Patterns of plant functional variation and specialization along secondary succession and topography in a tropical dry forest

The study area covers a 352 km² landscape of semi-deciduous tropical forest in the Yucatan Peninsula, Mexico, with warm sub-humid AW1 climate (García 1981), summer rains, a dry season from November to April, mean annual temperature of 26 °C and mean annual precipitation between 900 and 1100 mm. The landscape consists of flat areas with relatively deep (40–100 cm) clayey Luvisols and Cambisols and small limestone hills (elevations between 60 and 190 m asl) with shallow (5–20 cm) rocky Lithosols and Rendzines (Flores and Espejel 1994, Bautista-Zúñiga et al 2003). Soil fertility tends to increase with forest stand age and is higher on hills than on flat areas (Dupuy et al 2012b, López-Martínez et al 2013). The predominant land use has been traditional slash and burn agriculture, practiced by the Mayan people for over 2000 years (Rico-Gray and García-Franco 1991). For further details (see Dupuy et al 2012b).

A Spot 5 satellite image (year 2005) was used to conduct a supervised classification using the following vegetation classes (1) 3 to 8 years of secondary succession, (2) 9 to 15 years, (3) > 15 years on flat areas, and (4) > 15 years on hills (as flat sites are preferred by local farmers) (Dupuy et al 2012a). In the summers of 2008 and 2009, twenty-three 1 km² landscape units were selected across the landscape, and 12 sampling plots (three per vegetation class) were established in each unit (276 plots in total). Stand age of each plot was determined from interviews with local residents who owned or had used the land. Each plot consisted of a circular 200 m² area where all woody plants with a diameter at breast height (DBH, measured at 1.3 m height) > 5 cm were identified and measured (1–99 individuals per plot, mean = 33.3, 9129 individuals in total) (see Hernández-Stefanoni et al 2011 for further details).

Relative importance value (RIV) for all species in each vegetation class was calculated, as the sum of their relative abundance, frequency and basal area. For each vegetation class and separately for trees, lianas and shrubs, species that constituted at least 90% of the cumulative RIV were selected, yielding a total of 72 species. However, only 65 species were included for the analyses (those for which we could collect material from at least five individuals), representing 90% ± 10% (mean ± standard deviation) of the total richness per plot (Pakeman and Quested 2007). Of these species, 45 were trees, 11 shrubs, and 9 lianas.

Leaf samples were collected from at least five individuals per species (14 individuals per species on average, 934 in total) and wood samples from four individuals per species from August to November, 2011 and from September to November, 2012. We calculated total leaf area (LA), minimal photosynthetic unit (MPU, leaf area for entire leaves or leaflet area for compound leaves), leaf dry matter content (LDMC), specific leaf area (SLA) and wood specific gravity (WSG) (for details see online supplementary material 1). The following binary traits were determined from field specimens, parataxonomists, and bibliography: leaf pulvination (LPulv); leaf pubescence (LPb); plant exudates (Ex); plant spininess (Sp; with 0 = absent, 1 = present in all cases); leaf compoundness (LC; 0 = simple, 1 = compound); dispersal syndrome (Dis; 0 = abiotic, 1 = biotic) and deciduousness (LD; 0 = evergreen, 1 = deciduous). Finally, seed volume (SV) was obtained for 55 species as a proxy for seed mass by measuring at least three (and in most cases ten) seeds per individual from samples collected in the field, obtained from the Herbarium at the Centro de Investigación Científica de Yucatán, or from botanical descriptions. Seed volume (SV) was calculated using the formula of an ellipsoid.

