Fire as a Filter of Insect Assemblages in Algerian Cork Oak Forests: Two-Year Changes in Similarity, Composition and Diversity Across Four Suberaies

1. Introduction

Cork oak forests are among the most emblematic and economically valuable ecosystems of the western Mediterranean Basin. In Algeria they cover approximately 230,000 ha and support a rich invertebrate fauna that contributes to pollination, decomposition, nutrient cycling and pest regulation [1,2]. These forests are also among the most fire-prone landscapes of the region: recurrent summer wildfires, amplified by drought, fuel accumulation and human ignition, have accelerated markedly in recent decades [3].

Fire acts simultaneously as a destructive and a structuring agent. It removes canopy cover, consumes litter, alters soil microclimate and abruptly modifies the availability of trophic and nesting resources. For insects, these changes operate as an ecological filter: species with narrow habitat requirements are eliminated, while mobile, thermophilous or pyrophilous taxa may temporarily benefit from newly opened habitat [4,5]. The resulting assemblage is therefore not a depauperate subset of the pre-fire fauna but a compositionally distinct community.

Most work on post-fire insect responses in the Mediterranean has focused on Coleoptera in pine and holm oak systems, and on the first months following fire. Cork oak forests have received comparatively little attention, and studies that follow assemblages over successive post-fire years while maintaining paired unburnt controls remain scarce, particularly in North Africa. Without such controls it is impossible to distinguish the effect of fire from natural interannual turnover.

Here we address this gap using a paired burnt/unburnt design replicated across four Algerian suberaies and repeated in two consecutive post-fire years. Our objectives were: (i) to quantify the effect of fire history on species richness, abundance and diversity; (ii) to measure compositional similarity between burnt and unburnt stands using the Sorensen index; (iii) to test whether unburnt stands are temporally stable, thereby validating them as references; and (iv) to visualise and statistically test assemblage structure using ordination, set-overlap and chord representations. We hypothesised that fire history would explain more compositional variation than either site identity or year, and that burnt assemblages would show high interannual turnover consistent with early successional dynamics.

2. Materials and Methods

2.1. Study Area

The Collo Massif, located at coordinates 37° 0′ 23″ N, 6° 33′ 39″ E, is a densely forested moun-tain range located in the north-eastern region of Algeria, where it constitutes an integral part of the Tellian Atlas. It ranges in altitude from 980 to 1,183 metres and experiences a humid to sub-humid Mediterranean climate characterised by distinct seasonal variations. The region has a tem-perate and wet winter from December to March, followed by a hot, dry summer from June to Sep-tember. Three unburnt study sites were selected across diverse locations within the massif for this investigation. Cork is derived from the bark of the cork oak (Quercus suber L.), a species char-acterised by semi-evergreen foliage, moderate longevity, and considerable resistance drought and fire conditions, as noted by [6]. The study was conducted in four cork oak (Quercus suber L.) forests of north-eastern Algeria: Djebel El Goufi (locally Lemdjalba), El Maktoua, Oued Agouf and Ain Fegoune. The region has a sub-humid Mediterranean climate with hot, dry summers and mild, wet winters. Each suberaie included a stand burnt by wildfire and an adjacent unburnt stand of comparable altitude, slope, aspect and pre-fire structure, separated by a few hundred metres. This paired design allows the effect of fire to be isolated from broader landscape and climatic variation (Figure 1).

2.2. Sampling Design

Eight stations were surveyed (four suberaies x two fire-history classes: burnt, hereafter FI; unburnt, hereafter FNI) and each station was sampled during two consecutive post-fire years, 2015 and 2016, yielding 16 station-year units. Sampling combined complementary techniques (pitfall traps, coloured water traps, sweep netting and direct searching) applied with identical effort at every station and on every occasion, so that captures are directly comparable among stations and years. Specimens were preserved in 70% ethanol and identified to species whenever possible, otherwise to morphospecies within genus.

2.3. Data Analysis

All analyses were performed in R version 4.3 [7]. Species richness (S), total abundance (N), the Shannon index (H’), the Simpson index (1-D) and Pielou evenness (J’) were computed with vegan [8]. Compositional similarity between pairs of stations was quantified with the Sorensen index [9], Cs (%) = 2c / (a + b) x 100, where a and b are the numbers of species in the two samples and c the number of species common to both. Following common biocoenotic practice we treated Cs >= 60% as indicating assemblages belonging to the same faunal unit and Cs < 60% as indicating distinct assemblages [10]; values are reported so that the more permissive 50% threshold can also be applied.

