Effects of fire on ground‐dwelling arthropods in a shrub‐dominated grassland

Abstract Arthropods are abundant and diverse animals in many terrestrial food webs. In western Oklahoma, some shrublands are interspersed with discrete, dense thickets of tall, woody vegetation, known as mottes. Some of these shrublands are managed with prescribed burning. The goal of this study was to examine whether prescribed burning interacted with habitat type (i.e., shrubland versus mottes) to affect ground‐dwelling arthropod communities. Arthropods were collected in pitfall traps at four sampling locations in relation to mottes; in the center of mottes, and three plot location in shrublands; 1 m, 15 m, and 50 m away from the edge of the motte. There were three treatment levels for burning: one year postburn (burned in dormant months of 2017), two years postburn (burned in dormant months of 2016), and unburned (burned in dormant season of 2014 and prior). There were no significant interactions between prescribed burning and habitat type. Mottes had a different community of arthropods compared with the surrounding shrubland. Mottes also had lower overall abundance, but a higher diversity of arthropods. In terms of fires, arthropod communities one year after burning were different from those two or more years after burning. There was no effect of burning on overall arthropod abundance, but plots that were one year since burning had significantly lower diversity compared with plots that were two or more years postburn. The results of this study suggest that both fire and mottes can independently facilitate heterogeneity in arthropod communities, but they do not appear to interact with one another.

. In addition, estimates suggest that the overall biomass of arthropods is decreasing at an annual rate of 2.5% worldwide (Sánchez-Bayo & Wyckhuys, 2019). These declines in abundance and diversity of arthropods are alarming given their importance for food web and ecosystem function (Klink et al., 2020;Wagner, 2020).
Understanding the factors that influence the abundance and biomass of arthropods is critical for managing their populations and maintaining the ecosystem services that they provide.
Vegetation structure and composition are known to be important factors affecting arthropod abundance and diversity (Prather & Kaspari, 2019). In sandy soils in western Oklahoma, the prairie can consist of grassland interspersed with shinnery oak shrubs (Quercus havardii) and mottes, which are dense patches of taller (e.g., >2 m tall) oak trees (Peterson & Boyd, 2000;Weideman & Penfound, 1960). Mottes are unique in this landscape because they form small patches of trees dispersed within the shrublands. The vegetation structure provided by mottes can create thermal refugia to allow some species, including northern bobwhite (Colinus virginianus) and wild turkey (Meleagris gallopavo), to escape high temperatures experienced in the relatively open shrubland habitat (Carroll et al., 2015;Rakowski et al., 2019). For example, female northern bobwhite with broods will often use mottes and other areas of shade during the middle of the day to reduce the temperatures they experience by up to 10°C relative to other locations in the landscape (Carroll et al., 2015). Thermoregulation is also important for ectotherms, such as arthropods (May, 1979), although less is known of the response of arthropods to mottes. Mottes may support different arthropod communities relative to the surrounding shrubland, due to the different microclimates, and mottes may provide areas for more wide-ranging arthropods to avoid mid-day temperature extremes (Robertson et al., 1996).
Prescribed fire is a management tool that has historically been used in prairie ecosystems in the Great Plains to reduce woody plant encroachment. Prescribed fire has also been used to restore heterogeneity in grasslands by creating a shifting mosaic of different seral stages of vegetation across the landscape through the interaction of fire and grazing (Fuhlendorf et al., 2009). The ecological interaction between fire and grazing, known as pyric herbivory, can reduce the abundance of woody vegetation, change vegetation quality, and shift the successional stage of vegetation (Fuhlendorf & Engle, 2004).
Changes in vegetation, in turn, can affect invertebrate communities since invertebrate communities are often dependent on plant community composition and structure (Engle et al., 2008;Reed, 1997). In grassland and shrubland ecosystems, arthropod taxa are differentially affected by prescribed burning depending on their life stage during the burn, habitat use (e.g., aboveground versus belowground), dispersal ability, and other factors (Swengel, 2001;Warren et al., 1987). Arthropod communities can also shift with time since fire due to successional changes in plant communities and accumulation of litter (Swengel, 2001;Warren et al., 1987). By burning different patches at different times, insect diversity and abundance can be increased across landscapes (Doxon et al., 2011;Fuhlendorf & Engle, 2004;Fuhlendorf et al., 2006). However, while fire has been studied in a number of grassland systems, it remains less clear how arthropod communities in grasslands interspersed with shrubs and mottes respond to prescribed fire. In particular, arthropod communities in mottes may respond differently than arthropod communities in shrubland as the fire intensity, fuel loads, and soil moisture could differ significantly between the shrubs and mottes.
The goal of this study was to test whether ground-dwelling arthropod communities were affected by: (a) habitat type (i.e., mottes versus shrubland), (b) time since prescribe fire, and (c) the interaction between habitat type and fire. To achieve this, we examined arthropod communities across a gradient of two habitat types (i.e., mottes and shrubland at different distances from the mottes) that also included patches burned at different times since fire intervals. We hypothesized that ground-dwelling arthropod communities would differ between mottes and the surrounding shrubland. We examined arthropods over a gradient of distance from mottes as the mobility of some arthropod groups can be relatively limited. The ground layer of closed-canopy mottes is dominated by leaf litter and resources for detritivores, whereas the surrounding shrubland has abundant primary production that can support herbivores. We also predicted that there would be a significant interaction between habitat type and time since burning. Previous work has shown a higher abundance of invertebrates in grass and shrublands in the year after a burn, likely due to increased quantity or quality of primary production (Doxon et al., 2011;Engle et al., 2008). In contrast, in mottes there could be a decline in invertebrate abundance, especially of detritivores, if fires remove detrital resources that take time to accumulate.

