Soil organic carbon storage across Mediterranean mountain forests: a comparative assessment of forest ecosystem diversity and geological substrate in Ifrane and Talassemtane National Parks, Morocco

1 Introduction

Forest ecosystems play a fundamental role in the global carbon cycle, with soils representing the largest and most stable carbon reservoir; forest soils store more carbon than living biomass and the atmosphere combined, underscoring their strategic importance for climate mitigation and ecosystem resilience (1–7). This soil organic carbon (SOC) pool reflects the long-term balance between organic matter inputs – litter, woody debris, root turnover – and losses through mineralization, erosion and disturbance (8–10). Because climate, topography, vegetation, parent material and mineralogy jointly govern this balance (11, 12), SOC storage varies considerably across landscapes, an effect that is particularly pronounced in ecologically heterogeneous mountain regions.

Mediterranean mountain forests are among the most heterogeneous terrestrial ecosystems, combining steep gradients in elevation, climate, lithology and vegetation over short distances (13, 14). Within these landscapes, geological substrate is a fundamental yet frequently overlooked driver of SOC storage: parent material controls soil depth, mineralogy and water-holding capacity, and basalt-derived soils generally show greater mineral protection of organic matter than calcareous or dolomitic substrates (15, 16). Recent syntheses of lithologically contrasting landscapes confirm that geological setting can rival, or even exceed, vegetation as a control on SOC stability (17), yet substrate is still rarely incorporated explicitly into landscape-scale comparative studies of forest SOC.

Vegetation is likewise a primary determinant of SOC accumulation, since forest ecosystems differ in productivity, litter quality, rooting depth and microbial activity (18–21). Most biodiversity–SOC studies, however, have relied on species richness measured at the plot scale (22, 23), which does not capture the ecological complexity of protected mountain landscapes where multiple forest ecosystem types coexist as landscape mosaics. Consistent with recent evidence that tree species identity, rather than diversity per se, drives Mediterranean SOC dynamics (24, 25), the present study instead defines forest ecosystem diversity as the diversification and spatial variability of forest ecosystem types across the landscape, rather than tree species richness within individual plots.

Morocco’s Rif and Atlas Mountains host some of the most ecologically diverse Mediterranean forests in North Africa, providing essential ecosystem services including carbon sequestration (26). Ifrane National Park (Middle Atlas), with cedar, holm oak, zeen oak and maritime pine forests distributed over basaltic and dolomitic substrates, and Talassemtane National Park (Rif), dominated by fir, cork oak, holm oak, thuya and matorral ecosystems on limestone, offer contrasting ecological and geological settings well suited to this question. While previous Moroccan studies have documented SOC differences among forest types and soil classes (27–33), none has simultaneously evaluated forest ecosystem diversity and geological substrate across two contrasting protected landscapes.

Regional machine-learning-based mapping has further shown that SOC storage in Moroccan mountain catchments is jointly controlled by land cover and environmental covariates, including soil, climatic and topographic factors (34), reinforcing the need to consider multiple environmental dimensions jointly rather than land cover alone.

The objectives of this study were therefore to: (i) compare SOC stocks among contrasting forest ecosystem types in Ifrane and Talassemtane National Parks; (ii) quantify the contribution of forest ecosystem diversity to the spatial variability of SOC; (iii) evaluate the influence of geological substrate on SOC accumulation; and (iv) investigate how ecosystem diversity and geological substrate jointly shape SOC storage across Mediterranean mountain landscapes. We hypothesize that SOC storage is jointly, and only partly separably, controlled by forest ecosystem diversity and geological substrate, and that landscapes combining greater ecological and geological heterogeneity exhibit higher spatial variability and greater carbon storage potential than more homogeneous landscapes.

2 Materials and methods

2.1 Study area

The study was conducted in two representative Mediterranean mountain protected areas of Morocco: Ifrane National Park (INP), located in the Middle Atlas Mountains, and Talassemtane National Park (TNP), situated in the Rif Mountains (Figure 1). These parks constitute two of the country’s most important forest biodiversity hotspots and are recognized for their high ecological value, diverse forest ecosystems and strategic role in ecosystem service provision, particularly carbon sequestration, biodiversity conservation and watershed protection (35).

Ifrane National Park extends over approximately 125,000 ha and is characterized by a humid to sub-humid Mediterranean mountain climate with cold winters and relatively high annual precipitation (29, 31, 32, 36). For the Azrou forest within this park, El Mderssa et al. (29, 31, 32) report an Emberger rainfall coefficient (Q2) of 101–114, defining a humid Mediterranean bioclimate with a cold to sub-humid variant, based on a mean annual precipitation of approximately 982 mm and a January mean monthly minimum temperature of −1.31 °C over a 30-year record (37, 38, as cited in 29, 31, 32). The park encompasses a wide altitudinal gradient supporting a mosaic of cedar forests (Cedrus atlantica), holm oak (Quercus rotundifolia), zeen oak (Quercus faginea), maritime pine (Pinus pinaster) and mixed forest formations. Geologically, the park is dominated by basaltic volcanic formations and dolomitic carbonate formations, providing contrasting parent materials for soil development.

