Resilience of Amazonian forests : the roles of fire, flooding and climate

The Amazon has recently been portrayed as a resilient forest system based on quick recovery of biomass after human disturbance. Yet with climate change, the frequency of droughts and wildfires may increase, implying that parts of this massive forest may shift into a savanna state. Although the Amazon basin seems quite homogeneous, 14% is seasonally inundated. In my thesis I combine analyses of satellite data with field measurements and experiments to assess the role of floodplain ecosystems in shaping the resilience of Amazonian forests. First, I analyse tree cover distribution for the whole Amazon to reveal that savannas are relatively more common on floodplains. This suggests that compared to uplands, floodplains spend more time in the savanna state. Also, floodplain forests seem to have a tipping point at 1500 mm of annual rainfall in which forests may shift to savanna, whereas the tipping point for upland forests seems to be at 1000 mm of rainfall. Combining satellite and field measurements, I show that the higher frequency of savannas on floodplain ecosystems may be due to a higher sensitivity to fire. After a forest fire, floodplains lose more tree cover and soil fertility, and recover more slowly than uplands (chapter 2). In floodplains of the Negro river, I studied the recovery of blackwater forests after repeated fires, using field data on tree basal area, species richness, seed availability, and herbaceous cover. Results indicate that repeated fires may easily trap blackwater floodplains in an open-vegetation state, due the sudden loss of forest resilience after a second fire event (chapter 3). Analyses of the soil and tree composition of burnt floodplain forests, reveal that a first fire is the onset of the loss of soil fertility that intensifies while savanna trees dominate the tree community. A tree compositional shift happens within four decades, possibly accelerated by fast nutrient leaching. The rapid savannization of floodplain forests after fire implies that certain mechanisms such as environmental filtering may favor the recruitment of savanna trees over forest trees (chapter 4). In chapter 5, I experimentally tested in the field the roles of dispersal limitation, and environmental filtering for tree recruitment in burnt floodplain forests. I combine inventories of seed availability in burnt sites with experiments using planted seeds and seedlings of six floodplain tree species. Repeated fires strongly reduce the availability of tree seeds, yet planted trees thrive despite degraded soils and high herbaceous cover. Moreover, degraded soils on twice burnt sites seem to limit the growth of most pioneer trees, but not of savanna trees with deeper roots. Our results suggest a limitation of forest trees to disperse into open burnt sites. The combined evidence presented in this thesis support the hypothesis that Amazonian forests on floodplains are less resilient than forests on uplands, and more likely to shift into a savanna state. The lower ability of floodplains to retain soil fertility and recover forest structure after fire, may accelerate the transition to savanna. I also present some evidence of dispersal limitation of floodplain forest trees. Broad-scale analyses of tree cover as a function of rainfall suggest that savannas are likely to expand first in floodplains if Amazonian climate becomes drier. Savanna expansion through floodplain ecosystems to the core of the Amazon may spread fragility from an unsuspected place.