Climatic and anthropogenic drivers of northern Amazon fires during the 2015-2016 El Niño event

Abstract

The strong El Niño Southern Oscillation (ENSO) event that occurred in 2015-2016 caused extreme drought in the northern Brazilian Amazon, especially in the state of Roraima, increasing fire occurrence. Here we map the extent of precipitation and fire anomalies and quantify the effects of climatic and anthropogenic drivers on fire occurrence during the 2015-2016 dry season (from December 2015 to March 2016) in the state of Roraima. To achieve these objectives we first estimated the spatial pattern of precipitation anomalies, based on long-term data from the TRMM (Tropical Rainfall Measuring Mission), and the fire anomaly, based on MODIS (Moderate Resolution Imaging Spectroradiometer) active fire detections during the referred period. Then, we integrated climatic and anthropogenic drivers in a Maximum Entropy (MaxEnt) model to quantify fire probability, assessing (1) the model accuracy during the 2015-2016 and the 2016-2017 dry seasons; (2) the relative importance of each predictor variable on the model predictive performance; and (3) the response curves, showing how each environmental variable affects the fire probability. Approximately 59% (132,900 km² ) of the study area was exposed to precipitation anomalies ≤-1 standard deviation (SD) in January and ~48% (~106,800 km² ) in March. About 38% (86,200 km² ) of the study area experienced fire anomalies ≥1 SD in at least one month between December 2015 and March 2016. The distance to roads and the direct ENSO effect on fire occurrence were the two most influential variables on model predictive performance. Despite the improvement of governmental actions of fire prevention and firefighting in Roraima since the last intense ENSO event (1997-1998), we show that fire still gets out of control in the state during extreme drought events. Our results indicate that if no prevention actions are undertaken, future road network expansion and a climate-induced increase in water stress will amplify fire occurrence in the northern Amazon, even in its humid dense forests. As an additional outcome of our analysis, we conclude that the model and the data we used may help to guide on-the-ground fire-prevention actions and firefighting planning and therefore minimize fire-related ecosystems degradation, economic losses and carbon emissions in Roraima.

Keywords: anthropogenic ignition; climate; fire modeling; hot pixels; machine learning; multivariate ENSO index; savannas; tropical forests.