Talassemtane National Park covers approximately 58,950 ha in the western Rif Mountains. The park experiences a humid Mediterranean climate influenced by Atlantic air masses, with abundant rainfall and rugged mountainous terrain (39). Its vegetation is dominated by Moroccan fir (Abies maroccana) (40), cork oak (Quercus suber), holm oak (Quercus rotundifolia), thuya (Tetraclinis articulata), pine forests and Mediterranean matorral (41). Unlike Ifrane, the geological framework of Talassemtane is almost exclusively composed of limestone formations, resulting in relatively homogeneous calcareous soils.

The marked differences in vegetation composition and geological substrate between the two parks provide an excellent framework for investigating the combined influence of forest ecosystem diversity and parent material on soil organic carbon storage.

2.2 Data sources and harmonization

The present study combines two independent datasets derived from field investigations conducted in Ifrane National Park and Talassemtane National Park during the summer season of 2022. A total of 67 sampling observations were included in the analyses, comprising 48 observations from Ifrane National Park and 19 observations from Talassemtane National Park. The two datasets were harmonized before analysis by applying the same response variable definition, sampling depth and SOC stock calculation procedure.

For both datasets, the response variable was soil organic carbon stock (SOC stock, t C ha^-¹) calculated for the upper mineral soil layer using harmonized procedures. Variable names were standardized prior to statistical analyses to ensure direct comparison between datasets.

The forest ecosystems included in the analyses are presented in Figure 2.

2.3 Geological substrate characterization

Geological substrate was considered as a major explanatory variable because parent material strongly influences soil formation, mineralogy, nutrient availability and long-term stabilization of organic matter.

In Ifrane National Park, sampling sites were distributed over two major geological substrates:

• Basalt, corresponding to volcanic parent material;
• Dolomite, corresponding to carbonate parent material.

The distribution of the sampled ecosystems according to substrate was as follows:

Overall, four ecosystem types were sampled exclusively on one substrate (Zeen oak forest and Holm oak–zeen oak mixed forest on basalt; Maritime pine forest and Cedar–holm oak mixed forest on dolomite), whereas Cedar forest and Holm oak forest were each sampled on both basalt and dolomite (n = 6 per substrate), allowing a direct, ecosystem-controlled comparison of substrate effects for these two ecosystem types.

In Talassemtane National Park, all sampling sites were established on limestone (calcareous) substrate. Consequently, substrate variability could not be evaluated within Talassemtane, but limestone was considered when interpreting differences between the two parks.

2.4 Soil sampling and SOC determination

Soil samples were collected from the upper mineral soil layer (0–30 cm depth) at each sampling point, following the same protocol in both national parks. The 0–30 cm sampling depth was selected for both methodological and ecological reasons. First, it is consistent with the IPCC guidance for reporting mineral-soil organic carbon stocks (3, 42 Refinement) and is widely used in regional and national SOC inventories, thereby facilitating comparison with previous studies, including related regional studies (27). Second, this upper soil layer is particularly relevant for SOC dynamics because a substantial proportion of tree roots and associated carbon inputs are concentrated within the upper 30 cm of the soil profile. As reported by Chowdhury et al. (43) and supported by our field observations, this layer receives important carbon inputs through aboveground litter, root exudates, and fine-root turnover. Organic carbon concentration was determined using the Walkley–Black wet oxidation method. Soil organic carbon stocks (SOC, t C ha^-¹) were calculated for each sample using the equation:

SOC = 0.1 × E × DA × Ci (44, 45)

Where:

E is the depth of the sampled layer (cm),

DA is the bulk density of the sieved soil fraction (<2 mm) for that depth (g cm^-³),

Ci is the organic carbon concentration of the sieved soil fraction (<2 mm) for that depth (g kg^-¹);

the constant 0.1 converts the product to t C ha^-¹.

The same sampling depth, carbon determination method and SOC calculation formula were applied to both the Ifrane and Talassemtane datasets, supporting the comparability of SOC values reported across the two national parks. Bulk density, particle-size distribution, pH and total nitrogen were also determined for each sample but were not used as covariates in the present analysis.

Bulk density was determined using the core (cylinder) method, in which a fixed-volume metal ring was driven into the soil, and bulk density was calculated as the oven-dried (105 °C) soil mass divided by the known core volume (46). Soil pH was measured potentiometrically with a glass electrode in both distilled-water (pH-H₂O) and 1 M KCl (pH-KCl) suspensions, at a 1:2.5 soil:solution ratio. Particle-size distribution (clay, fine and coarse silt, and fine and coarse sand fractions) was determined using the Robinson pipette method, following pretreatment to remove organic matter (47). Total nitrogen was determined by titration following Kjeldahl digestion (48).

2.5 Forest ecosystem diversity

In this study, biodiversity was evaluated at the ecosystem level rather than at the species level.

