Carbon stocks and biodiversity diverge between seasonally flooded and semi-evergreen tropical forests in the Yucatan Peninsula: implications for conservation in a threatened dry landscape

Seasonally dry tropical forests (SDTF) host ecologically distinct vegetation types, with differences in forest structure, taxonomic diversity, and aboveground carbon stocks that remain insufficiently quantified. We compared a semi-evergreen tropical forest (SETF) and a seasonally flooded lowland forest (SFLF) in the Calakmul region of south-eastern Mexico. A total of 2,622 trees were recorded across 10 plots (40 subplots; n = 20 per type), measuring forest structure parameters such as basal area, canopy height, and clear bole height. We estimated above-ground carbon stocks, taxonomic diversity (Hill’s q0–q2 indices using iNEXT rarefaction/extrapolation), and floristic composition (NMDS, PERMANOVA). Basal area, canopy height, and carbon stocks were significantly higher in SETFs (33.4 m 2 ha −1, 15.3 m, 118.9 t C ha −1) compared to SFLFs (24.1 m 2 ha −1, 11.5 m, 63.6 t C ha −1), while stem density showed no difference between the two. A linear mixed model explained 97.4% of the variance in carbon stocks (marginal R 2 = 0.974), with basal area as the strongest single predictor (semi-partial R 2 = 0.255), partly reflecting the shared allometric basis of basal area and carbon estimates. Leaf area index made no independent contribution, suggesting that seasonal flooding reduces carbon storage primarily through smaller individual stem size rather than differences in stand density or canopy cover. Contrary to our prediction, Shannon and Simpson diversity indices were significantly higher in the SFLF at a standardised sample size, indicating that periodic flooding reduces competitive dominance and promotes more equitable abundance distributions, rather than decreasing species richness. Floristic composition differed entirely between the two types (PERMANOVA R 2 = 0.73). Given projected drought trends for the Yucatán Peninsula, shifts in flooding seasonality could drive landscape-scale carbon losses that low-resolution global models may miss. These findings underscore the ecological importance of seasonally flooded forests and the risk of systematically undervaluing them in carbon-based conservation strategies.

Tropical forests contain approximately two-thirds of the planet’s terrestrial biodiversity (Gardner et al., 2009), and are essential for maintaining ecosystem processes and services, such as carbon sequestration (Pan et al., 2011), water supply and flood control (Ellison et al., 2017), soil maintenance and formation (Lal, 2004), as well as for biodiversity conservation. However, deforestation and degradation continue to rise due to human activities, such as land-use change towards mechanised agriculture, the introduction of pasture for livestock, soya cultivation, and oil palm plantations, amongst others (Boucher et al., 2011; Kissinger et al., 2012; Feurer et al., 2025). Consequently, international commitments have been established for their protection—such as the Paris Agreement and the Bonn Challenge—which have been adopted by numerous countries (Anderegg et al., 2020; Bastin et al., 2019; Chazdon et al., 2016; Soto-Navarro et al., 2020). Furthermore, financial schemes have been developed to incentivise the conservation of forest cover, notably Payment for Environmental Services (PES) and carbon markets. However, their implementation requires information on vegetation’s potential to sequester carbon, as well as on sequestration dynamics.

Global estimates of the mitigation potential of forests exist. Anderegg et al. (2020), for example, estimated a global forest mitigation potential of ~7 Pg CO₂e year −1 by 2030, while a more recent, integrated global assessment estimated an additional 226 Gt of forest carbon potential outside agricultural and urban lands (Mo et al., 2023). However, these estimates remain highly uncertain for seasonally dry tropical forests (SDTFs). These uncertainties arise, in part, because large-scale models based on remote sensing often fail to capture the structural heterogeneity of fragmented landscapes (Réjou-Méchain et al., 2019), at different successional stages and combinations of plant communities resulting from local-scale variation in soil and topography (Canto-Sansores et al., 2024). Moreover, the land management that local people do in these lands also is barely captured in these remote sense large-scale models. Furthermore, a geographical bias has been observed in the research, with an over-representation of studies in tropical and temperate rainforests, at the expense of SDTFs (Becknell et al., 2012). This bias in the estimates is particularly significant, given that SDTFs, which account for ~42% of tropical forests (Murphy and Lugo, 1986), have lost more than 90% of their original cover (Banda et al., 2016).

