Local Tree Cover and Regional Climate Hierarchically Shape Ant Communities in Mediterranean Dehesas

Abstract

Mediterranean dehesas are open agroforestry systems in which tree configuration and climatic regime condition the organisation of biodiversity. In these landscapes, ants are commonly used as ecological indicators, although the relative importance of local versus regional factors in structuring their communities remains poorly defined. Ant assemblages were sampled using pitfall traps at 15 farms in southern Spain, and the influence of environmental variables defined at two spatial scales was analysed: microhabitat, distinguishing between areas under tree canopy and open areas, and farm as a unit representative of the regional context. The multivariate analyses applied (dbRDA, PERMANOVA and variance partitioning) reveal a hierarchical organisation of community assemblages. At the local scale, community variation was primarily explained by structural attributes of the tree layer, particularly canopy cover and distance to trees. At the farm scale, environmental predictors explained a modest proportion of community variation, with strong overlap among climatic, vegetation and structural variables. Overall, the structure of ant communities in dehesas follows a scale-dependent pattern, in which climate sets the regional framework and tree structure modulates assemblage organisation at a fine scale.

1. Introduction

Dehesas constitute one of the most extensive and representative agrosilvopastoral systems in southwestern Europe, particularly in the Iberian Peninsula, where they form landscapes of high ecological, productive and cultural value [1,2,3]. They are characterised by a dispersed tree canopy dominated by species of the genus Quercus and an herbaceous matrix subject to extensive livestock grazing and long-term historical management. This management consists of the combination of extensive grazing and the selection and management of trees, which has generated high structural and functional heterogeneity [4,5]. This complexity has enabled dehesas to maintain high levels of biodiversity compared with other ecosystems subjected to intensive agriculture, acting as refuges for numerous plant and animal species [6,7]. However, these ecosystems are currently threatened by processes of intensification, abandonment of traditional management and socio-economic changes, which are compromising their ecological functionality and their capacity to provide ecosystem services [6,8].

Dispersed trees constitute a key structural element in dehesas, as they directly modulate the microclimatic and edaphic conditions of the environment. The presence of trees reduces the radiation reaching the soil, buffers extreme temperatures, increases soil moisture, and promotes the accumulation of organic matter, generating mosaics of microhabitats that contrast with the surrounding matrix [9,10,11,12,13,14]. In human-modified landscapes, isolated trees have been identified as keystone structures [15]: structures whose contribution to ecosystem functioning and biodiversity maintenance is disproportionately high relative to the small area they occupy.

Assessing the ecological status of complex agroforestry systems such as dehesas requires the use of biological indicators that are sensitive to environmental gradients and management changes. Among epigeic arthropods, ants (Hymenoptera: Formicidae) stand out due to their high diversity, abundance, and functional relevance, as well as their predictable responses to environmental factors and disturbances, which has widely supported their use as bioindicators [16,17]. In addition, they play an essential role as ecosystem engineers as they alter soil structure and fertility, nutrient cycling and the redistribution of resources, with effects that extend beyond their own biomass [18,19,20].

In Mediterranean ecosystems, the composition and structure of ant communities have been closely linked to thermal gradients, resource availability and habitat structure, showing different patterns depending on the spatial scale considered [21,22,23]. At the local scale, factors such as vegetation cover, proximity to trees and immediate microclimatic conditions can modulate surface activity and species diversity [24,25]. At broader scales, regional environmental filters, particularly maximum temperatures during the period of activity and water availability, condition community presence and composition, acting as physiological and biogeographical limits [26,27].

Beyond their value as bioindicators, there is growing applied interest in the use of ants as a surrogate group for epigeic arthropods as a whole. In dehesa systems of northern Andalusia, a significant correspondence has been demonstrated between the patterns of diversity and composition of ant communities and those of the overall epigeic arthropod assemblage captured using pitfall traps, reinforcing their usefulness as an efficient ecological diagnostic tool in these landscapes [28]. Nevertheless, the effectiveness of surrogacy and the interpretation of observed patterns depend critically on the spatial scale at which both communities and their environmental predictors are evaluated [16,29].

In this context, the present study adopts a multiscale approach to evaluate the factors structuring ant communities in Mediterranean dehesa systems. Two main hypotheses are proposed: (H1) at the local (microhabitat) scale, vegetation cover and, in particular, dispersed trees act as generators of microenvironmental heterogeneity, modulating community diversity and composition; and (H2) at the landscape (farm) scale, the main determinants of community composition correspond to climatic filters, especially maximum temperatures during the period of activity and water availability, overriding local structural factors. Testing these hypotheses advances our understanding of the multiscale processes operating in dehesas and helps assess the potential of ants as ecological bioindicators in these agroforestry systems.