Forest ecosystem diversity was defined as the diversity and spatial heterogeneity of forest ecosystem types occurring across the landscape. This approach differs from traditional biodiversity metrics based on tree species richness because it integrates differences in vegetation structure, ecological functioning, productivity and environmental conditions among ecosystem types.

Landscape-scale ecosystem diversity was quantified using:

• Ecosystem richness (S);
• Shannon diversity index (H’);
• Simpson diversity index (1 − D);
• Pielou’s evenness index (J).

These indices were calculated using the relative frequency of each forest ecosystem type among the sampled plots within each national park (i.e., pi = number of sampling points in ecosystem i/total number of sampling points in that park), following standard practice for point-based vegetation sampling. Because sampling intensity was not proportional to the true areal extent of each ecosystem type on the ground, these indices describe the diversity and evenness of the sampled point data rather than a direct, area-weighted measure of landscape composition.

2.6 Statistical analyses

All statistical analyses were performed using R software (version 4.3.3; 49) Descriptive statistics (mean, standard deviation, minimum, maximum, median and coefficient of variation) were first calculated for each park, ecosystem type and geological substrate.

Differences in SOC stocks between the two national parks were evaluated using Welch’s t-test, while Mann–Whitney U tests were applied when assumptions of normality were not satisfied.

The influence of forest ecosystem type on SOC storage within each park was assessed using one-way analysis of variance (ANOVA) followed by Tukey’s Honestly Significant Difference (HSD) multiple comparison test.

The effect of geological substrate within Ifrane National Park was similarly evaluated using one-way ANOVA.

Because Cedar forest and Holm oak forest were each sampled on both basalt and dolomite substrates within Ifrane National Park (n = 6 per substrate), an additional, ecosystem-controlled comparison of SOC stocks between substrates was performed separately for these two ecosystem types using Welch’s t-test, providing the cleanest available test of the substrate effect independent of vegetation type.

To complement the fixed-effects GLM and better account for the unequal and non-independent sampling of ecosystem types across parks and substrates, a linear mixed-effects model was additionally fitted, with geological substrate (basalt, dolomite, limestone) as a fixed effect and forest ecosystem type as a random intercept, and estimated by restricted maximum likelihood (REML). Forest ecosystem, rather than sampling site or plot, was used as the random-effect grouping factor because each plot contributed a single, non-repeated SOC measurement, so a plot-level random effect would not have been statistically identifiable; ecosystem type was the natural grouping level at which sample sizes were unequal (ranging from n = 2 to n = 16) and for which partial pooling (shrinkage) of group means toward the overall mean was most informative, particularly for the most sparsely sampled ecosystems (Fir forest and Mediterranean matorral, n = 2 each).

Homogeneity of variances was verified using Levene’s test, whereas normality of residuals was assessed using the Shapiro–Wilk test.

To quantify the magnitude of ecosystem effects, eta-squared (η²) was calculated for each ANOVA.

Relationships among ecosystem types, geological substrate and SOC storage were further explored using:

• Principal Component Analysis (PCA);
• Hierarchical Cluster Analysis (HCA);
• Heatmap visualization of ecosystem-level SOC stocks.

Finally, a General Linear Model (GLM) was proposed to evaluate the combined effects of forest ecosystem type, geological substrate and national park on SOC storage according to the following model:

[SOC = \beta_0 + Ecosystem + Substrate + Park + Ecosystem \times Substrate + \varepsilon].

Because all Talassemtane observations occurred on limestone substrate, the interaction term was interpreted cautiously due to the partial confounding between park identity and geological substrate.

Throughout the study, statistical significance was accepted at P < 0.05.

A summary of the methodological approach adopted in this study is presented in Figure 3.

3 Results

3.1 Comparison of soil organic carbon stocks between the two National Parks

A total of 67 observations were analyzed to compare soil organic carbon (SOC) stocks between Ifrane National Park (INP) and Talassemtane National Park (TNP). Significant differences in SOC storage were observed between the two protected areas.

Mean SOC stock reached 173.99 ± 93.90 t C ha^-¹ in Ifrane National Park, which was approximately 2.5 times higher than the mean value recorded in Talassemtane National Park (69.44 ± 15.50 t C ha^-¹) (Table 1). SOC stocks also exhibited a much wider range in Ifrane (36.10–437.92 t C ha^-¹) than in Talassemtane (44.64–107.74 t C ha^-¹), reflecting greater landscape heterogeneity.

Table 1. Forest ecosystems and associated geological substrates in Ifrane National Park.

Welch’s t-test confirmed that the difference between the two parks was highly significant (P < 0.001).

The coefficient of variation reached 53.97% in Ifrane compared with only 22.33% in Talassemtane, indicating that SOC storage is considerably more heterogeneous in the Middle Atlas than in the Rif Mountains (Table 1).

3.2 Variation of SOC stocks among forest ecosystem types

Forest ecosystem type significantly influenced SOC storage in both national parks (Figure 4; Table 2).

Table 2. Mean SOC stocks according to forest ecosystem type.

