The research was conducted in the southern part of central Slovakia within the geomorphological unit of the Cerová vrchovina Upland (Fig. 1). The study area belongs to a warm climatic region (T), situated on the boundary of subregions T3 (warm, very dry, with mild winters) and T5 (warm, moderately dry, with cold winters) (Lapin et al. 2002). The grasslands occur predominantly on south- to southwest-facing slopes with shallow to moderately deep, nutrient-poor soils. Vegetation is dominated by xerothermic grassland communities corresponding to the Natura 2000 habitat types 6210 (Semi-natural dry grasslands and scrubland facies on calcareous substrates) and 6240 (Sub-Pannonic steppic grasslands) (Šuvada 2023) (Table 1). Species-rich xerothermic grasslands of the Festuco-Brometea class dominated by thermophilous grasses such as Festuca rupicola, Bromus erectus, Brachypodium pinnatum, Chrysopogon gryllus and Carex humilis, accompanied by a diverse assemblage of forbs including Salvia pratensis, Pimpinella saxifraga, Helianthemum grandiflorum, and Hippocrepis comosa are present. Long-term abandonment resulted in varying degrees of encroachment by Juniperus communis, Prunus spinosa, Rosa sp., Crataegus monogyna and others (Table 1). The sites are bordered by broadleaved forests and arable lands.

Prescribed burning was carried out during late February in 2023 and/or 2024, under the supervision of staff from the Administration of Cerová vrchovina Upland Protected Landscape Area and local land managers (Table 1). Burns were ignited manually using matches and were allowed to spread freely within patches delimited by a perimeter of approximately 50 cm of bare ground, which served as a firebreak preventing spread into adjacent unmanaged vegetation. At Gemerské Dechtáre, Nagymál-oldal, a single burn was conducted in 2024 covering approximately 2.0 ha. At Jesenské, burns were carried out in both years, covering approximately 2.3 ha in 2023 and approximately 1.3 ha in 2024. At Drňa, burns were similarly conducted in both years, covering 0.15 ha in 2023 and 1.0 ha in 2024. At Petrovce, a single burn was conducted in 2023 covering ca 1.5 ha. These burn sizes are characteristic of small-scale prescribed burning typical of fragmented Central European protected landscapes, where management units are constrained by landscape configuration and conservation objectives.

Control sites represent grassland and scrubland habitats that have been withdrawn from active management for periods ranging from approximately 8 to more than 25 years prior to sampling. These sites were treated as unmanaged controls in all analyses. We note, however, that occasional extensive grazing cannot be entirely excluded at the Petrovce and Gemerské Dechtáre control localities, as pastoral activity may sporadically extend into adjacent low-quality habitats during periods of high grazing pressure on nearby productive grasslands. Such events are considered rare and opportunistic, and are unlikely to have substantially altered the overall management history of these sites. Nevertheless, this potential source of minor disturbance is acknowledged in the interest of transparency.

Microhymenopterans were sampled using Malaise traps deployed within representative areas of each management regime across the nine study localities (Suppl. material 1). A single trap per locality was deployed in line with the standard protocol for landscape-scale Malaise sampling of small Hymenoptera (e.g. Karlsson et al. 2020), where logistical constraints and the high catch efficiency of Malaise traps in this taxon typically preclude within-locality replication; instead, replication is achieved across localities and through repeated sample collection. Collecting bottles filled with pure monopropylene glycol were emptied at regular two-week intervals, and the captured specimens were transferred into 95% ethanol and subsequently stored at –18 °C from early April to late July in 2024 (Fig. 1a–f). Traps were positioned centrally within each grassland management unit to sample the arthropod community representative of the local habitat (Fig. 1a–f). We nevertheless acknowledge that Malaise trap catches can vary with fine-scale trap positioning and habitat configuration (Steinke et al. 2021), and that individuals dispersing through the sampling area from surrounding habitats may contribute to the recorded catch. Because placement criteria were applied consistently across all localities and fire history categories, any such effect is expected to operate uniformly across treatments and should not systematically bias among-treatment comparisons.