2. Materials and Methods

2.1. Study Area

The study was conducted in the dehesa system of Sierra Morena (southern Iberian Peninsula), a Mediterranean agroforestry landscape characterised by a matrix of wooded grasslands mainly dominated by species of the genus Quercus. This system exhibits high structural and functional heterogeneity, resulting from the historical interaction between climatic, edaphic, and traditional management factors. The climate is Mediterranean, with hot, dry summers and mild winters, and marked interannual variability in precipitation.

A total of 15 farms were sampled (Figure 1,

Table 1. Geographic location and general characteristics of the dehesa farms included in the study.

Code	Farm	Location	Province	Lat	Long	Surface (ha)	Sampling Year
AP05	La Juanita	Alosno	Huelva	37.555913	−7.08225	191.14	2017
AP06	Paymoguillo	Paymogo	Huelva	37.751957	−7.330534	109.4	2016
AS02	El Palomar de la Morra	Pozoblanco	Córdoba	38.348277	−4.819241	96.69	2017
AS05	Lote de los Pérez	Cazalla de la Sierra	Sevilla	37.896797	−5.875586	107.7	2016
AS06	Las Morrillas	Pozoblanco	Córdoba	38.36055	−4.771079	157.42	2017
CO01	Las Ánimas	Aroche	Huelva	37.962325	−7.01239	77.62	2016
CO05	Monterrey y Carretero	Aroche	Huelva	37.905465	−7.049219	119.59	2017
CO08	Quebradahonda	Castillo de las Guardas	Sevilla	37.653961	−6.421098	114.14	2016
CO12	Majada del Indio	El Viso	Córdoba	38.545339	−4.986985	123.94	2016
EN04	Encinarejo	Alosno	Huelva	37.551441	−7.148217	281.33	2016
FA01	Las Hazas	Villanueva de Córdoba	Córdoba	38.40372	−4.603515	459.09	2016
FA05	La Panadera	Pozoblanco	Córdoba	38.381964	−4.758971	83.6	2017
FA11	Santa Clotilde	Cardeña	Córdoba	38.202356	−4.286978	292	2016
UP23	Oropesa	Fuente Ovejuna	Córdoba	38.299421	−5.463488	110.69	2016
UP24	Las Caras	Vilches	Jaén	38.229856	−3.544986	592.61	2016

), corresponding to those included in a LIFE project focused on the conservation and sustainable management of dehesas. Taken together, these farms encompass a large proportion of the environmental and structural variability characteristic of dehesa systems in Sierra Morena, including gradients of tree cover, vegetation productivity, and topographic conditions.

2.2. Sampling Design and Ant Sampling

At each farm, a total of 40 pitfall traps were established for the capture of epigeic ants. The traps were arranged along a single linear transect of approximately 1600 m in length and placed at intervals of at least 40 m (Figure S1). This distance is commonly used in studies of surface-active ant communities and is considered sufficient to ensure sample independence without compromising spatial representativeness. For field organisation, the traps were grouped into sets of 10, hereafter referred to as “lines”. This term is used only for logistical purposes and does not represent an analytical unit. Each trap consisted of a translucent plastic urine-sample container (mouth diameter 5.7 cm, base diameter 5 cm, depth 7.3 cm, capacity 150 mL; Ref. 409702, Deltalab SL, Rubí, Spain).

The traps consisted of plastic containers buried flush with the soil surface and partially filled (30–35 mL) with a 1% soapy solution, intended to reduce surface tension and increase capture efficiency [30]. The sampling period was selected to coincide with the peak of surface activity of ants in Mediterranean ecosystems, determined primarily by thermal constraints [31]. Traps remained active for 48 h, an exposure time that minimises the digging-in effect associated with their installation; in Mediterranean ecosystems, differences in community composition attributable to this effect disappear after 48 h [32].

Sampling was conducted in 2016 and 2017, with each farm surveyed in a single year (see Table 1). Consequently, year was treated as descriptive metadata rather than as a predictor, based on the interannual stability documented in previous studies [28]. Captured ants were separated from the rest of the sample and preserved in ethanol for identification to species level using specialised taxonomic keys. Other epigeic arthropods collected in the pitfall traps were not considered in the present study and were excluded from further analyses.

