Distribution Patterns of Burned Areas in the Brazilian Biomes: An Analysis Based on Satellite Data for the 2002–2010 Period

Relative fire-proneness of land cover types in the Brazilian Atlantic forest

Fires are increasingly affecting tropical biomes, where landscape-fire interactions remain understudied. We investigate the fire-proneness-the likelihood of a land use or land cover (LULC) type burning more or less than expected based on availability-in the Brazilian Atlantic Forest (AF). This biodiversity hotspot is increasingly affected by fires due to human activities and climate change. Using a selection ratio-based approach, we analyzed fire-LULC interactions in 40,869 fires over a 35-year period (1987-2022) across various ecoregions in the AF. Our findings revealed that secondary forests, forest areas that have regrown after major disturbances, burned 61% more than expected by chance, whereas old-growth forests, native forests that have developed over very long periods, burned 57% less than expected, highlighting a nearly inverse relationship in their fire-proneness. Interestingly, our data indicate that pastures in the AF are less prone to fire than expected, despite being considered among the land uses that burn the most in Brazil. Other LULCs showed variable fire-proneness, with some differences between ecoregions. Over time, the fire-proneness of secondary forests decreased, likely due to forest aging and changes in land management practices. We emphasize the necessity for tailored fire management strategies that address the unique vulnerabilities of secondary forests, particularly in the context of ongoing restoration efforts aimed at increasing native forests. Effective measures, including the implementation of 'fire-smart management' practices and enhancing the perceived value of secondary forests among local communities, are crucial for mitigating fire risks. Integrating these strategies with incentive-based approaches can bolster fire prevention, ensuring the long-term success of restoration programs. Our study provides a framework for understanding fire-landscape dynamics in tropical forests and offers actionable insights for practitioners working to safeguard these biomes from the escalating threat of wildfires.

Guiding seed movement: environmental heterogeneity drives genetic differentiation in Plathymenia reticulata, providing insights for restoration

Forest and landscape restoration is one of the main strategies for overcoming the environmental crisis. This activity is particularly relevant for biodiversity-rich areas threatened by deforestation, such as tropical forests. Efficient long-term restoration requires understanding the composition and genetic structure of native populations, as well as the factors that influence these genetic components. This is because these populations serve as the seed sources and, therefore, the gene reservoirs for areas under restoration. In the present study, we investigated the influence of environmental, climatic and spatial distance factors on the genetic patterns of Plathymenia reticulata, aiming to support seed translocation strategies for restoration areas. We collected plant samples from nine populations of P. reticulata in the state of Bahia, Brazil, located in areas of Atlantic Forest and Savanna, across four climatic types, and genotyped them using nine nuclear and three chloroplast microsatellite markers. The populations of P. reticulata evaluated generally showed low to moderate genotypic variability and low haplotypic diversity. The populations within the Savanna phytophysiognomy showed values above average for six of the eight evaluated genetic diversity parameters. Using this classification based on phytophysiognomy demonstrated a high predictive power for genetic differentiation in P. reticulata. Furthermore, the interplay of climate, soil and geographic distance influenced the spread of alleles across the landscape. Based on our findings, we propose seed translocation, taking into account the biome, with restricted use of seed sources acquired or collected from the same environment as the areas to be restored (Savanna or Atlantic Forest).

A soil productivity system reveals most Brazilian agricultural lands are below their maximum potential

Food production is extremely dependent on the soil. Brazil plays an important role in the global food production chain. Although only 30% of the total Brazilian agricultural areas are used for crop and livestock, the full soil production potential needs to be evaluated due to the environmental and legal impossibility to expand agriculture to new areas. A novel approach to assess the productive potential of soils, called "SoilPP" and based on soil analysis (0-100 cm) - which express its pedological information - and machine learning is presented. Historical yields of sugarcane and soybeans were analyzed, allowing to identify where it is still possible to improve crop yields. The soybean yields were below the estimated SoilPP in 46% of Brazilian counties and could be improved by proper management practices. For sugarcane, 38% of areas can be improved. This technique allowed us to understand and map the food yield situation over large areas, which can support farmers, consultants, industries, policymakers, and world food security planning.