Within Ifrane National Park, one-way ANOVA revealed significant differences among ecosystem types (P = 0.0155). Zeen oak forests exhibited the highest SOC stocks (251.54 t C ha^-¹), followed by Holm oak forests (219.62 t C ha^-¹) and Cedar forests (169.47 t C ha^-¹). Lower values were recorded in Maritime pine forests (134.06 t C ha^-¹), Cedar–Holm oak mixed forests (119.13 t C ha^-¹) and Holm oak–Zeen oak mixed forests (109.02 t C ha^-¹).

In Talassemtane National Park, ecosystem type also significantly affected SOC storage (P = 0.0076). Matorral exhibited the highest SOC stock (101.41 t C ha^-¹), followed by Cork oak (73.60 t C ha^-¹) and Fir forests (72.50 t C ha^-¹). Holm oak forests showed the lowest mean SOC stock (59.69 t C ha^-¹). Because Matorral and Fir forest were each represented by only n = 2 observations, their means, and their position relative to the better-replicated ecosystems, should be regarded as indicative rather than statistically robust; the linear mixed-effects model (section 3.6.1) partially corrects for this by shrinking estimates for these sparsely sampled ecosystems toward the overall mean.

Effect sizes and assumption checks for these one-way ANOVAs, and for the Ifrane basalt-versus-dolomite comparison discussed in section 3.3, are summarized in Table 3. Residual normality was violated for the Ifrane ecosystem-effect and substrate-effect ANOVAs (Shapiro–Wilk P < 0.01), and variance homogeneity was borderline violated for the Talassemtane ecosystem-effect ANOVA (Levene’s P = 0.044); the corresponding P-values should therefore be treated as approximate, and a non-parametric equivalent (Kruskal–Wallis) is reported alongside for the two tests with non-normal residuals.

Table 3A. Descriptive statistics of soil organic carbon (SOC) stocks in the two National Parks.

Table 3B. Site physicochemical characterization by National Park (mean ± SD).

Table 3C. Effect sizes and ANOVA assumption checks.

η² = eta-squared (proportion of variance explained). *Indicates a violated assumption (Levene P < 0.05 for homogeneity of variance; Shapiro–Wilk P < 0.05 for non-normal residuals). Kruskal–Wallis P is reported as a non-parametric equivalent for the two tests with non-normal residuals.

The range between ecosystem means reached 142.52 t C ha^-¹ in Ifrane but only 41.72 t C ha^-¹ in Talassemtane, confirming that ecosystem heterogeneity is considerably greater in Ifrane.

3.3 Influence of geological substrate on SOC storage

The incorporation of geological substrate into the analyses substantially improved the ecological interpretation of SOC variability.

Within Ifrane National Park, basalt-derived soils supported a mean SOC stock of 200.84 t C ha^-¹ (n = 24), whereas dolomitic soils averaged 147.14 t C ha^-¹ (n = 24), corresponding to an increase of approximately 37% on basalt in this pooled, ecosystem-confounded comparison (Figures 5, 6).

All Talassemtane sampling sites occurred on limestone substrate and exhibited a mean SOC stock of 69.44 t C ha^-¹.

These results indicate that geological substrate strongly modulates the capacity of forest ecosystems to accumulate and stabilize soil organic carbon.

3.3.1 Ecosystem-controlled test of the substrate effect (Cedar and Holm oak forests)

Cedar forest and Holm oak forest were each sampled on both basalt and dolomite within Ifrane National Park, providing a direct, vegetation-controlled test of the substrate effect. For Cedar forest, mean SOC did not differ significantly between basalt (178.1 t C ha^-¹, n = 6) and dolomite (160.9 t C ha^-¹, n = 6; Welch’s t-test, P = 0.680). For Holm oak forest, SOC was higher on basalt (264.7 t C ha^-¹, n = 6) than on dolomite (174.5 t C ha^-¹, n = 6), but the difference likewise did not reach significance (P = 0.143). These ecosystem-controlled comparisons, although limited by small sample sizes (n = 6 per group), do not provide statistically significant support for an independent substrate effect once vegetation type is held constant, and temper the strength of the substrate-driven interpretation given for the full, substrate-confounded dataset above.

A comparison is also available for Holm oak forest, the only ecosystem type present in both national parks and therefore sampled across all three geological substrates (basalt, n = 6; dolomite, n = 6; limestone, n = 4). A one-way ANOVA restricted to this single ecosystem showed a significant substrate effect (F = 6.778, P = 0.0096), with SOC decreasing from basalt (264.7 t C ha^-¹) to dolomite (174.5 t C ha^-¹) to limestone (59.7 t C ha^-¹). This comparison holds forest ecosystem type constant, but it does not hold National Park constant: the limestone group is drawn entirely from Talassemtane, while the basalt and dolomite groups are drawn entirely from Ifrane – the same collinearity between substrate and Park invoked elsewhere (section 3.6) to justify excluding Park from the main GLM. Consequently, this result should be read as a suggestive, but Park-confounded, pattern rather than a clean, vegetation-only test of the substrate effect: any difference between the limestone group and the other two may partly reflect Park-level differences unrelated to substrate. The two comparisons that do hold both National Park and ecosystem type constant – Cedar forest and Holm oak forest, basalt versus dolomite within Ifrane – remain the only tests in this dataset free of this particular confounding, and both were non-significant (P = 0.680 and P = 0.143; section 3.3.1).