We used a multinomial model based on estimated species relative abundances (Chazdon et al 2011) to identify generalist and specialist species, choosing a conservative threshold of k = 2/3, 40 points to define boundaries among groups and a p value of 0.005, as recommended by Chazdon et al (2011). Since the multinomial model classifies species only between two habitats, we performed two analyses, one considering the successional gradient, and another considering the topographical gradient. In the first analysis, abundances of 118 species were compared in 158 young (3–29 y-old) forest plots, versus 53 old-growth forest (30–70 y-old) plots, both groups on flat areas. The successional age limit for old-growth forest was based on a previous finding that abandoned agricultural fields in the Yucatan Peninsula recover a tree species community indistinguishable from extant mature forest in 25–30 years (Turner II et al 2001). The second analysis compared abundances of 119 species in forests on flat areas (106 plots) versus hills (63 plots). We excluded young forest plots from this second analysis (<13 years), because in our study site agriculture is established mainly on flat areas, and we could not find enough replicates of young forests on hills. The multinomial models were performed with the vegan package (Oksanen et al 2016) in R software (R Development Core Team 2015).

To characterize multivariate functional strategies of the 65 most abundant species (for which functional traits were determined), we performed a Principal Component Analysis (PCA) on a species x traits matrix (LA, MPU and PL were log10 transformed). We performed an ANOVA among specialists and generalists PCA scores and post hoc Tukey test to determine if they showed different functional strategies, and compared each trait among groups with Kruskall-Wallis tests for continuous traits and X² on absolute frequencies for binary traits (Zar 1999). We also calculated the relative abundance of specialists and generalists per plot and performed a Kruskall-Wallis analysis to determine if generalists dominated across forest stand ages and topographic positions.

To analyze the turnover of dominating functional strategies over the successional and topographical gradients, we calculated community-weighted means (hereafter CWM) for each trait, defined as the mean of values present in the community (i.e. plot) weighed by the relative abundance of taxa bearing each value (Lavorel et al 2008), using the LD package (Laliberté et al 2014) in R software. We then performed a PCA on a plot x traits CWM matrix. Since we had seed volume data for only 55 species, we excluded this trait from both PCA analyses. To determine if there were significant differences in trait dominance (CWM) among successional age and topographic position classes, we performed non-parametric Kruskal-Wallis tests and paired comparisons among three vegetation classes (1) 3 to 15 years of secondary succession, (2) 16 to 24 years and (3) ≥ 25 years) and two topographic positions (flat areas and hills) in InfoStat software (Di Rienzo et al 2013). To evaluate how functional dominance of plant strategies varied along secondary succession, we used linear regression of the first two axes of the CWM PCA on successional age. Finally, we analyzed continuous variation of traits (CWM) with successional age by using heteroskedasticity-consistent standard error (HCSE) regressions (HC3 version) separately for plots on flat sites and on hills (online supplementary figure S1, available at stacks.iop.org/ERL/12/055004/mmedia), using the SPSS macro of Hayes and Cai (2007) in SPSS v. 16.0.

Our results show that patterns of plant functional variation in the landscape are driven mostly by generalist species as they dominate (in species number and relative abundance) across successional ages and topographic positions. This agrees with our first prediction and may be due to the long history of disturbance in this region (Rico-Gray and García-Franco 1992, Mizrahi et al 1997, Dupuy et al 2012b), and partly also due to the prevalence of earlier stages of forest succession –158 out of 211 plots (75%) were in young forests.

As described in previous studies for several plant communities (Wright et al 2004, Chave et al 2009, Reich 2014), and particularly in dry forests (Méndez-Alonzo et al 2012, Lohbeck et al 2013, Buzzard et al 2016), species lifestyles are distributed along axes of functional variation between acquisitive and conservative strategies of resource use. Specialists in this forest appear to be highly variable functionally, with the exception of secondary forest specialists, which were all bipinnate legumes. Reyes-García et al (2012) found that large legumes in a TDF in Yucatan had lower whole-tree water use than non-legume species due to higher allocation to non-conductive heartwood in legumes. Powers and Tiffin (2010) found in TDF of Costa Rica that, compared to leaf-habit functional types, legume tree species had a different suite of traits, including high wood density, leaf nitrogen and leaf carbon. In our study, legumes had higher WSG (mean ± SE: 0.784 ± 0.16) than non-legumes (0.694 ± 0.204), but the difference was not significant. Contrary to our second prediction, generalist species were not functionally closer to acquisitive or conservative strategies; instead, they displayed the entire range of variation between both extremes of the resource use continuum (figure 1). This is intriguing and its generality should be explored including forests with higher levels of water limitation, where generalists are expected to be conservative and drought resistant species.