The study is focused on microhymenopterans (Parasitoida sensuPeters et al. 2017), because they represent a highly diverse and ecologically important component of terrestrial arthropod communities and may respond sensitively to environmental disturbances, including changes in habitat structure following fire. A total of 52 samples were collected across the sampling season. Microhymenopterans (Parasitoida sensuPeters et al. 2017), together with Evanidae, were sorted to superfamily level, while Chalcidoidea as our main study group was sorted at family level using Goulet and Huber (1993) and Heraty and Woolley (2025). Species-level identification was beyond the scope of the present study. Microhymenopterans are extraordinarily labour-intensive and require extensive specialist expertise across multiple families; family-level analysis is a standard first-step approach in microhymenoptera ecology and captures ecologically meaningful guild-level variation.

We analysed the abundance of microhymenoptera across six superfamilies: Platygastroidea, Chalcidoidea, Diaprioidea, Ceraphronoidea, Cynipoidea, and Evanoidea. Malaise trap exposure varied from 14 to 18 days among observations; accordingly, log-transformed trap exposure time was included as an offset term in all models to standardise abundance estimates.

All analyses were conducted in the R statistical environment (R Core Team 2020). To assess the effects of fire regime and seasonality on abundance, we fitted generalized linear mixed models (GLMMs) using the glmmTMB package (Brooks et al. 2017). A negative binomial error distribution (nbinom2) was specified to account for overdispersion in count data. For each superfamily and for total abundance, the response variable was specimen count, with fire regime and sampling period (month) as fixed effects, locality as a random intercept, and log-transformed trap exposure as an offset. The overall contribution of each fixed effect to model fit was assessed using likelihood ratio tests (LRTs) via the drop1 function, which sequentially removes each term and compares the resulting model to the full model using a chi-square test.

To verify that the five control sites could be treated as a single unmanaged control category, we conducted a complementary analysis restricted to the control subset (n = 28 samples). Specifically, we tested (i) whether community composition differed among the five control sites using PERMANOVA on a Hellinger-transformed family-level community matrix with homogeneity of multivariate dispersion assessed via betadisper; (ii) whether time since abandonment (ranging from 8 to over 25 years) predicted total microhymenoptera abundance via a negative binomial GLMM including sampling month and log-transformed trap exposure as covariates; and (iii) whether absolute abundance differed among control localities. The outcome of this preliminary analysis is reported at the beginning of the Results section and informed our decision to treat the five control sites as a single category in all subsequent models.

To examine fire history effects at finer taxonomic resolution within Chalcidoidea, we repeated the modelling approach at the family level restricted to families exceeding 100 collected specimens: Aphelinidae, Encyrtidae, Eulophidae, Eurytomidae, Mymaridae, and Pteromalidae. Model structure was identical to the superfamily-level analyses. Model coefficients for fire regime terms were exponentiated to yield abundance ratios relative to unburnt controls, with 95% confidence intervals derived from the asymptotic normal approximation on the log scale. The overall significance of the fire regime term in each family-level model was evaluated using LRTs via drop1. Significance of individual fire regime coefficients was assessed using Wald z-tests without multiplicity correction, as each coefficient represents a pre-specified comparison against the unburnt baseline.

Estimated marginal means (EMMs) for fire regimes were calculated with the emmeans package and back-transformed to the response scale (Lenth et al. 2026). Pairwise contrasts among fire regimes were performed on the log scale using Wald z-tests, with p-values adjusted for multiple comparisons using Tukey’s method. Compact letter displays summarizing pairwise differences were generated using Sidak-adjusted p-values, as required when all pairwise comparisons are computed simultaneously. Model fit and distributional assumptions were assessed for all models using simulation-based residual diagnostics implemented in the DHARMa package, including tests for uniformity, dispersion, and outliers (Hartig 2016).

