Description: The Cerrado biome in Brazil covers approximately 24% of the country. It is one of the richest and most diverse savannas in the world, with 23 vegetation types (physiognomies) consisting mostly of tropical savannas, grasslands, forests and dry forests. It is considered as one of the global hotspots of biodiversity because of the high level of endemism and rapid loss of its original habitat. This dataset includes maps of the vegetation in the Cerrado in two different hierarchical levels of physiognomies. These physiognomies were defined by Ribeiro and Walter (2008) and consist in a hierarchical classification structure. The first hierarchical level (referred as level-1) consists on three classes: grassland, savanna and forest; which are further split in a total of 12 sub classes in level-2. The maps were produced under the scope of the project "Development of systems to prevent forest fires and monitor vegetation cover in the Brazilian Cerrado” (WorldBank Project #P143185) – Forest Investment Program (FIP) - in collaboration with the Earth Observation Lab from the Humboldt University. The methodological approach was published at: doi:10.5194/isprs-archives-XLIII-B3-2020-953-2020, 2020. The goal was to analyze the potential of Landsat Analysis Ready Data (ARD) in combination with different environmental data to classify the vegetation in the Cerrado in two different hierarchical levels. The field data used for training and validation are included in this dataset. The classification accuracy was assessed using Monte Carlo simulation, in which 1000 simulations were carried out by randomly selecting 70% of the samples to train the random forest (RF) classification model, while the remaining 30% were used for validation. In each iteration, a confusion matrix was calculated, and the average confusion matrix was used to derive the overall accuracy and the class-wise f1-scores. On the first hierarchical level, with the three classes savanna, grasslands and forest, our model results reached f1-scores of 0.86, 0.87 and 0.85 leading to an overall accuracy of 0.86. In the second hierarchical level, we differentiated a total of 12 vegetation physiognomies with an overall accuracy of 0.77. The following class f1-scores for the vegetation classes in the second hierarchical level were: Campo limpo: 0.687, Campo rupestre: 0.528, Campo sujo: 0.851, Cerradao: 0.658, Cerrado rupestre: 0.847, Cerrado sensu stricto: 0.815, Ipuca: 0.830, Mata riparia: 0.743, Mata seca: 0.611, Palmeiral: 0.907, Parque de Cerrado: 0.966, Vereda: 0.364. The following data sets are provided here: (a) the classified maps in compressed TIFF format (one per hierarchical level) at 30-meters spatial resolution, (b) a QGIS style file for displaying the data in the QGIS software, (c) a csv file with the training data set (2,828 ground samples).

Description: Phenology is the study of reoccurring events during a year or season. It can be linked to the behavior of animals, such as phases of mating, breeding, or movement and to events such as green-up, bud burst, flowering, or senescence when referring to vegetation, as a response to changing environmental factors throughout a season. While these changes can be tracked on the level of individual species, their observation is usually restricted to small spatial extents. To broaden the extent of the observed area remote sensing data have been proven useful. As remote sensing data capture the seasonal change rather on a pixel than on a species level, they enable to analyze the phenology of the observed vegetation on a different scale, which is known as land surface phenology. Land surface phenological metrics that can, for example, be derived from time series of vegetation indices, allow to analyze the observed spatial and temporal patterns in relation to ecosystem processes (e.g., primary productivity). Subsequently, the derived metrics can be grouped based on their similarities into land surface phenological archetypes (LSP), defined as areas with comparable phenologies. However, the spatial resolution of the data used is crucial, which becomes even more critical when looking at heterogeneous ecosystems such as the Brazilian savanna, known as the Cerrado. The Cerrado covers an extent of approximately 2 mio. km², hosts many endemic species and is considered as a biodiversity hotspot that provides several ecosystem services of national and even global importance. However, due to a lack of extensive conservation regulations the Cerrado is prone to land cover changes for agricultural expansion, highlighting the need for detailed mapping and monitoring approaches. To reveal and analyze the spatial patterns of the remaining share of natural vegetation based on their land surface phenology, we analyzed a dense 8-day time series of combined enhanced vegetation data derived from Landsat 7 ETM+ and Landsat 8 images. Data gaps that were due to cloud contamination or sensor errors were filled using a radial basis convolution filter, enabling to subsequently derive phenological metrics for the season 2013/2014 using TIMESAT (Eklundh and Jönsson 2017). As these variables, such as start and end of season, amplitude or the base value, relate to the seasonality and primary productivity of the observed vegetation, we clustered them based on their similarities and defined 8 land surface phenological archetypes (LSP) of the Cerrado. The GeoTiff file contains the 8 LSPs that are explained in detail in Schwieder et al. in prep. For further questions please contact Marcel Schwieder. Class labels: 0 = Unclassified 1 = FORMBMS 2 = SAFORMS 3 = FORHBLS 4 = GLSAVLB 5 = GLSAVHB 6 = FORHBHS 7 = VEGINMS 8 = VEGINLS