Mooney et al. (1995) defined SDTFs as forests found in regions with significant seasonal water availability, which causes distinct variations in structure and composition (Jaramillo et al., 2003). Locally, variability in soil type and topography creates a patchwork of contrasting plant communities. For example, the semi-evergreen tropical forest (SETF) grows on well-drained soils with water shortages during the dry season, while the Seasonally Flooded Lowland Forest (SFLF) occurs in depressions with poorly drained soils, where periodic waterlogging and root hypoxia pose the main stresses during the wet season. This difference in water regimes —drought versus waterlogging— serves as an environmental filter that influences the composition, functional traits, and structure of vegetation (Keddy, 1992; López and Kursar, 2003; Engelbrecht et al., 2007). Recent evidence reinforces this filtering role and its species-specific nature: in Amazonian riparian forests, soil flooding reduces taxonomic and phylogenetic diversity and generates more randomly assembled tree lineages along hydrographic gradients (Souza et al., 2024), while experimental work shows that flooding disproportionately suppresses recruitment in native tree species without equally affecting invasive plants, indicating that flood tolerance is unevenly distributed across taxa (Herrera-Bevan and Ibáñez, 2024). Despite this, no empirical quantification has shown how these differences in water regimes translate into variations in aboveground carbon stocks. Recent evidence, however, indicates that dry forests might store more carbon than previously thought (Mesa-Sierra et al., 2022; dos Santos et al., 2024), highlighting the importance of understanding how carbon sequestration varies among different plant communities. Therefore, measuring and comparing the carbon stored in these vegetation types is crucial for reducing uncertainties in national inventories and developing conservation and restoration strategies that maximise ecosystem services in human-altered tropical landscapes.

The rainy season across the Yucatan Peninsula is already shifting in onset, duration, and total volume, according to recent station-based records (Romero and Alfaro, 2024), and climate scenarios project a further increase in the frequency and intensity of droughts by the middle and late 21st century (IPCC, 2021; Cavazos et al., 2020). Since the flooding regime is the primary factor shaping floristic composition and forest structure in this karst landscape, and because the two vegetation types differ by nearly a factor of two in their above-ground carbon stocks, changes in seasonality and flooding regimes could lead to landscape-scale carbon losses that low-resolution global models might not detect. Therefore, understanding how flooding influences forest structure and diversity is essential for predicting how these ecosystems will respond to climate change and for developing effective conservation strategies in the region. The aim of this study was to characterise and compare forest structure, taxonomic diversity, and aboveground carbon stocks in two plant communities, SETF and SFLF, within a tropical landscape in southern Mexico. Based on the principle of environmental filtering, we hypothesised that contrasting flooding regimes (drought stress versus waterlogging) are the key factors driving the differences between the two communities. Accordingly, we predict that the SFLF, due to the physiological constraints caused by periodic waterlogging on species growth and survival, will display (a) lower species diversity, (b) a shorter forest (height, diameter) and vegetation complexity (Canto-Sansores et al., 2024), and (c) consequently, a significantly smaller above-ground carbon pool compared to the SETF.