Assemblage structure was ordinated by principal component analysis (PCA) of Hellinger-transformed abundances, a transformation that renders Euclidean-based ordination appropriate for species count data by down-weighting dominant taxa and avoiding the double-zero problem [11]. Differences among fire-history classes and years were tested by permutational multivariate analysis of variance (PERMANOVA, adonis2, Bray-Curtis distance, 9999 permutations) [12] and by ANOSIM; homogeneity of multivariate dispersion was checked with PERMDISP (betadisper). Patterns of species sharing among the eight stations were visualised with an UpSet plot [13], which generalises the Venn diagram beyond five sets, and with chord diagrams [14] in which ribbon width is proportional to the number of species shared by a pair of stations. Figures were produced with ggplot2 [15] and base graphics.

3. Results

3.1. Abundance, Richness and Diversity

Across the 16 station-year units, 402 insect species were recorded. Unburnt stands accumulated 364 species and burnt stands only 111; 73 species occurred in both classes, 291 were exclusive to unburnt stands and 38 to burnt stands. Species richness per station was consistently lower in burnt stands (8-57 species) than in the paired unburnt stands (57-153 species), and the contrast widened in 2016 (Table 1). Shannon diversity followed the same pattern (burnt 1.79-3.42; unburnt 3.11-4.28), whereas Pielou evenness remained comparatively high in burnt stands, indicating assemblages that are species-poor but not strongly dominated.

3.2. Sorensen Similarity

Similarity between paired burnt and unburnt stands was low in every suberaie and in both years (Table 2). In 2015, Cs ranged from 10.53% (Ain Fegoune) to 26.59% (El Maktoua), mean 18.42%. In 2016 values declined further (6.41-16.51%, mean 10.82%), because unburnt stands continued to accumulate species while burnt stands remained depauperate. Pooling both years raised the values only slightly (16.11-28.57%, mean 20.24%), confirming that the low similarity is not an artefact of incomplete sampling within a single season.

The two temporal comparisons behaved in opposite ways. Unburnt stands compared between 2015 and 2016 retained a large shared core (41-90 species) and reached a mean similarity of 61.87%, with El Maktoua (77.51%) and Ain Fegoune (66.42%) exceeding the 60% threshold and El Goufi (51.90%) and Oued Agouf (51.66%) falling just below it. Burnt stands, by contrast, shared only four to six species between years and averaged 22.51%, revealing near-complete turnover of the assemblage from one post-fire year to the next.

3.3. Ordination of Assemblages

PCA of Hellinger-transformed abundances separated the 16 station-year units primarily according to fire history (Figure 2). PC1 explained 22.1% and PC2 14.6% of the total variance (36.7% cumulatively, a value typical of species-rich assemblage matrices). All unburnt units occupied positive PC1 scores (0.249 to 0.343) whereas burnt units were displaced towards negative scores (-0.636 to 0.038). Burnt units were also more widely scattered than unburnt units, reflecting the idiosyncratic composition of each post-fire assemblage.

Ordinations computed separately for each year confirmed this structure and showed that it was already fully expressed in the first post-fire year and maintained in the second (Figure 3). In 2015, PC1 explained 30.6% and PC2 15.9% of the variance (46.5% cumulatively), with the four burnt stands occupying exclusively negative PC1 scores (-0.459 to -0.326) and the four unburnt stands exclusively positive scores (0.282 to 0.471). In 2016 the separation was equally clear and slightly stronger on the first axis (PC1 33.2%, PC2 15.5%, 48.7% cumulatively), burnt stands ranging from -0.658 to 0.243 and unburnt stands from 0.307 to 0.420. In neither year did a burnt station fall within the unburnt group. Site-level ordinations conducted separately for each suberaie gave the same result (Supplementary Figures S1–S4).

3.4. Patterns of Species Sharing

The UpSet analysis of all eight stations (Figure 4) showed that the largest intersections were those uniting unburnt stations, and that single-station (exclusive) species were far more numerous in unburnt stands. In 2015, exclusive species numbered 15, 28, 34 and 65 in the unburnt stands of El Goufi, El Maktoua, Oued Agouf and Ain Fegoune respectively, against 4, 17, 4 and 3 in the corresponding burnt stands; the pattern was even more pronounced in 2016 (46, 28, 78 and 50 versus 6, 3, 1 and 5; Figure 5). Only two species, the ants Cataglyphis viaticus and Pheidole pallidula, were present in all eight stations in 2015, and a single species, the dung beetle Sisyphus schaefferi, in 2016.

