Vegetation recovery in a desert landscape after wildfires: influences of community type, time since fire and contingency effects


Correspondence author. E-mail:


1. Theories of plant succession are poorly developed in arid lands, hindering our understanding of how long communities may take to recover after disturbances such as fire. In desert landscapes vulnerable to fire, information about vegetation recovery is important when deciding whether land managers should facilitate vegetation recovery. The deserts of the southwestern USA are increasingly subject to unprecedented fires, facilitated by fuel from exotic grasses, yet management strategies are unclear.

2. We evaluated post-fire recovery patterns of perennial plant species richness and diversity, compared the rate and direction of succession between two major community types, and explored the relationship of time since fire (TSF) and other environmental factors with vegetation recovery. We sampled perennial plant communities and environmental variables (e.g. soil N) on 32 burns, ranging from 2 to 29 years TSF and each paired with their own unburned area, within a 1·8 million ha landscape in the Mojave Desert, USA.

3. Species richness, diversity and composition exhibited different post-burn recovery patterns, and recovery rates differed between community types. Specifically, diversity in Coleogyne ramosissima communities was greater in burned than unburned areas, but diversity did not differ in Larrea tridentata communities. Species composition in Larrea communities exhibited trajectories that indicate convergence with unburned community composition after 19 years TSF. Conversely, burned and unburned Coleogyne communities lacked convergence irrespective of TSF. Environmental variables (e.g. soil texture and P) accounted for 79–83% of the variation in burned species composition, suggesting environmental characteristics in part control recovery patterns.

4.Synthesis and applications. The results indicate that geographically similar vegetation types within the same landscape can have markedly different post-disturbance successional rates and trajectories. Furthermore, the persistence of fire effects varied depending on the community measure, with fire effects on species composition more long-lasting than the effects on species diversity. This work supports (i) the use of post-disturbance successional analyses for helping to prioritize management where it is most needed (e.g. communities not recovering naturally) and (ii) the need to assess whether persistent, early successional desert communities meet functional management objectives.


As anthropogenic impacts on the biosphere intensify, ecological communities are increasingly exposed to novel disturbances that were uncommon or absent in their evolutionary history (Aronson et al. 2004). Understanding the responses of these communities to disturbances is essential for developing management strategies to conserve biodiversity and maintain ecosystem services over the coming decades (Seastedt, Hobbs & Suding 2008). The arid lands of the western United States offer a dramatic example of exposing ecosystems to a novel disturbance, in this case fire. Landscape-scale invasion by exotic plants, such as the annual grasses Bromus spp., has augmented indigenous fuel loads to facilitate fire spread in these arid lands (Esque & Schwalbe 2002; Steers & Allen 2010; Rao, Steers & Allen 2011). For instance, more than 3850 km2 (2·5% of the entire land area) of the Mojave Desert burned in 2005 alone (Brooks & Matchett 2006). Considering the recent several millennia, the landscape-scale fires of the 1900s and 2000s are unprecedented in this biome (Brown & Minnich 1986). Our understanding of the responses of desert communities to these novel disturbances is further complicated by poorly developed general principles regarding succession after any type of disturbance in arid lands (McAuliffe 1988).

In many temperate biomes, general patterns of plant succession after disturbances are reasonably well understood. Upon forest clearing in the deciduous forest biome in the eastern USA, for example, succession proceeds through annual plant, perennial grass and forb, shrub, young forest and mature forest seres within a time frame of decades to hundreds of years (e.g. Inouye et al. 1987). No such well-established successional trajectories exist for southwestern deserts, where ecologists have even debated the existence of generalizable successional pathways (Abella 2010). Furthermore, aridity causes the rate of perennial species establishment to be slower than in more mesic biomes, complicating research efforts (Cody 2000). Some ecologists have postulated that communities such as those dominated by Coleogyne ramosissima (hereafter Coleogyne) may require hundreds of years or millennia to reestablish after severe disturbances (Callison, Brotherson & Bowns 1985; Abella 2009). The stressful environment of arid lands offers a unique opportunity to examine whether successional theories that were developed in more mesic biomes apply to other biomes where these principles have been little evaluated.