3.4 Combined influence of forest ecosystem and geological substrate

The interaction between forest ecosystem and geological substrate further emphasized the importance of environmental heterogeneity.

Among basalt-derived ecosystems, Zeen oak and Holm oak forests consistently exhibited the highest SOC stocks, whereas Cedar and Maritime pine forests developed on dolomitic substrates showed lower carbon accumulation.

Similarly, all ecosystems occurring on limestone in Talassemtane displayed comparatively lower SOC stocks despite differences in vegetation composition.

These findings demonstrate that vegetation effects cannot be interpreted independently from parent material.

3.5 Ecosystem diversity and SOC heterogeneity

Both parks exhibited identical ecosystem richness, with six ecosystem types recorded in each protected area. Shannon and Simpson indices were also very close, indicating that the sampled ecosystem types were relatively evenly represented. However, the carbon response to ecosystem diversification differed strongly between parks. Ifrane had a much greater SOC coefficient of variation at the plot level (53.97%) than Talassemtane (22.33%). The range of SOC stocks was also much larger in Ifrane (401.81 t C ha^-¹) than in Talassemtane (63.09 t C ha^-¹).

At the ecosystem-mean level, the range among ecosystem types reached 142.52 t C ha^-¹ in Ifrane, compared with 41.72 t C ha^-¹ in Talassemtane (Figure 7). This indicates that Ifrane presents stronger ecosystem-level differentiation in carbon storage potential. Therefore, ecosystem diversity in Ifrane is associated with a more pronounced spatial heterogeneity of SOC storage, while Talassemtane shows a more homogeneous carbon profile among ecosystems.

3.6 Statistical modeling: GLM, mixed-effects model and multivariate analyses

The General Linear Model confirmed that forest ecosystem type and geological substrate both contributed to SOC variability, with substrate as the stronger determinant.

Among these explanatory variables, geological substrate represented one of the strongest determinants of SOC storage.

Model formula: SOC stock ~ Forest ecosystem + Geological substrate + Ecosystem × Substrate.

The GLM (Type II sums of squares, each main effect adjusted for the other) indicated a significant effect of geological substrate (F = 8.933, P < 0.001) and a marginal effect of forest ecosystem (F = 1.897, P = 0.066) on SOC stocks once each was adjusted for the other. The exploratory ecosystem × substrate interaction term, added last and tested against both main effects, was not significant (F = 1.534, P = 0.221); this result was robust to the order in which model terms were entered, and should nonetheless not be over-interpreted given the unbalanced and partially confounded design.

3.6.1 Linear mixed-effects model

The linear mixed-effects model (SOC ~ Substrate, with forest ecosystem as a random intercept; REML) estimated SOC on basalt at 200.5 t C ha^-¹ (SE = 20.1), with dolomite averaging 53.7 t C ha^-¹ lower (SE = 24.2, approximate t = –2.22) and limestone averaging 139.0 t C ha^-¹ lower than basalt (SE = 27.0, approximate t = –5.15). The between-ecosystem variance component (σ² = 947) represented approximately 15% of the total variance in SOC (intraclass correlation = 0.154), the remaining 85% corresponding to residual, within-ecosystem variability. The model’s best linear unbiased predictions (BLUPs) shrank the estimated deviation of each ecosystem type toward the overall mean in proportion to its sample size; shrinkage was most pronounced for the two most sparsely sampled ecosystems (Fir forest and Mediterranean matorral, n = 2 each), whose adjusted deviations from the mean were small despite comparatively extreme raw group means, illustrating how the mixed-effects approach mitigates the influence of small, potentially unreliable strata rather than treating them as equivalent to well-replicated ecosystem groups. Because software implementing exact Satterthwaite or Kenward–Roger denominator degrees of freedom was not available for this analysis, the t-values above should be regarded as approximate rather than exact.

3.6.2 Multivariate analyses (PCA and HCA)

Principal Component Analysis (PCA) clearly separated basalt-derived ecosystems from dolomitic and limestone ecosystems along the first principal component, which explained most of the total variance.

Hierarchical Cluster Analysis produced a similar grouping pattern, indicating that geological substrate strongly structured ecosystem-level SOC storage (Figure 8).

4 Discussion

4.1 Forest ecosystem diversity and geological substrate as drivers of SOC storage

The present study demonstrates that forest ecosystem diversity strongly influences the spatial distribution of soil organic carbon (SOC) across Mediterranean mountain landscapes. Significant differences in SOC stocks were observed among ecosystem types in both Ifrane and Talassemtane National Parks, confirming that ecosystem identity is an important determinant of belowground carbon accumulation. However, the magnitude of ecosystem effects differed considerably between the two protected areas, with ecosystem-level heterogeneity being substantially greater in Ifrane than in Talassemtane.