A total of 3,975 microhymenoptera specimens belonging to six superfamilies (Ceraphronoidea, Cynipoidea, Diaprioidea, Evanoidea, Chalcidoidea, Platygastroidea) were analysed. Total abundance of Hymenoptera families differed between burnt and control plots, with consistently higher values recorded in control sites for most taxa (Fig. 2). The most abundant taxonomic groups were Mymaridae, Platygastroidea, Eulophidae, and Encyrtidae respectively. In all these groups the abundance was markedly higher in control plots, with Mymaridae showing the highest overall counts (431 in control vs. 212 in burnt plots), followed by Platygastroidea (425 vs. 311) and Eulophidae (376 vs. 185). A similar pattern was observed for Diaprioidea (326 vs. 199) and Ceraphronoidea (270 vs. 160), as well as for Pteromalidae (140 vs. 55) and Cynipoidea (120 vs. 49). In contrast, some families showed comparable or only slightly different abundances between treatments, such as Evanidae (62 vs. 60) and Eurytomidae (90 vs. 44), although control plots still tended to have higher values. Less abundant families, including Eupelmidae, Torymidae, and Trichogrammatidae, were also generally more numerous in control plots, while several rare groups (Chalcididae, Tetracampidae, Ormyridae, Megastigmidae, Perilampidae, and Metapelmatidae) were recorded in very low numbers, often with minimal differences between treatments. Overall, the data show that most Hymenoptera families exhibited higher total abundance in control plots compared to burnt plots, with differences being most pronounced in the dominant taxa.

Preliminary analysis confirmed that the five secondary succession sites could be treated as a homogeneous unmanaged control category. Community composition did not differ significantly among control sites (PERMANOVA F_4,18 = 0.64, p = 0.887; homogeneous multivariate dispersion: F_4,23 = 0.50, p = 0.736), and time since abandonment did not predict total microhymenoptera abundance (LRT χ² = 1.83, df = 1, p = 0.177). Although absolute abundance differed among control sites (LRT χ² = 13.57, df = 4, p = 0.009), this variation was uncorrelated with successional time and is absorbed by the locality random intercept in our main models. On this basis, the five control sites were pooled into a single unmanaged control category for all subsequent analyses.

Sampling month was a highly significant predictor of total microhymenoptera abundance (χ² = 62.98, df = 5, p < 0.001), with a consistent decline from April through July across all fire history categories (Fig. 3). The omnibus likelihood ratio test for fire regime approached but did not reach significance (χ² = 6.18, df = 3, p = 0.103); however, individual contrasts revealed biologically and statistically meaningful differences among fire history categories.

Sites burned one year prior to sampling showed the highest total estimated abundance (EMM = 85.4, 95% CI: 53.6–136.0 individuals per standardized trap-day), followed by unburnt controls (69.4, 55.7–86.6), same-year burn sites (45.2, 28.2–72.3), and repeatedly burned sites (40.8, 29.2–57.1). Repeatedly burned sites showed significantly lower abundance than unburnt controls (β = –0.530 ± 0.206 SE, z = –2.58, p = 0.010; pairwise ratio = 0.59, p = 0.049), while one-year post-fire sites showed a positive but non-significant trend relative to controls (β = 0.207 ± 0.263 SE, z = 0.79, p = 0.431).

The overall fire regime term approached but did not reach significance in Chalcidoidea (χ² = 7.33, df = 3, p = 0.062) and Cynipoidea (χ² = 6.33, df = 3, p = 0.097) and was non-significant in the remaining four superfamilies (all p > 0.10).

Despite the absence of significant likelihood ratio test effects, consistent directional patterns emerged across superfamilies (Fig. 4). Sites sampled one year after burning (2023) showed the highest estimated marginal means in four of the six superfamilies, suggesting positive response to recent prescribed burning, whereas sites where burning occurred within the same year as sampling regardless of whether they had been burned once (same-year burn) or twice (both years) consistently showed the lowest abundance estimates. This pattern was most pronounced in Chalcidoidea, where one year post-fire sites reached an estimated abundance of 55.5 individuals per standardized trap-day (95% CI: 30.1–102.2), compared to 36.4 (27.4–48.4) at unburnt controls, 19.5 (10.4–36.6) at same-year burn sites, and 17.8 (11.4–27.8) at sites burned in both years. Pairwise contrasts confirmed that both categories of recently burned sites had significantly lower abundance than one-year post-fire sites and unburnt controls: unburnt controls versus repeatedly burned sites (ratio = 0.49, z = –2.65, p = 0.040) and one year post-fire and repeatedly burned sites (ratio = 0.32, z = –2.94, p = 0.017). Individual model coefficients further supported the one year post-fire positive trend in Chalcidoidea (β = 0.420 ± 0.344 SE, z = 1.22, p = 0.222) and Platygastroidea (β = 0.220 ± 0.338 SE, z = 0.65, p = 0.515), though neither reached significance.