| Study site
This study was conducted at Packsaddle Wildlife Management Area (hereafter, Packsaddle WMA) in Ellis County, Oklahoma. It is a 6,475-ha shrub-dominated grassland with elevations ranging from 579 to 762 m above mean sea level (Townsend et al., 2001).
Sand shinnery oak shrubs range between 0.25 and 1.5 m tall (Harrell & Fuhlendorf, 2002) and rarely exceed 1.5 m in height, while mottes were primarily comprised of hybrid post-shinnery oaks and identified as a distinct patch of trees with heights averaging 2 m or greater (Peterson & Boyd, 2000).
Packsaddle WMA has been managed using prescribed burns since the 1990s with areas burned once every two to five years. The size of patch burns is highly variable and depends on weather conditions, other habitat management activities, and available personnel. Within the boundaries of Packsaddle WMA, several units are burned every two to three years, weather permitting. Many areas within Packsaddle WMA are also grazed by cattle during the growing season with a stocking rate of about seven ha per animal.

| Sampling design
Our overall sampling design focused on comparing arthropod abun- shrubland" plot and at least two meters away from the other, corresponding center plot. For each motte, data from the corresponding plot locations were averaged, such that for each motte there was one data point each for the center, 1, 15, and 50 m plot locations.
This allowed us to observe whether mottes contained different arthropod orders relative to shrubland and whether mottes had an effect on arthropods in the surrounding landscape.

| Vegetation measurements
To determine potential factors influencing the arthropod communities, we collected vegetation data in May and July of 2018.
Vegetation sampling included woody shrub canopy cover and percent ground cover composition at each plot location. A line intercept method was used to quantify the canopy composition of woody shrubs. This method used a 20 m transect that crossed through each sampling plot within which was measured the horizontal linear length of each shrub that intercepted the line. Percent ground cover was determined using a Daubenmire frame (20 cm × 50 cm microplot marked in 10% classes) (Daubenmire, 1959). Daubenmire cover classes for grasses, forbs, bare ground, litter, and rock were recorded at three points along the vegetation transect, at each end and in the center of the study plot. Ground cover was described as a range of six cover classes including 0%-5%, 5%-25%, 25%-50%, 50%-75%, 75%-95%, and 95%-100%.