Unlike most previous studies that assessed biodiversity through tree species richness (22, 23), the present work adopted a landscape perspective by considering biodiversity as the diversity and spatial variability of forest ecosystem types. This approach captures not only differences in floristic composition but also variations in vegetation structure, litter production, rooting patterns, microclimate and ecological functioning. Consequently, ecosystem diversity provides a more realistic representation of the mechanisms governing SOC storage in heterogeneous mountain environments.

The high SOC stocks observed in zeen oak and holm oak forests illustrate how ecosystem-specific characteristics regulate carbon inputs to the soil. Broadleaved deciduous and evergreen forests generally produce abundant litter with relatively favorable chemical composition, promoting the continuous incorporation of organic matter into mineral soils (21). Differences in root biomass, microbial activity and decomposition rates further contribute to the contrasting SOC levels observed among ecosystem types (10, 19).

4.1.1 Geological substrate is associated with, but not yet confirmed as an independent driver of, SOC accumulation

One of the most important findings of this study is that the influence of forest ecosystem diversity cannot be interpreted independently of geological substrate. The comparison between basaltic, dolomitic and limestone-derived ecosystems clearly indicates that parent material exerts a major control on SOC storage.

Basalt-derived ecosystems exhibited the highest average SOC stocks (200.84 t C ha^-¹, n = 24), followed by dolomitic ecosystems (147.14 t C ha^-¹, n = 24), whereas limestone ecosystems in Talassemtane stored only 69.44 t C ha^-¹ (n = 19). This gradient is present in the pooled, ecosystem-confounded comparison, but the two available ecosystem-matched tests (Cedar forest and Holm oak forest, basalt vs dolomite; section 3.3.1) were both non-significant (P = 0.680 and P = 0.143), so it should be read as a suggestive association rather than confirmed evidence that geological substrate independently influences the capacity of soils to accumulate and stabilize organic carbon.

Basaltic rocks weather relatively rapidly and generate clay-rich soils containing abundant Fe and Al oxides capable of protecting organic matter through organo-mineral associations (12, 15). These reactive mineral phases enhance aggregate formation and reduce microbial accessibility to organic substrates, thereby increasing long-term SOC stabilization.

In contrast, calcareous and dolomitic substrates frequently develop shallower soils with lower water-holding capacity and weaker mineral protection mechanisms (16, 50). Consequently, organic matter decomposition may proceed more rapidly, resulting in lower SOC accumulation despite comparable forest cover.

These findings indicate that geological substrate represents a fundamental ecological filter controlling ecosystem carbon storage.

4.2 Interaction between vegetation and parent material

Rather than acting independently, forest ecosystem type and geological substrate were largely confounded in the present sampling design, particularly within Ifrane National Park, where four of the six ecosystem types occurred exclusively on a single substrate; only Cedar forest and Holm oak forest were sampled on both substrates, offering a partial, though limited, opportunity to test the substrate effect independently of vegetation (section 3.3.1). Consequently, the formal ecosystem × substrate interaction term could not be statistically validated (F = 1.534, P = 0.221; Table 4) and should not be interpreted as evidence of a genuine interaction effect. The patterns discussed below are therefore descriptive rather than confirmatory, and are presented to guide, not to conclude, future sampling designs capable of formally separating these two sources of variation.

Table 4. General linear model (GLM) evaluating the effects of forest ecosystem, geological substrate and their interaction on soil organic carbon (SOC) stocks.

SOC stocks are expressed in t C ha^-¹. The model was fitted using Type II sums of squares (each main effect tested after adjusting for the other main effect), with the ecosystem × substrate interaction added last and tested against both main effects. National Park was not retained as a separate term because it was perfectly collinear with, and therefore statistically redundant given, the combination of Forest ecosystem and Geological substrate already in the model (all Talassemtane observations are on limestone, and Ifrane’s forest ecosystems are largely substrate-specific); including Park in addition to Ecosystem and Substrate left it with zero identifiable degrees of freedom. Degrees of freedom sum to 66 (n − 1 = 67 − 1), confirming the model is fully and correctly specified. The ecosystem × substrate term must be interpreted cautiously because, within Ifrane National Park, most ecosystem types occur on only one of the two substrates, so ecosystem and substrate effects are themselves substantially confounded and cannot be fully disentangled with the present data.

