Valdete Duarte’s research while affiliated with National Institute for Space Research and other places

Deforestation and fire play a critical role in shaping the Brazilian Amazon, yet accurate automated long-term monitoring remains a challenge. This article presents a method for monitoring deforestation (1980–2022) and burned areas (2001–2022) in the state of Rondônia, a deforestation hotspot region in the Brazilian Amazon. We applied the Linear Spectral Mixing Model (LSMM) to Landsat (MSS, TM, and OLI) datasets to derive vegetation, soil, and shade fraction images, effectively reducing data volume while highlighting relevant target features. A threshold-based classification method was then applied to produce annual classification maps, showing forest, non-forest, deforestation, and water bodies. Burned areas were obtained from the MODIS MCD64A1 product. Our deforestation estimates (2008–2022) showed strong agreement with PRODES (Project for Monitoring Deforestation in the Legal Amazon) and GFC (Global Forest Change), achieving an overall accuracy of 89% compared to PRODES. The results reveal a significant decline in forest cover: from 86% in 1986 to 70% in 2000, and just 52% in 2022. Considering the use of fire during the deforestation process, fire activity was also extensive, with approximately 4.5 million hectares burned during the study period—2.9 million hectares within forested areas, of which only 894 thousand hectares remained as forest after burning. These findings provide critical insights into land-use dynamics and fire-related forest loss in the Amazon, offering valuable support for conservation policies and climate change mitigation strategies.

This work aims to develop a new method to map Land Use and Land Cover (LULC) classes in the São Paulo State, Brazil, using Landsat-8 Operational Land Imager (OLI) data. The novelty of the proposed method consists of selecting the images based on the spectral and temporal characteristics of the LULC classes. First, we defined the six classes to be mapped in the year 2020 as forest, forest plantation, water bodies, urban areas, agriculture, and pasture. Second, we visually analyzed their variability spectral characteristics over the year. Then, we pre-processed these images to highlight each LULC class. For the classification, the Random Forest algorithm available on the Google Earth Engine (GEE) platform was utilized individually for each LULC class. Afterward, we integrated the classified maps to create the final LULC map. The results revealed that forest areas are primarily concentrated in the eastern region of São Paulo, predominantly on steeper slopes, accounting for 19% of the study area. On the other hand, pasture and agriculture dominated 73% of all São Paulo’s landscape, reaching 39% and 34%, respectively. The overall accuracy of the classification achieved 89.10%, while producer and user accuracies were greater than 84.20% and 76.62%, respectively. To validate the results, we compared our findings with the MapBiomas Project classification, obtaining an overall accuracy of 85.47%. Therefore, our method demonstrates its potential to minimize classification errors and offers the advantage of facilitating post-classification editing for individual mapped classes.

Este estudo busca identificar estádios fenológicos de plantios de Eucalipto por meio de ferramentas de Sensoriamento Remoto orbital. Este artigo utiliza como método proposto o Modelo Linear de Mistura Espectral para identificar ciclos fenológicos em plantios florestais de Eucalipto. Foram utilizados mosaicos mensais de imagens Landsat-8 (OLI) no período entre maio de 2013 a abril de 2014 para verificar o comportamento dos estágios fenológicos da espécie de acordo com o ciclo de rotação. O método proposto permite identificar por meio das imagens fração os períodos de plantio e colheita em períodos de mudanças. Os resultados são importantes para subsidiar futuros estudos que levem em consideração estágios fenológicos em florestas plantadas em diferentes escalas. Palavras-chave-modelo linear de mistura espectral, imagem fração, eucalipto, plantio florestal, processamento de imagem. ABSTRACT This study aims to identify phenological stages of Eucalypt plantations using orbital Remote Sensing tools. This article uses as a proposed method the Linear Spectral Mixture Model to identify phenological cycles in Eucalypt Forest plantations. Monthly mosaics of Landsat-8 images (OLI) were used in the period between May 2013 and April 2014 to verify the behavior of the phenological stages of the species according to the rotation cycle. The proposed method makes it possible to identify through the fraction images the planting and harvesting periods. The results are important to support future studies that take into account phenological stages in planted forests at different scales.

