Recently, remote sensing image time series analysis has being widely used to investigate the dynamics of environments over time. Many studies have combined image time series analysis with machine learning methods to improve land use and cover change mapping. In order to support image time series analysis, analysis-ready data (ARD) image collections have been modeled and organized as multidimensional data cubes. Data cubes can be defined as sets of time series associated with spatially aligned pixels. Based on lessons learned in the research project e-Sensing, related to national demands for land use and cover monitoring and related to state-of-the-art studies on relevant topics, we define the requirements to build Earth observation data cubes for Brazil. This paper presents the methodology to generate ARD and multidimensional data cubes from remote sensing images for Brazil. We describe the computational infrastructure that we are developing in the Brazil Data Cube project, composed of software applications and Web services to create, integrate, discover, access, and process the data sets. We also present how we are producing land use and cover maps from data cubes using image time series analysis and machine learning techniques.

We address the task of mapping deforested areas in the Brazilian Amazon. Accurate maps are an important tool for informing effective deforestation containment policies. The main existing approaches to this task are largely manual, requiring significant effort by trained experts. To reduce this effort, we propose a fully automatic approach based on spatio-temporal deep convolutional neural networks. We introduce several domain-specific components, including approaches for: image preprocessing; handling image noise, such as clouds and shadow; and constructing the training data set. We show that our preprocessing protocol reduces the impact of noise in the training data set. Furthermore, we propose two spatio-temporal variations of the U-Net architecture, which make it possible to incorporate both spatial and temporal contexts. Using a large, real-world data set, we show that our method outperforms a traditional U-Net architecture, thus achieving approximately 95% accuracy.

We investigated the dynamics of savanna clearing and land degradation in the MATOPIBA region, the newest agricultural frontier over the Cerrado in Brazil. For geospatial analysis of the “big data,” we used algorithms available on the Google Earth Engine (GEE) platform along with different satellite products. For the detection of vegetation clearing in the 1985–2018 period, we applied the LandTrendr algorithm to the normalized difference vegetation index (NDVI) obtained from Landsat satellite instruments. In the evaluation of land degradation (2001–2015), we used the Trends.Earth algorithm to analyze NDVI data calculated from the Moderate Resolution Imaging Spectroradiometer (MODIS). By combining different datasets into the analysis, we discussed the relationships among savanna clearings, land degradation, topographic attributes, and fire frequency. The results showed that the largest proportions of savanna clearing in the 1985–1990 period were detected in the south of MATOPIBA (mesoregions Ocidental do Tocantins and Extremo Oeste Baiano). The savanna clearing progressed toward the mesoregions Oriental do Tocantins, Sudoeste Piauiense, and Sul Maranhense between 1991 and 2005. Variations over time in cleared areas ranged from 0.086 Mha to 2.146 Mha per year, showing well-defined peaks in 1986, 1991, and 2012. The clearing events followed the soybean expansion, especially over flat terrains at high elevations. In addition to the spillover effects of the Amazon Soy Moratorium on the savannas, other political and economic factors promoted changes in cleared areas of MATOPIBA. Compared to the decade preceding the signature of the Amazon Soy Moratorium in 2006, the cleared area of MATOPIBA increased 41% in the 10-year post-moratorium period. In the other analysis, our Trends.Earth results showed that the largest proportions of land degradation areas, as proxied by the MODIS NDVI, were the Sudoeste Piauiense and Centro Maranhense. These mesoregions presented distinct patterns of precipitation and soil composition. The number of fire events increased with vegetation clearing. However, the amount of cleared areas correlated moderately with that of burned areas due to the multi-use of fire in the region. Finally, we observed that the amount of land-degraded areas increased with fire frequency over agricultural areas, but remained relatively stable over native vegetated areas. Results highlight the importance of satellite sensors and cloud-computing platforms to inspect long-term trajectories of land occupation and degradation over broad savanna areas.

We investigated the water use of Caatinga vegetation, the largest seasonally dry forest in South America. We identified and analysed the environmental phenological drivers in woody species and their relationship with transpiration. To monitor the phenological evolution, we used remote sensing indices at different spatial and temporal scales: normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), and green chromatic coordinate (GCC). To represent the phenology, we used the GCC extracted from in-situ automated digital camera images; indices calculated based on sensors included NDVI, SAVI and GCC from Sentinel-2A and B satellites images, and NDVI products MYD13Q1 and MOD13Q1 from a moderate-resolution imaging spectroradiometer (MODIS). Environmental drivers included continuously monitored rainfall, air temperature, soil moisture, net radiation, and vapour pressure deficit. To monitor soil water status and vegetation water use, we installed soil moisture sensors along three soil profiles and sap flow sensors for five plant species. Our study demonstrated that the near-surface GCC data played an important role in permitting individual monitoring of species, whereas the species’ sap flow data correlated better with NDVI, SAVI, and GCC than with species’ near-surface GCC. The wood density appeared to affect the transpiration cessation times in the dry season, given that species with the lowest wood density reach negligible values of transpiration earlier in the season than those with high woody density. Our results show that soil water availability was the main limiting factor for transpiration during more than 80% of the year, and that both the phenological response and water use are directly related to water availability when relative saturation of the soil profile fell below 0.25.