Conference Agenda

The growing pressure on the ecosystems of the Amazon Biome highlights the need for continuous monitoring of vegetation cover, especially in sensitive areas such as Indigenous lands and Agrarian reform settlements. This study aims to identify and analyze forest change trajectories in the municipality of São Félix do Xingu, with an emphasis on the Trincheira Bacajá indigenous lands reservation during the period from 2018 to 2023. Two methodologies using Landsat time series were compared: the LandTrendr algorithm on the Google Earth Engine platform, and annual land cover maps from the MapBiomas project. Despite differences in sensitivity and detection criteria, both approaches revealed consistent spatial patterns of vegetation suppression, with over 91.8% overlap in identified disturbance areas. The comparison between the methodologies showed that, although they differ in sensitivity and detection criteria, both identify consistent patterns of forest suppression areas in the region. LandTrendr enables a more detailed analysis, while the MapBiomas-based approach is more straightforward. The high spatial convergence of the results reinforces the importance of integrating multiple approaches for forest monitoring in Indigenous territories and land reform settlements, contributing to the support of public policies aimed at conservation and environmental protection.

llegal gold mining in remote regions of the Brazilian Amazon poses great environmental and security threats. Developing detection systems for mining dredges is hampered by limited labeled data, often due to adverse weather. Despite the urgency, few studies have explored automated approaches using SAR imagery under low-data regimes. This work presents a automated pipeline for detecting mining dredges in ICEYE SAR images. The method begins by splitting full-scene GRD images into chips, classified using a Prototypical Network trained via Few-Shot Learning. A local range filter is then applied to enhance contrast, followed by threshold-based blob detection to localize potential dredges. The approach achieved recall rates between 79.7% and 93.2%, demonstrating competitive performance. Although it cannot yet distinguish dredges from other small vessels, its full automation and scalability make it suitable for large-scale monitoring of regions at risk of illegal mining.

The increase in built-up areas within metropolitan perimeters has far-reaching negative effects. The increase in buildings, suburbs, and roads leads to reductions in native environment areas, contributing to the phenomenon of urban heat-islands (UHIs). According to the World Bank (World Bank Group, 2023), the temperature of south-eastern Asian cities is on average 1.6 to 2.0 degrees warmer than their surroundings, with some cities reaching 5.9 degrees warmer than the surrounding areas. Extreme urban heat has negative consequences including reducing city GDP and increasing the demand for electric energy. Furthermore, it can lead to the accumulation of greenhouse gases in the local area and can cause heat-related health issues, which can even result in death (Environmental Protection Agency, 2008). Measuring the decrease of natural areas within cities is a complicated task which involves government monitoring and accurate mapping. This work aims to accurately detect the reduction in natural areas within the city of Rio de Janeiro using semantic segmentation machine learning techniques. The models used were able to detect vegetation and urban areas with over 90% accuracy.

This paper presents a multimodal deep learning framework for deforestation detection that integrates satellite-based deforestation alerts (DETER) with weather and atmospheric variables (WF). While we hypothesized that WF could provide complementary signals for short-term deforestation prediction, our experiments show that fusion models provide only isolated and modest gains, with no consistent improvement over DETER-only baselines. The WF-only model highlights structurally vulnerable regions but lacks precision in identifying which specific pixels will be deforested—an ability retained by the alert-based models. Our findings confirm the dominant role of satellite alerts for precise deforestation monitoring, with WF signals offering limited added value for operational systems.

Illegal mining poses significant environmental and socio-political challenges in the Brazilian Amazon, particularly within protected areas and indigenous territories. While optical remote sensing has been widely employed for detecting mining activities, its effectiveness is often hindered by persistent cloud cover. The aim of this paper is to investigate the use of C-band Synthetic Aperture Radar (SAR) imagery from Sentinel-1 combined with a lightweight convolutional neural network (CNN) to detect illegal mining sites under challenging atmospheric conditions. The model was trained on seven Sentinel-1 scenes from the Tapajós basin (Pará) and evaluated within the training region as well as on an independent test set from the Yanomami Indigenous Territory (Roraima), using reference annotations sourced from the Amazon Mining Watch project. A total of 2,394 labelled patches supported the supervised training. The CNN achieved balanced classification performance in the Tapajós area (F1-score: 0.676 at 0.80 threshold) and demonstrated the generalization capabilities in the unseen Yanomami region (F1-score: 0.630 at 0.90 threshold). Detection errors were mainly related to peripheral mining structures and small-scale disturbances, indicating challenges in identifying low-density mining patterns. These findings highlight the promise of SAR-based deep learning methods for monitoring illegal mining in cloud-prone Amazonian regions. Future work could improve detection accuracy by integrating terrain variables—such as elevation and proximity to watercourses—given the common occurrence of mining activities along narrow streams (igarapés) closely tied to local topography.

For nearly a decade, ALOS-2/PALSAR-2 has been monitoring the Amazon basin with a nominal revisit frequency of 42 days, achieving comprehensive coverage using its wide-swath (350 km) dual-polarization (HH+HV) ScanSAR mode. While the Long-Term Amazon Monitoring (LTAM) data archive provides amplitude data only due to processing and storage constraints, the coherent time-series data enables a range of forest applications with exceptional reliability. This paper highlights recent advancements in deforestation detection with the operational JJ-FAST early warning system, a powerful tool for mitigating tropical deforestation. Rigorous evaluation during the 2024 deforestation season confirmed JJ-FAST’s efficacy, with false alarm rates below 20% for deforestation sites larger than 10 hectares, establishing it as a scientifically validated and broadly applicable solution for reducing large-scale biomass loss in the Amazon. Leveraging ALOS-2 time-series forest/non-forest (TS-FNF) maps, we also track annual forest cover changes in the Brazilian Legal Amazon (BLA) offering improved accuracy compared to traditional optical remote sensing methods. Results indicate a net forest loss of approximately 2% in the BLA from 2018 to 2024. Furthermore, we explore the potential of ALOS-2 and ALOS-4 for monitoring forest regrowth, which is considered as one natural climate solution due to its CO2 sequestration capacity. These developments underscore the critical role of L-band SAR in supporting sustainable forest management and climate change mitigation in the Amazon rainforest.