Dispersal limitation and fire feedback maintain mesic savannas in Madagascar: Reply

Savannas, characterized by a continuous flammable C4 grassy layer interspersed with woody vegetation, are widespread across the tropics, ranging from open grasslands to grassy woodlands (Oliveras & Malhi, 2016; Veldman et al., 2015). Despite their ubiquity, the origin and maintenance of savannas are fiercely debated. This is particularly true in the Central Highlands of Madagascar. Some argue that these grassy ecosystems can climatically support closed-canopy forests and, consequently, are likely a product of anthropogenic deforestation over the last two millennia (Humbert, 1949; Joseph et al., 2024). Others challenge the deforestation hypothesis (Bond et al., 2008, 2023; Phelps et al., 2022; Solofondranohatra et al., 2020; Vorontsova et al., 2016), arguing that paleoecological and phylogeographic data suggest that these high-rainfall grassy ecosystems might be ancient and predate human arrival. To reconcile the scientific evidence supporting these opposing viewpoints, we proposed an ecological mechanism—range pinning—to explain how the savannas in Madagascar could have been naturally maintained in mesic regions that can climatically support forests (Goel, Van Vleck, et al., 2020). In particular, we showed that the combination of dispersal limitation and fire-vegetation feedback could have restricted Eastern rainforests from expanding onto the Central Highlands and other mesic regions of Madagascar. Joseph and Seymour (2023) (hereafter referred to as J&S) criticized our paper (1) for making what they characterize as incorrect model assumptions, (2) for drawing inferences based on sparse paleoecological data, and (3) for citing papers critiqued by J&S elsewhere (Joseph & Seymour, 2021, 2022). Here, we address the first two concerns. Bond et al. (2023) have published a reply to address J&S's third concern, showing that most of their claims are not supported by accumulating published evidence. In Goel, Van Vleck, et al. (2020), we initialized our model simulations based on the present-day distribution of savanna and forest in Madagascar (Mayaux et al., 2000), in which forests are restricted to the eastern side of Madagascar. We used this initial condition because it roughly reflects the reconstructed biome patterns from the Holocene and the Last Glacial Maximum (LGM) (Adams & Faure, 1997a, 1997b). J&S critiqued this choice, arguing that the reconstructed LGM biome patterns from Adams and Faure (1997a, 1997b) show a mix of forest and savannas on the eastern side, with forest occupying the far northern part and savanna dominating in the south (Figure 1A). J&S argued that, since there is no dispersal barrier separating the two biomes at the outset of the simulation period, range pinning could not explain the distribution of savannas and forests in Madagascar. J&S further claimed that, since some forest patches occur west of the escarpment barrier, range pinning cannot explain mesic savannas on the Central Highlands. While J&S may be right that forests potentially occurred in small areas east of Madagascar's escarpment during the LGM (Adams & Faure, 1997a), this observation does not mean that range pinning played no role in maintaining the distribution of mesic savannas in Madagascar. To support this assertion, here we present simulations of savannas and forests in Madagascar using three initial conditions—reconstructed distribution of biomes during (a) LGM (14,000–22,000 14C years before present [BP]), (b) end of Youngers Dryas (10,000 14C years BP), and (c) mid-Holocene (5000–6000 14C years BP) (Adams, 2025) (Figure 1A–C). All three initial conditions predate extensive anthropogenic fire activity and land use change (Burney, 1999; Douglass et al., 2019). Our simulations consistently show that, for all initial conditions, forests fail to expand onto the Central Highlands (Figure 1E–G). These results are best interpreted within the context of global paleoclimate change since the LGM. At the LGM, the Earth's climate was relatively cold and dry, which considerably reduced the extent of forests in Madagascar, restricting them to the northeastern part of the island (Figure 1A). Following the LGM, as the climate became warmer and wetter, forests expanded globally (Adams & Faure, 1997a). However, in Madagascar, forest expansion was restricted to the eastern side, while the Central Highlands maintained open-canopy ecosystems even though these regions could climatically support forests (see Figure 3A in Goel, Van Vleck, et al. [2020]). Our simulations suggest that these open-canopy ecosystems