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.

Tropical forests regulate the global water and carbon cycles and also host most of the world’s biodiversity. Despite their importance, they are hard to survey due to their location, extent, and particularly, their cloud coverage. Clouds hinder the spatial and radiometric correction of satellite imagery and also diminishing the useful area on each image, making it difficult to monitor land change. For this reason, our purpose is to identify the cloud detection algorithm best suited for the Amazon rainforest on Sentinel–2 images. To achieve this, we tested four cloud detection algorithms on Sentinel–2 images spread in five areas of the Amazonia. Using more than eight thousand validation points, we compared four cloud detection methods: Fmask 4, MAJA, Sen2Cor, and s2cloudless. Our results point out that FMask 4 has the best overall accuracy on images of the Amazon region (90%), followed by Sen2Cor’s (79%), MAJA (69%), and S2cloudless (52%). We note the choice of method depends on the intended use. Since MAJA reduces the number of false positives by design, users that aim to improve the producer’s accuracy should consider its use.

The agricultural frontier expansion in the Cerrado biome made Brazil a leader in commodity exports and is changing its landscape. Hence, efforts to accurate land use and land cover (LULC) monitoring in this region are strategic, due to its role in Brazil’s food, environmental, and economic security policy. Thinking on planning and technical sovereignty in the spatial sector, the China–Brazil Earth Resources Satellite (CBERS) Program was launched to provide useful data for decision-makers to manage the Brazilian territory independently of external policies. Their data, especially from CBERS-4 Wide-Field Imager (CBERS-4/WFI), are largely applied in deforestation monitoring by remote sensing specialists but less applied than data from other image providers for machine learning-based LULC mapping due to the small number of spectral bands and limitations related to clouds and shadows detection. However, with advances in orbital data analysis, data cubes enabled storing and accessing large spatio-temporal analysis-ready data. Within this scope, the Brazil Data Cube Project (BDC) creates multidimensional data cubes from orbital sensors’ data for all Brazilian territory. We applied BDC CBERS-4/WFI data cubes to generate LULC classifications for the Extremo Oeste Baiano agricultural belt correspondent to the 2017/2018 and 2019/2020 harvest periods, at two levels of detail: broad and crop type, incorporating ground truth samples, crop calendar knowledge, and vegetation indices to a dense time series analysis approach. Overall Accuracies were equal to 0.87 and 0.89 for broad, and 0.91 and 0.94 for crop type classifications. The results indicate CBERS-4/WFI data cubes as a useful tool for improving crop monitoring in the Cerrado biome based on machine learning.

The ecosystems of savanna cover $\sim 20\%$ of the Earth’s terrestrial area. Especially in tropical regions, they are rich in biodiversity¹ and water resources² and play an important role in carbon stock due to their content of above and below-ground biomass.³ The Brazilian Savanna, also known as Cerrado, is considered a global hotspot for biodiversity conservation, containing around 4800 endemic species.^1,4 In addition, its water resources feed the three largest watersheds in South America: the Amazon, Prata, and São Francisco. The Cerrado comprises 24% of the Brazilian territory, being the second largest biome in the country. However, despite its ecological relevance, only 8.6% of its native vegetation is in protected areas, i.e., specific regions established to protect biodiversity, water bodies, and other environmental resources.⁵ Approximately 47% of the Cerrado native vegetation have already been converted to other land uses, especially pasture (29%) and annual agriculture (9%).⁶ Moreover, from 2008 to 2019, the deforestation rates in the Cerrado have been higher than in the Amazon biome in 9 of these 12 years interval.⁷ This heavy loss of native vegetation has severe environmental consequences, such as vegetation fragmentation, habitat loss,⁸ and reduction of water yield and carbon stocks.^9,10 In this scenario, accurate mapping of Cerrado vegetation is essential to support policies against deforestation and, consequently, to maintain the provision of ecosystem services, since these maps are crucial to assess biodiversity, improve carbon stock estimation within the biome, and guide conservation policies.

Land use and land cover (LULC) mapping initiatives are essential to support decision making related to the implementation of different policies. There is a need for timely and accurate LULC maps. However, building them is challenging. LULC changes affect natural areas and local biodiversity. When they cause landscape fragmentation, the mapping and monitoring of changes are affected. Due to this situation, improving the efforts for LULC mapping and monitoring in fragmented biomes and ecosystems is crucial, and the adequate separability of classes is a key factor in this process. We believe that combining multidimensional Earth observation (EO) data cubes and spectral vegetation indices (VIs) derived from the red edge, near-infrared, and shortwave infrared bands provided by the Sentinel-2/MultiSpectral Instrument (S2/MSI) mission reduces uncertainties in area estimation, leading toward more automated mappings. Here, we present a low-cost semi-automated classification scheme created to identify croplands, pasturelands, natural grasslands, and shrublands from EO data cubes and the Surface Reflectance to Vegetation Indexes (sr2vgi) tool to automate spectral index calculation, with both produced in the scope of the Brazil Data Cube (BDC) project. We used this combination of data and tools to improve LULC mapping in the Brazilian Cerrado biome during the 2018–2019 crop season. The overall accuracy (OA) of our results is $88\%$, indicating the potential of the proposed approach to provide timely and accurate LULC mapping from the detection of different vegetation patterns in time series.