2.3. Data Treatment and Quality Control

Traps with no captures were excluded from the analyses when sampling incidents were identified (e.g., trampled traps, incorrectly placed traps, or traps that had been disturbed). To rule out potential ecological bias, the environmental conditions of excluded traps were visually compared with those retained, and no systematic differences likely to affect the results were detected.

2.4. Environmental Variables

Environmental variables were compiled at two spatial scales (trap and farm) and grouped into different ecological blocks, including climate, vegetation/productivity, topography, and tree structure (Table S1). Climatic variables corresponded to the 19 bioclimatic variables (BIO1–BIO19) derived from the CHELSA v2.1 database [33]. Given the spatial resolution of the CHELSA dataset (~1 km), climatic values did not vary among traps within the same farm and therefore represent the broader climatic context of each site rather than local microhabitat conditions. Vegetation and productivity variables were obtained as spectral indices (NDVI, EVI, NDMI, BSI, NBR, and SAVI) calculated from Landsat [34] and Sentinel-2 imagery [35], processed using Google Earth Engine [36]. In addition, NDVI metrics derived from MODIS [37] were included to characterise peak productivity. Topographic variables (elevation, slope, and aspect) were obtained from the 5 m resolution Digital Elevation Model of the PNOA [38].

At the trap scale, vegetation and tree structure variables were calculated within a circular buffer with a radius of 10 m, an appropriate scale for capturing microhabitat conditions relevant to epigeic ants, which influence their activity and foraging [25]. Local tree structure metrics included canopy cover within a 10 m buffer around each trap (TREE_COV10_PCT), calculated from classified high-resolution PNOA orthophotographs refined by manual crown delineation, and distance to the nearest tree (TREE_DIST), together with topographic variables.

2.5. Tree Sampling and Dehesa Canopy Cover

Different tree-related variables were measured between spring and autumn of 2014 and 2015. At each farm, two transects approximately 120 m in length and 20 m in width were established, including a total of 20 trees. In farms with low tree density, transects were extended until 20 trees were reached or a maximum length of 300 m was attained.

Trees were marked with numbered tags and the species of each individual was recorded. For each tree, dasometric measurements were obtained, including trunk circumference at breast height (CBH, measured at 1.3 m above ground level), total height, and crown diameter. Height and crown diameter were measured using a Christen ruler, while trunk circumference was measured with a measuring tape. Mean values per farm were calculated from these variables and used to characterise tree structure at the farm scale.

2.6. Statistical Analyses

To examine the relationship between ant communities and the environment, multivariate analyses were conducted at two spatial scales (trap and farm). In all cases, analyses focused on community-level patterns.

2.6.1. Response Matrices and Dissimilarities

At the trap scale, the community matrix was constructed from species abundances per trap. At the farm scale, community response was defined as species occupancy, expressed as the number of traps occupied and the corresponding percentage relative to the total number of traps sampled at each farm. Community dissimilarities were calculated using the Bray–Curtis index (for abundance or percentage occupancy data) and, additionally at the trap scale, using the Jaccard index based on presence–absence data.

2.6.2. Selection of Environmental Predictors

Environmental variables were grouped into ecological blocks (climate, vegetation/productivity, topography, and tree structure). Within each block, collinearity was assessed using Spearman correlations, and highly correlated predictors (r ≥ 0.7) were removed, retaining a non-redundant subset of variables.

The BIOENV procedure [39], implemented in the vegan package [40], was applied to the previously filtered set of variables. BIOENV identifies the subset of predictors showing the highest Spearman correlation with community dissimilarities through an exhaustive search of possible combinations. At the trap scale, BIOENV was run for both Bray–Curtis and Jaccard dissimilarities, and the union of variables selected under both criteria was used in subsequent analyses. At the farm scale, BIOENV was applied to Bray–Curtis dissimilarities calculated from percentage occupancy data.

In all cases, BIOENV was applied to the original (untransformed) community matrices, as the method is rank-based and robust to data scale.

2.6.3. Community–Environment Relationships

To explore patterns of community structure and their relationship with environmental predictors, we applied a complementary multivariate framework combining constrained ordination (dbRDA), unconstrained ordination (NMDS), and permutation-based hypothesis testing (PERMANOVA).

The association between community composition and selected environmental predictors was evaluated using distance-based redundancy analysis (dbRDA), employing the capscale function of the vegan package. For these analyses, community matrices were first Hellinger-transformed in order to operate in a Euclidean space, and model significance was subsequently assessed using permutation tests.