In this study, we evaluated three questions about secondary (post-fire) succession in arid lands by examining the responses of plant communities to fire on an eastern Mojave Desert landscape. First, we examined to what extent fire influenced species richness and diversity and how long burn effects persisted. The general pattern of post-disturbance species diversity in temperate biomes has been described as an initial decline after disturbance to below pre-disturbance levels, then a sharp increase followed by a leveling off as succession proceeds (e.g. Oosting 1942). A complicating factor is that while removing a dominant species is purported to enhance diversity in some ecosystems by increasing the resources available to other species, in arid lands the dominant late-successional species reduced by disturbance are often nurse plants for other species (McAuliffe 1988; Padilla & Pugnaire 2006). Removing these nurse species could hinder recruitment of other species, making it difficult to predict overall changes in species richness and diversity following disturbance in deserts. Secondly, we assessed whether the rate and direction of succession vary among plant community types. We evaluated this by comparing succession in two of the most dominant communities of the Mojave Desert, Larrea tridentata (hereafter Larrea) and Coleogyne shrublands. Succession in these communities could differ because of several factors, including the geographic distribution of the communities where Coleogyne inhabits higher elevations that are slightly moister and cooler than Larrea (Carpenter, Barbour & Bahre 1986). The reduced stress of the Coleogyne environment could be hypothesized to hasten succession, although some authors have suggested that Coleogyne communities may never recover after severe disturbance in today’s climate (Webb, Belnap & Thomas 2009). Thirdly, we assessed whether environmental contingency effects corresponded with post-fire plant community composition across different communities and time since fire (TSF). The roles of contingency effects are considered potentially important but poorly understood in influencing ecological change (Bakker & Moore 2007), and the aridity of deserts could temper site-based contingency effects such as soil texture. We tested the contingency effects by examining the relationships of soil properties and geographic location with variation in post-fire succession.

Managers are grappling with uncertainty about how long communities take to recover following disturbances such as fire; if natural succession will result in vegetation equivalent to unburned areas; and if active revegetation should be attempted to promote recovery. Our study sought to examine theoretical aspects of plant succession in arid lands while informing management of recovery processes in increasingly burned landscapes. Although numerous disturbances now occupy desert landscapes, many existing studies are site-specific addressing only a single disturbed area (Abella 2010). Little work has been carried out on generalizing patterns of succession across the desert landscape, encompassing varying environmental conditions and plant community types. Specifically, we addressed three predictions about the responses of desert plant communities to fire: (i) Perennial species richness and diversity will follow general post-disturbance patterns described for temperate regions, by declining below unburned levels on burns with short TSFs, then increasing and leveling off on burns with longer TSFs. (ii) The rate and direction of succession differ between community types. (iii) Species composition of successional communities is related to TSF and contingency effects including soil and topographic variation, resulting in different successional trajectories across the landscape.

Materials and methods

Study area

We performed this study on federal land managed by the US Bureau of Land Management, Fish and Wildlife Service, and National Park Service within a 1·8 million ha area in the eastern Mojave Desert in Clark County, southern Nevada, and Mohave County, northwestern Arizona, in the southwestern USA (Fig. S1, Supporting information). Over the last 70 years, the Las Vegas, Nevada, weather station recorded an average of 11 cm year−1 of precipitation, a daily low January temperature of 1 °C and a daily high July temperature of 40 °C (1937–2009 records; The Mojave is classified as a hot desert and receives most of its precipitation in winter, with about 60% falling in the winter months of October–April in the Las Vegas area. Major landforms include mountain ranges, alluvial fans, flat plains, washes forming intermittent stream drainages and volcanic landforms. Aridisols and Entisols constitute the major soil orders (Lato 2006). Communities containing Larrea, often with Ambrosia dumosa, dominate elevations below 1300 m, while Coleogyne communities occupy elevations from c. 1300–1800 m (Rowlands et al. 1982).

Identifying burn areas

We obtained digital polygons of post-1999 fires from a geospatial database (BLM 2010). We digitized perimeters of fires from 1980 to 1988 (1980 marked the earliest date of extensive fire occurrence record keeping) from paper records kept by the Bureau of Land Management, Southern Nevada District Office (Las Vegas, NV) and Lake Mead National Recreation Area (Boulder City, NV), and aerial photographs taken in the 1980s. We sampled fires that occurred in Larrea or Coleogyne communities and that were at least 8 ha in size. In total, we sampled 32 fires, 15 in Larrea and 17 in Coleogyne communities, ranging in size from 12 to 33 064 ha and with burn dates spanning 1980–2007 (TSF = 2–29 years prior to sampling; Table S1, Fig. S1, Supporting information).

