Soils on exposed Sunda Shelf shaped biogeographic patterns in the equatorial forests of Southeast Asia

Our results provide evidence that the exposed sea-bed soils may have played a role as dispersal barrier in central Sundaland. The special character and potential impact on plant species distributions of the currently submerged soils of central Sundaland have been noted before (4, 12) but have not yet been taken seriously as a dispersal barrier. A soil dispersal barrier also makes sense in a historical perspective because, unlike the hypothesized savanna corridor, the soil conditions would have been present in central Sundaland whenever sea levels dropped enough to expose the current shelf area. During the Pleistocene, this was the case for hundreds of thousands of years (2), more than enough time to cause a detectable biogeographic signal. Based on our reinterpretation of recent data (26), a likely vegetation type for the former soils of the submerged Sunda Shelf would have been heath forest on well-drained sandy soils, with kerapah peat swamps on poorly drained interfluves, and with herb-dominated swamp vegetation for areas that experienced some degree of climatic seasonality to the north, e.g., in the vicinity of Natuna. The region of the Java sea is more likely to have borne seasonal climate vegetation based on the palynological record from the southern part of the Makassar Straits, which shows the presence of widespread grass-dominated vegetation throughout the last glacial period (marine isotope stages 2–4, 16–74 ka) with extensive burning, as shown by charcoal records, from ∼16 to 37 ka (27, 28). Heath forests and kerapah peats in the equatorial region between Sumatra and Borneo, herbaceous swamps in low-lying areas of the Natuna region, and seasonal climate vegetation in the Java Sea area could have acted as a dispersal barrier for plants and animals not adapted to such soils or vegetation types. Indeed, current heath and swamp forests are characterized by a limited mix of tree species of the surrounding forests, resulting in much reduced diversity and productivity compared with lowland forests on fine-textured and richer soils (22, 29).

On the other hand, our study provides no support for a continuous savanna corridor in central Sundaland, which is in accordance with most coupled vegetation–climate models (16–20), although the most recent historical vegetation reconstruction for this region (4) includes a savanna corridor in some models. However, this outcome could only be achieved by a priori forcing of a continuous corridor into these models (4). These outcomes do not mean that the savanna corridor did not exist because there is strong evidence for drier, savanna-like or seasonal conditions north and south of central Sundaland in the Pleistocene (10, 12–15, 30, 31). However, if it existed, it left no clear floristic evidence among current tree populations. One reason for this lack of evidence could be that the wetter climates that prevailed during interglacial periods have led to the disappearance of the drought-tolerant taxa from Borneo, Sumatra, and Peninsular Malaysia, thus erasing any floristic evidence for a savanna corridor in these areas. Another reason could be that, even if there was a continuous savanna corridor in central Sundaland, it probably occurred during conditions of maximum sea-level retreat. This situation existed for only short periods of time during the Pleistocene (2), making it unlikely that it is solely responsible for the observed biogeographic patterns in Sundaland.

Another interesting outcome of our analysis is the high floristic diversity of Borneo compared with other Sundaland areas, with Borneo harboring 10 of the 11 identified floristic clusters (5 endemic), Sumatra 5 (none endemic), and Peninsular Malaysia only 3 (1 endemic). Even when Sumatra and Peninsular Malaysia are combined, they still only harbor 6 of the 11 identified floristic clusters, 5 of which are shared with Borneo. This pattern, which has recently also been demonstrated for bird species (6), probably largely reflects the longevity of everwet climate forests on Borneo (4, 28). Glacial–interglacial forest expansions and contractions seem to have had the most negative impact on lowland forests of Sumatra and Peninsular Malaysia because, compared with their present day potential extent, their size was small during glacial periods (3, 4). This small forest size might have resulted in the disappearance of some unique lowland floristic associations. Additionally, for parts of Sumatra and especially Peninsular Malaysia, considerably drier conditions existed during glacial periods (10, 12–15), which might have negatively affected their lowland everwet forest types. Borneo, on the other hand, has had a relatively stable everwet climate at its core and in most of the lowland coastal areas with the exception of the south and northeast, which might explain why it maintained so many unique lowland forest types (4, 12, 15, 28, 32).

Another possibility explaining the low floristic diversity of lowland forests on Sumatra could be related to the underrepresentation of eastern Sumatran locations, which might have affected our results because this area is one of the largest lowland forest regions within Sundaland. Unfortunately, it is also one of the most deforested areas within Sundaland, which means that getting inventory data from undisturbed forests of this region is almost impossible. Although older inventory studies probably exist, they are difficult to access because they are published in local reports. Therefore, Sumatra might eventually turn out to harbor more floristic diversity than is detected by our study.