could have been maintained by range pinning. Additionally, range pinning is also robust to small spatial perturbations, such as the establishment of forest patches in the Central Highlands. To show this, we simulated vegetation patterns in Madagascar, starting with the present-day biome distribution, including small forest patches in areas currently occupied by savanna (Figure 1D). Our simulations show that, due to critical patch size requirements (Goel, Guttal, et al., 2020; Lewis & Kareiva, 1993), small forest patches within a savanna matrix are insufficient to trigger a shift to a forest-dominated state, even though the landscape is climatically suited to support forests (Figure 1H). In Goel, Van Vleck, et al. (2020), our synthesis of paleoecological records from multiple sites across Madagascar showed that peaks in charcoal particles and Poaceae pollen pre-dated the onset of widespread human landscape transformation, typically placed around 2000 14C years BP (Burney, 1999) (see Figure 1A in Goel, Van Vleck, et al. [2020]), despite evidence of earlier human presence (Hansford et al., 2018). This suggested that fire- and grass-dominated ecosystems were already significant ecological components before intensive anthropogenic disturbance. J&S criticized our study for insufficient replication, arguing that only two of the 11 sites in our analysis were located in the Central Highlands and that the remaining nine sites from other phytogeographic regions cannot inform historical fire activity and its relationship with vegetation in the region. However, J&S's assertion misses a key point: our study aimed to mechanistically explain how mesic savannas persist in Madagascar despite climatically similar regions in the eastern lowlands supporting forests. Therefore, paleodata from other mesic phytogeographic regions within Madagascar (e.g., Northwestern Madagascar; see Figure 1B in Goel, Van Vleck, et al. [2020]) provide valuable context for understanding the interplay of climate and human activity in shaping the historical distribution of high-rainfall grassy biomes in Madagascar. Nevertheless, we agree that more paleodata are always better for understanding how climate and human activity have shaped vegetation in the Central Highlands. Recently published paleoecological records from Central Madagascar (Broothaerts et al., 2023; Razafimanantsoa et al., 2024) and previously published snapshots of sedimentary records (not included in our paper) (Samonds et al., 2019; Straka, 1996) suggest that vegetation patterns around these Central Highlands sites are broadly consistent with those inferred from paleoecological data considered in our paper (Gasse & Van Campo, 2001). We outline their main findings below. Data cited in our original paper from Gasse and Van Campo (2001) suggested that during the early Holocene, when the climate was wet and warm and human impact on vegetation was low or absent, the sedimentary record from Lake Tritrivakely indicates an increase in Poaceae pollen alongside tree components, including Celtis and Macaranga (Figure 2A). This suggests a dynamic mosaic of grassland and woodland rather than a closed forest landscape (Gasse & Van Campo, 2001). Despite the presence of woody components, multiple charcoal peaks (at 10,470 14C years BP; 8400 14C years BP; and 3800 14C years BP; Burney (1987)) indicate that fire was present in the landscape, with charcoal morphology reflecting grassy fuel type. A new record from Razafimanantsoa et al. (2024) documents a similar trend at Lake Dangovavy, where grass pollen (>30%) persisted from 6000 calibrated years BP onward (Figure 2B). Thus, paleo evidence supports the idea that fires were present in these grassy ecosystems long before the onset of rapid anthropogenic habitat transformation around 2000 14C years BP (Figure 2). While human activity likely has intensified fire regimes and landscape openness in Madagascar's Central Highlands over the last two millennia, assuming that Malagasy grassy ecosystems were entirely forested before human settlement contradicts both empirical and theoretical studies of savanna–forest coexistence. Instead, these grassy ecosystems exhibit resilience to forest expansion across millennia, perhaps in part due to range pinning. These ancient ecosystems are worth our attention and care. Nikunj Goel conducted biome simulations. Julie C. Aleman collated paleo evidence. All authors wrote the manuscript. This work was supported by NSF DMS-1615531. The authors declare no conflicts of interest. Data and code (Goel, 2025) are available in Zenodo at https://doi.org/10.5281/zenodo.16385547.