ALBERT

Description: These data sets include yearly maps of land cover classification for the state of Mato Grosso, Brazil, from 2001 (2000-09-01 to 2001-08-31) to 2017 (2016-09-01 to 2017-08-31), based on MODIS image time series (collection 6) at 250-meter spatial resolution (product MOD13Q1). Ground samples consisting of 1,892 time series with known labels are used as training data for a support vector machine classifier. We used the radial basis function kernel, with cost C=1 and gamma = 0.01086957. The classes include natural and human-transformed land areas, discriminating among different agricultural crops in state of land cover change maps for Mato Grosso State in Brazil. The results provide spatially explicit estimates of productivity increases in agriculture as well as the trade-offs between crop and pasture expansion. --- The correlation coefficients between the agricultural areas classified by our method and the estimates by IBGE (Brazil's Census Bureau) for the harvests from 2001 to 2017, were equal to 0.98. At the state level the soybean, cotton, corn and sunflower areas had a correlation equal 0.97, 0.85, 0.98 and 0.80. --- The areas classified as forest were compared with the Hansen et al. (2013, doi:10.1126/science.1244693) mapping for the year 2000. In order to separate the forest areas, we examined the areas with more than 25% tree cover on the Hansen et al. (2013, doi:10.1126/science.1244693) map. We found that 98% of the pixels classified as forest match the pixels indicated by Hansen et al. (2013) as having more than 25% tree cover. When we joined the cerrado and forest classes, 83% of the pixels match the pixels by Hansen et al. (2013) as having more than 25% tree cover. --- The pixels labeled as pasture were compared to the pasture mapping done by Parente et al. (2017, doi:10.1016/j.jag.2017.06.003). We found that 80% of the pixels classified as forest match the pixels indicated by Parente et al. (2017, doi:10.1016/j.jag.2017.06.003) for the state of Mato Grosso. --- In the land cover change maps for Mato Grosso State in Brazil version 3, we applied a methodology to deal with trajectories in classified maps. This methodology for reasoning about land-use change trajectories, called LUC Calculus, has been discussed in previous work (Maciel et al., 2018, doi:10.1080/13658816.2018.1520235). For reducing the temporal variability, we use the entire history of the study area considered as a set of land-use trajectories (from 2001 to 2017). For reasoning about this, we adopt the reference date 2001 and we used two-step post-processing, first applying masks and rules on the initial classified map (2001) and then land-use rules using LUC Calculus for the all years (2001-2017). The first-step post-processing was performed on the initial classified map (2001). We applied the forest mask to the classified map of the year 2001. This forest mask comes from the PRODES Project (http://www.obt.inpe.br/OBT/assuntos/programas/amazonia/prodes). In the non-forest, the appearance of secondary vegetation is not mapped. An additional set of rules was applied on the initial map using two sets of maps: PRODES map of the year 2001 and Cerrado map of 2000 (http://www.obt.inpe.br/cerrado). This mask of the Cerrado biome depicts the Cerrado within two classes: Anthropized Cerrado and Non-Anthropized Cerrado. The second-step post-processing was carried on the entire years from the classified map (2001-2017) using the LUC Calculus method. First, we elaborate a set of rules defined by experts in Amazon and Cerrado biomes. These rules express information about different trajectories of land-use change in MT that represent an irregular transition between classes. The rules used was: Forest (F), Cerrado (C), Pasture (P) and Soybean (S) 1. C -〉 F* to C -〉 C* 2. C -〉 C -〉 P* -〉 C to C -〉 C -〉 C* -〉 C 3. C -〉 C -〉 S* -〉 C to C -〉 C -〉 C* -〉 C 4. P -〉 P -〉 C* -〉 C* -〉 P to P -〉 P -〉 P* -〉 P* -〉 P 5. F -〉 C* -〉 F -〉 F to F -〉 F* -〉 F -〉 F 6. F -〉 F -〉 C* -〉 F to F -〉 F -〉 F* -〉 F 7. F -〉 C* -〉 F to F -〉 F* -〉 F 8. F -〉 C* to F -〉 F* 9. F -〉 F -〉 P -〉 F* to F -〉 F -〉 P -〉 SV* 10. P -〉 P -〉 F* -〉 P to P -〉 P -〉 SV* -〉 P The sequential application of the rules is able to ensure the temporal consistency among classes over the years. The class changed is highlighted with "*". From rule 1 to 8 we assume the reference date, 2001, as the starting point to find the class to will be changed. Rules 9 and 10 exemplify scenery where new class secondary vegetation (SV) occurs. The trajectory methodology enables us to include a new class called 'secondary vegetation'. This class represents a significant portion of the deforestation areas that have fallen into disuse or abandoned and have regrown as secondary forest. --- The following data sets are provided: (a) The classified maps in compressed TIFF format (one per year) at MODIS resolution. (b) A QGIS style file for displaying the data in the QGIS software (c) An csv file with the training data set (1,892 ground samples). --- The software used to produce the analysis is available as open source on https://github.com/e-sensing. --- Note: The TIFF raster files use the Sinusoidal Projection, which is the same cartographical projection used by the input MODIS images. When opening the TIFF raster maps in QGIS, to ensure correct navigation please use the Sinusoidal Projection, by selecting in QGIS projection menu, the following option: "Generated CRS (+proj=sinu +lon_0=0 +x_0=0 +y_0=0 +a=6371007.181 +b=6371007.181 +units=m +no_defs)"