Deforestation and fires in the Brazilian Amazon from 2001 to 2020: Impacts on rainfall variability and land surface temperature

Deforestation and fires in the Amazon are serious problems affecting climate, and land use and land cover (LULC) changes. In recent decades, the Amazon biome area has suffered constant fires and deforestation, causing severe environmental problems that considerably impact the land surface temperature (LST) and hydrological cycle. The Amazon biome lost a large forest area during this period. Thus, this study aims to analyze the deforestation and burned areas in the Amazon from 2001 to 2020, considering their impacts on rainfall variability and LST. This study used methods and procedures based on Google Earth Engine for analysis: (a) LULC evolution mapping, (b) vegetation cover change analysis using vegetation indices, (c) mapping of fires, (d) rainfall and LST analyses, and (e) analysis of climate influence and land cover on hydrological processes using the geographically weighted regression method. The results showed significant LULC changes and the main locations where fires occurred from 2001 to 2020. The years 2007 and 2010 had the most significant areas of fires in the Brazilian Amazon (233,401 km² and 247,562 km², respectively). The Pará and Mato Grosso states had the region's largest deforested areas (172,314 km² and 144,128 km², respectively). Deforestation accumulated in the 2016-2020 period is the greatest in the period analyzed (254,465 km²), 92% higher than in the 2005-2010 period and 82% higher than in the 2001-2005 period. The study also showed that deforested areas have been increasing in recent decades, and the precipitation decreased, while an increase is observed in the LST. It was also concluded that indigenous protection areas have suffered from anthropic actions.

Temporal and spatial patterns of fire activity in three biomes of Brazil

The trade-off between conservation of natural resources and agribusiness expansion is a constant challenge in Brazil. The fires used to promote agricultural expansion increased in the last decades. While studies linking annual fire occurrence and rainfall seasonality are common, the relationship between fires, land use, and land cover remains understudied. Here, we investigated the frequency of the fires and performed a trend analysis for monthly, seasonal, and annual fires in three different biomes: Cerrado, Pantanal, and Atlantic Forest. We used burned area and integrated models in distinct scales (interannual, intraseasonal, and monthly) using Probability Density Functions (PDFs). The best fitting was found for Generalized Extreme Values (GEV) distribution at all three biomes from the several PDFs tested. We found the most fire in the Pantanal (wetlands), followed by Cerrado (Brazilian Savanna) and Atlantic Forest (Semideciduous Forest). Our findings indicated that land use and land cover trends changed over the years. There was a strong correlation between fire and agricultural areas, with increasing trends pointing to land conversion to agricultural areas in all biomes. The high probability of fire indicates that expanding agricultural areas through the conversion of natural biomes impacts several natural ecosystems, transforming land cover and land use. This land conversion is promoting more fires each year.

Long-Term Landsat-Based Monthly Burned Area Dataset for the Brazilian Biomes Using Deep Learning

Fire is a significant agent of landscape transformation on Earth, and a dynamic and ephemeral process that is challenging to map. Difficulties include the seasonality of native vegetation in areas affected by fire, the high levels of spectral heterogeneity due to the spatial and temporal variability of the burned areas, distinct persistence of the fire signal, increase in cloud and smoke cover surrounding burned areas, and difficulty in detecting understory fire signals. To produce a large-scale time-series of burned area, a robust number of observations and a more efficient sampling strategy is needed. In order to overcome these challenges, we used a novel strategy based on a machine-learning algorithm to map monthly burned areas from 1985 to 2020 using Landsat-based annual quality mosaics retrieved from minimum NBR values. The annual mosaics integrated year-round observations of burned and unburned spectral data (i.e., RED, NIR, SWIR-1, and SWIR-2), and used them to train a Deep Neural Network model, which resulted in annual maps of areas burned by land use type for all six Brazilian biomes. The annual dataset was used to retrieve the frequency of the burned area, while the date on which the minimum NBR was captured in a year, was used to reconstruct 36 years of monthly burned area. Results of this effort indicated that 19.6% (1.6 million km2) of the Brazilian territory was burned from 1985 to 2020, with 61% of this area burned at least once. Most of the burning (83%) occurred between July and October. The Amazon and Cerrado, together, accounted for 85% of the area burned at least once in Brazil. Native vegetation was the land cover most affected by fire, representing 65% of the burned area, while the remaining 35% burned in areas dominated by anthropogenic land uses, mainly pasture. This novel dataset is crucial for understanding the spatial and long-term temporal dynamics of fire regimes that are fundamental for designing appropriate public policies for reducing and controlling fires in Brazil.

