.G. (2007) Restoring heterogeneity on the Tallgrass Prairie Preserve: applying the 450 fire-grazing interaction model. Proceedings of the 23rd Tall Timbers Fire Ecology 451 Conference: Fire in Grassland and Shrubland Ecosystems pp. 163–169. Allen Press, Tall 452 Timbers Research Station, Tallahassee, Florida, USA. 453 22 Harrell, W.C. & Fuhlendorf, S.D. (2002) Evaluation of habitat structural measures in a shrubland 454 community. Journal of Range Management, 55, 488–493. 455 Harrison, F. (2011) Getting started with meta-analysis. Methods in Ecology and Evolution, 2, 1–456 10. 457 Holechek, J.L., Pieper, R.D. & Herbel, C.H. (2003) Range Management Principles and 458 Practices. Prentice Hall Inc., Upper Saddle River, NJ. 459 IEM. (2011) Iowa Environmental Mesonet, Iowa State University Department of Agronomy: 460 Lamoni, IA. Available at 461 http://mesonet.agron.iastate.edu/climodat/index.phtml?station=IA4585&report=17 462 (Accessed 31 May 2011). 463 Kendall, M.G. (1938) A new measure of rank correlation. Biometrika, 30, 81–93. 464 Kerby, J.D., Fuhlendorf, S.D. & Engle, D.M. (2007) Landscape heterogeneity and fire behavior: 465 scale-dependent feedback between fire and grazing processes. Landscape Ecology, 22, 466 507–516. 467 Kopp, S.D., Guthery, F.S., Forrester, N.D. & Cohen, W.E. (1998) Habitat selection modeling for 468 northern bobwhites on subtropical rangeland. The Journal of Wildlife Management, 62, 469 884–895. 470 Kotliar, N.B. & Wiens, J.A. (1990) Multiple scales of patchiness and patch structure: a 471 hierarchical framework for the study of heterogeneity. Oikos, 59, 253–260. 472 Leonard, S., Kirkpatrick, J. & Marsden-Smedley, J. (2010) Variation in the effects of vertebrate 473 grazing on fire potential between grassland structural types. Journal of Applied Ecology, 474 47, 876–883. 475 Limb, R.F., Fuhlendorf, S.D., Engle, D.M., Weir, J.R., Elmore, R.D. & Bidwell, T.G. (2011) 476 Pyric-herbivory and cattle performance in grassland ecosystems. Rangeland Ecology and 477 Management, 64, 659–663. 478 McGranahan, D.A. (2011) Species richness, fire spread, and structural heterogeneity in tallgrass 479 prairie. Dissertation, Iowa State University, Ames (IA), USA. 480 McGranahan, D.A., Engle, D.M., Fuhlendorf, S.D., Miller, J.R. & Debinski, D.M. (2012) An 481 invasive cool-season grass complicates prescribed fire management in a native warm-482 season grassland. Natural Areas Journal, 32, 208–214. 483 O’Connor, T.G., Kuyler, P., Kirkman, KP & Corcoran, B. (2010) Which grazing management 484 practices are most appropriate for maintaining biodiversity in South African grassland? 485 African Journal of Range and Forage Science, 27, 67–76. 486 OK Mesonet. (2011) Oklahoma Mesonet. Available at http://www.mesonet.org (Accessed 23 487 June 2011). 488 23 Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., O’Hara, R.G., Simpson, G.L., Solymos, P., 489 Stevens, M.H.H. & Wagner, H. (2011) vegan: Community ecology package. Available at 490 http://cran.r-project.org/package=vegan (Accessed 25 January 2011). 491 Ostfeld, R.S., Pickett, S.T., Shachak, M. & Likens, G.E. (1997) Defining the scientific issues. 492 The Ecological Basis of Conservation: Heterogeneity, Ecosystems, and Biodiversity pp. 493 3–10. Chapman and Hall. 494 Pillsbury, F.C., Miller, J.R., Debinski, D.M. & Engle, D.M. (2011) Another tool in the toolbox? 495 Using fire and grazing to promote bird diversity in highly fragmented landscapes. 496 Ecosphere, 2, art28. 497 Polasky, S., Nelson, E., Lonsdorf, E., Fackler, P. & Starfield, A. (2005) Conserving Species in a 498 Working Landscape: Land Use With Biological and Economic Objectives. Ecological 499 Applications, 15, 1387–1401. 500 R Development Core Team. (2011) R: A language and environment for statistical computing. R 501 Foundation for Statistical Computing, Vienna, Austria. Available at http://www.R-502 project.org/ (Accessed 22 May 2011). 503 Robel, R.J., Briggs, J.N., Dayton, A.D. & Hulbert, L.C. (1970) Relationships between visual 504 obstruction measurements and weight of grassland vegetation. Journal of Range 505 Management, 23, 295–297. 506 Samson, F. & Knopf, F. (1994) Prairie conservation in North America. BioScience, 44, 418–421. 507 Senft, R.L., Coughenour, M.B., Bailey, D.W., Rittenhouse, L.R., Sala, O.E. & Swift, D.M. 508 (1987) Large herbivore foraging and ecological hierarchies. BioScience, 37, 789–799. 509 Ter Braak, C.J.F. (1986) Canonical Correspondence Analysis: A new eigenvector technique for 510 multivariate direct gradient analysis. Ecology, 67, 1167–1179. 