Study site

Sampling was carried out in the municipality of Calakmul, located in Campeche state, south-southeastern Mexico, specifically in the ejidos of Centenario and Hopelchén (Figure 1; Table 1). The region’s climate corresponds to Aw (tropical, dry-winter climate) under the internationally standardised Köppen-Geiger system (Beck et al., 2018), described for Mexico as warm sub-humid (Aw) under García's (1988) modification of the Köppen system, which further distinguishes a south-to-north precipitation gradient comprising three humidity subtypes based on the precipitation/temperature (P/T) ratio: Aw0 (P/T < 43.2, least humid), Aw1 (43.2 ≤ P/T < 55.3, intermediate), and Aw2 (P/T ≥ 55.3, most humid). At the nearest long-term meteorological station (Zoh Laguna, 18.59° N, 89.42° W; CONAGUA-SMN, 1991–2020 climate normal), mean annual precipitation is 909.4 mm (monthly range: 18.1 mm in March to 144.1 mm in September) and mean annual temperature is 23.6 °C (P/T = 38.5, Aw0 subtype) (CONAGUA, 2022). The geology is of karstic origin (Bauer-Gottwein et al., 2011), and the soils exhibit an altitudinal distribution over Eocene-Palaeocene carbonate rocks. The dominant soil types contain abundant organic matter but have low mineral fertility and are adapted to the karstic relief and variable water regimes; they include lithosols and rendzinas on the hills, as well as deep vertisols and gleysols on the plateaus (Miranda and Hernández, 1963; Bautista et al., 2011). The dominant vegetation is a medium-sized evergreen forest at various stages of ecological succession. The plant composition consists of species of high commercial value, such Swietenia macrophylla King, Cedrela odorata L., Lysiloma latisiliquum [L.] Benth., Metopium brownei Jacq., (Brosimum alicastrum Swartz, Manilkara zapota [L.] Van Royen, Bursera simaruba L., Platymiscium yucatanum Standl., Cordia dodecandra D. C., Coccoloba spicata, Piscidia piscipula [L] Sarg., and Caesalpinia mollis S.W.) (Ellis et al., 2015). To a lesser extent, but widely distributed across the region, there is also the seasonally flood-prone lowland forest, which is characterised by its occurrence in enclosed, flat, poorly drained valleys (polje), typical of karst landscapes. This plant community is characterised by the presence of species such as Haematoxylum campechianum L and Terminalia buceras (L.) C. Wright, both of which are forest species, as well as Gymnopodium floribundum Rolfe, Croton reflexifolius Kunth, Byrsonima bucidifolia Standl and Byrsonima crassifolia (L.) Kunth (honey plants). Furthermore, it plays an important role as a wildlife refuge and a source of floristic germplasm (Perez-Salicrup, 2004; Tun-Dzul et al., 2008; Ellis et al., 2017).

Sampling design

Sampling was carried out during February 2023, characterised by outbreaks of continental polar/arctic air (known as “Nortes”) that affect the Gulf of Mexico and the Yucatán Peninsula, bringing sporadic rainfall, cooler temperatures, and strong winds (Kurczyn et al., 2021). Ten 20 × 50 m (1,000 m 2) plots were established across two types of vegetation communities: five belonging to the SETF and five to SFLF (Figure 1; Table 1). Each plot was subdivided into four subplots measuring 12.5 m × 20 m (20 subplots per vegetation type, 40 in total), and the following variables were recorded for all individuals (trees, shrubs and palms) with a diameter at breast height (DBH) ≥ 5 cm: taxonomic identity, DBH, clear-stem height and total height. Furthermore, hemispheric photographs of the canopy were taken at the centre of each subplot to calculate the leaf area index (LAI), which quantifies the total leaf area per unit of ground area (m 2 m −2). This index uses the gap fraction method, based on the Beer–Lambert extinction model: LAI = −ln(Pgap)/k, where Pgap is the light transmittance of the canopy and k is the extinction coefficient (spherical distribution of leaf angles: k ≈ 0.5). Litter was collected from a 1 m 2 area to calculate biomass (Figure 1).