In addition, permutational analysis of variance (PERMANOVA; adonis2 function) was applied using Bray–Curtis dissimilarities, performing both sequential and marginal tests. At the trap scale, permutations were restricted by farm to account for the hierarchical sampling structure, using restricted permutation schemes defined with the permute package.

To verify that differences detected by PERMANOVA were not influenced by heterogeneity in multivariate dispersion among groups, homogeneity of dispersions was assessed using PERMDISP (betadisper function, vegan package), with permutation tests (permutest), considering farms as groups.

As a complement to constrained multivariate analyses, a non-metric multidimensional scaling (NMDS) ordination was performed based on Bray–Curtis dissimilarities calculated from species percentage occupancy at the farm scale. Environmental variables selected for this scale were fitted to the ordination using envfit (999 permutations), and only those showing significant associations with the ordination configuration (p < 0.05) were represented.

2.6.4. Variance Partitioning

Finally, the relative contribution of different predictor blocks to variation in community composition at the farm scale was quantified using variance partitioning (varpart). Predictor sets included climate, vegetation/productivity, topography, and space, with the spatial component represented by the geographic coordinates (XM, YM) of farm centroids. Variance partitioning was conducted exclusively at the farm scale; therefore, local tree structure variables derived at the trap scale were not included. The significance of individual fractions was evaluated using partial redundancy models (pRDA) using the capscale function.

All statistical analyses were performed in the R environment [41], using primarily the vegan package for ecological community analyses [40] and the permute package for defining restricted permutation schemes [42].

3. Results

3.1. Trap Scale: Local Variation in Community Composition

A total of 568 traps out of the 600 deployed were analysed. The 32 traps not included in the analyses corresponded to traps that were filled with soil, broken, or displaced (see Table 1). A total of 49 ant species belonging to 17 genera were recorded (

Table 2. Percentage of traps occupied by each ant species in each farm, and number of farms in which each species was recorded; values represent the percentage of analysed traps occupied by each species within each farm; a trap was considered occupied when at least one worker of the species was recorded; the final column (“P/F”) indicates the number of farms in which each species was detected, and the last row (“N_SPECIES”) shows the total number of species recorded per farm.