511 Towne, E.G., Hartnett, D.C. & Cochran, R.C. (2005) Vegetation trends in tallgrass prairie from 512 bison and cattle grazing. Ecological Applications, 15, 1550–1559. 513 Winter, S.L., Fuhlendorf, S.D., Goad, C.L., Davis, C.A., Hickman, K.R. & Leslie Jr, D.M. 514 (2011) Fire tolerance of a resprouting Artemisia (asteraceae) shrub. Plant Ecology, 212, 515 2085–2094. 516 Winter, S.L., Fuhlendorf, S.D., Goad, C.L., Davis, C.A., Hickman, K.R. & Leslie Jr, D.M. 517 (2012) Restoration of the fire-grazing interaction in Artemisia filifolia shrubland. Journal 518 of Applied Ecology, 49, 242–250. 519 Supporting information 520 Additional supporting information may be found in the online version of this article: 521 24 Appendix S1. Description of data included in rangeland heterogeneity analysis 522 523 524 525 526 527 528 529 530 25 Table 1: Precipitation, vegetation, and grazing information for five experimental locations comparing heterogeneously-applied fire management with homogeneous fire regimes. Refer to Methods and Appendix S1 for information about experimental design, data collected, and years included. Locations are listed geographically from west to east Study location Cooper a Klemmeb Stillwater c TGPPd GRG e Annual precipitation (cm) Long-term mean 57 78 83 88 91 Study period range 41-77 51-82 61-99 59-109 97-147 Vegetation type Artemisia shrubland- mixed prairie Midgrass prairie Tallgrass prairie Tallgrass prairie Tallgrass prairie Stocking rate f 26 Prior to study period Moderate Heavy Moderate Moderate -light Severe Study period (Animal-Unit-Months ha-1) 0.8 (Moderate) 1.6 (Moderate) 4.3 (Moderat e) 3.2 (Moderat e-light) 3.1 (Heavy) Grazing season 1 April - 15 Sept. 15 Mar. - 15 Sept. 1 Dec. - 1 Sept. 15 Apr. - 20 Jul. 1 May - 1 Oct. Pasture area ( ha) 406-848 ca. 50 45-65 400-900 15 - 31 Annual primary productivity g (kg ha-1) 1500 2000 5600 6000 6700 aHal and Fern Cooper Wildlife Management Area. (Gillen & Sims 2004; Winter et al. 2012) b Marvin Klemme Experimental Research Range. (Gillen, Eckroat, & McCollum 2000; Limb et al. 2011) c Stillwater Research Range. (Gillen, Rollins, & Stritzke 1987; Fuhlendorf & Engle 2004; Limb et al. 2011; OK Mesonet 27 2011) d Tallgrass Prairie Preserve. (Hamilton 2007; Coppedge et al. 2008; OK Mesonet 2011) e Grand River Grasslands. (IEM 2011; Pillsbury et al. 2011) fStocking rate categories expressed in relation to local recommendations from the USDA Natural Resource Conservation Service. g Estimated annual primary productivity of native vegetation not recently disturbed by grazing or fertilization. Published data were used for Cooper (Gillen & Sims 2004), Klemme (Gillen, Eckroat, & McCollum 2000), and Stillwater (Gillen, Rollins, & Stritzke 1987). Unpublished data on end-of-season biomass one year after fire from at least one year within the study period included here were used to estimate annual primary productivity at the TGPP and the GRG. 28 Figure 1: Proportion of total variance in vegetation structure contributed by the patch term in nested, spatially hierarchical sampling measures patch contrast at five experiments comparing management for heterogeneity (blue triangles) to management for homogeneity (orange circles). Data are plotted for each pasture replicate within each of the five locations. Locations are arranged along a general west-to-east geographical gradient (western Oklahoma – south-central Iowa), which corresponds to a precipitation gradient. Asterisks represent results of Student’s t tests for differences in means of management groups: “ ** ” P < 0.01; “ * ” P ≤ 0.05. Figure 2: Range of RDA1 scores measures variance in plant functional group composition at five experiments comparing management for heterogeneity (blue triangles) to management for homogeneity (orange circles). Data are plotted for each pasture replicate within each of the five locations. Locations are arranged along a general west-to-east geographical gradient (western Oklahoma – south-central Iowa), which corresponds to a precipitation gradient. Asterisks represent results of Student’s t tests for differences in means of management groups: “ * ” P ≤ 0.05. Figure 3: Effect size of patch contrast (Y axis) plotted against effect size of variance in plant functional group composition (X axis), with corresponding 95% confidence intervals, for five rangeland experiments comparing management for heterogeneity against management for homogeneity. Effect sizes are calculated with the meta-analysis statistic Cohen’s d (see Methods for equation) and are plotted on a log scale. 29 Fig. 1 30 Fig. 2 31 Fig. 3