Effects of Drought on Mortality in Macro Urban Areas of Brazil Between 2000 and 2019

A significant fraction of Brazil's population has been exposed to drought in recent years, a situation that is expected to worsen in frequency and intensity due to climate change. This constitutes a current key environmental health concern, especially in densely urban areas such as several big cities and suburbs. For the first time, a comprehensive assessment of the short-term drought effects on weekly non-external, circulatory, and respiratory mortality was conducted in 13 major Brazilian macro-urban areas across 2000-2019. We applied quasi-Poisson regression models adjusted by temperature to explore the association between drought (defined by the Standardized Precipitation-Evapotranspiration Index) and the different mortality causes by location, sex, and age groups. We next conducted multivariate meta-analytical models separated by cause and population groups to pool individual estimates. Impact measures were expressed as the attributable fractions among the exposed population, from the relative risks (RRs). Overall, a positive association between drought exposure and mortality was evidenced in the total population, with RRs varying from 1.003 [95% CI: 0.999-1.007] to 1.010 [0.996-1.025] for non-external mortality related to moderate and extreme drought conditions, from 1.002 [0.997-1.007] to 1.008 [0.991-1.026] for circulatory mortality, and from 1.004 [0.995-1.013] to 1.013 [0.983-1.044] for respiratory mortality. Females, children, and the elderly population were the most affected groups, for whom a robust positive association was found. The study also revealed high heterogeneity between locations. We suggest that policies and action plans should pay special attention to vulnerable populations to promote efficient measures to reduce vulnerability and risks associated with droughts.

Putting fire on the map of Brazilian savanna ecoregions

The Brazilian savanna (Cerrado) is considered the most floristically diverse savanna in the world, home to more than seven thousand species. The region is a mosaic of savannas, grasslands and forests whose unique biophysical and landscape attributes are on the basis of a recent ecoregional map, paving the way to improved region-based strategies for land management actions. However, as a fire-prone ecosystem, Cerrado owes much of its distribution and ecological properties to the fire regime and contributes to an important parcel of South America burned area. Accordingly, any attempt to use ecoregion geography as a guide for management strategies should take fire into account, as an essential variable. The main aim of this study is to complement the ecoregional map of the Cerrado with information related to the fire component. Using remotely sensed information, we identify patterns and trends of fire frequency, intensity, seasonality, extent and scar size, and combine this information for each ecoregion, relying on a simple classification that summarizes the main fire characteristics over the last two decades. Results show a marked north-south fire activity gradient, with increased contributions from MATOPIBA, the latest agricultural frontier. Five ecoregions alone account for two thirds of yearly burned area. More intense fires are found in the Arc of Deforestation and eastern ecoregions, while ecoregions in MATOPIBA display decreasing fire intensity. An innovative analysis of fire scars stratified by size class shows that infrequent large fires are responsible for the majority of burned area. These large fires display positive trends over many ecoregions, whereas smaller fires, albeit more frequent, have been decreasing in number. The final fire classification scheme shows well defined spatially-aggregated groups, where trends are found to be the key factor to evaluate fire within their regional contexts. Results presented here provide new insights to improve fire management strategies under a changing climate.

Observations of Emissions and the Influence of Meteorological Conditions during Wildfires: A Case Study in the USA, Brazil, and Australia during the 2018/19 Period

Wildfires can have rapid and long-term effects on air quality, human health, climate change, and the environment. Smoke from large wildfires can travel long distances and have a harmful effect on human health, the environment, and climate in other areas. More recently, in 2018–2019 there have been many large fires. This study focused on the wildfires that occurred in the United States of America (USA), Brazil, and Australia using Cloud-Aerosol Lidar with Orthogonal Polarisation (CALIOP) and a TROPOspheric Monitoring Instrument (TROPOMI). Specifically, we analyzed the spatial-temporal distribution of black carbon (BC) and carbon monoxide (CO) and the vertical distribution of smoke. Based on the results, the highest detection of smoke (~14 km) was observed in Brazil; meanwhile, Australia showed the largest BC column burden of ~1.5 mg/m2. The meteorological conditions were similar for all sites during the fires. Moderate temperatures (between 32 and 42 °C) and relative humidity (30–50%) were observed, which resulted in drier conditions favorable for the burning of fires. However, the number of active fires was different for each site, with Brazil having 13 times more active fires than the USA and five times more than the number of active fires in Australia. However, the high number of active fires did not translate to higher atmospheric constituent emissions. Overall, this work provides a better understanding of wildfire behavior and the role of meteorological conditions in emissions at various sites.