Species	AP05	AP06	AS02	AS05	AS06	CO01	CO05	CO08	CO12	EN04	FA01	FA05	FA11	UP23	UP24	P/F
Aphaenogaster dulcineae Emery, 1924	0	0	21	11	3	3	0	3	0	0	8	3	5	5	5	10
Aphaenogaster gibbosa (Latreille, 1798)	11	8	21	13	5	3	5	5	16	14	8	5	39	3	5	15
Aphaenogaster iberica Emery, 1908	0	18	73	42	58	3	0	28	50	17	54	66	56	10	18	13
Aphaenogaster senilis Mayr, 1853	66	20	3	0	0	55	50	0	3	43	0	0	0	45	13	9
Camponotus cruentatus (Latreille, 1802)	0	0	46	0	28	29	34	3	37	20	41	58	8	10	0	11
Camponotus fallax (Nylander, 1856)	0	0	3	0	0	0	0	0	0	0	0	5	0	8	0	3
Camponotus foreli Emery, 1881	13	0	0	0	0	3	0	0	8	3	0	3	3	3	0	7
Camponotus lateralis (Olivier, 1792)	3	3	0	3	3	0	0	0	0	0	0	0	0	0	0	4
Camponotus micans (Nylander, 1856)	0	0	0	0	0	5	0	0	5	0	3	0	0	0	0	3
Camponotus piceus (Leach, 1825)	0	0	0	0	0	0	0	0	0	0	0	0	5	0	0	1
Camponotus pilicornis (Roger, 1859)	3	5	6	13	3	0	0	3	11	3	8	0	5	5	0	11
Camponotus sylvaticus (Olivier, 1792)	0	3	0	3	0	0	0	3	0	6	0	0	0	5	5	6
Cataglyphis hispanica (Emery, 1906)	40	68	88	95	73	18	68	65	90	66	97	97	87	85	69	15
Cataglyphis iberica (Emery, 1906)	13	10	33	0	0	11	0	0	0	3	0	16	0	15	0	7
Cataglyphis rosenhaueri Santschi, 1925	0	0	0	16	3	0	0	10	0	0	0	3	8	23	13	7
Colobopsis truncata (Spinola, 1808)	0	0	0	5	0	0	0	0	0	0	0	5	0	0	0	2
Crematogaster auberti Emery, 1869	18	0	0	0	0	5	3	3	5	6	8	8	10	5	0	10
Crematogaster scutellaris (Olivier, 1792)	16	10	9	11	28	13	16	5	0	11	11	3	10	15	3	14
Crematogaster sordidula (Nylander, 1849)	0	0	0	0	0	0	0	3	3	0	0	0	0	0	0	2
Formica cunicularia Latreille, 1798	0	0	0	0	3	0	0	0	0	0	0	0	0	0	0	1
Formica gerardi Bondroit, 1917	0	0	3	0	3	0	0	0	0	0	0	0	0	0	0	2
Goniomma baeticum Reyes & Rodriguez, 1987	0	0	12	0	3	5	0	3	0	3	0	3	0	0	3	7
Goniomma hispanicum (André, 1883)	3	0	9	3	10	16	0	3	0	3	0	11	0	10	13	10
Goniomma kugleri Espadaler, 1986	3	3	0	0	0	0	0	0	0	9	0	0	0	0	0	3
Iberoformica subrufa (Roger, 1859)	21	75	49	82	35	50	53	85	37	34	51	42	74	73	21	15
Lasius grandis Forel, 1909	0	0	0	0	0	0	0	0	0	0	0	0	8	0	0	1
Lasius lasioides (Emery, 1869)	3	13	3	5	10	3	5	5	0	3	16	8	3	20	0	13
Messor barbarus (Linnaeus, 1767)	37	50	94	66	85	58	21	33	29	26	62	87	46	78	69	15
Messor bouvieri Bondroit, 1918	3	0	3	0	0	0	0	0	0	0	8	0	3	0	18	5
Messor celiae Reyes, 1985	0	0	0	0	0	5	0	0	0	0	0	3	0	0	0	2
Messor hispanicus Santschi, 1919	3	0	0	0	10	3	0	8	0	0	0	8	3	5	5	8
Oxyopomyrmex saulcyi Emery, 1889	3	0	3	5	0	5	3	18	0	6	3	5	3	0	3	11
Pheidole pallidula (Nylander, 1849)	13	20	0	0	0	3	3	0	21	6	0	3	0	0	3	8
Plagiolepis pygmaea (Latreille, 1798)	26	18	18	34	8	11	18	5	24	31	14	3	39	15	10	15
Plagiolepis schmitzii Forel, 1895	18	10	0	5	0	13	5	25	13	6	3	0	3	5	0	11
Proformica ferreri Bondroit, 1918	0	0	0	0	0	0	0	0	0	0	0	0	13	0	0	1
Solenopsis sp.	5	0	3	24	10	3	5	3	8	0	5	8	10	8	3	13
Tapinoma nigerrimum complex (Nylander, 1856)	47	10	49	55	30	50	16	53	76	31	22	74	46	68	3	15
Temnothorax alfacariensis Tinaut & Reyes-López, 2000	0	0	0	0	0	0	0	0	0	0	0	5	0	0	0	1
Temnothorax angustulus (Nylander, 1856)	0	0	0	0	0	0	0	0	0	0	5	0	0	3	0	2
Temnothorax racovitzai (Bondroit, 1918)	5	0	0	8	3	0	0	0	16	6	3	5	3	0	0	8
Temnothorax recedens (Nylander, 1856)	5	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1
Temnothorax tyndalei (Forel, 1909)	0	0	0	0	0	0	0	0	3	0	0	0	0	0	0	1
Tetramorium caespitum (Linnaeus, 1758)	0	8	0	0	0	0	5	0	3	0	0	0	0	5	0	4
Tetramorium forte Forel, 1904	21	5	61	29	58	34	26	40	5	6	14	71	23	15	33	15
Tetramorium semilaeve André, 1883	45	15	49	40	15	18	37	40	13	29	32	50	41	13	21	15
N_SPECIES	26	20	23	22	23	26	18	24	22	25	22	28	26	27	21

).

The ant community was dominated by thermophilous species characteristic of open Mediterranean environments, particularly Cataglyphis hispanica (Emery, 1906), Messor barbarus (Linnaeus, 1767), Iberoformica subrufa (Roger, 1859), and the Tapinoma nigerrimum species complex (Nylander, 1856), which were present in most of the sampled farms. In addition, species associated with more structurally complex microhabitats or higher tree cover were recorded, especially representatives of the genus Aphaenogaster and several species of Temnothorax, reflecting the structural heterogeneity characteristic of dehesa ecosystems. The coexistence of a high number of generalist species together with the presence of specialist taxa suggests a community assemblage influenced by both regional thermal conditions and local microenvironmental variability.