Our results did not support our third prediction: that young forest stands and hills would share conservative functional strategies since they represent the driest habitats in the landscape. Patterns of trait CWM showed that, unlike other dry forest (Buzzard et al 2016), acquisitive species dominate in the youngest plots (figures 2–4). Although dominant species of young forests showed functional traits associated with an efficient use of water (such as large divided leaves with small leaflets, leaf pulvini or leaf deciduousness), they also showed traits (long petioles and large leaf area) associated with an increased efficiency and area for light harvest (Takenaka 1994, Niinemets et al 2004), as found previously in another Mexican TDF (Lohbeck et al 2013). We concur with Lohbeck et al (2013) who suggested that the number of months without leaves and root depth may explain the survival of species with cheap and large leaves in tropical dry forests (Pineda-García et al 2013).

In contrast, species that dominated on hills showed traits associated with a more water-limiting environment, such as the smallest leaves (both in terms of LA and MPU) with high LDMC, and the maximum WSG. This latter trait has been directly associated with high resistance to cavitation, low water storage capacity, and low hydraulic conductivity (Santiago et al 2004, Ishida et al 2008, Markesteijn et al 2011, Pineda-García et al 2013). These distinct strategies of species dominating on hills vs young plots suggest that conditions and resources differ between these portions of the landscape; microenvironmental studies are needed to assess this.

Late successional plots on flat sites also showed more conservative species with small perennial leaves of high LDMC. However, WSG was remarkably similar along succession on flat sites (figures 2 and S1), as also reported for TDF in Costa Rica (Becknell and Powers 2014), but contrary to the lower WSG found at older stands in other Mexican TDF (Lebrija-Trejos et al 2010, Lohbeck et al 2015). It is possible that, in our study area, a fine scale heterogeneity of water availability in the porous and soil-filled cavities in limestone bedrock may allow local niche partitioning of water (Querejeta et al 2007), and consequently the coexistence of low and high wood density species with different rooting strategies across succession, although this also requires further investigation.

Even though our results show that considering CWM some ecological strategies are more abundant in certain parts of the landscape, when considered individually most of the species abundances are not significantly different among successional ages or topographic positions. These results indicate that there are multiple ways to be a generalist, and suggest that environmental conditions across the successional and topographical gradient are not sufficiently different compared to the range of conditions in which generalist species can persist and become abundant. However, we cannot discard the possibility that the most influential factors affecting species filtering do not respond to spatial environmental gradients. Temporal variation of environmental resources and conditions could have a more important role determining plant reproduction and recruitment (i.e. the storage effect; Warner and Chesson 1985), which could result in the coexistence of distinctive functional strategies independently of successional age or topographical position.

Furthermore, according to local farmers, all 65 species included in our analyses are capable of sprouting after slash-and-burn agriculture. Sprouting is an important regeneration mechanism (Kammesheidt 1999), especially in a context of recurrent, but spatially patchy and low-intensity shifting cultivation, as practiced for millennia in our study area. It remains to be investigated if sprouting capacity is a key trait for being a generalist, since sprouts have a well-developed root system packed with reserves, which gives them a competitive advantage over seedlings (Miller and Kauffman 1998, Kennard et al 2002).

Finally, more sensitive statistical methods to determine habitat specialization for rare species are needed. It is possible that rare species could be highly sensitive to environmental heterogeneity, while generalist species may be largely tolerant and perhaps common for that reason (Pitman et al 2001). Although the multinomial method has been shown to be more sensitive than other methods in classifying specialist species, and also detects generalists statistically (Chazdon et al 2011), it was unable to classify rare species that represent over 50% of the species in our landscape.