Plot location and sampling

Locations for sample plots within paired burned and unburned polygons were generated using the ‘random point’ generator in the AlaskaPak toolkit, version 2·2 (Alaska Support Office, National Park Service, Anchorage, AK, USA), in version 9.3 of the ArcMap Geographic Information System. Unburned plots were located within polygons drawn adjacent to fire perimeters that encompassed topography similar to burns. We sampled 6–12 plots, each 10 × 10 m (0·01 ha), for each fire, equally split between burned and unburned areas and with more plots in the larger fires (Table S1, Supporting information). Plots were sampled from November to June, 2008–2010.

Within each plot, we established six 1-m2 subplots located at each of the four plot corners and centred at 5 m along the north and south plot lines. We categorized the areal cover of all live perennial plant species rooted within each subplot, and all measures for analysis are based on the perennial plant community. The abundance of annual plants sharply varies among years, with some years supporting few to no annuals (Rao, Steers & Allen 2011). Annuals may be important in post-fire recovery, warranting comparisons of annual plant composition sampled within the same year across the chronosequence in future research. Cover in whole plots was recorded for species not detected in the subplots. We used Peet, Wentworth & White (1998) cover classes: 1 = 0·1%, 2 = 0–1%, 3 = 1–2%, 4 = 2–5%, 5 = 5–10%, 6 = 10–25%, 7 = 25–50%, 8 = 50–75%, 9 = 75–95% and 10 = >95%. Nomenclature follows NRCS (2010). We also recorded location [expressed as a Universal Transverse Mercator (UTM) coordinate], elevation, aspect and slope gradient for each plot. The location variable could help capture unmeasured environmental variability and historical factors such as species dispersal.

Near the southwestern and northeastern corners of plots in interspaces (≥1 m from the nearest perennial plant), we collected a total of 700 cm3 of the upper 5 cm of soil for analysis, excluding surface litter, and 240 cm3 for estimating bulk density. Analytical samples were air dried, sieved through a 2-mm sieve and analysed for texture (hydrometer method) following Tan (2005); pH and electrical conductivity (1 : 1 soil:water); available P (Olsen sodium-bicarbonate extraction); CaCO3 (manometer method); total C, N and S (dry combustion, CNS analyzer); organic C (subtraction of inorganic C from total C); NO3, SO4 and Cl (ion chromatography); and the water-soluble concentrations of the elements Na, K, Mg, Ca, Mn, Fe, Ni, Cu, Zn, Co, B, Mo, Pb and Cd (1 : 3 soil:water extracts, inductively coupled plasma mass spectroscopy) following Burt (2004). Bulk density was estimated by sieving samples through a 2-mm sieve and oven drying at 70 °C for 24 h. Bulk density of the <2-mm fraction was then calculated with the volume (estimated by water displacement) of coarse fragments ≥2 mm included.

Data analysis

Within each fire, plots were averaged (separately for burned and unburned areas) to form an independent replicate for the landscape-scale focus of this study, resulting in a sample size of 32 fires each paired with their unburned area. As is typical of temporal patterns of fire occurrence in southwestern deserts (Brown & Minnich 1986), more fires occurred after years of high precipitation, resulting in a clumped temporal distribution of fires (Table S1, Supporting information). We therefore grouped fires by decade (1980s, 1990s, 2000s), resulting in natural breaks between groups of more active fire years. We categorized fires as occurring in Coleogyne communities if the mean relative cover of Coleogyne was ≥35% in unburned areas (determined by natural breaks in the data), and as Larrea communities for all others. Relative percent cover, which we used in all multivariate community analyses, was calculated using the midpoints of cover classes for each species on each plot as: (cover of speciesi/∑ cover of all species) × 100.