Chalcidoidea was the most abundant superfamily in our dataset, comprising 2,306 specimens. Of these, 2,162 specimens (93.7%) belonged to the six families exceeding the 100-specimen threshold for family-level analysis: Mymaridae (n = 643), Eulophidae (n = 561), Encyrtidae (n = 525), Pteromalidae (n = 175), Aphelinidae (n = 134), and Eurytomidae (n = 124). The remaining 144 specimens (6.3%) were distributed across less abundant families that were excluded from family-level modelling.

The overall fire regime term was statistically significant in Aphelinidae (Fig. 5) (χ² = 20.76, df = 3, p < 0.001) and Mymaridae (χ² = 8.36, df = 3, p = 0.039), and marginally non-significant in Eulophidae (χ² = 6.86, df = 3, p = 0.076); it was not significant in Encyrtidae, Pteromalidae, or Eurytomidae (all p > 0.13) (Fig. 5). Within families showing a globally supported fire effect, the most consistent pattern was suppression of abundance at sites burned in both years relative to unburnt controls. In Mymaridae, sites burned in both years showed significantly reduced abundance (ratio = 0.45, z = –2.83, p = 0.005), as did sites burned only in 2024 (ratio = 0.38, z = –2.63, p = 0.009). In Aphelinidae, abundance at sites burned in both years was approximately half that of unburnt controls (ratio = 0.54, z = –2.09, p = 0.037). In Eulophidae and Pteromalidae, the both years coefficient was nominally significant (ratio = 0.42, z = –2.72, p = 0.006 and ratio = 0.34, z = –2.74, p = 0.006 respectively). Sites burned only in 2023 showed consistently positive point estimates across all families relative to unburnt controls, suggesting a potential short-term positive response to recent burning. Seasonality was a dominant driver, with most taxa showing peak abundance in early season and sharp declines towards mid and late summer.

Within the superfamilies Chalcidoidea, Platygastroidea, Ceraphronoidea, and Cynipoidea, we observed a consistent decline in abundance at repeatedly burned sites and at sites burned in the year of sampling (i.e. in 2024). This pattern did not extend to sites burned only in 2023, where higher abundances were recorded, potentially showing gradual recovery of host populations during the year following fire. In contrast, Diaprioidea exhibited a less consistent response, with relatively higher abundances at same-year burn sites and at sites burned in both years. This pattern may be explained by a short-term post-fire increase in dipteran host populations, driven by elevated availability of decomposing organic material, reduced interspecific competition immediately after disturbance, and dispersal-mediated community assembly in which highly mobile dipteran taxa are among the earliest colonisers of recently burned patches (Moretti et al. 2006). Such dispersal-driven early dominance of mobile taxa is consistent with the recolonisation dynamics discussed above for microhymenopterans, and mirrors the broader pattern in disturbance ecology whereby mobile species disproportionately occupy initially empty post-disturbance environments (Fukami 2015). A comparable post-fire increase in dipteran abundance has been documented by Zumr et al. (2025). Similarly, Evanoidea did not follow the general declining trend and showed relatively stable or slightly elevated abundances at same-year burned plots. This may reflect their association with more disturbance-tolerant hosts, such as cockroaches (Stange 2006; Tee and Lee 2017), which can persist or quickly recolonise post-fire environments, as well as a generally higher tolerance of these microhymenopterans to open and disturbed habitats.