Measurement of vegetation structure and diversity

Vegetation structure was quantified using the dasometric data collected in each subplot. The basal area (BA) of each individual was calculated as BA = π · (DBH/2) 2 and subsequently converted to square metres per hectare (m 2 ha −1). Stem density, dominant height (P95), maximum height and clear-stem height were also determined. Aboveground forest biomass was estimated using species-specific allometric equations where available, or general equations based on DBH and wood density (Table 2; Puc-Kauil et al., 2019). Litter biomass was calculated from a 1 m 2 area at the centre of each subplot, where litter was collected, weighed excluding the decomposition layer, and dried to constant weight at 70 °C. Both biomass components were extrapolated to Mg ha −1 and converted to carbon equivalents by multiplying by 0.5, following IPCC recommendations IPCC (Intergovernmental Panel on Climate Change) (2006). Task force on national greenhouse gas inventories (TFI). Taxonomic diversity was quantified using Hill numbers of orders q = 0, 1 and 2 for each subplot, calculated with the R package iNEXT (Hsieh et al., 2016). These orders differ in the weight assigned to rare versus dominant species: q0 equals observed species richness; q1 is the exponential of Shannon entropy, representing the effective number of common species; and q2 is the inverse of Simpson’s dominance index, representing the effective number of dominant species. Finally, to assess the relative importance of each species within the community, determine the structural and compositional complexity of the communities, and compare them, the value importance relative (VIR) was calculated. This index was calculated using the following three attributes: (i) relative abundance, defined as the proportion of individuals of a species relative to the total number of individuals; (ii) relative frequency, defined as the proportion of plots in which the species is found relative to the total number of plots; and (iii) relative basal area, defined as the proportion of the species’ basal area relative to the total basal area.

Data analysis

Differences in structural metrics (basal area, stem density, mean square diameter, dominant height, clear-stem height), canopy metrics (LAI) and aboveground carbon stocks between vegetation types (SETF and SFLF) were assessed using linear mixed models (LMM), with vegetation type as a fixed effect and plot as a random effect. This approach accounts for the hierarchical structure of the data, incorporates within-subplot variability, and avoids pseudoreplication. For the vegetation-type contrast specifically, Kenward-Roger degrees of freedom converge to eight, matching the expected value for a between-plot factor tested across five plots per vegetation type (Luke, 2017). Post hoc pairwise comparisons between vegetation types were obtained using estimated marginal means with no adjustment for multiple comparisons, given that only a single contrast was evaluated (emmeans package; Lenth and Piaskowski, 2024).

To explore the mechanisms underlying differences in aboveground carbon stocks, we fitted an LMM with carbon stock as the response variable and vegetation type, basal area, dominant height (P95), LAI, and species richness (q0) as fixed effects, with plot as a random effect.

We standardised all continuous predictors (z-scores) to compare effect sizes. We estimated the semi-partial R 2 of each predictor using the partR2 package (Stoffel et al., 2021) with 1,000 bootstrap replicates, decomposing the model’s marginal R 2 into each variable’s unique contribution. Degrees of freedom for all fixed-effect tests were approximated using both the Kenward-Roger (pbkrtest package) and Satterthwaite (lmerTest package) methods; multicollinearity among predictors was assessed via variance inflation factors (VIF; performance package), and marginal/conditional R 2 and the intraclass correlation coefficient (ICC) were computed to quantify the contribution of the plot-level random effect (Nakagawa et al., 2017). Because the full model additionally includes covariates that vary within as well as between plots, the degrees of freedom for the vegetation-type contrast in this model are higher (Kenward-Roger and Satterthwaite df = 25.9) than in the univariate model above, reflecting the combined within- and between-plot information used by the full model rather than treating subplots as independent replicates for vegetation type itself. Because vegetation type itself does not vary within plots, we additionally verified this contrast using cluster-robust standard errors (10 clusters, CR2 bias-reduced linearization; Pustejovsky and Tipton, 2018) and by refitting the model on plot-level means (n = 10, no subplots).