The presence of a savanna corridor in central Sundaland has been used as an argument for the rapid dispersal of early humans (between ca. 60,000 and 45,000 y ago) from mainland Southeast Asia and central Sundaland to Java and then onward to eastern Indonesia, Papua New Guinea, and Australia (12). However, the presence of swamps and heath forests in central Sundaland, as suggested by our study, would not favor this human-dispersal route because swamps and heath forests, aside from being hard to traverse, are generally low-productivity systems with limited wildlife and other edible products for hunter-gatherers (22). It would be more likely that humans used the coastal routes along Sundaland to reach Java and beyond (33, 34), especially because, during this period, large parts of central Sundaland would have been submerged by sea, leaving only a small land area in central Sundaland as a land connection (2).

Our study underlines the conservation importance of lowland forests because they contain floristic associations that are unique and among the most threatened in SoutheastAsia. In Borneo, only a fraction of the identified lowland forest types are protected (35). Moreover, they are mostly excluded from the Heart of Borneo program, which is especially focused on hill and montane forests. In Sumatra and Peninsular Malaysia, the ongoing loss of forest is critical (36, 37). In special need of conservation are kerapah swamps, which form one of the oldest plant communities in the region and can be followed back in time to the Oligocene. Throughout the region, there is considerable pressure to convert forests to oil palm, pulp, and other industrial plantations even though many deforested areas, especially in Sumatra, remain unused (38, 39). Additionally, existing lowland forest reserves, especially in Indonesia, provide no guarantee for good forest protection because remote-sensing studies have shown that serious forest degradation, fires, and even clearance are occurring within their boundaries (40, 41). Our results strongly underline the need to focus conservation priorities in Southeast Asia toward better protection of lowland forest systems before most of them have disappeared.

We used tree inventory data for 111 locations across Sundaland (Fig. 2 and Dataset S2), which were centered on an underlying grid of ∼10 × 10 km that corresponded to our climate and soil layers (see below). Each of our locations thus represents the pooled trees from plots that fell within such grid cells. Most of these locations contained many morphospecies, making direct comparisons between locations based on species-level identifications impossible. Therefore, we based our floristic analysis on genera only, which not only improves the identification accuracy but has also been shown to reflect species-level floristic patterns well (42). For each location, we ranked genera from high to low abundance and selected the 30 most abundant genera for the floristic analysis because selection of a fixed number of genera makes floristic comparisons depend less on sample size. In cases where more than 30 genera were selected because of equal genus abundances, we randomly excluded genera from the lowest abundance class until we had selected 30 genera. In some cases, there were fewer than 30 genera present in a location. We decided to retain these locations because they usually represented low-diversity forest types such as heath, peat swamp, and montane forests. Overall, 30 genera represented between 54.8% and 100% of individuals from each inventory [77.7 ± 11.3% (mean ± SD)]. We used relative abundances of genera (number of individuals divided by total number of individuals in each inventory) to make the locations comparable. The final matrix, containing all locations and the relative abundances of their genera, was used in a cluster analysis (data were log 10-transformed to reduce impact of abundant genera) using minimal-variance clustering.

To determine the characteristic genera for each cluster in the floristic analysis, we used an indicator method (43). This method calculates an indicator value (IV) for each genus in predefined clusters (like the clusters identified by a floristic analysis). Only genera that have a high mean abundance and are present in the majority of locations of a cluster (and have low abundance and frequency outside of that cluster) will score a high IV. To test whether the observed IV of a genus in a cluster was significantly higher than could be expected based on a random distribution of individuals over the locations, the observed IV was compared with 999 randomly generated IVs. These random IVs were generated by a reallocation procedure in which the number of individuals per genus was randomly reshuffled over the locations (43). If the observed IV of a genus in a cluster fell within the top 5% of the random IVs (sorted in decreasing order), it was considered to deviate significantly from the random distribution. Indicator taxa were mapped on the dendrogram (Fig. S2).

Drought and coarse-soil tolerance of genera were determined by sorting the 111 study locations by their annual rainfall (WorldClim, http://www.worldclim.org) and subsoil texture values taken from the Food and Agriculture Organization of the United Nations’ TERRASTAT: Global Land Resources GIS Models and Databases for Poverty and Food Insecurity Mapping (21) (Fig. S3). Drought tolerance per genus was calculated as the abundance in the 50% driest sites divided by total abundance over all sites. Similarly, coarse-soil tolerance per genus was calculated as the abundance in the 50% coarsest soil sites divided by total abundance over all sites. Therefore, the higher the value (scaled between 0 and 1), the more tolerant a genus is to drought or coarse soil (the whole list is in Dataset S1). We then tested (one-way ANOVA) whether the means of the drought and coarse-soil tolerance values of the indicator taxa differed significantly between floristic clusters that crossed the dispersal barrier versus those that did not cross. Some indicator taxa occurred in several clusters. In those cases, we used only the data (cross or not cross central Sundaland) from the cluster where it scored its highest IV.