In this dissertation, I evaluated the spatiotemporal dynamics of anthropogenic fire activity in a neotropical rainforest-savanna agricultural frontier. Given its fast and affordable nature, fire is used around the globe as a cost-effective way to clear and manage lands. This scenario is especially common in tropical regions experiencing high deforestation rates. The Amazon-Cerrado transition zone has been subject to the highest deforestation rates in Brazil over the last four decades. In this ecotone, fire is mainly used to clear natural vegetation lands and to manage encroachment of shrubs and trees in pasturelands; often, these fires spread accidentally. Nonetheless, the dynamics of the human-fire interaction are still not fully understood.
To assess this relationship at a fine-scale, I developed a semi-automatic burned area mapping algorithm in Google Earth Engine that applies spectral mixture analysis to time-series of Landsat imagery: Burned Area Spectral Mixture Analysis (BASMA). Using BASMA, I generated annual burned area maps for a 32-year time-series (1985 to 2017), testing whether spectral mixture analysis is a robust means for mapping fire scars that is stable over time and space. Results showed that BASMA successfully identified char fractions and delineated burned area in Landsat imagery for a 36 million hectares study region. Accuracy assessment performed against independent burned area products returned high Dice coefficients (0.86 on average) demonstrating that BASMA is an effective algorithm to map fire scars for a large extent over long time-series analysis.
The pyric transition, as proposed by Pyne (2001), suggests that human-driven fire activity increases at early stages of human occupation in a given region, and then decreases as the area is further industrialized. In my third chapter, I developed a conceptual model that combines the BASMA-derived burned area maps with a land use and land cover (LULC) database produced by the MapBiomas Project to evaluate the validity of the pyric transition hypothesis. Merging these two fine-scale datasets spanning a long time-series allowed me to quantitatively characterize spatiotemporal changes in fire activity occurring in parallel to human occupation dynamics. Two pyric phase transitions were observed: from ‘wildland anthropogenic fire’ to ‘agricultural anthropogenic fire’, and then to ‘fire suppression and wildfires’ phase.
In my final chapter, I evaluated spatiotemporal patterns of traditional burnings in remnants areas of the Cerrado biome within four indigenous lands, assessing whether fire frequency can be modeled by statistical models. Three probability distribution models were tested: continuous and discrete two-parameter Weibull and the discrete lognormal. Results agreed with previous studies, finding a mean fire interval of 3 years, similar to metrics estimated in other protected areas of the Cerrado. However, the parameters estimated for the probability distribution models showed that the study area does not have a homogeneous fire regime, indicating that further studies must be conducted to better quantify the fire return interval in these fire-prone ecosystems. Thus, we suggest that subdividing the fire frequency modelling by each of the indigenous lands could potentially return better results. Ultimately, this dissertation clearly demonstrates the advantages of having fine-scale burned area products covering long time-series over large extents.