We monitored a tropical forest for 20 years in order to evaluate the long-term dynamics of a moist forest outlier at the southern fringes of Amazonia and to explore influence of periods of prolonged drought on the woody vegetation. We observed four periods of drought during the 20 years of monitoring that are in agreement with the information reported by other authors . In the last two decades, Brazil has experienced severe large-scale drought events, during 2005 , 2010  and 2016 . Our results show, first of all, that this is intrinsically a very dynamic forest, with high rates of mortality and recruitment even in the site which wasn’t burned and even in those intervals without major droughts. These measurements thus extend and confirm the earlier conclusion drawn from sites further east of here that the southern borders of Amazonia contain some of the most dynamic of all tropical forests – these are truly ‘hyperdynamic’ in terms of stem turnover rates  with some of the fastest tree turnover rates recorded anywhere in the tropics.
We noted that in addition to having remarkably high background values of between 3 and 5% each year, mortality rates have also been increasing in the woody vegetation of our study site. These occurred in both the burned and unburned forests. An increase in tree mortality has long been observed generally in many tropical forests [14,62,63], including in Brazil . However, the ultimate drivers and proximal mechanisms responsible for this increasing tree mortality remain unknown . Tree mortality often represents a low magnitude disturbance to the forest structure, helping to trigger local processes of forest succession [65,66], promoting the opening of canopy space which enables increased recruitment [12,65,66]. The trend of increasing rates of mortality and recruitment over time has been more pronounced in periods of more prolonged droughts in the tropical region . However, in the site with forest fire, the availability of new space did not result in higher recruitment rates than mortality, which resulted in a net reduction in the number of stems. We can discount stochastic flooding events as a confounding driver of change here, since over the 20 years of study, we have not registered any flooding process in the sampling units, including those positioned near the watercourse, and did not observe any physical consequences of over-banking such as sediment deposition or erosion.
In addition to the long-term increase across out study landscape, our data suggest that mortality rates can also reflect interactions between and forest fire (‘cascade effect’). Thus, during periods of prolonged droughts , mortality itself increases the probability of subsequent tree death by opening up the canopy, with environmental changes within the forest favouring the occurrence of fires [31,32]. The opening in the canopy favours the entry of more light inside the forest, with leaves and other combustible material both drying and increasing in abundance [31,32]. This drought-fire interaction has been identified elsewhere as responsible for tree mortality in some tropical forests . In part because of this potential for positive feedback, researchers have focused on the impacts of extreme droughts and climate change on tropical forests . Climate change drivers are consistent with the overall increase in mortality rates of tropical forests in general, and especially with the specific patterns of increased mortality of large trees  and the fact that the floristic composition of extensive regions of tropical forests has slowly changed to favour those species which have greater resistance to drought [30,69]. If severe drought events continue to increase mortality rates in tropical forests a more open forest structure could result, which provides conditions for more frequent occurrence of forest fires.
In our forests, the stock of biomass (and therefore carbon, which in these forests is equivalent to AGB x 0.47  generally increased until the most severe drought and the burning of some of our plots (in 2010). A general increase in tropical forest biomass has also been reported widely (in Amazonia, Africa, and Southeast Asia [28,71,72]. Indeed the vegetation of the whole terrestrial surface has acted as a strong carbon sink in recent decades, with a substantial fraction of this sink probably located in the tropics, particularly in the Amazon . Overall, structurally intact tropical forests were responsible for half of global terrestrial carbon uptake between 1990 and 2007, so removing ~15% of anthropogenic CO2 emissions [21,71]. This widespread increase in tropical forest biomass is often interpreted as a response to the increase accumulation in atmospheric CO2 [73,74], which over long-time scales should favour an increase of productivity biomass in conserved tropical forests . In our site, the increase of biomass reflects the growth of trees already present in the area, since the number of stems did not differ notably over time, a pattern that appears to hold more generally across Amazonia too [14,70].
In Chapada dos Guimarães National Park, net biomass change during monitoring intervals was strongly positive early on, but from 2006, one year after the first major drought of the 21st century , and became negative by the time of the 2010 drought – a pattern remarkably consistent with the larger South American trend based on an independent dataset [14,15,28]. Although tropical forests remain carbon sinks, their capacity to absorb atmospheric CO2 appears to be declining . One of the causes of this reduction is climate extremes, which exert a strong effect on biomass when evaluated on a short time scale [75,76]. Indeed, the sensitivity of tropical forests to environmental changes, especially drought, has already been documented by observational data from permanent plots networks, flux towers, remote sensing and greenhouse gas measurements [28,77]. Thus, the influence of current climate change – as well of course as more direct human intervention – is significantly impacting the carbon balance of the tropical land surface.
Long periods of drought seem to reduce primary productivity, number of trees, biomass production and increase tree mortality, especially when forest fires occur. We observed that these characteristics began in 2006 and remained during the monitoring period of 2010 (year of the forest fire) and 2016. During these periods, prolonged drought events were also recorded [24–26]. Even in our plots that were not affected by the forest fire, we observed similar behaviors of reduction in biomass production, recruitment and increased mortality. The occurrence of fire in tropical forests is one of the consequences of periods of drought  and serves as a catalyst for the reduction of accumulated biomass and productivity, and increase mortality. During years of severe drought, forest fires in the Amazon are typically destructive, killing up to 64% of trees where they occur . Overall, drought periods can significantly affect the structure of the woody vegetation, and especially so when there are fires.
As well as impacts on biomass carbon balance in general, droughts in tropical and temperate forests frequently have greatest impact on larger trees . In our study, large trees experienced increasing mortality, and this is associated with prolonged drought events. The main hypotheses to explain the mortality of trees with drought events invoke either hydraulic failures or carbon ‘starvation’ [56,64]. Hydraulic failure risks increase in proportion to tree height and tree crown exposure to light and heating, so they are more intensely experienced by larger trees . The hydraulic and carbon balance risks may be associated - in response to the water deficit provided by drought events, and potentially therefore to avoid the risk of hydraulic failure, plants close stomata in their leaves, but in this process the tree may suffer from carbon deficiency and "starve to death" . Regardless, drought-driven mortality of large trees in forests initiates processes of local succession, with the opening of the canopy and the entrance of light [2,3,7], and so can increase the temperature of the local microclimate , resulting in an even more drought-sensitive environment more prone to forest fire.
Potentially cyclical fluctuations between periods of disturbance and forest reconstruction can trigger longer-term recurrent outcomes, as large trees tend to suffer more from drought than smaller trees . Long-term tropical forest records, including ours, are consistent with greater growth due to increased atmospheric CO2 and this fertilization-induced stimulated especially of the growth of larger trees. These, in turn, are precisely those which experience the highest mortality rates during periods of drought, and so a loss of biomass stock. Thus, even without the aggravating effects of fire, contemporary, 21st century tropical forests may be especially susceptible to prolonged drought periods, leading to deeper changes in the structure and functioning of these ecosystems than was previously the case.
Lastly, we note that long-term research is clearly crucial to understand and measure the effects of natural and anthropogenic changes on forests. However, these studies are particularly demanding, requiring persistence often across generations of academic careers, and a great deal of standardization and data collection in the field. The present study was no exception. It required 20 years of monitoring as well as considerable financial support and the involvement more than a dozen different graduate students, three master's dissertations and two doctoral theses. More sustained, long-term initiatives like these are urgently needed so that the effects of climate change on tropical forest formations can be assessed accurately.