This study presents the most comprehensive and consistent reconstruction of wildfire history in the Sierras de Córdoba, Argentina, spanning the period 1986–2024. By integrating multiple satellite data sources (Landsat, Sentinel-2) and applying hybrid classification techniques in Google Earth Engine, a spatially explicit burned area dataset was developed. A total of 11,568 wildfire events were recorded,over the 39-year period studied, accumulating more than 2.6 million hectares (ha). While 83% of the fires were smaller than 100 ha, large fires (>1,000 ha) accounted for nearly 75% of the total burned area. Among these, megafires—defined as events exceeding 10,000 ha—represented only 0.35% of all fires but were responsible for over 43% of the total area burned. Fire activity showed strong seasonality: over 60% of the total burned area occurred between August and October, with September and October being the most critical months. This long-term dataset provides a critical tool for understanding fire regime dynamics, supporting land management, and assessing the ecological and socio-economic impacts of wildfires in the context of climate change and expanding Wildland-Urban interface.

Early detection and rapid response are critical in forest fire scenarios to minimize environmental and economic damage. This work explores the use of the YOLO (You Only Look Once) version 11 neural network for detecting three key classes: dispersed smoke, dense smoke, and fire. These classes were defined to support a tracking strategy aimed at identifying the fire source. In a typical forest fire, dispersed smoke is often visible from a distance, followed by dense smoke as one approaches the origin, and eventually, the flames themselves. A dataset composed of images from the Atlantic Forest was collected for training. The best model achieved an 85% mask accuracy with a recall of 76%. The network successfully segmented the target classes with few false positives, even on data not seen during training.

This study maps burned areas in Brasília National Park (BNP) and Chapada dos Veadeiros National Park (CVNP), including their 10 km buffer zones, from 2000 to 2020. A supervised classification was applied using the Random Forest algorithm in Google Earth Engine (GEE) with Landsat imagery and spectral indices. The combination of NDVI and NBR2 achieved the highest classification accuracy in 65% of the evaluated years compared to NDWI and NBR2. NDVI proved effective for detecting burned areas, especially for capturing structural changes in vegetation. Annual mosaics generated from the standard deviation of minimum and maximum pixel values improved the detection of spectral variability related to fire dynamics. The largest burned areas were observed in 2007 and 2010, with 310.37 km² and 433.45 km² in BNP, and 1,566.59 km² and 1,457.42 km² in CVNP. Spatial patterns differed between parks: fires were more frequent in the buffer zone of BNP, while in CVNP most fires occurred within park boundaries. Field validation in 2019 confirmed the accuracy of the mapped areas but also revealed limitations in detecting small burned scars. The historical data highlights burned area variation over time, supporting discussions on integrated fire management and strategies to reduce large wildfires.

This work presents a methodology for anticipating the emergence of fire events using hot spot products. It is based on the high temporal resolution and ultra-real-time availability of data from the ABI sensor of the geostationary satellites GOES-16 and GOES-19. The scope involves predicting the formation of events in Brazil and in the Amazon biome of the Amazon Cooperation Treaty Organization (ACTO) countries. It uses the K-Means algorithm for classifying clusters of recurrent alerts, formed by the spatio-temporal grouping of hot spot detections and considering the averages of: a) brightness temperature; b) estimated area; and c) radiative power. The data were processed through min-max normalization and the Euclidean distance from alerts to clusters was calculated. The differential of the approach lies in assigning a confidence estimate to each alert, indicating the probability that it anticipates the emergence of a fire event within up to 12 hours. The results obtained suggest that the methodology can contribute significantly to optimizing monitoring and directing actions, especially in remote regions, where early detection is crucial.

Biomass burning in South America contributes to regional aerosol loading, impacting air quality, climate, and public health. Accurate monitoring of AOD during fire events is critical for assessing these impacts. Satellite products such as MODIS-MAIAC offer reliable AOD retrievals but are limited to one or two observations per day. In contrast, the EPIC aboard the DSCOVR/NASA satellite provides near-hourly observations from the L1 Lagrange point, offering new opportunities to capture intra-daily aerosol variability. This study presents an analysis of EPIC-MAIAC AOD over southeastern South America during intense fire seasons in 2020 and 2021. An intercomparison with MODIS-MAIAC was performed. Results show higher values during spring (mean AOD = 0.2) and in known fire hotspots, though retrievals near water surfaces and coastal areas remain uncertain due to surface reflectance challenges. Standard deviation maps support the stability of EPIC retrievals, with low variability (standard deviation between 0 and 0.1) due to temporal averaging across multiple daily observations. Intra-daily analysis of AOD retrieved from EPIC instrument revealed peak values around midday. Time series analysis demonstrates overall agreement between EPIC-MAIAC and MODIS-MAIAC, with EPIC underestimating peak AOD values in some instances. This work highlights EPIC’s potential as a complementary tool for aerosol monitoring during fire events, particularly in capturing short-term fluctuations that are crucial for understanding aerosol transport mechanisms and supporting air quality assessments.