Chord diagrams conveyed the same structure in a pairwise form (Figure 6). Ribbons connecting unburnt stations were markedly wider than those linking burnt to unburnt stations: in 2015, unburnt pairs shared 25 to 61 species, whereas burnt-unburnt pairs shared only 5 to 23. In 2016 the contrast increased further, unburnt pairs sharing 26 to 46 species and burnt-unburnt pairs no more than nine (Figure 7).

3.5. Statistical Tests

PERMANOVA on Bray-Curtis distances confirmed the ordination: fire history explained a significant and substantial fraction of compositional variation (R2 = 0.252, F = 4.97, p = 0.0001), whereas the effect of year (R2 = 0.077, p = 0.099) and the fire x year interaction (R2 = 0.065, p = 0.195) were not significant. ANOSIM gave a high separation statistic (R = 0.724, p = 0.0003). PERMDISP indicated that multivariate dispersion also differed between classes (F = 8.75, p = 0.010), burnt assemblages being the more heterogeneous; part of the PERMANOVA signal may therefore reflect this greater dispersion in addition to a shift in centroid position. Because the ordination shows a clear centroid displacement along PC1 and the two effects act in the same ecological direction, we interpret the result as a genuine compositional shift accompanied by increased variability among burnt stands.

4. Discussion

Our results show that wildfire restructures cork oak insect assemblages far more strongly than either site identity or interannual variation. The similarity between burnt and unburnt stands never approached the threshold conventionally used to recognise a common faunal unit, and the compositional distance between the two classes was significant in every multivariate test applied. The magnitude of the contrast, 364 species in unburnt stands against 111 in burnt stands, with 291 species found only in unburnt stands, indicates that fire does not simply thin the community but replaces it.

The most informative element of the design is the temporal control. Because unburnt stands retained a mean similarity of 61.87% between the two years, the low burnt-unburnt similarity cannot be attributed to sampling noise or to natural turnover: it is attributable to fire. Conversely, burnt stands shared only four to six species between consecutive years, a turnover characteristic of early successional systems in which pioneer colonists of bare ground are replaced as herbaceous and shrubby vegetation re-establishes and modifies resources and microhabitats [4].

The widening of the gap in 2016 deserves emphasis, because it runs counter to a simple expectation of progressive recovery. It arises from two divergent trajectories: unburnt stands continued to accumulate species, whereas burnt stands remained impoverished. Oued Agouf illustrates the extreme case, with eight species in the burnt stand against 148 in the unburnt control. Two years after fire, therefore, the entomofauna had not begun to converge on the reference assemblage. This slow trajectory is consistent with reports that saproxylic and litter-dependent taxa, which depend on structural attributes that take decades to redevelop, are the last guilds to return [5,15].

The composition of the residual shared fauna supports a filtering interpretation. The few species present in every station were ants of the genera Cataglyphis and Pheidole and the dung beetle Sisyphus schaefferi, all mobile, thermotolerant and trophically generalist taxa. Such species buffer the loss of specialists in indices based on presence alone, which is why abundance-based ordination and set-overlap analyses provide a more complete picture than similarity indices used in isolation.

Two limitations should be acknowledged. First, the design comprises eight spatial units per year, so the statistical tests, although clearly significant, rest on a modest number of replicates, and the significant PERMDISP result means that heterogeneity and location effects cannot be fully separated. Second, sampling covers only the first two post-fire years; longer series will be required to determine when, and whether, convergence begins. Standardising effort across stations and retaining paired controls, as done here, remains the most effective safeguard against these constraints.

From a management perspective, the persistence of a distinct and impoverished assemblage two years after fire argues for protecting unburnt patches within and around burnt perimeters, since they act as refuges and as sources for recolonisation. Retaining deadwood and litter where safety permits, and avoiding salvage operations that remove the residual structural complexity of burnt stands, would be expected to accelerate the return of the specialised fauna.

5. Conclusions

In four Algerian cork oak forests, fire history was the dominant determinant of insect assemblage composition. Burnt and unburnt stands shared only a small fraction of their species, similarity between them declined rather than increased between the first and second post-fire year, and unburnt stands proved temporally stable while burnt stands turned over almost completely. Ordination, set-overlap and chord analyses converged on the same conclusion, and PERMANOVA attributed a significant quarter of the compositional variance to fire history. Two years is therefore too short a period for the entomofauna of cork oak forests to recover, and the conservation of unburnt refuges should be treated as a priority in post-fire management.