The study area (33°30′-34°28′N, 00°10′W-01°06′E; Fig. 1) is located in the northern part of the El Bayadh Province, Algeria. It serves as a transition zone between the Tell Atlas and the Sahara (El Zerey et al., 2009). The altitudes range from 900 to 1400 m (Negadi et al., 2014). The climate is semi-arid, characterized by cold winters and hot and dry summers (González et al., 2009). Precipitation is low, with an annual average of 230 mm recorded between 1990 and 2020. The mean annual temperature is 18.0°C (-0.9°C in winter and 36.7°C in summer) (Slimani and Aidoud, 2018). The soil is alkaline with a high calcium carbonate content. It contains low organic matter and humidity, and its texture varies from sandy to sandy-loamy and loamy.

The vegetation of the El Bayadh steppe is well-adapted to arid and semi-arid conditions. In the 1980s, the dominant species included Stipa tenacissima and Artemisia herba-alba, however, since the 1990s, desertification has significantly altered the plant composition (Slimani et al., 2010). The current plant community primarily consists of psammophilous species such as Thymelaea microphylla, Lygeum spartum, Stipagrostis pungens, and Atractylis serratuloides (Slimani and Aidoud, 2018). This steppe provides diverse and essential ecosystem services, including forage for grazing livestock, cropland productivity, and carbon storage, which are critical for both human livelihoods and environmental stability (Nedjraoui et al., 2016; Nasrallah et al., 2020).

The land use changes in the study area from 1980 to 2020 were analyzed using ArcGIS to quantify the decline of steppe vegetation cover and identify various ecological patterns. A total of 50 plots, each measuring 64 m×64 m, were selected for this study. We organized the sampling process based on the aridity gradient at the overall scale of the study area. Furthermore, at the community level, sampling was refined by considering vegetation cover density, management type (grazed or ungrazed), and topographical characteristics. Specifically, 19 plots were established in ungrazed areas, including Kef Lahmar (8), Tousmouline (2), Rogassa (6), Bougtob (1), and El Kreither (2), while 31 plots were designed in the grazed areas Kef Lahmar (8), Tousmouline (9), Rogassa (3), Bougtob (6), and El Kreither (5). Data collection on vegetation occurrence in the sampling plots was conducted in April and October, 2023, corresponding to the optimal periods for plant growth in the Algerian steppe (El Zerey et al., 2009). In each plot, species abundance was estimated using the point intercept method. This involves systematically placing a small-diameter sampling pole (or pins) along diagonal transects within the plot and recording the plant species that come into direct contact with the sampling points (Bonham, 1989). This method is preferred over the quadrat technique due to its suitability for the low vegetation density characteristic of steppe regions (Frenette, 2013).

We selected six functional traits from a core list relevant to understanding plant responses to ecological filters (Table 1). Leaf area and specific leaf area were chosen to assess plant responses to abiotic factors. Root length, clonality, and life history were selected to assess plant responses to biotic filters. Seed mass was the only trait related to dispersal (Weiher et al., 1999). Trait values were measured in the field at the plot level on species representing at least 80% of the vegetation cover in each plot (Pakeman and Quested, 2007; Lavorel et al., 2011). Individuals randomly selected from each species were used to measure traits, and the average was used to represent the mean trait and its variability for each species. A tape measure was employed to record root length. The plant was carefully removed from the soil to measure root length accurately and minimize measurement errors.

The sampling and treatment of leaves used for measuring specific leaf area and leaf area were conducted according to the protocol described by Garnier et al. (2001). The leaves were harvested from 25 individuals of each species and placed in bottles filled with deionized water before being transported to the laboratory in a cooler (Garnier et al., 2001). The leaf surface was obtained by analyzing pictures of fresh leaves with ImageJ software v.1.8 (National Institutes of Health, Bethesda, USA). Then, the leaves were dried at 60.0°C to a constant weight, and the specific leaf area was calculated (Yu et al., 2023). Seed mass data were obtained from the work of Frenette (2013) and Kouider and Maifi (2018). Clonality and life history were collected by observation in the field. A phylogenetic tree was created by transforming the taxonomy data in R v.4.4.2 using the phylo feature from the 'ape' package (Paradis and Schliep, 2019).

