Studying fire propagation in the Amazonian jungle based on geospatial imagery analytics.

The Amazon rainforest is a critical reservoir of global biodiversity and a vital carbon sink that is currently facing growing threats from wildfires, intensified by climate change and human expansion (Malhi et al., 2008). Early identification of fire-prone regions is critical for ecological conservation and policy intervention (Bowman et al., 2009). This study uses geospatial imagery analytics to present a data-driven framework for fire risk prediction in the Brazilian Amazon. We developed a spatiotemporal dataset for predictive modeling that includes remote sensing data, vegetation indices such as NDVI and EVI, meteorological variables such as temperature, precipitation, and wind speed, and anthropogenic indicators such as land cover and human modification index. Our analysis ranges from 2022 to 2024 and focuses on the state of Pará, one of Brazil’s most fire-affected regions in recent years. We employed and tuned ensemble machine learning models (Random Forest, LightGBM, and XGBoost) and a deep learning approach (ConvLSTM) to assess fire risk across different modeling scenarios. The best-performing models achieved strong predictive accuracy in classifying fire and non-fire events. In addition to the binary classification task, we used the probabilistic outputs of the trained models to generate weekly dynamic fire risk maps for the year 2024. This study reveals important insights about fire dynamics across Pará. In fact, fire risk patterns are driven by complex interactions among vegetation stress, seasonal climatic patterns, and landscape modifications in the study region. The proposed approach supports real-time weekly fire forecasting at the grid level. It provides us with insights from which we can derive actions regarding high-risk hotspots and potential fire locations. This enables government agencies and environmental stakeholders to strengthen early warning systems, allocate resources efficiently, and take timely mitigation actions in the Amazon forest ecosystems.