Within Ifrane National Park, the highest SOC stocks were observed in broadleaved ecosystems developed on basalt, while lower stocks occurred in ecosystems developed on dolomite. However, because ecosystem type and substrate co-vary almost perfectly within this park, this pattern cannot be attributed to vegetation, to substrate, or to their interaction individually; it reflects their combined, non-separable influence. A cleaner test of the vegetation effect alone is provided by Talassemtane National Park, where all sampling sites occur on a single, homogeneous limestone substrate. There, ecosystem type still significantly affected SOC storage (P = 0.0076; section 3.2), providing more direct evidence that vegetation exerts an effect on SOC storage independently of substrate, even though the magnitude of this effect (range of ecosystem means = 41.72 t C ha^-¹) was considerably smaller than the ecosystem-level range observed in the substrate-confounded context of Ifrane (142.52 t C ha^-¹, Table 5). Consistent with this partial disentangling, the direct, ecosystem-controlled comparison presented in section 3.3.1 found no statistically significant SOC difference between basalt and dolomite within either Cedar forest (P = 0.680) or Holm oak forest (P = 0.143), even though Holm oak forest showed a large numerical difference (90 t C ha^-¹) in the expected direction. This does not rule out a genuine substrate effect — the comparison is likely underpowered with only six replicates per group — but it means the substrate effect emphasized throughout this section should be regarded as suggestive rather than statistically confirmed at the ecosystem level.

Table 5A. Ecosystem diversity indices and SOC variability metrics.

Table 5B. Ecosystem-level heterogeneity in SOC storage potential.

“Range of ecosystem means” is the difference between the highest and lowest ecosystem-level mean SOC stock within each park (i.e., max minus min of the ecosystem means listed in Table 2), used here as a simple, non-parametric summary of ecosystem-level heterogeneity in SOC storage; “SD of ecosystem means” is the standard deviation of these same ecosystem-level means.

This asymmetry is consistent with the conceptual framework proposed by Jenny (1941), in which soil development results from the joint action of climate, organisms, relief, parent material and time, none of which can be fully isolated from the others under field conditions. It also explains why the ecosystem effect appeared strongly significant in the within-park ANOVA (Ifrane: P = 0.0155; Talassemtane: P = 0.0076; section 3.2) but only marginal once geological substrate was included as a covariate in the GLM (F = 1.897, P = 0.066; Table 4): part of the apparent ecosystem effect estimated by the simple ANOVA is statistically inseparable from the substrate effect, and is absorbed once substrate is controlled for. The complementary linear mixed-effects model (section 3.6.1) attributed only about 15% of the total variance in SOC to systematic differences among ecosystem types once substrate was accounted for (intraclass correlation = 0.154), while confirming a significant reduction in SOC on limestone and a comparatively weaker, borderline reduction on dolomite, both relative to basalt. Together, these results suggest that geological substrate, rather than ecosystem identity, accounts for most of the systematic (non-residual) variability in SOC storage captured by the present design.

This borderline dolomite-versus-basalt effect in the mixed-effects model (approximate t = –2.22) is not fully consistent with the two direct, ecosystem-matched comparisons in section 3.3.1 (Cedar and Holm oak forest, basalt vs dolomite), which found no significant difference (P = 0.680 and P = 0.143). We do not consider this tension resolved by the present data, and offer two non-exclusive explanations rather than adjudicating between them. First, the mixed model’s pooled substrate estimate borrows statistical weight, through its random-intercept structure, from ecosystems that exist on only one substrate each (Zeen oak forest and Holm oak–Zeen oak mixed forest on basalt only; Maritime pine forest and Cedar–Holm oak mixed forest on dolomite only), for which substrate and ecosystem identity are perfectly aliased; its borderline significance could therefore partly reflect residual ecosystem-level confounding rather than a genuine, vegetation-independent substrate effect. Second, and equally plausible, the two fully matched comparisons (n = 6 per substrate within each of two ecosystems) may simply be underpowered to detect a real effect of the modest size implied by the pooled estimate. Because the present dataset cannot distinguish confounding from insufficient power as the explanation for this discrepancy, we treat the substrate effect as an open question rather than as either confirmed or refuted, and calibrate the Abstract and section headings accordingly.

Therefore, differences among Mediterranean forest ecosystems in this dataset cannot be interpreted solely in terms of floristic composition. Geological substrate should be regarded as an environmental filter and covariate that constrains and partly co-determines the vegetation effect, rather than as a fully independent, competing driver whose separate contribution can be quantified with confidence from the present sampling design.

This combined influence, rather than a confirmed statistical interaction, explains why ecosystem-level SOC heterogeneity was much greater in Ifrane than in Talassemtane. The coexistence of basaltic and dolomitic formations in Ifrane creates contrasting pedological environments that co-vary with ecosystem type, amplifying the apparent range of SOC values among ecosystems. In contrast, the uniform limestone substrate in Talassemtane removes this source of confounding, so the smaller but still significant differences among ecosystems observed there more plausibly reflect vegetation effects alone.

4.3 Implications for carbon management and forest conservation

The present findings have important implications for ecosystem-based forest management under climate change.

Current carbon accounting approaches often prioritize aboveground biomass while paying comparatively little attention to belowground carbon pools. Our results demonstrate that soil carbon storage varies substantially among ecosystems and geological substrates, emphasizing the need to incorporate pedological information into national carbon inventories.

From a conservation perspective, ecosystems characterized by both high biodiversity and high SOC storage should receive priority in restoration and protection programs. In particular, basalt-derived holm oak and zeen oak forests represent strategic carbon reservoirs that contribute disproportionately to long-term carbon sequestration.