| Arthropod collection
We used pitfall traps to sample arthropod communities once a month from May through August 2018. Each sampling plot con- concentration and a few drops of clear, odorless dish soap (Seventh Generation, Inc.). This level of fluid was sufficient to submerge arthropods while avoiding the potential for the cup to overflow following rain or for arthropods to escape.
After 48 hr, all five cups at each 1-m 2 plot were consolidated into one sample for each sampling point along the transect. Samples were removed from the field and transferred into 70% ethanol the same day. Pitfall samples remained stored in ethanol until identified and counted in the laboratory. Following collection, traps were covered with a lid and left closed until the next month's sampling.
Following sorting of each sample, we identified arthropods to order and counted them after which samples were placed back in ethanol and stored. In some circumstances, arthropods could not be identified with complete confidence, often as a result of individuals being too damaged. These were classified as "Other." Additionally, some orders were encountered relatively infrequently (i.e., 1% or less of the total arthropod abundance) and therefore did not represent a significant component of the arthropod community. These were also classified as "Other."

| Data analysis
Since each motte had two transects, the data from corresponding plots in each transect were averaged such that there was only one value per plot location (center, 1, 15, or 50 m) per motte. However, 8 of the 504 pitfall samples were disturbed, and, in these cases, we used the value for the corresponding undisturbed pitfall location rather than the average of the two locations.
The data were square root transformed to reduce the effect of highly abundant taxa while considering lesser represented orders as well. The square root transformed abundance data were visualized using a multivariate ordination procedure, nonmetric multidimensional scaling (nMDS). This analysis was done using Bray-Curtis distances in the program R using the vegan package burn year were individually analyzed as separate variables affecting arthropod abundance and biomass in nMDS. Tests for significance were then determined using a nonparametric multivariate statistical test, permutational multivariate analysis of variance (PERMANOVA) using the adonis function in R. Motte was used as a blocking factor using the strata option. To determine the percent dissimilarity seen in the nMDS and PERMANOVA results, we performed a SIMPER analysis using the PRIMER software (version 7, Anderson et al., 2008). The SIMPER analysis identifies which taxa of arthropods primarily contributed to the differences in community composition between treatments. For this analysis, all orders were included but we only report on those orders that contributed to the top 70% of the total dissimilarity.
Abundance of the top five most abundant arthropod orders (Collembola, Coleoptera, Hymenoptera, Acari, and Diptera) and Shannon's diversity index (Lande, 1996) were analyzed with mixed model nested ANOVAs using the software program JMP (version 14, SAS Institute, 2018). Shannon's diversity index was included as an additional metric of how the community responded to treatments. Differences in diversity among treatments were likely driven more by differences in evenness than by differences in richness, as we used arthropod order as our level of identification and most orders were recorded at most sites. These ANOVAs included motte nested within burn treatment as a random effect to include proper degrees of freedom for testing the burn treatment effect. The ANOVA models included burn year, distance from motte, and time separately and in all interactions. All abundance data were log(x + 1) transformed for the ANOVA analysis because log-transformed data better approximated a normal distribution relative to other transformations. For the vegetation data, we first combined all of the variables into an overall analysis using principal components and then analyzed each vegetation variable to determine which variables contributed to the overall differences.

| RE SULTS
We collected 206,477 arthropods from 504 pitfall traps during our study (Table A1). Overall, we collected individuals from 15 taxonomic groups of arthropods plus one group of "Other" that included all other orders that represented less than 1.0% composition, including Lepidoptera, Blattodea, Neuroptera, Isopoda, Psocoptera, Thysanoptera, and the subphylum Myriapoda.

| Community composition
Collembola were the most numerous arthropods collected, representing 50% of all individuals, with Hymenoptera being the next most abundant, representing 30% of the total community (Table A1).
All other arthropod orders represented 5% or less of the total abundance.
There were significant effects of burn year and distance from motte on the arthropod community, using data from all months combined and for the individual months (Table 1). August was the only month without a significant effect of burn year, and June was the only month without a significant effect of distance from a motte, using our more conservative alpha value. For burn year, arthropod communities in 1 year since burn plots appeared different from those in the control and 2 years since burn-in nMDS plots, especially in all months combined and the months of May and July ( Figure 1, Figure A1). For distance from motte, the motte location appeared more separated in space from all of the shrubland locations, which overlapped each other broadly, for all months combined and for each individual month in nMDS plots (Figure 1, Figure A1). (Table A2) show that Collembola contributed the most (30%-45%) to the dissimilarity between comparisons of each burn treatment. Hymenoptera contributed between 17% and 21% of the dissimilarity across all burn years. Taxa affecting less than 11% of the dissimilarities in burn treatments observed include Diptera, Acari, and Other. Differences in community composition between plot locations were most explained by Collembola (45%-47%).