Climatic data, soil characteristics, topography, and land use types (grazing and exclosure) were collected for the 50 plots. We selected 12 climatic variables from 1990 to 2020 from the TerraClimate dataset (http://www.climatologylab.org/terraclimate.html). TerraClimate is a dataset that provides monthly climate and climatic water balance data for terrestrial surfaces worldwide (Abatzoglou et al., 2018). These variables encompass annual and seasonal precipitation (winter, autumn, summer, and spring), potential evapotranspiration, Palmer Drought Severity Index, average annual temperature, and both maximum and minimum temperatures. To mitigate the risks of multicollinearity and model overfitting, we performed a principal component analysis (PCA) to reduce the number of climatic variables (Seoane et al., 2023). The precipitation-related variables were reduced to two principal components, P1 and P2, which explained 90% of the total variance in precipitation within the study area. Component P1 (explaining 70% of the total variance in precipitation) represented a north-south humidity gradient, while Component P2 (explaining 20% of the total variance in precipitation) captured seasonal variability. Additionally, temperature variables were synthesized into a single principal component, T1 (explaining 60% of the total variance in temperature), reflecting a general temperature gradient. Furthermore, a Spearman correlation test was conducted between species occurrence and the climatic principal components (P1, P2, and T1) to select the most explanatory variables (Pratt et al., 2022). P1 and P2 were found significantly correlated with species occurrence, while T1 was not significantly correlated. Based on this analysis, we used P1 and P2 as the climatic data in the Species Distribution Model (SDM).

Additionally, we analyzed eight soil property variables. The chemical variables included pH, soil moisture, carbon-to-nitrogen ratio, salinity, electrical conductivity, and calcium carbonate (CaCO₃). The physical variables were granulometry and sand height. We reduced the chemical variables to two principal components. The first component (S1) captured a gradient related to salinity and electrical conductivity, while the second component (S2) represented a gradient of fertility and moisture. The soil samples applied to analyze the physical and chemical properties were collected from designated plots used for vegetation sampling. Soil pH, salinity, and electrical conductivity were measured using an Oakton WD-35634-35 pH meter (Oakton Instruments, Vernon Hills, USA). Soil moisture content was determined by calculating the difference between the fresh and dry masses of the soil (Barbagelata, 2006). The CHN analyzer (Flash 2000, Thermo Fisher Scientific, Waltham, USA) was used to determine the soil's carbon and nitrogen contents (Telek and Marshall, 1974). CHN analyzers combust the soil samples at high temperatures, converting the carbon and nitrogen into gases that are then quantified (Telek and Marshall, 1974). The proportion of soil CaCO₃ was determined by the soil's reaction to hydrochloric acid (HCl). The intensity and duration of effervescence provide qualitative insights into the amount of CaCO₃ in the soil (Paltseva, 2024). To determine soil texture, we used the sedimentation method, which relies on the principle that larger, heavier particles settle more rapidly in water than smaller, lighter particles (Yudina et al., 2020). Sand height was measured in the field using tape (Wang et al., 2021). We measured the height of the sand from the level of the firm ground as a reference up to the maximum height of the sand present in the plot. Elevation was identified as the most relevant topographic variable in this study, as it showed the strongest Spearman correlation with species occurrence among all topographic variables. It was measured in the field using the Trimble Geo 7X GPS (Trimble Inc., Sunnyvale, USA).

HMSC is a joint species distribution model featuring a hierarchical structure that enables us to evaluate how environmental and biotic filters influence community assembly (Ovaskainen and Abrego, 2020). HMSC also accounts for the study design and the geographical coordinates of the study area to quantify stochastic processes at the local level (plots). HMSC is a multivariate hierarchical generalized linear mixed model fitted with Bayesian inference (Ovaskainen et al., 2017; Ovaskainen and Abrego, 2020). Thus, for plot i, the distribution of species j (Y_ij) can be expressed as: Y_ij~D(L_ij, σ_j²), where D is a statistical distribution (Probit, normal, or Poisson) suitable for the type of data used in Y_ij; L_ij is the linear predictor; and σ_j² is a variance term for the distribution of species j. The linear predictor L_ij is modelled from a fixed part L_F, and a random part L_R. L_F represents the distribution of species j that is explained by the environmental covariates. It can be expressed as a linear regression: L_ij=∑(x_ikβ_jk), where x_ik represents the value of covariate k in the plot i and β_jk represents the response of species j to covariate k. β_jk depends on the value of the functional traits and phylogeny and is modelled as follows: β_jk~Normal(u_j, V⨂[ρC+(1-ρ)I]), where u_j represents the expected fundamental niche of species j; V is the variation around this expectation; ⨂ represents the Kronecker product; ρ measures the strength of the phylogenetic signal; and C and I represent the phylogenetic relationship matrix and the identity matrix, respectively. The paramater u_j accounts for the trait's effect on plant response to covariate k and can be modelled as: u_j=∑(t_jlγ_lk), where t_jl represents the value of trait l for species j and γ_lk represents the effect of trait l on the species response to covariate k. ρ ranges from 0 to 1; when ρ=0, the residual variance is independent among species, indicating that closely related species do not have more similar environmental niches than distantly related ones (Ovaskainen and Abrego, 2020). As ρ approaches 1, the species' environmental niches are fully structured by their phylogeny, with closely related species exhibiting greater niche similarity than expected by chance, reflecting niche conservatism (Ovaskainen and Abrego, 2020). The random term L_R enables modelling the variations in species occurrence and co-occurrence that cannot be attributed to species responses to the environmental covariates. L_R can be expressed as follows: L_R=ε_ij^S, where ε refers to a random effect that operates at the level of the sampling unit S. These random effects will be modelled as: ε_ij^S~Normal(0, Ω^𝑆), where Ω^𝑆 is a species-by-species residual variance-covariance matrix.