Taxonomic diversity (Hill numbers of orders q = 0, 1 and 2) was compared between vegetation types using separate LMMs with the same random structure. Community-level diversity was compared using individual-based rarefaction and extrapolation curves (iNEXT package; Hsieh et al., 2016) standardised to n = 1,214 individuals, corresponding to the smallest observed community (SFLF), with 200 bootstrap replicates and 95% confidence intervals (Chao et al., 2014). Floristic composition was assessed using non-metric multidimensional scaling (NMDS) based on Bray–Curtis dissimilarities at the subplot level. Differences in composition between vegetation types were tested using PERMANOVA (adonis2; 999 permutations) stratified by plot to account for the nested structure of subplots within plots. Homogeneity of multivariate dispersion (betadisper) was assessed to verify that PERMANOVA results reflected differences in composition rather than differences in internal variability between groups, and was also interpreted as an indicator of floristic heterogeneity within each vegetation type. Finally, the relative importance of each species was quantified using the Relative Importance Value (VIR), synthesised from relative density, relative frequency and relative dominance calculated at the subplot level. All analyses were conducted in R version 4.5.0 (R Core Team, 2024).

A total of 2,622 individuals and 68 tree species belonging to 33 families were recorded. In the SETF, 1,408 individuals and 49 species distributed across 28 families were recorded, with Lonchocarpus yucatanensis being the most dominant species (VIR = 38.0); in the SFLF, 1,214 individuals of 42 species across 23 families were recorded, with Haematoxylum campechianum being the most dominant species (VIR = 36.7). In terms of species richness, Fabaceae was the best-represented family in both communities (SETF: 9 spp.; SFLF: 11 spp.), followed by Polygonaceae (SETF: 4 spp.; SFLF: 3 spp.) and Euphorbiaceae (SFLF: 4 spp.). In terms of abundance, Fabaceae was also the family with the most individuals in SETF (273 ind., 19.4% of the total), whilst in SFLF the most abundant families were Fabaceae (197 ind.), Euphorbiaceae (194 ind.) and Polygonaceae (192 ind.), with a notably more even distribution amongst the dominant families than in SETF—a pattern consistent with the greater evenness at species level detected by Hill numbers (Table 3).

Forest structure

Both plant communities showed differences in all structural variables, but not in individual density or leaf area index (Figure 2). Basal area was significantly greater in SETF than in SFLF (p = 0.001), as were dominant height (p < 0.001), maximum height (p < 0.001), clear bole height (p < 0.001) and mean square diameter (p = 0.034). Stem density showed a marginal trend without reaching significance (p = 0.073). The leaf area index did not differ between the types (p = 0.240). Carbon stocks were almost double in the SETF compared to the SFLF in raw terms (Table 3: 118.9 versus 63.6 t C ha −1). After adjusting for basal area, dominant height, LAI and species richness in the full mechanism model, the vegetation-type contrast was reduced (emmean: 98.6 versus 83.9 t C ha −1) and nominally significant under the Kenward-Roger/Satterthwaite approximation (t = 2.59, df = 25.9, p = 0.016; Figure 2d). However, because vegetation type does not vary within plots, we verified this contrast using cluster-robust standard errors (10 clusters, CR2: t = −1.79, df = 5.63, p = 0.126; Pustejovsky and Tipton, 2018) and by refitting the model on plot-level means (n = 10, df = 4: t = −1.41, p = 0.231); neither approach supported the adjusted vegetation-type contrast as statistically significant. This indicates that basal area statistically accounts for nearly all of the raw carbon difference between vegetation types, consistent with our interpretation that flooding acts on carbon largely through its effect on stem size. The full model explained 96.8% of the variance in carbon stocks (marginal R 2 = 0.968). Conditional R 2 was 0.977, and the plot-level random effect had an adjusted ICC of 0.274 (unadjusted ICC = 0.009, matching the difference between conditional and marginal R 2 by construction), indicating a moderate degree of between-plot clustering once fixed effects were accounted for. Collinearity was low (VIF < 2) for basal area, LAI, and species richness, but moderate for vegetation type and dominant height (VIF = 4.2–4.8), consistent with their expected structural covariation. It was found that basal area was by far the strongest single predictor of carbon stocks (semi-partial R 2 = 0.255; 95% CI [0.165; 0.343]), followed by vegetation type (R 2 = 0.031) and species richness (R 2 = 0.012), whilst dominant height and LAI made negligible unique contributions (R 2 = 0.006 and 0.000, respectively). Subplots with higher species richness tended to store less carbon, independently of basal area and vegetation type (p = 0.013); however, this effect was modest (semi-partial R 2 = 0.012) and, given the observational design and the single-landscape sample, we treat it as an exploratory, hypothesis-generating result rather than a robust ecological relationship (Figures 2a–c).