The set of environmental variables selected by BIOENV to explain variation in community composition at this scale included two climatic variables (BIO2_trap, mean diurnal temperature range; BIO3_trap, isothermality), two tree-structure variables (TREE_COV10_PCT, tree canopy cover within a 10 m radius; TREE_DIST, distance to the nearest tree), and one topographic variable (ELEV_M, elevation) (

Table 3. Environmental variables used in the final analyses at trap and farm scales.

Scale	Variable	Block	Description	Source
Trap	BIO2_trap	Climate 1	Mean diurnal temperature range	CHELSA
Trap	BIO3_trap	Climate 1	Isothermality	CHELSA
Trap	TREE_DIST	Tree structure 5	Distance from each trap to the nearest tree (m)	Remote sensing
Trap	TREE_COV10_PCT	Tree structure 5	Tree canopy cover within a 10 m buffer around each trap (%)	Remote sensing
Farm	BIO3_farm	Climate 1	Isothermality	CHELSA
Farm	BIO10_farm	Climate 1	Mean temperature of the warmest quarter	CHELSA
Farm	NDVIsd_farm	Vegetation productivity 3	Temporal variability of NDVI during 2016–2017 (50th percentile)	Remote sensing
Farm	CBHm_farm	Tree structure 4	Mean trunk circumference at breast height (cm)	Field-derived
Farm	ELEVm_farm	Topography 2	Elevation (50th percentile)	Remote sensing

¹ Climatic variables were obtained from the CHELSA v2.1 bioclimatic dataset and extracted at trap locations or aggregated at farm scale using spatial zonal statistics. ² Topographic variables were derived from digital elevation models. ³ Vegetation indices were calculated from multispectral satellite imagery processed using the Google Earth Engine platform. ⁴ Tree-structure variables at farm scale were obtained from standardised field-based dendrometric measurements. ⁵ Tree-structure variables derived from remote sensing were obtained from the SIOSE (Sistema de Información sobre Ocupación del Suelo de España) land-use database and complemented with high-resolution orthophotographs from the PNOA (Plan Nacional de Ortofotografía Aérea) for spatial verification and refinement.

).

The distance-based redundancy analysis (dbRDA) indicated that the global model was significant (F = 12.389, p = 0.001). In the sequential analysis, four of the five predictors contributed significantly to explaining community variation (BIO3_trap, TREE_COV10_PCT, ELEV_M, and TREE_DIST; p = 0.001), whereas BIO2_trap did not show a significant effect (p = 0.257). In the marginal analysis, which evaluates the unique contribution of each variable, only tree canopy cover (TREE_COV10_PCT; p = 0.001), elevation (ELEV_M; p = 0.002), and distance to the nearest tree (TREE_DIST; p = 0.001) showed significant effects (Figure 2).

PERMANOVA analyses showed a similar pattern, with all terms significant in the sequential analysis (bio2: F = 9.1147, R² = 0.0147, p = 0.001; bio3: F = 20.5492, R² = 0.0332, p = 0.001; TREE_COV10_PCT: F = 16.5282, R² = 0.0267, p = 0.001; ELEV_M: F = 6.9171, R² = 0.0112, p = 0.003; TREE_DIST: F = 3.3054, R² = 0.0053, p = 0.001), whereas in the marginal analysis only TREE_COV10_PCT (F = 6.5619, R² = 0.0106, p = 0.001), ELEV_M (F = 6.9915, R² = 0.0113, p = 0.004), and TREE_DIST (F = 3.3054, R² = 0.0053, p = 0.001) remained significant. The PERMDISP test revealed significant heterogeneity in multivariate dispersion among farms (F = 3.6716, p = 0.001), indicating that differences detected by PERMANOVA may reflect both shifts in centroid position and variation in within-group dispersion. Dispersion diagnostics showed that farms EN04, AP05, CO05, AP06, and CO01 had the greatest mean distances to the centroid (0.569–0.604), compared with an overall range of 0.493–0.604 across all farms.

3.2. Farm Scale: Landscape-Level Variation

At the farm scale, community composition was characterised using the percentage occupancy of the 49 species across the 15 sampled farms. The set of selected environmental variables included two climatic variables (BIO3_farm, isothermality; BIO10_farm, mean temperature of the warmest quarter), one vegetation variable (NDVIsd_farm, temporal variability of NDVI), one topographic variable (ELEVm_farm, median elevation), and one structural variable (CBHm_farm, mean trunk circumference at breast height).