HMSC can be conceptually linked to the assembly rules framework (Ovaskainen et al., 2017). The fixed effects (L_F) component of HMSC models environmental filtering—that is, how the interplay between species' niches and environmental heterogeneity influences species occurrence and abundance. The random effects (L_R) part of HMSC models biotic filtering, namely how the ecological interactions among species influence their occurrences, particularly their co-occurrences. A key parameter of this part of HMSC is the species-to-species association matrix Ω, which describes those species pairs that are found together either more or less often than expected by chance (Ovaskainen et al., 2017; Ovaskainen and Abrego, 2020). Dispersal limitation also causes variations in species occurrence that environmental filtering cannot explain, resulting, for example, in a species being absent from an area where the environmental conditions are suitable. Therefore, the influences of dispersal limitation will also be reflected in the random effects part of HMSC (Ovaskainen and Abrego, 2020).

Two models were developed to evaluate whether incorporating functional traits and phylogeny improves model accuracy and performance (Ovaskainen and Abrego, 2020). The full model (PA) included all input data, such as species occurrence (presence/absence), environmental covariates, functional traits, the phylogenetic tree, and the study design, while the environmental model (PA.Env) excluded functional traits and the phylogenetic tree. We built a probit regression model (with species presence/absence as the response variable) for the occurrences of 24 species, including the 8 environmental covariates described above (P1, P2, S1, S2, soil texture, elevation, sand height, and land use type) as fixed effects. The model also incorporated random effects for spatial variation among the sampling plots. Additionally, we included the 7 species traits described in Table 1. To determine if a phylogenetic signal was present in our data, we integrated a phylogenetic tree created by transforming the taxonomy data in R version 4.4.2 using the 'phylo' feature of the 'ape' package (Paradis and Schliep, 2019). We fitted the two models using the R package 'Hmsc', employing the default prior distributions (Ovaskainen et al., 2017; Ovaskainen and Abrego, 2020). The posterior distribution was sampled using two Markov Chain Monte Carlo (MCMC) runs, each run for 600,000 iterations. The convergence of MCMC was evaluated using the potential scale reduction factors (PSRFs) (Gelman and Rubin, 1992). Additionally, we performed two-fold cross-validation on models thinned by a factor of 200. We used both coefficients Tjur's R² and Area Under the Curve (AUC) as the indice of model performance. Tjur's R² is asymptotically equivalent to the coefficient of determination (R²) and AUC summarizes model accuracy (Zhang et al., 2020a). The units of AUC and Tjur's R² are different; a model that behaves as well as by chance will yield an AUC of 0.50 and a Tjur's R² of 0.00, whereas a model that perfectly discriminates empty and occupied sampling units will have an AUC of 1.00 and a Tjur's R² of 1.00 (Ovaskainen et al., 2017; Ovaskainen and Abrego, 2020).

The fitted model PA (which incorporates all the input data) demonstrated strong explanatory power (AUC=0.95 and Tjur's R²=0.51) and moderate predictive power (AUC=0.63 and Tjur's R²=0.16). Similarly, PA.Env (which excludes functional traits and phylogeny) exhibited good explanatory power (AUC=0.93 and Tjur's R²=0.30) but lower predictive power (AUC=0.59 and Tjur's R²=0.07). These findings suggested that PA outperforms PA.Env, indicating that incorporating functional traits and phylogeny into the model enhances its overall performance. Therefore, we further presented the results related to PA below.