Species richness and diversity

We calculated perennial species richness and Shannon’s diversity index at the 0·01-ha plot scale using pc-ord software (McCune & Mefford 1999). To evaluate the effects of fire and community type on species richness and diversity, we used a partially hierarchical analysis of variance (anova) model in which each fire was the subject, plant community type and burn year group were the between-subjects effects, and burn status (burned, unburned) was a within-subject effect. We performed this analysis using proc mixed in sas software (SAS Institute 2001). Statistical conclusions were consistent with or without the inclusion of the single 1980s fire available for Larrea communities, so all data were included in the models. The minimum adequate model was chosen using the Akaike information criterion corrected for small sample size. Based on these criteria, the full model was used to evaluate species richness, and the minimum adequate model needed to evaluate diversity excluded the three-way fixed interactions.

Direction and rate of community change

We ordinated community composition (based on relative cover) using non-metric multidimensional scaling (NMS; slow and thorough mode with Sørensen distance) and joint plots with species and environmental variables as vectors (the length and direction of vectors indicate the strength and direction of their relationship with community composition) using pc-ord software (McCune & Mefford 1999). Within NMS, we computed the successional vectors, portraying the relative deviance in species composition between burned and unburned areas on each fire. Vector lengths, indicative of rates of successional change, were calculated using Euclidean distance and compared with a one-way anova among TSF categories within Coleogyne and with a two-tailed t-test within Larrea communities (only 1990s and 2000s Larrea burns were compared as only one 1980s Larrea fire was available).

Lengths of the successional vectors were compared by translating vectors to a common (unburned) origin, calculating the length of each vector, and comparing means of each community type with a one-way anova (Coleogyne communities) or t-test (Larrea communities). Direction of change between burned and unburned communities was tested by standardizing the length of each vector. Vector angles were then compared using a multiresponse permutation procedure (McCune & Grace 2002).

Community composition and species-specific relationships

Species composition among groups (burned/unburned status, TSF groups, community types and all combinations within) was compared using permutational multivariate analysis of variance (permanova, Anderson 2001) implemented in ‘distlm’ software (Anderson 2004). We further assessed species that distinguished each TSF and community category using indicator species analysis (Dufrêne & Legendre 1997), with 1000 Monte Carlo randomizations for assessing significance at = 0·05 (McCune & Mefford 1999). We calculated indicator values hierarchically first by overall burned/unburned status, followed by each community type within burned/unburned categories, then by burn age group within community types. Indicator species analyses show faithfulness and exclusivity of species occurrence to particular groups (Dufrêne & Legendre 1997).

Contingency effects

We assessed the relationships of environmental variables with burned species composition using a regression tree analysis with Axis 1 and 2 NMS scores as dependent variables in jmp software (SAS Institute 2004). Regression trees are nonparametric models that partition data into increasingly homogenous subsets and provide dichotomous keys to estimate a dependent variable at different values of explanatory variables (Breiman et al. 1984). We stopped splitting when adding more explanatory variables increased r2 by <0·05 or when the minimum node size (= 5) was reached. We employed JMP’s k-fold cross-validation (= 5) to compute a cross-validated r2 [1 – (cross-validated sum of squares error/corrected sum of squares)].


Species richness and diversity

Across all sites, we recorded 129 perennial species in 35 families. Perennial species richness was greater in Coleogyne (10·2 ± 0·8 species 100 m−2) than in Larrea communities (7·1 ± 1·2). This pattern was consistent across all TSF units and between burned and unburned communities (Table 1). There was a TSF and burn status (burned or unburned) interaction within the most recent burns, where species richness was lower in burned than in unburned sites (6·5 ± 0·9 vs. 8·7 ± 0·9 species 100 m−2) within burns that occurred in the 2000s. Burn status did not affect species richness in any of the other year groups. Fire affected diversity in Coleogyne communities, but not in Larrea communities (community type × burn status; = 0·001). Diversity was greater in burned than in unburned plots within Coleogyne communities (1·6 ± 0·1 and 1·4 ± 0·1, respectively) but did not differ within Larrea communities. Species diversity did not differ across all other main and interactive effects.

Table 1.   Summary of statistical results for analyses examining differences among plant community types, burning, and time since fire for Coleogyne and Larrea communities that burned in the 1980s, 1990s, and 2000s, Mojave Desert, USA
EffectResponse variable and statistics
  1. *Partially hierarchical analysis of variance model with Shannon’s diversity index and richness (number of species per 100 m2 plot).

  2. Lengths of vectors were compared with a one-way anova (Coleogyne communities) or t-test (Larrea communities). Statistics correspond with Fig. 1.