SIMPER results
Hymenoptera accounted for 13%-22% of the differences across all distances from motte. The remaining orders, Diptera, Acari, and Coleoptera, contributed less than 14% to the dissimilarities between plot locations.

| Diversity
There were significant effects of burn year and distance from a motte on Shannon's diversity index for all arthropod data (Figure 2, Table 2). Analyzing data by month, there were significant effects of burn year, distance from motte, and time ( Figure 2, Table 2). Diversity was significantly higher in mottes compared with all locations in the shrubland, and diversity was significantly lower the first year after burning compared with two or more years after burning, especially toward the last two sampling dates. There were no interactions between burn year and distance from a motte on Shannon's diversity index.

| Individual orders
There were significant effects of distance from motte and time on total abundance of arthropods, summed across all 4 months, but there was no burn year by distance from motte interactions for any of the arthropod taxa. (Table A3, Figure 3). For Collembola, there were significant effects of distance from motte and time since fire on abundance. Collembola were most abundant in the 1 year since burn treatment plots relative to the other two burn treatments and they were least abundant in the center plot compared with the other distances from motte ( Figure 3). For Hymenoptera, there were significant effects of distance from motte and the interaction of burn year and month sampled on abundance (Table A3). Hymenoptera were least abundant in the plots in the center of a motte relative to the other plot locations (Figure 3). For Coleoptera, there were significant effects of distance from motte and time on abundance (Table A3). Coleopterans were most abundant in the center plot relative to other plot locations (Figure 3). For Acari and Diptera, there were only significant effects of time on abundance (Table A3).

| Vegetation
For vegetation composition, the six habitat measurements were combined into two principal components. Principal component 1 had relatively high loading for shrub cover, bare ground, and litter (Table A4). Principal component 2 had relatively high loading for grass cover and forb cover. For both principal components 1 and 2, there were significant effects of distance from motte on vegetation structure (Table 3). For both principal component axes, the center plot location was different in vegetation structure relative to the other three distances from a motte (Figure 4). When analyzing the individual habitat components, there were significant effects of burn year for the percent grass composition and the percent forb composition (Table A5). Forb cover in the 1 year since burn treatment was significantly different from control burn forb composition (Table 4). Grass cover in 1 year since burn was significantly different from 2 years since burn, but neither were significantly different from the control treatment. There were significant effects of distance from motte on shrub cover, grass cover, bare ground, and litter (Table A5). Shrub cover was only significantly different between the 1 m and 15 m plot locations (Table 4). There were significant differences in grass cover, bare ground, and litter in the center plots compared with all three of the open shrubland plots.
Grass cover was the only variable that showed a significant effect of time (Table A5).