Persistent fire foci in all biomes undermine the Paris Agreement in Brazil

Brazil is one of the world’s biggest emitters of greenhouse gases (GHGs). Fire foci across the country contributes to these emissions and compromises emission reduction targets pledged by Brazil under the Paris Agreement. In this paper, we quantify fire foci, burned areas, and carbon emissions in all Brazilian biomes (i.e., Amazon, Cerrado, Caatinga, Atlantic Forest, Pantanal and Pampa). We analyzed these variables using cluster analysis and non-parametric statistics to predict carbon and CO₂ emissions for the next decade. Our results showed no increase in the number of fire foci and carbon emissions for the evaluated time series, whereby the highest emissions occur and will persist in the Amazon and Cerrado biomes. The Atlantic Forest, Pantanal, Caatinga and Pampa biomes had low emissions compared to the Amazon and Cerrado. Based on 2030 projections, the sum of emissions from fire foci in the six Brazilian biomes will exceed 5.7 Gt CO2, compromising the national GHG reduction targets. To reduce GHG emissions, Brazil will need to control deforestation induced by the expansion of the agricultural frontier in the Amazon and Cerrado biomes. This can only be achieved through significant political effort involving the government, entrepreneurs and society as a collective.

Performance Analysis of Deep Convolutional Autoencoders with Different Patch Sizes for Change Detection from Burnt Areas

Fire is one of the primary sources of damages to natural environments globally. Estimates show that approximately 4 million km2 of land burns yearly. Studies have shown that such estimates often underestimate the real extent of burnt land, which highlights the need to find better, state-of-the-art methods to detect and classify these areas. This study aimed to analyze the use of deep convolutional Autoencoders in the classification of burnt areas, considering different sample patch sizes. A simple Autoencoder and the U-Net and ResUnet architectures were evaluated. We collected Landsat 8 OLI+ data from three scenes in four consecutive dates to detect the changes specifically in the form of burnt land. The data were sampled according to four different sampling strategies to evaluate possible performance changes related to sampling window sizes. The training stage used two scenes, while the validation stage used the remaining scene. The ground truth change mask was created using the Normalized Burn Ratio (NBR) spectral index through a thresholding approach. The classifications were evaluated according to the F1 index, Kappa index, and mean Intersection over Union (mIoU) value. Results have shown that the U-Net and ResUnet architectures offered the best classifications with average F1, Kappa, and mIoU values of approximately 0.96, representing excellent classification results. We have also verified that a sampling window size of 256 by 256 pixels offered the best results.

Assessment of the Characteristics of Recent Major Wildfires in the USA, Australia and Brazil in 2018–2019 Using Multi-Source Satellite Products

This study analysed the characteristics of the recent (2018–2019) wildfires that occurred in the USA, Brazil, and Australia using Moderate Resolution Imaging Spectroradiometer (MODIS) active fires (AF), fire radiative power (FRP, MW) and burned area (BA) products. Meteorological and environmental parameters were also analysed. The study found various patterns in the spatial distribution of fires, FRP and BA at the three sites, associated with various vegetation compositions, prevailing meteorological and environmental conditions and anthropogenic activities. We found significant fire clusters along the western and eastern coasts of the USA and Australia, respectively, while vastly distributed clusters were found in Brazil. Across all sites, significant fire intensity was recorded over forest cover (FC) and shrublands (SL), attributed to highly combustible tree crown fuel load characterised by leafy canopies and thin branches. In agreement, BA over FC was the highest in the USA and Australia, while Brazil was dominated by the burning of SL, characteristic of fire-tolerant Cerrado. The relatively lower BA over FC in Brazil can be attributed to fuel availability and proximity to highly flammable cover types such as cropland, SL and grasslands rather than fuel flammability. Overall, this study contributes to a better understanding of wildfires in various regions and the underlying environmental and meteorological causal factors, towards better wildfire disaster management strategies and habitat-specific firefighting.