  3. The direction (trajectory) of change between unburned and burned communities were compared using a multi-response permutation procedure (T = test statistic, A = chance-corrected within group agreement; McCune & Grace 2002).

  4. ¶Burned and unburned community composition (CR = Coleogyne, LT = Larrea) was compared using a permutational multivariate analysis of variance with distance-based pseudo-F statistics and permutation-derived P-values.

Diversity and richness response across year, community type, and burn*
Year group0·290·7527260·860·435126
Community type1·410·2460264·870·036426
Burn status2·940·0976260·080·923426
Year × community0·750·4806262·290·142526
Year × burn status3·160·0578263·580·042226
Community type × burn status13·220·0011260·600·444326
Year × community type × burn0·410·666260·000·996626
Length of vectors between burned and unburned plots†
 FP > Fd.f.   
Coleogyne communities0·530·6022   
 tP > td.f.   
Larrea communities−2·80·01612   
Successional trajectories across year groups‡
 TAP > T   
Coleogyne communities0·52−0·040·647   
Larrea communities0·06−0·0040·395   
Effects of burning on plant community composition¶
Burn (overall) 15·520·001   
Burn (within group comparisons)CR 20007·030·001   
 CR 19905·270·001   
 CR 19807·570·001   
 LT 20004·460·002   
 LT 19901·080·35   

Direction and rate of community change

The direction of successional change (shown by the spread of points within successional vectors) did not differ among TSF groups within Coleogyne or Larrea communities (> 0·05, Fig. 1, Table 1), indicating that sites were either recovering along similar trajectories or the range of establishing communities was broad enough to mask similarities in successional direction within TSF groups. However, within Larrea communities, the rate of change (length of the vectors between burned and unburned plots) differed between fires in the 1990s and 2000s, indicating a greater difference between burned and unburned plots in the 2000s fires than in the 1990s fires (= 0·02). The rate of change between burned and unburned plots did not differ significantly among burn year groups in the Coleogyne communities. Therefore, older burns in Coleogyne communities were no more similar to unburned plots than the younger burns.

Figure 1.

 Successional vectors from non-metric multidimensional scaling ordination of perennial plant composition between unburned (origin; diamond) and burned sites for each sampled fire within Coleogyne (top) and Larrea (bottom) communities for burns occurring in different decades. Distances between plot designations and the origin are proportional to the difference in community composition between burned and unburned sites.

Community composition and species-specific relationships

Unburned plots within Coleogyne and Larrea communities differed in species composition (F1,27 = 6·63, < 0·001); therefore, all additional analyses were conducted separately between the two community types. Within Coleogyne communities, burned and unburned communities differed from each other within fires from the 1980s (Fig. 2, Table 1). Conversely, while burned and unburned composition differed within Larrea communities that burned in the 2000s, burned and unburned composition within 1990s fires did not differ.

Figure 2.

 Non-metric multidimensional scaling ordination of plant community composition in unburned and burned areas for each of 32 sites in the Mojave Desert, USA. Ordinations are grouped by decade in which burns occurred (columns) and by the community types Coleogyne (top two rows) and Larrea (bottom two rows). For each time and community type grouping, there are two ordinations, one showing joint plots of species associations (top) and one showing environmental correlates (bottom). Joint plots show only those species or environmental variables with r2 > 0·3. Only one fire was available from a Larrea community in the 1980s, so those data are not ordinated. WPLive, live whole plot cover; SG%, % slope gradient; UTMx and UTMy, UTM coordinates for x (easting) and y (northing); sand, % sand; silt, % silt; EC, electrical conductivity; and species codes are in Table S3 (Supporting information).

Species that were strongly associated (by jointplots) with unburned communities were those that typified the community types, such as Coleogyne and Larrea (Fig. 2). Overall, burned sites were dominated by Sphaeralcea ambigua, Baileya multiradiata and Gutierrezia sarothrae, all forb or sub-shrubs that typically respond well and increase in abundance in disturbed areas (Figs. 2 and 3, Table S2, Supporting information). Larrea communities only differed between burned and unburned communities in the 2000s burns, and in that case, Sphaeralcea ambigua was the only species directly associated with burned communities.

Figure 3.