| D ISCUSS I ON
Our results do not support the hypothesis that fire and mottes interact to affect ground-dwelling arthropod communities. Rather, Mottes also had lower overall abundance, but a higher diversity of arthropods. In terms of fires, arthropod communities one year after burning were different from those two or more years after burning. There was no effect of burning on overall arthropod abundance, but plots that were one year since burning had significantly lower diversity compared with plots that were two or more years postburn. Hence, the findings of this study suggest that both fire and mottes can independently facilitate heterogeneity in arthropod communities, but they do not appear to interact with one another.
The transition from mottes to shrubland is fairly discrete in terms of overall vegetation structure as mottes have a closed canopy of trees >2 m tall while shrubland has much shorter vegetation. Our ground cover data support the discrete change in vegetation with mottes having lower grass, lower bare ground, and higher litter cover compared with the surrounding shrubland. The principal components analysis of vegetation also shows mottes as being distinct from shrubland plots. Shannon's diversity index, total arthropod abundance, and the abundances of Collembola and Hymenoptera also showed fairly distinct differences between motte and shrubland plots. This should be expected given that arthropod populations and communities often vary with vegetation structure, due to its influence on microclimate and food availability (Engle et al., 2008;Prather & Kaspari, 2019). Hence, the arthropod community, including the most abundant arthropods, shows a relatively discrete response to mottes versus shrubland.
Differences in arthropod communities between mottes and shrubland could be due to differences in the types of plants available as food as well as the effects of vegetation on microclimate.
The shrublands in western Oklahoma can reach high temperatures during summer afternoons (Carroll et al., 2015). Mottes can serve as thermal refugia for some birds and other vertebrates (Carroll et al., 2015;Rakowski et al., 2019), and could also serve as thermal refugia for invertebrates. Our pitfall data are unable to evaluate temporal variation in use of mottes, as traps were open for two full days and nights. Opening pitfalls only for certain time periods (e.g., day or night) could help test if arthropods vary their use of mottes during the day. However, ground-dwelling arthropods may be less likely to move in and out of mottes as they might have to move large distances to find mottes relative to other microhabitats (e.g., litter or the thick bases of clump grass or shrubs) that may provide thermal refugia. More mobile, flying arthropods (e.g., adult Orthoptera, Diptera, Hymenoptera, Hemiptera, and Lepidoptera) might be more likely to use mottes as thermal refugia.
Fire only appeared to have short-term effects on vegetation cover and arthropod orders at our study site, as the only differences we found were between the plots burned 1 year prior and plots burned longer ago. For vegetation, there were no overall effects of prescribed burning on the vegetation principal components. But, there were some effects of fire on ground cover as plots burned 1 year prior had higher grass and forb cover compared with older burns. Fire often stimulates vegetation growth in prairie and shrubland for a year or more due to the release of nutrients when detritus is burned (Allred et al., 2011). For invertebrates, there were significant effects of fire on arthropod community composition (PERMANOVA) and diversity, with lower diversity in the plots burned 1 year prior. Of the 5 most abundant arthropod taxa, Collembola was the only one to respond to fire with higher abundance in plots 1 year since burning. Prior studies have shown that Collembola species respond differently to fire and sexually reproducing species with fast and active dispersal are more likely to recover quickly following fire (Brand, 2002;Malmström, 2012). A more detailed understanding of which species of Collembola increased in abundance and the preferred diets of these species would be needed to confirm this hypothesis. Other studies have also shown that certain arthropod taxa or communities in prairie and shrubland often recover from fire within the first year or two after a burn ( F I G U R E 3 Effects of distance from a motte (0/center, 1, 15, and 50 m) and time since burning (control, 1 year, and 2 years) on the mean ± 1 SE abundance of (a) Collembola, (b) Hymenoptera, (c) Coleoptera, and (d) total arthropod abundance across the 4 months of the study (May-August). Collembola, Hymenoptera, and Coleoptera represent the 3 most abundant orders of arthropods collected Swengel, 2001;Warren et al., 1987). The relatively rapid recovery of the habitat following prescribed fire in western Oklahoma may be due to the rapid growth of plants in the shrubland combined with the low intensity of the fires, due to low fuel load, which often does not kill trees in mottes or the belowground parts of shrubs in the shrubland (Malmström et al., 2008;Smit et al., 2010).  (Fuhlendorf & Engle, 2004;Fuhlendorf et al., 2009). In our study, we did not find support for an interaction between mottes and fire. Although this study was done in a landscape that is regularly maintained by prescribed fire, we cannot preclude the possibility that there could be interactions between fire and mottes when there are longer intervals without burning.
Better understanding the potential interactions of prescribed fire with other factors in the landscape is key to using this management technique more effectively. Note: SIMPER analyses, similarity percentages, break down the contribution of each order to the observed dissimilarity between samples for the PERMANOVA analyses. Total % explained shows the cumulative percentage of the average dissimilarity that is explained by all orders in each row. Only orders that contributed to the top 70% of the total dissimilarity were considered. Data were square root transformed.

TA B L E A 3
Summary of repeated measures ANOVAs testing the effects of burn year and plot location (distance from motte) on abundance of the five most abundant orders analyzed in each month separately and with all 4 months combined