Description: This dataset contains the yearly maps of land use and land cover classification for Amazon biome, Brazil, from 2000 to 2019 at 250 meters of spatial resolution. We used image time series from MOD13Q1 product from MODIS (collection 6), with four bands (NDVI, EVI, near-infrared, and mid-infrared) as data input. A deep learning classification MLP network consisting of 4 hidden layers with 512 units was trained using a set of 33,052 time series of 12 known classes from both natural and anthropic land covers. Quality assessment using 5-fold cross-validation of the training samples indicates an overall accuracy of 99.22% and the following user's and producer's accuracy for the land cover classes: ProdAcc UserAcc Forest 99.80% 99.86% Pasture 98.72% 98.04% Soy_Corn 98.92% 99.06% Soy_Cotton 99.23% 99.25% Fallow_Cotton 95.74% 96.43% Millet_Cotton 100.00% 97.98% Soy_Fallow 99.76% 99.09% Savanna2 99.94% 99.47% Savanna1 98.18% 99.06% Wetlands 99.31% 98.19% Soy_Millet 76.67% 84.66% Soy_Sunflower 84.62% 78.57%

Description: This data set includes deforestation maps, located in the border between the west of Brazil and the north of Bolivia (corresponding to Sentinel-2's tile 20LKP). The source images for this dataset came from ESA's Sentinel-2A satellite. They were processed from top of the atmosphere to surface reflectance using the Sen2Cor 2.8 software and their clouds were masked using the algorithm Fmask 4.0. The K-Fold technique was used to select the best Random Forest (RF) model varying different combinations of Sentinel-2A bands and vegetation indices. The RF models were trained using the time series of 481 samples included in this data set. The two selected models that presented the highest median of F1 score for the Deforestation class were: 1) the combination of the blue, bnir, green, nnir, red, swir1, and swir2 bands (hereafter Bands); and 2) the combination of Enhanced Vegetation Index, Normalized Difference Moisture Index, and Normalized Difference Vegetation Index (hereafter Indices). Each RF model produced a deforestation map. During training, we used RF models of 1000 trees and the full depth of the Sentinel-2A time series, comprising 36 observations ranging from August 2018 to July 2019. To assess the map's accuracy, good practices were followed [1]. To determine the validation data set size (n), the user accuracy was conjectured using a bootstrapping technique. Two validation data sets (n=252) were collected independently to assess the maps' accuracy. For Deforestation, the Bands classification model has the highest values of the F1 score (93.1%) when compared with the Indices model (91.9%). The Forest and Other classes had better results of the F1 score using the Indices (85.8% and 82.2%, respectively) than using the Bands (85.3% and 78.7%, respectively). Our classifications have an overall accuracy of 88.9% for Bands and 84.9% for Indices, and the following user's and producer's accuracy for the models: Accuracy of classification using Bands: Deforestation: UA - 97.4% PA - 89.2% Forest: UA - 80.8% PA - 90.4% Other: UA - 80.2% PA - 77.3% Accuracy of classification using Indices: Deforestation: UA - 96.1% PA - 88.0% Forest: UA - 88.6% PA - 83.3% Other: UA - 77.0% PA - 88.0% --- To produce the maps, the R package sits was used. The sits is an open source software that provides tools for time series analysis and classification. The sits packages can be found at GitHub (https://github.com/e-sensing/sits), and the scripts used in work can be found at http://doi.org/10.5281/zenodo.3932013. --- The following data sets are provided: (a) The classified map in compressed GeoTIFF format (10-meter resolution) using the Bands model. (b) The classified map in compressed GeoTIFF format (10-meter resolution) using the Indices model. (c) CSV file with the training data set. (d) CSV file with the validation data set for the Bands model. (e) CSV file with the validation data set for the Indices model. (f) A QGIS style file for displaying the data in the QGIS software. --- Note: The GeoTIFF raster files use the UTM Projection, which is the same cartographical projection used by the input Sentinel-2 images. When opening the GeoTIFF raster maps in QGIS, to ensure correct navigation please use the UTM zone 20S projection (EPSG:32720). The projection string parameters are: "+proj=utm +zone=20 +south +datum=WGS84 +units=m +no_defs +ellps=WGS84 +towgs84=0,0,0"