 Indicator species hierarchy for perennial plant species partitioned by community type (Coleogyne or Larrea), burn status (burned or unburned) and decade of burn. The numbers after each species represent the indicator value of that species within that subgroup. Only species statistically significant are shown: *< 0·05; **< 0·01; ***< 0·001.

Compared to Coleogyne communities, there were few indicator species for burned or unburned Larrea communities in the hierarchical indicator species analysis (Fig. 3). Within Larrea communities, indicator species for 1990s burns were characterized by the shrub Ambrosia dumosa and two other shrub species (Ephedra nevadensis and Krameria erecta) that also indicated unburned plots. No indicator species were associated with the 2000s burns. Species unique to the 1980s Coleogyne burns included woody species such as Prunus fasciculata, Fallugia paradoxa and Purshia stansburiana, the grass Achnatherum speciosa, and the cactus Echinocerus engelmannii (also indicative of unburned plots). Additionally, Gutierrezia sarothrae, an overall indicator species for burns across decades and within Coleogyne communities, was also an indicator of 1980s burns. The few indicator species for the 2000s fires consisted of Thymophylla pentachaeta, Ambrosia dumosa and Ephedra nevadensis.

Contingency effects

More environmental variables were correlated with TSF across Coleogyne than Larrea communities (Fig. 2). Furthermore, different suites of variables were associated with different TSF categories within community types. For example, silt, total N and Mg were among the variables correlated with 2000s burned Coleogyne communities, whereas suites of soil micronutrients such as Fe and Zn, but not soil texture, were correlated with 1980s burned Coleogyne communities. Geographic location (expressed as UTM coordinates) was associated with species composition in 1980s and 1990s Coleogyne and the 1990s Larrea communities.

A regression tree model using environmental variables and community type, built for relationships with burned species composition, explained 79% of the variability in NMS ordination Axis 1 (cross-validated r2 = 0·63) and 83% of the variation in Axis 2 (cross-validated r2 = 0·76; Fig. 4). Across the two axes, species composition within burned sites corresponded with community type, location (elevation and latitude), soil texture, N, P and Na. Elevation formed the first division within Axis 1, showing separation of Coleogyne (higher elevation) and Larrea (lower elevation) communities, followed by divisions related to soil Na, silt and N. Community type formed the first and most influential division in Axis 2, followed by UTMx (easting), P and clay.

Figure 4.

 Burned community composition across all community types (NMS ordination plot), associated joint plots and representations of regression trees for ordination axes 1 and 2. Joint plots show only those species with r2 > 0·25. For regression trees, mean axis scores (diamonds) and standard deviations are presented for each hierarchical split. Splitting variables and associated values (directly related to axis scores) for each rule are shown adjacent to the associated means and deviations. Closed diamonds represent the range of data explained along an axis that is associated with values greater than the given splitting value, and open diamonds represent data associated with values less than the split. Cumulative r2 with each addition to the model are given in parentheses. The first branch on the y-axis is the mean % cover of Larrea minus the % cover of Coleogyne (LT-CR) of the unburned sites for each fire (representing community type). UTMx is the coordinate for easting. The species vectors are abbreviated according to Table S3 (Supporting information). Branching structure along axis 1 is: elevations <1118 m are parent to Na, elevations ≥1118 m are parent to silt and silt <34% is parent to total N. For axis 2, LT-CR ≥ −2 is parent to P and P ≥ 25 μg g−1 is parent to clay.


As fires become more prevalent in arid lands, the scars across the landscape are more noticeable, and sensitive plant and animal habitats are lost at an increasing rate (Brooks & Matchett 2006). With increasing fire in an ecosystem where fires (especially large fires) are historically rare and the ecosystem is poorly adapted, the relative impact of fires on biological and ecosystem services may be more dramatic than in ecosystems better adapted to fire. However, the lack of knowledge about successional patterns in arid lands may hinder the ability of land managers to make appropriate decisions when tasked with revegetating existing burned areas with desirable species and attempting to reduce the prevalence of fire effects. The results of this study provide variable support for expectations of post-disturbance species richness and diversity patterns derived from theory developed primarily in wetter ecosystems. However, the results strongly supported the hypothesis that successional patterns differed between community types. Additionally, the strength and importance of contingency effects varied among contingency variables, TSF and community type.

