In tropical dryland ecosystems, fire, rainfall, and soil composition determine proportional tree cover, which in turn differentiates among grassland, savanna, and forest biomes. Nearly 10% of the terrestrial tropics, however, experience the further disturbance that is periodic flooding. For many tree species, inundation threatens survival by creating anoxic soil conditions that kill roots, starving trees of water and nutrients. Nonetheless, while some tropical floodplains are grasslands largely devoid of trees (e.g., the Okavango Delta), others are savannas with intermediate tree cover (the Pantanal) or even forests (Amazonian varzea) with near-complete tree cover. One existing hypothesis to explain this variability is derived from conceptual theory on Amazon ecosystem function and posits that more temporally predictable disturbances will select more strongly for adaptation; consequently, trees are predicted to persist in regions with more predictable floods. Testing this possibility and finding the strong drivers of tree cover in tropical floodplains generally, will help predict how altered flood regimes will impact future biome distributions. We used GIS datasets, including a monthly inundation product spanning 15 years, to test the predictability hypothesis and how 11 hydrological, climatic, edaphic, and topographic variables affect local proportional tree cover in 13 tropical floodplains ecosystems on four continents.
Results/Conclusions
Random forest regression models offered only weak support for the predictability hypothesis, and primarily from the Amazon region. In the Llanos, tree cover declined with increasing predictability of annual flood duration and seasonality, and elsewhere there were weak or no correlations between tree cover and hydrological predictability.
Rather, annual climatic water balance (CWB; rainfall-evapotranspiration), elevation, soil sand content, and in some regions, fire frequency, were the strongest determinants of tree cover. As in drylands, CWB was generally positively associated with tree cover, suggesting its influence may override some flood-imposed constraints. Frequent fire reduced tree cover, and higher-elevation areas had more tree cover, possibly because they experienced shorter-term flooding (which was also negatively associated with tree cover).
Although we found little signature of flood regime predictability on an annual time scale, trees may adapt to predictability over longer time periods not captured by our inundation dataset. That floodplain tree cover is more strongly influenced by climate and soil than by even the best-performing hydrological predictor (mean annual flood duration) is striking given the oft-cited selective force that is inundation. Our models can help predict vegetation structure, biodiversity, and carbon stocks changes, as dams, climate change, and habitat conversion alter flood regimes.