Respiratory Diseases, Malaria and Leishmaniasis: Temporal and Spatial Association with Fire Occurrences from Knowledge Discovery and Data Mining

The relationship between the fires occurrences and diseases is an essential issue for making public health policy and environment protecting strategy. Thanks to the Internet, today, we have a huge amount of health data and fire occurrence reports at our disposal. The challenge, therefore, is how to deal with 4 Vs (volume, variety, velocity and veracity) associated with these data. To overcome this problem, in this paper, we propose a method that combines techniques based on Data Mining and Knowledge Discovery from Databases (KDD) to discover spatial and temporal association between diseases and the fire occurrences. Here, the case study was addressed to Malaria, Leishmaniasis and respiratory diseases in Brazil. Instead of losing a lot of time verifying the consistency of the database, the proposed method uses Decision Tree, a machine learning-based supervised classification, to perform a fast management and extract only relevant and strategic information, with the knowledge of how reliable the database is. Namely, States, Biomes and period of the year (months) with the highest rate of fires could be identified with great success rates and in few seconds. Then, the K-means, an unsupervised learning algorithms that solves the well-known clustering problem, is employed to identify the groups of cities where the fire occurrences is more expressive. Finally, the steps associated with KDD is perfomed to extract useful information from mined data. In that case, Spearman's rank correlation coefficient, a nonparametric measure of rank correlation, is computed to infer the statistical dependence between fire occurrences and those diseases. Moreover, maps are also generated to represent the distribution of the mined data. From the results, it was possible to identify that each region showed a susceptible behaviour to some disease as well as some degree of correlation with fire outbreak, mainly in the drought period.

Characterization and Trends of Fine Particulate Matter (PM2.5) Fire Emissions in the Brazilian Cerrado during 2002–2017

Fire occurrence is a major disturbance in the Brazilian Cerrado, which is driven by both natural and anthropogenic activities. Despite increasing efforts for monitoring the Cerrado, a biome-scale study for quantifying and understanding the variability of fire emissions is still needed. We aimed at characterizing and finding trends in Particulate Matter with diameter less than 2.5 µm (PM2.5) fire emissions in the Brazilian Cerrado using the PREP-CHEM-SRC emissions preprocessing tool and Moderate Resolution Imaging Spectroradiometer (MODIS) active fires datasets for the 2002–2017 period. Our results showed that, on average, the Cerrado emitted 1.08 Tg year−1 of PM2.5 associated with fires, accounting for 25% and 15% of the PM2.5 fire emissions in Brazil and South America, respectively. Most of the PM2.5 fire emissions were concentrated in the end of the dry season (August, 0.224 Tg month−1 and September, 0.386 Tg month−1) and in the transitional month (October, 0.210 Tg month−1). Annually, 66% of the total emissions occurred over the savanna land cover; however, active fires that were detected in the evergreen broadleaf land cover tended to emit more than active fires occurring in the savanna land cover. Spatially, each 0.1° grid cell emitted, on average, 0.5 Mg km−2 year−1 of PM2.5 associated with fires, but the values can reach to 16.6 Mg km−2 year−1 in a single cell. Higher estimates of PM2.5 emissions associated with fires were mostly concentrated in the northern region, which is the current agricultural expansion frontier in this biome. When considering the entire Cerrado, we found an annual decreasing trend representing -1.78% of the annual average PM2.5 emitted from fires during the period analyzed, however, the grid cell analysis found annual trends representing ± 35% of the annual average PM2.5 fire emissions.

Spatiotemporal Patterns and Phenology of Tropical Vegetation Solar-Induced Chlorophyll Fluorescence across Brazilian Biomes Using Satellite Observations