Variable richness and diversity recovery patterns

Perennial species richness on burns was significantly lower than in unburned areas until ≥10 years TSF, contrasting with models from semi-arid woodlands and mesic forests where richness increases within a shorter time period following disturbance to levels greater than prior to disturbance (Fulbright 2004). However, while we did not detect time-dependant oscillations in diversity, the response of diversity to burning was dependent upon the disturbed community type.

In Coleogyne communities, diversity in burned communities was greater than in unburned communities. This is probably because of the establishment of perennial shrubs resulting in more evenly distributed species abundances than in the undisturbed communities where Coleogyne strongly dominates (Keeler-Wolf 2007). Coleogyne responds poorly to disturbances of any type (Callison, Brotherson & Bowns 1985; Webb, Steiger & Newman 1988; Brooks & Matchett 2003; Vamstad & Rotenberry 2010). Eliminating this strong dominant from the system may provide opportunities for the establishment of other species (Emery & Gross 2007). Brooks & Matchett (2003) also reported that richness increased with decreasing Coleogyne abundance when sampling three Mojave Desert sites burned 6–14 years earlier. They found that the establishment of species other than the dominant Coleogyne enriches evenness and enhances species diversity within similar richness levels.

Community identity affects post-burn recovery rates

Mojave Desert landscapes are composed of varying community types that may be governed by different ecological principles within a relatively small geographic region (Keeler-Wolf 2007). We found that recovery rate was dictated less by geography (plots were scattered within the same geographic region) and more by initial community type. Despite occupying lower elevation sites with higher temperatures and lower precipitation, Larrea community composition (although not cover) recovered more rapidly than in Coleogyne, with species composition similar to pre-fire levels in <20 years. Composition in the Larrea communities became more similar to that of unburned communities relatively quickly, but there was no indication of convergence of perennial species composition between burned and unburned sites in Coleogyne communities.

Successional development and composition of burned communities

We did not observe clear successional pathways within a community type and instead observed similar species trajectories across all TSF groupings. While post-fire Larrea communities were becoming more compositionally similar to each other, Coleogyne burned communities continued to comprise species assemblages that differed from those on even the oldest burned sites. Within burned Coleogyne sites, composition was more variable than within undisturbed sites. Sørensen distances were greater among burned sites than among unburned sites, indicating that burned communities had more variation in vegetative composition than unburned plots (Fig. 2).

Overall, burns were differentiated from unburned communities by the presence of species that typically establish in disturbed sites (e.g. Baileya multiradiata, Gutierrezia sarothrae, Sphaeralcea ambigua; Abella 2010). Older burns were typically defined by species unique to their TSF stage, while younger burns had few indicator species, suggesting that the early colonizing species (Table S3, Supporting information) were not unique to recently disturbed areas. This may differ from successional processes in more mesic systems where early colonizers may be shaded out as the community develops. In the case of desert systems, early colonizers may serve as nurse plants (McAuliffe 1988) that encourage the recruitment of native shrubs, and these early colonizers may be present even in mature ecosystems (albeit at low abundance).

Larrea communities vary widely in subdominant perennial species composition among unburned communities (Wells & Hunziker 1976). Therefore, we should recognize that while the total amount of variation (ordination ‘clouds’ of points) between burned and unburned sites completely overlapped, we cannot necessarily infer that each community returned to its unburned state. Rather, the burned community composition fell somewhere within the range of sampled unburned Larrea communities. In contrast, Lathrop & Archibold (1980) found that anthropogenic land clearing in Larrea communities in California resulted in differences in community composition even 65 years after disturbance. This could result from a more severe and lasting disturbance type than fire, or it could reflect the fact that in our study, the variation in unburned Larrea communities across the landscape subsumes the range of possible outcomes. Greater variation within Larrea communities should be further examined to isolate whether subgroups of composition (e.g. variation in Larrea abundance, presence or absence co-dominant species) and environmental variation may affect the recovery rate or act in a deterministic manner to predict post-disturbance community composition.

After c. 30 years of recovery, species composition in burned Coleogyne communities was highly variable. This may indicate that each burn exhibits a unique successional sequence that is not necessarily replicated among disturbances within similar environments and initial communities. Coleogyne communities were drastically different from unburned communities; therefore, fires may induce a shift in successional trajectory such that entirely different communities form in disturbed areas.