The responses of individual families within the superfamily Chalcidoidea likely reflect differences in their ecological requirements and the post-fire availability of their hosts. As noted in previous studies (Tavella et al. 2025), fire substantially alters the overall composition of insect communities, which is in turn reflected in the responses of their parasitoids. However, because microhymenoptera groups differ in their host preferences, the effects of fire on these groups vary accordingly. Several distinct patterns were observed within Chalcidoidea. Because most chalcidoid parasitoids are strong fliers, it is plausible that both one-year post-fire sites and control sites contained sufficient resources to attract individuals from the surrounding landscape (Thompson et al. 2022). In the most abundant family, Mymaridae, which are obligate egg parasitoids of various insect orders (Goulet and Huber 1993; Usmani 2012), a marked decline in abundance was observed in both years and in same-year burned sites. This decline may be attributed to reduced availability of suitable host egg clutches following fire-induced modification of vegetation, on which many hemipteran, coleopteran, and other host taxa deposit eggs, as well as to physical destruction of oviposition substrates during burning (Hill et al. 2019). In Pteromalidae, whose hosts span a broad taxonomic range including Coleoptera, Diptera, Lepidoptera, and Hymenoptera, a significant decline was detected at sites burned in both years, while at same-year burn sites the decline was not statistically significant, although a trend towards increased abundance was suggested at one-year post-fire sites. A similar pattern was observed in Eulophidae, a family that primarily parasitises larval and pupal stages of various insect orders and in which hyperparasitism is also common (Goulet and Huber 1993). The relatively small differences among fire regimes suggest that host populations across most treatments were sufficient to sustain stable Eulophidae abundances (Rameshkumar et al. 2020). In Encyrtidae, differences among fire regimes were not statistically significant. Nevertheless, abundance declines were observed at both years and same-year burned treatments, while a slight increase was recorded at sites one year post-fire, indicating a response consistent with the patterns observed in other chalcidoid families. The host spectrum of Encyrtidae is distinct, encompassing Coccoidea, Aphidoidea, and Psylloidea (Kapranas and Tena 2015), with limited dietary overlap with other chalcidoid families. Their abundance therefore depends indirectly on vegetation condition and on the persistence of vegetation-associated hosts. A comparable response was observed in Aphelinidae, many of which parasitise hemipteran hosts, including aphids, whiteflies and scale insects (Goulet and Huber 1993). Aphelinidae abundance declined significantly at sites burned in both years, likely mirroring insufficient time for aphid host populations to recover between successive winter burns (Quicke 2014). Finally, no significant effect of fire regime was detected in Eurytomidae, indicating that this family showed no directional response to prescribed burning under the conditions of our study. A finer taxonomic resolution would be required to detect responses within this ecologically heterogeneous family (Goulet and Huber 1993).

Our results indicate that the compatibility of prescribed burning with microhymenopteran conservation depends critically on the time interval between successive burns and on the season of sampling relative to the most recent fire. Sites sampled one year after prescribed burning consistently showed the highest microhymenopteran abundances, exceeding both unburnt long-term abandoned sites and sites where burning had occurred within the sampling year. This pattern suggests that occasional prescribed burning, followed by at least one full growing season for community recovery, can support microhymenopteran populations at levels comparable to or exceeding those in long-term unmanaged grasslands, likely by opening vegetation structure, enhancing floral and host resource availability, and slowing successional accumulation of litter and shrub cover (Bonanomi et al. 2022; Page-Dumroese et al. 2023). Similar short-term positive responses of Hymenoptera to fire have been reported by Bogusch et al. (2015). In contrast, sites where burning had occurred within the sampling year whether burned once or in both preceding years showed consistently reduced abundances, which our data attribute to insufficient time for recolonisation and host-population recovery.

Our findings therefore support the use of prescribed burning as a conservation-compatible management tool in high-nature-value semi-natural grasslands (Bonanomi et al. 2022; Page-Dumroese et al. 2023; Stočes et al. 2025). Frequent burning at short intervals, in which successive fires prevent communities from re-establishing between disturbance events, appears detrimental to microhymenopteran abundance and should be avoided where possible (Podgaiski et al. 2018, Hamřík et al. 2025). Our findings, together with previous studies, indicate that maintaining a mosaic of burned and unburned habitats within the landscape may best support persistence and recolonisation of arthropod populations, by preserving regional source populations and sustaining spatial and temporal heterogeneity (Engle et al. 2008).

Future research should focus on species-level analyses and functional traits to achieve a more precise understanding of microhymenoptera responses to disturbance. Although family-level responses provide valuable insights into broad ecological patterns, they may obscure species-level variability and functional differences among taxa.