The global dbRDA was significant (F = 1.8133, p = 0.001). In the sequential analysis, the climatic variables BIO3_farm (F = 3.8247, p = 0.001) and BIO10_farm (F = 1.8811, p = 0.036) contributed significantly to explaining community variation, whereas NDVIsd_farm (F = 1.3799, p = 0.156), ELEVm_farm (F = 1.0689, p = 0.358), and CBHm_farm (F = 0.9121, p = 0.507) did not show significant effects. In the marginal analysis, none of the predictors showed a significant unique contribution (BIO3_farm: F = 1.0188, p = 0.400; BIO10_farm: F = 1.1592, p = 0.282; NDVIsd_farm: F = 1.3489, p = 0.165; ELEVm_farm: F = 1.0287, p = 0.392; CBHm_farm: F = 0.9121, p = 0.536). PERMANOVA results reflected the same pattern, with significant sequential effects for BIO3_farm (F = 4.0560, R² = 0.2128, p = 0.001) and BIO10_farm (F = 2.2334, R² = 0.1172, p = 0.019), but no significant marginal effects.

The NMDS ordination at the farm scale (stress = 0.157) showed a community structure mainly associated with broad climatic gradients and environmental heterogeneity among farms. Environmental variables fitted using envfit showed significant associations with isothermality (BIO3_farm), elevation (ELEVm_farm), and temporal variability of NDVI (NDVIsd_farm), whereas other variables considered at this scale, such as mean temperature of the warmest quarter (BIO10_farm) or mean tree cover, did not show significant relationships with the ordination configuration (Figure S2; Figure 3).

Variance partitioning (VARPART) showed that the four predictor sets considered (climate, vegetation, topography, and space) jointly explained 21.8% of the variation in community composition. However, partial redundancy analyses (pRDA) indicated that none of the individual fractions were significant, including the pure climate fraction (F = 0.8927, p = 0.651) and the vegetation fraction (F = 0.9933, p = 0.481), reflecting strong overlap among predictor blocks and a predominance of unexplained variation (78.2%).

4. Discussion

This study addresses a central question in the ecology of Mediterranean agroforestry systems: whether the clearly detectable local-scale influence of dispersed trees on ant communities is maintained when the analysis is expanded to the farm level, or whether regional climatic gradients instead predominantly structure assemblages. The distinction between local and regional processes, widely formalised in spatial ecology [43,44], is particularly relevant in dehesas, an intensively managed system where structural heterogeneity coexists with a strong anthropogenic imprint. Our results show that the relative importance of environmental factors is scale-dependent, revealing a hierarchical pattern in community structuring.

4.1. Local Processes: Tree Structure and Microenvironmental Heterogeneity

At the trap scale, local tree structure was consistently associated with ant community composition. Both tree canopy cover within a 10 m radius and distance to the nearest tree emerged as relevant predictors, outweighing the influence of local climatic variables. Genera such as Aphaenogaster and Temnothorax, which are more frequently associated with shaded and structurally complex microhabitats, were preferentially linked to tree-influenced traps. This pattern is consistent with the recognition of dispersed trees as keystone structures in open systems, as they generate differentiated microenvironments that directly modify the thermal, hydric, and structural conditions of the soil [10,12,15]

The presence of trees buffers thermal extremes, reduces incident radiation, and promotes litter accumulation and substrate complexity [3,13], thereby affecting surface activity, nest placement, and the coexistence of species with different physiological tolerances and ecological strategies in Mediterranean ants [22,45]. Overall, our results reinforce evidence that fine-scale structural heterogeneity plays a key role in the organisation of assemblages in semi-natural landscapes [46]. Similar patterns have also been reported in other European wood–pasture systems, where scattered trees and microhabitat heterogeneity promote shifts in ant community composition and diversity [14,47].

In the specific context of dehesas, holm oaks have been described as generators of “islands of diversity” whose influence extends beyond the canopy [10,12,14,28,48]. The detected association between tree structure and community composition confirms that this heterogeneity is ecologically functional at the local scale.

4.2. Climate and Topography at the Local Scale

Climatic variables considered at the trap scale showed limited and non-independent effects in marginal analyses, suggesting that climate sets a general environmental framework within which other, more organism-proximate factors operate. This result may partly reflect the coarse spatial resolution of the CHELSA dataset (~1 km), which captures regional climatic conditions rather than microhabitat variation among traps. This pattern is consistent with hierarchical frameworks of community structuring in ants across spatial scales. Against this climatic background, habitat structure and topography appear to modulate the effective distribution of species, in line with studies highlighting the interaction between broad environmental filters and local modulators in ant assemblages [49]. Accordingly, numerous studies show that temperature imposes broad physiological limits on ants [23,27], but that its effects can be attenuated by local structural heterogeneity [45].