Our results indicated that the two soil principal components of soil, S1 (salinity and electrical conductivity) and S2 (fertility and moisture), contributed the most to explaining the variance (28%). This was followed by the two precipitation principal components, P1 and P2 (22%), elevation (16%), sand height (11%), soil texture (10%), and land use type (7%). The community-level random effect accounted for 7% of the variance proportions (Fig. 2).

Regarding soil variables (Fig. 3), statistical support (with at least 75% posterior probability) indicated a positive response to S1 in 21% of species, specifically Atriplex halimus and Salsola kali, while 46% of species showed a negative response, including Atractylis serratuloides and Artemisia herba-alba. For S2, 29% of species demonstrated exclusively negative responses, such as Thymelaea microphylla. Similarly, soil texture significantly influenced species responses, with 21% of species responding positively to sandy texture, while 25% of species exhibited a negative response. In terms of sandy-loamy texture, 12% of species responded positively, whereas 8% of species, including Artemisia herba-alba, showed a negative response.

Sand height negatively affected 45% of species (Peganum harmala, Thymelaea microphylla, Hammada scoparia, Atractylis carduus, Salsola gemmascens, and Atractylis serratuloides), with only 8% showing a positive response. Exclosure positively affected 50% of species (Peganum harmala, Atriplex halimus, Artemisia herba-alba, Atractylis carduus, and Atractylis serratuloides), while only 8% of species showed a negative response. For precipitation-related principal components, 33% of species (Noaea mucronata, Salsola kali, and Lygeum spartum) showed a positive response to P2, whereas 8% of species exhibited a negative response. In contrast, P1 excited positive responses in 8% of species and negative responses in 17% of species. Elevation positively affected 46% of species, while 12% of species had a negative response.

Regarding phylogeny, we found a weak signal (ρ=0.19). This indicated that closely related species do not show similar patterns of responses to environmental factors (Fig. 4). Additionally, our analysis indicated that all functional traits respond differently to various explanatory variables (Fig. 5). For instance, seed mass increased with P1, S2, and soil texture (sandy and sandy loam). However, seed mass decreased with S1, P2, and exclosure. Additionally, leaf area responded positively to P1, sand height, and sandy texture but decreased with S1. Specific leaf area increased with P2 but decreased with factors such as S2, elevation, and sandy texture. Root length increased with S1 and both sandy and sandy loam textures decreased with exlosure and sand height. Finally, clonality decreased with increasing P1 and elevation, while it responded positively to S1, sandy texture, and exclosure. Perennial plants responded positively to P1 and elevation but negatively to salinity and sand height.

The species-to-species association matrix provides insight into the interspecific relationships within the plant community by illustrating the strength and direction of pairwise correlations between species (Fig. 5). Most correlations were relatively weak or close to zero, with the mean correlation coefficient of 0.0026 and variance correlation coefficient of 0.0056, suggesting that strong interspecific interactions such as competition or facilitation are limited. However, some species pairs, such as Onopordon arenarium and Atractylis serratuloides, displayed moderate positive correlations, which could reflect coexistence through niche complementarity. In contrast, several species pairs, such as Atractylis serratuloides with Astragalus armatus and Noaea mucronata with Stipa tenacissima, exhibited moderate negative correlations, potentially indicating competition for resources.

In this study, our primary objective was to identify the ecological filters driving community assembly at the local scale in the El Bayadh steppe. Our results indicated that soil properties were the most influential factor, explaining 38% of the variance, followed by precipitation (22%), elevation (16%), sand height (11%), and land use type (7%). In contrast, species-to-species associations revealed that most species were not significantly associated with one another, suggesting that biotic filters—such as competition—played a minimal role in shaping species distribution in the El Bayadh steppe.