RESUMO O projeto PRODES realiza o monitoramento do desmatamento por corte raso na Amazônia Legal desde 1988, disponibilizando inicialmente somente informações na forma de tabelas, sem os mapas da distribuição espacial como atualmente. Dessa maneira, o objetivo deste trabalho é apresentar um método para gerar a distribuição espacial das áreas mapeadas anualmente pelo projeto PRODES para o estado de Rondônia. Nesta descrição do método, a série temporal anual foi agregada a cada 5 anos. Baseado no código de legenda do mapa PRODES de 2021 e nos resultados obtidos pelo método proposto, foi possível recuperar os dados para os anos de 1988 a 2021 referentes ao desmatamento acumulado desde o início do projeto. Além do desmatamento, as classes mapeadas são de hidrografia, não floresta e floresta. As áreas de hidrografia e não-floresta foram mantidas e foram recuperadas as áreas de desmatamentos para os períodos analisados. Dessa maneira, as distribuições das áreas de desmatamento foram recuperadas para a composição da série temporal de áreas desmatadas para o estado de Rondônia. Os resultados obtidos mostram que o método proposto é consistente, podendo ser expandido para toda região Amazônica e assim obter uma série temporal de áreas desmatadas para ser usada como referência. Palavras-chave-Modelo linear de mistura espectral, áreas de desflorestamento, Amazônia, floresta tropical. ABSTRACT PRODES project has been monitoring deforestation by clear cut in the Legal Amazon since 1988, initially making available only information in the table form, without maps of spatial distribution as currently available. Thus, the objective of this work is to present a method to generate the spatial distribution of the areas mapped annually by the PRODES project to the state of Rondônia. In this method description, the annual time series was aggregated to every 5 years period. Based on the legend code of the 2021 PRODES map and on the results obtained by the proposed method, it was possible to recover the data for the years 1988 to 2021, referring to the deforestation accumulated since the beginning of the project. In addition to deforestation, the mapped classes are hydrography, non-forest and forest. Areas of hydrography and non-forest were maintained and the deforestation areas were recovered for the periods analyzed. Thus, the distributions of deforestation areas were recovered for the composition of the time series of deforested areas for the state of Rondônia. The results obtained show that the proposed method is consistent, and can be expanded to the entire Amazon region and thus obtaining a time series of deforested areas to be used as a reference.

This article presents a method, based on orbital remote sensing, to map the extent of forest plantations in São Paulo State (Southeast Brazil). The proposed method uses the random forest machine learning algorithm available on the Google Earth Engine (GEE) cloud computing platform. We used 30 m annual mosaics derived from Landsat-5 Thematic Mapper (TM) images and from Landsat-8 Operational Land Imager (OLI) images for the 1985 to 1995 and 2013 to 2021 time periods, respectively. These time periods were selected based on the planted areas’ rotation, especially the eucalypt’s short rotation. To classify the forest plantations, green, red, NIR, and MIR spectral bands, NDVI, GNDVI, NDWI, and NBR spectral indices, and vegetation, shade, and soil fractions were used for both sensors. These indices and the fraction images have the advantage of reducing the volume of data to be analyzed and highlighting the forest plantations’ characteristics. In addition, we also generated one mosaic for each fraction image for the TM and OLI datasets by computing the maximum value through the period analyzed, facilitating the classification of areas occupied by forest plantations in the study area. The proposed method allowed us to classify the areas occupied by two forest plantation classes: eucalypt and pine. The results of the proposed method compared with the forest plantation areas extracted from the land use and land cover maps, provided by the MapBiomas product, presented the Kappa values of 0.54 and 0.69 for 1995 and 2020, respectively. In addition, two pilot areas were used to evaluate the classification maps and to monitor the phenological stages of eucalypt and pine plantations, showing the rotation cycle of these plantations. The results are very useful for planning and managing planted forests by commercial companies and can contribute to developing an automatic method to map forest plantations on regional and global scales.

This article presents a method to map the extent of forest plantation in an area located in the São Paulo State (Brazil). The proposed method applies the Linear Spectral Mixing Model (LSMM) to Landsat Thematic Mapper (TM) datasets to derive annually vegetation, soil and shade fraction images for local analysis. We used 30 m annual mosaics of TM images during the 1985 to 1995 time period. These fraction images have the advantage to reduce the volume of data to be analyzed highlighting the target characteristics. Then, we generated only one mosaic for each fraction images for TM dataset computing de maximum value through this period, facilitating the classification of areas occupied by forest plantation. The proposed method allowed to classify two forest plantation classes: Eucalypt and Pine. In addition, it allowed to monitor the phenological stages of Eucalypt according to its growth cycle. The results are very important for planning and management by the commercial companies and can contribute to develop an automatic method to map forest plantation areas in a regional and global scales.