Elevation showed a consistent effect, probably due to its integrative nature across multiple environmental gradients. In Mediterranean ecosystems, these results support the idea that regional climatic gradients define the general environmental framework, while microenvironmental factors determine the fine-scale organisation of communities [22,50].

The significant heterogeneity in multivariate dispersion among farms detected by PERMDISP provides additional insight into the internal variability of the system. This result suggests that differences observed among farms may involve both shifts in centroid position and patterns of internal heterogeneity in community composition. Given that multivariate dispersion can be interpreted as a measure of community heterogeneity [51], the observed differences likely reflect real contrasts in the internal ecological complexity of farms. These differences may be associated with management factors not quantified in this study, such as grazing pressure or management history, a hypothesis that would require specific experimental designs to test.

4.3. Farm Scale: Shared Climatic Signal and Residual Variance

When the analysis was scaled up to the farm level, the environmental signal was dominated by broad climatic variables, particularly isothermality, together with elevation and temporal variability in vegetation productivity. The significant association with isothermality suggests that annual thermal stability constitutes a relevant filter at this scale, in agreement with studies emphasising the importance of broad thermal gradients in structuring Mediterranean ant communities [23,52]. Likewise, elevation may act as an integrative variable encompassing multiple environmental gradients, including temperature, moisture, and productivity, and has been widely documented as a key determinant of ant community composition at regional scales [53,54]. The observed relationship with temporal variability in NDVI indicates that interannual differences in productive dynamics among farms may indirectly modulate community structure, in line with studies linking heterogeneity in resource availability to arthropod community assembly patterns [54,55]. In contrast, other structural or thermal variables did not show independent associations in the unconstrained ordination, reinforcing, within the context of this study, the idea that, at this scale, climatic and productivity filters dominate over purely structural habitat effects.

In human-modified systems such as dehesas, this residual variance should not necessarily be interpreted as an explanatory weakness, but rather as a reflection of the historical and management singularity of each farm. Unmeasured site-specific and historical characteristics, not captured by the environmental predictors considered, may contribute to this residual variation, diluting the signal of general environmental gradients [6,56].

The NMDS ordination supports this interpretation, showing moderate differentiation among farms without a clear dominance of a single gradient (Figure 3), in agreement with regional studies reporting broad climatic responses in ant communities modulated by spatial and biogeographical context [21,26].

4.4. Multiscale Integration

Overall, the results support a hierarchical model in which climate acts as a first-order filter, defining the regional pool of potential species, while local tree structure modulates effective assemblage composition at finer scales. This framework is consistent with conceptual models proposed for ant communities in heterogeneous landscapes, which emphasise the need to integrate multiple spatial scales to understand community organisation [16,45].

This scale-dependent response highlights that the interpretation of ecological patterns in dehesas requires explicitly multiscale approaches.

4.5. Implications for Bioindication, Management, and Climate Change

The sensitivity of ant communities to both local structural heterogeneity and regional climatic filters reinforces their usefulness as bioindicators in Mediterranean agroforestry systems. Previous studies have demonstrated their value as a surrogate group for epigeic arthropods in dehesas [28]; our results extend this evidence by showing that ant communities integrate environmental signals operating at different spatial scales. This reinforces the applicability of ant-based indicators for management-oriented monitoring in dehesa systems and provides a practical basis for the design of long-term monitoring schemes focused on key structural and climatic variables identified at complementary spatial scales.

From a management perspective, the relevance of local tree structure suggests that landscape homogenisation processes, whether through tree removal or excessive densification, could reduce the diversity of microhabitats that facilitate species coexistence. In this system, trees do not represent merely a structural element, but rather a functional component that conditions key ecological processes [57].

At the regional scale, the observed climatic influence highlights the potential vulnerability of dehesas to climate change, particularly in the Mediterranean region, which has been identified as one of the main global warming hotspots [58,59].

5. Conclusions

The results demonstrate a hierarchical and scale-dependent structuring of ant communities in Mediterranean dehesas. The interaction between regional climatic filters and local structural modulators shapes assemblages and highlights the need for explicitly multiscale approaches. This framework provides a robust ecological basis for integrating the conservation of structural heterogeneity into sustainable management strategies and reinforces the value of ants as diagnostic tools in complex agroforestry systems. Future studies incorporating additional site-specific and historical information may help disentangle the substantial unexplained variation observed at the farm scale.