Description: 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.

Description: In recent years, Earth observation (EO) satellites have generated big amounts of geospatial data that are freely available for society and researchers. This scenario brings challenges for traditional spatial data infrastructures (SDI) to properly store, process, disseminate and analyze these big data sets. To meet these demands, novel technologies have been proposed and developed, based on cloud computing and distributed systems, such as array database systems, MapReduce systems and web services to access and process big Earth observation data. Currently, these technologies have been integrated into cutting edge platforms in order to support a new generation of SDI for big Earth observation data. This paper presents an overview of seven platforms for big Earth observation data management and analysis—Google Earth Engine (GEE), Sentinel Hub, Open Data Cube (ODC), System for Earth Observation Data Access, Processing and Analysis for Land Monitoring (SEPAL), openEO, JEODPP, and pipsCloud. We also provide a comparison of these platforms according to criteria that represent capabilities of the EO community interest.

Description: The physical phenomena derived from an analysis of remotely sensed imagery provide a clearer understanding of the spectral variations of a large number of land use and cover (LUC) classes. The creation of LUC maps have corroborated this view by enabling the scientific community to estimate the parameter heterogeneity of the Earth’s surface. Along with descriptions of features and statistics for aggregating spatio-temporal information, the government programs have disseminated thematic maps to further the implementation of effective public policies and foster sustainable development. In Brazil, PRODES and DETER have shown that they are committed to monitoring the mapping areas of large-scale deforestation systematically and by means of data quality assurance. However, these programs are so complex that they require the designing, implementation and deployment of a spatial data infrastructure based on extensive data analytics features so that users who lack a necessary understanding of standard spatial interfaces can still carry out research on them. With this in mind, the Brazilian National Institute for Space Research (INPE) has designed TerraBrasilis, a spatial data analytics infrastructure that provides interfaces that are not only found within traditional geographic information systems but also in data analytics environments with complex algorithms. To ensure it achieved its best performance, we leveraged a micro-service architecture with virtualized computer resources to enable high availability, lower size, simplicity to produce an increment, reliable to change and fault tolerance in unstable computer network scenarios. In addition, we tuned and optimized our databases both to adjust to the input format of complex algorithms and speed up the loading of the web application so that it was faster than other systems.

Description: In their comments about our paper, the authors remark on two issues regarding our results relating to the MACCS-ATCOR Joint Algorithm (MAJA). The first relates to the sub-optimal performance of this algorithm under the conditions of our tests, while the second corresponds to an error in our interpretation of MAJA’s bit mask. To answer the first issue, we acknowledge MAJA’s capacity to improve its performance as the number of images increases with time. However, in our paper, we used the images we had available at the time we wrote our paper. Regarding the second issue, we misread the MAJA’s bit mask and mistakenly labelled shadows as clouds. We regret our error and here we present the updated tables and images. We corrected our estimation and, consequently, there is an increment in MAJA’s accuracy in the detection of clouds and cloud shadows. However, these increments are not enough to change the conclusion of our original paper.

Description: 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.