The lack of clear successional pathways and predictable species composition indicates the need for further research with the aim of understanding secondary successional patterns and mechanistic processes that drive the trajectories of increasingly disturbed desert communities. For example, some perennial species are known to resprout after light or moderate disturbances (Table S2, Supporting information), but the importance of resprouting in influencing recovery is not well known in deserts (Abella 2009). With many of the species, it is difficult to visually diagnose resprouting without prior knowledge, and coupled with little to no information on the fire severity, we were not able to segregate resprouted from newly established individuals. Additionally, while this study only included perennial species composition in the community analyses, assessing the annual species communities, which shift much more drastically through space and time than perennial species (including significant inter-annual variation within a site; Beatley 1974), would provide further perspective of post-disturbance recovery. Native annual communities may be linked to the establishment of native perennial communities and could play a crucial role in plant community re-establishment (Steers & Allen 2010).

Influence of environmental contingency effects on post-fire community composition

The strongest environmental correlates of burned community composition across sites and burn year consisted of landscape position variables (elevation and latitude), key soil elemental nutrients (Na, P and total N) and soil texture (Fig. 4.). Soil chemistry, particularly soil N and P, has previously been found to be a key secondary predictor (after water availability) of plant community composition in arid lands (Rao et al. 2009). There were few environmental associations within the Larrea community aside from the relatively weak relationships of Fe and NO3 with burned plot composition (Fig. 2.). Within Coleogyne communities, joint plots were generally associated mostly with burned plots, and many soil and environmental variables explained relatively equal amounts of the variation. Therefore, within Coleogyne communities, the data suggest that soil nutrients play a secondary role in determining post-fire composition, but the relationships are relatively weak. Additionally, the regression trees do not support stronger associations of environmental variables with one community type over another, indicating that post-fire community composition is dictated primarily by pre-disturbance community identity.

Management implications

The aim of this work was to inform post-fire management of arid landscapes, to address questions such as how rapidly communities recovered or whether active revegetation may be needed to hasten recovery. Recovery differed between community types and among the components (species richness, diversity, and composition) characterizing the community recovery. For instance, Larrea communities recovered more readily than Coleogyne communities. We observed that fire increased perennial plant diversity in Coleogyne communities, although this increase was at the cost of converting mature shrublands to early successional forb-shrub communities. These communities were persistent for longer than 29 years and showed little trend for convergence with unburned species composition. The data suggest that direct intervention, i.e. seeding or planting, may be necessary in Coleogyne communities to restore keystone shrub species. Additionally, if Coleogyne communities are considered vital habitat, managers may want to consider actions aimed at fire prevention in these habitats. Where resources for fire suppression are limited, priority for management actions could be allocated based on how long the communities take to recover from fire if they were allowed to burn. Given the observed recovery patterns, an important next step is to determine how well the persistent, successional plant communities meet functional management objectives, such as for wildlife habitat, conservation of iconic species (e.g. the Joshua Tree, Yucca brevifolia) or promoting ecosystems that are resistant to invasion by exotic species (in this case, exotic annual grasses such as in the genus Bromus). This study illustrates the importance of a landscape-scale perspective for understanding post-fire recovery, as conclusions about recovery differed sharply between plant community types across different geographic settings. A further understanding of the anticipated recovery trajectories combined with knowledge of the functional benefits for the range of possible post-disturbance communities could assist fire management and revegetation planning.


Cooperative agreements between the University of Nevada Las Vegas (UNLV) and the Bureau of Land Management [Southern Nevada District (SND)] and National Park Service [Lake Mead National Recreation Area (LMNRA)] supported this study. We thank Christina Lund (formerly of SND), Kevin Oliver and Nora Caplette of SND, and Alice Newton (LMNRA), for facilitating work under these agreements; Tim Rash (formerly of SND) for supplying fire records; Adria DeCorte, Teague Embry, Kate Prengaman, Chris Roberts and Sarah Schmid for help with field sampling; and Cheryl Vanier for guidance on anova and permanova analyses. Soil analyses were conducted by the Environmental Soil Analysis Laboratory at UNLV. We thank Stan Smith, the editor and two anonymous reviewers for comments that greatly improved the quality of this manuscript.