Solar-induced fluorescence (SIF) has been empirically linked to gross primary productivity (GPP) in multiple ecosystems and is thus a promising tool to address the current uncertainties in carbon fluxes at ecosystem to continental scales. However, studies utilizing satellite-measured SIF in South America have concentrated on the Amazonian tropical forest, while SIF in other regions and vegetation classes remain uninvestigated. We examined three years of Orbiting Carbon Observatory-2 (OCO-2) SIF data for vegetation classes within and across the six Brazilian biomes (Amazon, Atlantic Forest, Caatinga, Cerrado, Pampa, and Pantanal) to answer the following: (1) how does satellite-measured SIF differ? (2) What is the relationship (strength and direction) of satellite-measured SIF with canopy temperature (Tcan), air temperature (Tair), and vapor pressure deficit (VPD)? (3) How does the phenology of satellite-measured SIF (duration and amplitude of seasonal integrated SIF) compare? Our analysis shows that OCO-2 captures a significantly higher mean SIF with lower variability in the Amazon and lower mean SIF with higher variability in the Caatinga compared to other biomes. OCO-2 also distinguishes the mean SIF of vegetation types within biomes, showing that evergreen broadleaf (EBF) mean SIF is significantly higher than other vegetation classes (deciduous broadleaf (DBF), grassland (GRA), savannas (SAV), and woody savannas (WSAV)) in all biomes. We show that the strengths and directions of correlations of OCO-2 mean SIF to Tcan, Tair, and VPD largely cluster by biome: negative in the Caatinga and Cerrado, positive in the Pampa, and no correlations were found in the Pantanal, while results were mixed for the Amazon and Atlantic Forest. We found mean SIF most strongly correlated with VPD in most vegetation classes in most biomes, followed by Tcan. Seasonality from time series analysis reveals that OCO-2 SIF measurements capture important differences in the seasonal timing of SIF for different classes, details masked when only examining mean SIF differences. We found that OCO-2 captured the highest base integrated SIF and lowest seasonal pulse integrated SIF in the Amazon for all vegetation classes, indicating continuous photosynthetic activity in the Amazon exceeds other biomes, but with small seasonal increases. Surprisingly, Pantanal EBF SIF had the highest total integrated SIF of all classes in all biomes due to a large seasonal pulse. Additionally, the length of seasons only accounts for about 30% of variability in total integrated SIF; thus, integrated SIF is likely captures differences in photosynthetic activity separate from structural differences. Our results show that satellite measurements of SIF can distinguish important functioning and phenological differences in vegetation classes and thus has the potential to improve our understanding of productivity and seasonality in the tropics.

Deforestation and forest fires in Roraima and their relationship with phytoclimatic regions in the northern Brazilian Amazon

Deforestation and forest fires in the Brazilian Amazon are a regional-scale anthropogenic process related to biomass burning, which has a direct impact on global warming due to greenhouse gas emissions. Containment of this process requires characterizing its spatial distribution and that of the environmental factors related to its occurrence. The aim of this study is to investigate the spatial and temporal distribution of deforested areas and forest fires in the State of Roraima from 2000 to 2010. We mapped deforested areas and forest fires using Landsat images and associated their occurrence with two phytoclimatic zones: zone with savanna influence (ZIS), and zone without savanna influence (ZOS). Total deforested area during the interval was estimated at 3.06 × 10(3) km(2) (ZIS = 55 %; ZOS = 45 %) while total area affected by forest fires was estimated at 3.02 × 10(3) km(2) (ZIS = 97.7 %; ZOS = 2.3 %). Magnitude of deforestation in Roraima was not related to the phytoclimatic zones, but small deforested areas (≤17.9 ha) predominated in ZOS while larger deforestation classes (>17.9 ha) predominated in ZIS, which is an area with a longer history of human activities. The largest occurrence of forest fires was observed in the ZIS in years with El Niño events. Our analysis indicates that the areas most affected by forest fires in Roraima during 2000-2010 were associated with strong climatic events and the occurrence these fires was amplified in ZIS, a sensitive phytoclimatic zone with a higher risk of anthropogenic fires given its drier climate and open forest structure.

Early recruitment responses to interactions between frequent fires, nutrients, and herbivory in the southern Amazon

Understanding tropical forest diversity is a long-standing challenge in ecology. With global change, it has become increasingly important to understand how anthropogenic and natural factors interact to determine diversity. Anthropogenic increases in fire frequency are among the global change variables affecting forest diversity and functioning, and seasonally dry forest of the southern Amazon is among the ecosystems most affected by such pressures. Studying how fire will impact forests in this region is therefore important for understanding ecosystem functioning and for designing effective conservation action. We report the results of an experiment in which we manipulated fire, nutrient availability, and herbivory. We measured the effects of these interacting factors on the regenerative capacity of the ecotone between humid Amazon forest and Brazilian savanna. Regeneration density, diversity, and community composition were severely altered by fire. Additions of P and N + P reduced losses of density and richness in the first year post-fire. Herbivory was most important just after germination. Diversity was positively correlated with herbivory in unburned forest, likely because fire reduced the number of reproductive individuals. This contrasts with earlier results from the same study system in which herbivory was related to increased diversity after fire. We documented a significant effect of fire frequency; diversity in triennially burned forest was more similar to that in unburned than in annually burned forest, and the community composition of triennially burned forest was intermediate between unburned and annually burned areas. Preventing frequent fires will therefore help reduce losses in diversity in the southern Amazon's matrix of human-altered landscapes.