Taxonomic diversity

Rarefaction and extrapolation curves based on n = 1,214 individuals (sample coverage: SETF = 0.995, SFLF = 0.996) showed that SETF had greater species richness than SFLF at the observed sample size (q0: 49 versus 42 species), although the confidence intervals overlapped (SETF: 48.1 [44.5, 51.8]; SFLF: 42.0 [38.8, 45.2]; Figure 3). The pattern was reversed for the other diversity orders: SFLF showed higher Shannon diversity (q1: 22.1 [20.9, 23.3]) and Simpson diversity (q2: 15.9 [14.8, 17.0]), whilst SETF showed q1 = 18.1 [17.0, 19.2] and q2 = 12.2 [11.4, 13.0]; in neither case did the confidence intervals overlap. At the subplot level, however, none of the Hill diversity orders differed significantly between the types (q0: p = 0.865; q1: p = 0.985; q2: p = 0.665), suggesting that the pattern detected at the community level reflects a difference in the distribution of abundances between species, rather than a consistent signal within each plot.

Floristic composition

We found that species composition differed markedly between the two vegetation types. The NMDS ordination of subplots (stress = 0.061) showed no overlap between SETF and SFLF along the NMDS1 axis (Figure 4). The PERMANOVA analysis confirmed a significant difference in floristic composition (R 2 = 0.73, F₁,₈ = 21.6, p = 0.01; Bray–Curtis distance, 999 permutations). Multivariate dispersion differed between the types at the subplot level (betadisper: F₁,₃₈ = 7.08, p = 0.016), indicating greater internal compositional variability in the SFLF subplots compared to those of SETF.

Value importance relative

The two communities differed in the identity and dominance mechanism of their most important species (Figure 5). In SETF, Lonchocarpus yucatanensis (VIR = 38.0) was the most important species, combining high relative abundance (RA = 15.6%) and relative basal area (RBA = 17.0%). Gymnopodium floribundum ranked second (VIR = 27.9) also due to its relative abundance (RA = 14.9%), whilst Manilkara zapota ranked third mainly due to its relative basal area (RBA = 18.8%, RA = 3.3%). In the SFLF, Haematoxylum campechianum was clearly dominant (VIR = 36.7), and its importance was almost entirely due to its basal area (RBA = 22.5%, RBA = 8.9%). Coccoloba cozumelensis ranked second in the SFLF (VIR = 24.7) due to its relative abundance (RA = 11.7%). Manilkara zapota was the only species among the top five in both community types (VIR = 27.0 in SETF, 20.4 in SFLF).

