Change detection methods are often incapable of accurately detecting changes within time series that are heavily influenced by seasonal variations. Techniques for de-seasoning time series or methods that apply the spatial context have been used to improve the results of change detection. However, few studies have explored Landsat’s shortwave infrared channel (SWIR 2) to discriminate between seasonal changes and land use/land cover changes (LULCC). Here, we explored the effectiveness of Operational Land Imager (OLI) spectral bands and vegetation indices for detecting deforestation in highly seasonal areas of Brazilian savannas. We adopted object-based image analysis (OBIA), applying a multidate segmentation to an OLI time series to generate input data for discrimination of deforestation from seasonal changes using the Random Forest (RF) algorithm. We found adequate separability between deforested objects and seasonal changes using SWIR 2. Using spectral indices computed from SWIR 2, the RF algorithm generated a change map with an overall accuracy of 88.3%. For deforestation, the producer’s accuracy was 88.0% and the user’s accuracy was 84.6%. The SWIR 2 channel as well as the mid-infrared burn index presented the highest importance among spectral variables computed by the RF average impurity decrease measure. Our results give support to further change detection studies regarding to suitable spectral channels and provided a useful foundation for savanna change detection using an object-based method applied to Landsat time series.

There is currently a lot of interest in determining the state of Brazilian grasslands. Governmental actions and programs have recently been implemented for grassland recovery in Brazilian states, with the aim of improving production systems and socioeconomic indicators. The aim of this study is to evaluate the vegetative growth, temporal vigor, and long-term scenarios for the grasslands in Zona da Mata, Minas Gerais State, Brazil, by integrating phenological metrics. We used metrics derived from the normalized difference vegetation index (NDVI) time series from moderate resolution imaging spectroradiometer (MODIS) data, which were analyzed in a geographic information system (GIS), using multicriteria analysis, the analytical hierarchy process, and a simplified expert system (ES_S). These temporal metrics, i.e., the growth index (GI) for 16-day periods during the growing season; the slope; and the maximum, minimum, and mean for the time series, were integrated to investigate the grassland vegetation conditions and degradation level. The temporal vegetative vigor was successfully described using the rescaled range (R/S statistic) and the Hurst exponent, which, together with the metrics estimated for the full time series, imagery, and field observations, indicated areas undergoing degradation or areas that were inadequately managed (approximately 61.5%). Time series analysis revealed that most grasslands showed low or moderate vegetative vigor over time with long-term persistence due to farming practices associated with burning and overgrazing. A small part of the grasslands showed high and sustainable plant densities (approximately 8.5%). A map legend for grassland management guidelines was developed using the proposed method with remote sensing data, which were applied using GIS software and a field campaign.

A cafeicultura é uma das principais culturas agrícolas do Brasil e realizar o mapeamento e monitoramento desta cultura é fundamental para conhecer sua distribuição espacial. Porém, mapear estas áreas utilizando imagens de Sensoriamento Remoto não é uma tarefa fácil. Sendo assim, este trabalho foi realizado com o objetivo de comparar o uso de diferentes variáveis e algoritmos de classificação para o mapeamento de áreas cafeeiras. O trabalho foi desenvolvido em três áreas diferentes, que são bastante significativas na produção de café. Foram utilizados 5 algoritmos de aprendizagem de máquinas e 7 combinações de variáveis: espectrais, texturais e geométricas, associadas ao processo de classificação. Um total de 105 classificações foram realizadas, 35 classificações para cada uma das áreas. As classificações que não usaram variáveis espectrais não resultaram em bons índices de acurácia. Nas três áreas, o algoritmo que apresentou as melhores acurácias foi o Support vector machine, com acurácia global de 85,33% em Araguari, 87% em Carmo de Minas e 88,33% em Três Pontas. Os piores resultados foram encontrados com o algoritmo Random Forest em Araguari, com acurácia global de 76,66% e com o Naive Bayes em Carmo de Minas e Três Pontas, com 76% e 82% de acerto. Nas três áreas, variáveis texturais, quando associadas às espectrais, melhoraram a acurácia da classificação. O SVM apresentou o melhor desempenho para as três áreas