Conversely, ecosystems developed on calcareous substrates should not be regarded as low-value systems. Although their SOC stocks are lower, they provide unique biodiversity, hydrological regulation and habitat functions that complement carbon storage services. Consequently, conservation planning should balance carbon sequestration objectives with broader ecosystem service considerations.

The results further suggest that geological substrate should be incorporated into forest restoration planning. Restoration efforts focusing exclusively on tree species selection may overlook important pedological constraints that determine long-term soil carbon accumulation. Integrating vegetation and substrate information could therefore improve the effectiveness of ecosystem-based climate adaptation strategies in Mediterranean mountain landscapes.

4.4 Limitations and future research

Several limitations should be acknowledged when interpreting these results. First, sampling intensity differed markedly between parks (48 observations in Ifrane versus 19 in Talassemtane), and several ecosystem types were represented by very few observations (n = 2 for Matorral and Fir forest in Talassemtane); means for these ecosystems should be regarded as indicative rather than statistically robust, and the linear mixed-effects model (section 3.6.1) was used specifically to mitigate, though not eliminate, this imbalance through partial pooling. Second, geological substrate and forest ecosystem type remain substantially confounded within Ifrane National Park, since four of the six ecosystem types occurred on only one substrate; only Holm oak forest was sampled across all three substrates, and only this ecosystem-controlled comparison (section 3.3.1) provided a statistically significant, vegetation-independent test of the substrate effect. Third, this study did not incorporate climate variables (e.g., mean annual temperature and precipitation), topographic variables (aspect, slope, elevation) or direct measurements of soil mineralogy (e.g., iron and aluminium oxide content), all of which are known to influence SOC storage and stabilization; the mechanistic interpretation given in section 4.2, attributing greater SOC stabilization on basalt to reactive Fe and Al oxide phases, is therefore based on established pedogenic theory (12, 15) rather than on mineralogical analyses of the present soils, and should be confirmed by future work including direct measurements of extractable Fe and Al oxides. Fourth, soil sampling was restricted to the upper 0–30 cm mineral layer; while this depth captures the most biologically active, carbon-rich horizon and is consistent with common practice in comparative regional SOC studies, it may underestimate total profile SOC storage, particularly in deeper soils on gentler dolomitic or basaltic slopes, and future work should extend sampling to greater depths where feasible. Fifth, forest ecosystem diversity indices (section 2.5) were calculated from the relative frequency of ecosystem types among sampling plots rather than from an area-weighted landscape inventory, and should be interpreted as describing the sampled dataset rather than the true areal composition of each park. Future studies would benefit from a balanced sampling design in which the same ecosystem types are represented across multiple geological substrates, combined with climatic, topographic and mineralogical covariates and, where possible, site- or plot-level replication, to more fully disentangle the relative contributions of vegetation and parent material to SOC storage in Mediterranean mountain landscapes.

5 Conclusion and perspectives

This study demonstrates that soil organic carbon (SOC) storage in Mediterranean mountain forests is jointly influenced by forest ecosystem diversity and geological substrate. The comparison between Ifrane and Talassemtane National Parks revealed marked differences in carbon sequestration potential, with SOC stocks being substantially higher in Ifrane, reflecting greater ecological and geological heterogeneity. The results further highlight the dominant role of parent material, with basalt-derived ecosystems exhibiting the highest SOC storage, followed by dolomitic and limestone ecosystems. These findings emphasize that ecosystem diversity alone cannot fully explain SOC variability and that geological conditions must be integrated into ecosystem-based carbon accounting, forest restoration and conservation strategies. Although the present study provides new insights into the mechanisms controlling SOC storage in Mediterranean forests, future research should extend the analysis to a wider range of Moroccan forest ecosystems using balanced sampling designs capable of disentangling vegetation and substrate effects. Integrating additional environmental variables (e.g., soil texture, mineralogy, topography and climate), biological drivers such as microbial communities and litter quality, together with advanced statistical approaches, remote sensing and digital soil mapping, will improve the spatial prediction of SOC stocks and support more effective climate change mitigation, biodiversity conservation and sustainable forest management (51, 52).

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Generative artificial intelligence was used during the preparation of this manuscript to rephrase and improve the clarity, fluency, and conciseness of selected passages from the original text while preserving the scientific meaning, the structure of the arguments, and all original references. It also provided support for editorial tasks, including improving the consistency of the document layout and the presentation of figures, tables, and references. All AI-generated revisions were carefully reviewed, revised where necessary, and approved by the authors before being incorporated into the final version of the manuscript. The study design, data collection and processing, statistical analyses, interpretation of the results, and scientific conclusions were entirely conducted, critically reviewed, and validated by the authors. The authors take full responsibility for the accuracy, integrity, and originality of the content presented in this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Sofia Maria Muscarella, Università degli Studi di Palermo, Italy

Reviewed by: Vladimir Tabunshchik, A.O. Kovalevsky Institute of Biology of the Southern Seas of RAS, Russia

Giorgio Galluzzi, University of Verona, Italy