The dominance of soil properties as the primary drivers of species distribution in the El Bayadh steppe indicated that most species at the local scale are selected based on their ability to tolerate the region's edaphic constraints. These results align with those found by Yang et al. (2020), who reported that soil physical and chemical properties are the most important factors modulating plant diversity in arid and semi-arid ecosystems in China. In the El Bayadh steppe, as in most arid and semi-arid ecosystems, plants face multiple soil-related challenges, including high salinity, particularly in the northern part of the study area with low fertility, limited moisture, and sandy-limestone texture (Kargar-Chigani et al., 2017; Alliouche and Kouba, 2023). High salinity stresses plants by reducing water uptake and causing ion toxicity, which limits the number of species that can thrive in these areas (Koull and Chehma, 2016; Bidalia et al., 2019). This phenomenon is evident in the negative response of 45% of species to salinity observed in our study (Fig. 3a). However, some species, such as Atriplex halimus, exhibited a positive response to salinity, illustrating their adaptation to these harsh conditions. Additionally, the low soil fertility in the region presents another challenge, as only plants capable of efficiently acquiring and utilizing nutrients can survive and reproduce (Hag et al., 2021). Interestingly, 29% of species exhibited a negative response to fertility (Fig. 3a), which contrasts with our expectation of a positive response. This suggested that a moderate increase in fertility may serve as a stress factor. This phenomenon can be attributed to the fact that in areas with high salinity—corresponding to regions with greater nutrient content—the moderate increase in soil fertility fails to benefit plant establishment, as salinity remains the primary limiting factor. Additionally, the soil texture in El Bayadh region, like much of the soil in the Algeria steppes, is predominantly sandy and sandy limestone, which impacts water retention capacity and further restricts plant growth (Alliouche and Kouba, 2023). However, certain species, such as Lygeum spartum, responded positively to these soil textures (Fig. 3a). These results are consistent with previous studies in the El Bayadh steppe, which observed the replacement of dominant species like Stipa tenacissima by psammophilous species such as Lygeum spartum, Thymelaea microphylla, and Stipagrostis pungens (Slimani et al., 2010; Slimani and Aidoud, 2018).

Although soil properties were identified as the most important drivers, precipitation significantly influenced species distribution in the El Bayadh steppe (accounting for 28% of the variance explained). This finding is consistent with Slimani et al. (2010) and Slimani and Aidoud (2018), who demonstrated a significant correlation between temporal variations in species composition and long-term annual precipitation (during 1976-2016) in the El Bayadh steppe. Their work suggested that seasonal and spatial precipitation variability are critical drivers of species distribution in the El Bayadh steppe.

Furthermore, our results revealed that elevation and sand height were less influential, accounting for 16% and 11% of the variance in species distribution in the El Bayadh steppe, respectively. The limited impact of elevation can be attributed to the low topographical variability within the study area (Hamlat et al., 2024). Potential threshold effects can explain the weak influence of sand height. Specific threshold levels of sand height may be critical for influencing species distribution; if most measurements fall below this threshold, the overall impact of sand height may appear minimal (Bouallala et al., 2022, 2023). Future studies could concentrate on this aspect to enhance understanding of the impact of sand encroachment. In this study, 45% of species responded negatively to sand height, while only 12% of species responded positively. This indicated that sand encroachment presents a significant issue for the El Bayadh steppe, and conservation efforts should take this factor into account. These results align with those of Bouallala et al. (2023) that sand encroachment induced significant changes in plant community variables, including reductions in plant density, species rarity, species composition, morphological types, and adaptations to aridity.

The lack of significant correlation (negative or positive) among species (Fig. 5) suggested that competition (negative correlation among species) is not an important driver of community assembly in this environment. In areas under highly stressful conditions, like the El Bayadh steppe, environmental filtering tends to overshadow biotic interactions such as competition (Wang et al., 2021; Hu et al., 2022; Zhao et al., 2023; Tu et al., 2024). Additionally, land use type, recognized as an important driver of species distribution in the El Bayadh steppe, was found to have the weakest effect (7% of variance), indicating no significant difference between grazed and ungrazed areas in terms of species composition. These findings raise questions about the influence of exclosure on steppe rehabilitation. Exclosure involves removing the constraint of livestock to allow for the regeneration of plants (Wang et al., 2019). However, in this study, exclusion did not lead to a significant change in species composition, suggesting that the expected vegetation recovery may not have occurred. This implies that, beyond grazing exclusion, other ecological or environmental factors—such as improving soil quality or moisture availability—should be considered to achieve effective rehabilitation of Algerian steppe ecosystems.

The phylogenetic signal was not strong (ρ=0.19), indicating that functional trait responses to environmental change differed among closely related species. This suggested that traits associated with specific ecological strategies, such as stress tolerance or ruderal characteristics, are not consistently shared among closely related species. Each species appears to evolve its traits in response to ecological pressures from environmental filters. These results align with those of Dong et al. (2019), who found a weak environmental effect on community phylogeny. This phenomenon occurs particularly when environmental change happens rapidly (Dong et al., 2019). From a conservation perspective, this suggests that predicting a species' trait response solely based on its phylogenetic relatedness may generate species occurrence patterns that do not reflect field reality, which can affect the effectiveness of established conservation strategies.