The differences in basal area, canopy height, and clear bole height between SETF and SFLF are consistent with a pattern of physiologically limited tree growth associated with seasonal flooding, although our observational design cannot establish this as a direct causal mechanism. In particular, this pattern may reflect constraints on vertical and radial growth associated with seasonal flooding. Saturated soils reduce oxygen availability in the rhizosphere, with negative consequences for root respiration and nutrient uptake (Kozlowski, 1997). In this regard, we found that the seasonal flooding regime primarily affects canopy structure rather than relative density, which did not differ significantly between vegetation types (p = 0.073). This has been observed in other seasonally flooded tropical forests of the Neotropics; for example, in the transition forests of the Araguaia River (Brazil), stem density and species richness did not respond to the flooding gradient, but canopy height did, which was negatively correlated with the level of flooding (Kurzatkowski et al., 2015). This suggests that lower carbon stocks in SFLFs directly reflect structural constraints, rather than differences in species composition or stand density. This is consistent with the general observation that aboveground carbon in tropical forests is determined primarily by stem structure, rather than by stem number (Matsuo et al., 2025). Variance partitioning suggests that basal area accounted for 25.5% of the unique variance in carbon stocks (semi-partial R 2 = 0.255, 95% CI [0.165, 0.343]), compared to 3.1% for vegetation type and 1.2% for species richness. This pattern is consistent with an association between the flooding regime and carbon storage that may partly reflect constraints on stem growth, in line with the dominant role of structural attributes in determining aboveground carbon stocks across tropical forests (Poorter et al., 2015; Slik et al., 2013). It is important to note that basal area and carbon stocks are both derived from DBH measurements taken on the same trees—basal area as π(DBH/2) 2 and carbon via DBH-based allometric equations (Chave et al., 2014; Puc-Kauil et al., 2019; Puc-Kauil et al., 2020)—so part of the semi-partial R 2 attributed to basal area may reflect this shared measurement basis rather than solely an independent ecological mechanism. The non-significant contribution of LAI to carbon stocks (semi-partial R 2 = 0.000) suggests that canopy area is maintained despite reductions in height and basal area. In other words, trees in SFLFs are shorter and slimmer, but no less numerous, and their total leaf area is similar. This is relevant for remote sensing applications, as LAI-based indices may not reliably reflect structural differences between the two forest types. Regarding woody species diversity, we found greater equality of abundances in SFLF compared to SETF, which is reflected in significantly higher values of Shannon and Simpson diversity when standardising for sample size. This suggests that the water regime may influence which species can establish themselves and their abundance. For example, in the SETF, a few highly competitive species concentrate the majority of individuals under free-draining conditions, resulting in a markedly asymmetric distribution of abundances. On the other hand, the seasonal flooding of the SFLF appears to reduce the competitive advantage of these dominant species, allowing a greater number of species to share available resources more equitably. This result is consistent with the hypothesis that abiotic stress gradients moderate the intensity of interspecific competition, redistributing dominance without necessarily reducing the number of species present (Grime, 1977; Connell, 1978). A similar pattern has been documented in Neotropical secondary forests, where rarefied diversity was lower in more humid environments precisely because the numerical dominance of a few pioneer species compressed community evenness (van der Sande et al., 2024). Furthermore, the complete separation of SETF and SFLF in the NMDS ordination (stress = 0.061) and the PERMANOVA result (R 2 = 0.73, F₁,₈ = 21.6, p = 0.01) suggest that the seasonal flooding regime is the main determinant of floristic composition at this spatial scale, consistent with the environmental filtering hypothesis—the hydrological regime may act as a selective filter influencing which species can establish and persist in each vegetation type (Keddy, 1992; Engelbrecht et al., 2007). The significant result from betadisper (F₁,₃₈ = 7.08, p = 0.016) indicates that SFLF subplots are more variable in composition than those of SETF, which may reflect small-scale variation in flood intensity across the SFLF, where the microtopographic variation of karst soils generates drainage gradients that modulate the local coexistence of species. The more uniform composition of the SETF is consistent with the relatively homogeneous drainage conditions of the upland semi-evergreen forest on shallow calcareous soils. For example, Haematoxylum campechianum, a species strongly associated with flood-prone soils throughout the Yucatán Peninsula, was the most important species in the SFLF (VIR = 36.7), achieving this position almost exclusively through relative basal area (RBA = 22.5%), with a comparatively modest relative abundance (RA = 8.9%). This contrasts with the dominant species in the SETF, where Lonchocarpus yucatanensis achieves a high VIR through a combination of abundance and basal area. The dominance of H. campechianum through basal area, rather than abundance, is consistent with this species’ physiological tolerance to seasonal flooding and the high density of its wood, which supports slow but steady diameter growth under anaerobic soil conditions (Chablé-Vega et al., 2019). The species defines the tintal vegetation type throughout the Yucatán Peninsula, where it frequently forms almost monospecific stands in poorly drained depressions (Plasencia-Vázquez et al., 2025). Manilkara zapota was the only species among the five most important in both community types (VIR = 27.0 in SETF, 20.4 in SFLF), indicating greater ecological tolerance in relation to the hydrological gradient.