Our analysis also showed that all functional traits responded to multiple explanatory variables. Seed mass increased under favorable conditions, as shown by its positive relationship with precipitation and S2, and decreased in response to harsher environmental conditions such as S1 and P2 (Fig. 5). This aligns with the established relationship between seed size and stress (Hallett et al., 2011; Gao et al., 2023). In more stressful environments, plants tend to produce smaller seeds, likely due to the reduced availability of resources, which limits investment in each offspring. Conversely, in areas with higher water and nutrient availability, plants can invest in larger seeds, which provide greater energy reserves and enhance offspring survival, particularly in competitive environments (Qi et al., 2015). Furthermore, the increase in leaf traits, such as leaf area and specific leaf area with P1, is consistent with plant adaptation to water availability (De Micco and Aronne, 2012). Plants develop larger, thinner leaves in favorable conditions to optimize resource acquisition. Conversely, under stress, such as water scarcity, plants typically have smaller leaves to minimize water loss through transpiration (Iqbal et al., 2020; Marasco et al., 2022). This could explain the negative correlation between leaf traits and salinity, as saline environments often induce water stress (Koull and Chehma, 2016; Bidalia et al., 2019). However, the decrease in specific leaf area with soil fertility and elevation (Fig. 4) appears to contradict the ecological theory that plants should increase specific leaf area in response to greater resource availability (De Micco and Aronne, 2012). This could be explained by the possibility that plants in these environments may adopt a more conservative strategy. In this case, plants might limit resource loss to ensure long-term survival, even when water and nutrients are relatively abundant. This strategy may help plants maintain resilience in environments where conditions are prone to fluctuations or where the long-term stability of resources is uncertain (Fan et al., 2024). Further research should investigate this phenomenon in greater detail to enhance our understanding of plant adaptation strategies in the El Bayadh steppe and other semi-arid steppe ecosystems. The positive correlation of root length with S1 and sandy texture supports the expectation that plants with deeper or more extensive root systems thrive in areas with limited water availability (Iqbal et al., 2020). This adaptation helps them reach deeper water reserves. Additionally, the negative correlation between root length and exclosure may be attributed to increased water availability in exclosure areas, as livestock generally decrease water access through soil compaction (Abdalla et al., 2018). However, the negative correlation between root length and sand height raises important questions about how plants respond to sand encroachment. According to the literature, plants generally develop deeper root systems in stressful environments to improve water acquisition (Iqbal et al., 2020). Sand encroachment generally reduces the soil's water retention capacity, which should, in theory, prompt plants to deeper roots (Bouallala et al., 2023). These findings underscore the complexity of plant responses to sand encroachment, indicating that the relationship between sand and root systems may not be straightforward. Additionally, our results demonstrated a negative correlation between clonality and both precipitation and elevation. This observation aligns with plant strategy adaptation to harsh conditions, where clonality assists plants in coping with the challenges posed by arid environments (Zhang and Wu, 2014). By reproducing vegetatively through methods like rhizomes and stolons, clonal plants can effectively access localized resources and enhance their chances of survival (Liu et al., 2007). Additionally, this strategy allows for resilience, as clonal populations can recover from stressors such as drought, even if parts of the plant die (O'Brien et al., 2014). Finally, the positive response of perennial plants to P1 and elevation but the negative response to S1 and sand height are consistent with the general ecological understanding that perennials thrive in more stable environments with consistent water availability.

The findings of this study suggested that PA outperforms PA.Env, indicating that incorporating functional traits and phylogeny allows the model to better capture the underlying ecological processes shaping community assembly. This aligns with ecological literature emphasizing the importance of functional traits and phylogeny in species establishment and persistence within a community (Ovaskainen and Abrego, 2020; Keddy and Laughlin, 2021). However, we observed moderate predictive power for both models (PA and PA.Env), suggesting that while functional traits and phylogeny enhance our understanding of community structure, the current model may not fully capture additional factors influencing species distribution. These factors could include unmeasured environmental variables, species interactions, or stochastic events. Furthermore, the predictive power could potentially improve by increasing the sample size, which would provide more robust data for the model to detect finer-scale patterns (Crossa et al., 2022). Future research should explore the identification and integration of these additional variables to strengthen the model's predictive power and more effectively inform conservation strategies in the El Bayadh steppe.