Taken together, these results suggest that the seasonal flooding regime is associated with differences between these two plant communities through two complementary, non-exclusive patterns: one consistent with physiological limitation of individual tree growth—reflected in lower basal area, canopy height and carbon stocks in the SFLF—and one consistent with an environmental filtering effect on species composition, evidenced by the complete floristic turnover between types (PERMANOVA R 2 = 0.73) and the greater internal heterogeneity of the SFLF. The fact that diversity did not differ at the subplot level but did at the community level further suggests that the flooding gradient modulates species abundance distributions rather than reducing the total number of species present.

Limitations

Our study has some limitations that should be taken into account for future research. The study is limited to aboveground carbon and does not account for belowground biomass, deadwood or soil carbon, components that may differ between these two forest types due to differences in soil drainage and the accumulation of organic matter under periodic waterlogging. Finally, sampling was carried out during a single period, so temporal dynamics, including the effect of interannual variation in flood duration on structure and diversity, cannot be assessed. The small number of plots per vegetation type (n = 5) limits the statistical power of comparisons between vegetation types; because vegetation type does not vary within plots, Kenward-Roger degrees of freedom for this contrast converge to eight regardless of subplot sample size (Table 3), and the use of subplots (n = 40) increases precision for covariates that vary within plots but does not increase the effective degrees of freedom for the vegetation-type contrast itself. Furthermore, the two vegetation types differ not only in flooding regime but also, to some degree, in soil type, topography and associated species composition; because our design is observational and these covarying factors were not experimentally manipulated or statistically controlled, we cannot fully rule them out as alternative or contributing explanations for the structural and compositional differences reported here. We therefore interpret the flooding regime as the most parsimonious proximate driver consistent with our data, rather than as an isolated, demonstrated cause.

Implications for conservation

These findings have direct relevance for carbon-focused conservation and financing initiatives already underway in the study region. For example, numerous forestry projects are currently being carried out in the region through REDD+ initiatives certified to Climate, Community and Biodiversity (CCB) implemented with the ejidos in the buffer zone of the Calakmul Biosphere Reserve, under which conservation incentives are allocated to communities explicitly based on estimates of above-ground carbon stocks. More generally, the Yucatán Peninsula’s jurisdictional REDD+ strategy, coordinated since 2010 between the states of Campeche, Yucatán and Quintana Roo with the support of Mexico’s National Forestry Commission (CONAFOR) and The Nature Conservancy (Varns et al., 2018), allocates climate finance in accordance with landscape-level carbon estimates. Given that these programmes rely on carbon stock estimates to define priority areas and payment levels, the systematic difference we document between SETFs and SFLFs—linked primarily to basal area rather than canopy diversity or structure—suggests that seasonally flooded forests may be under-prioritised in carbon-based conservation funding, despite harbouring floristically distinct and, according to our diversity results, more balanced species communities. We recommend that carbon-focused conservation and payment schemes operating in the region incorporate forest-type-specific carbon estimates, rather than relying on landscape-level average values that may mask this divergence.