MGBAS2 reference data:reference fire perimeters obtained from Sentinel-2 imagery over Madagascar for the years 2019 and 2021

The MGBAS2 reference data was derived from Sentinel-2 images over a set of 6 tiles (sampling units)sampled following a design customized for the Sentinel tiling grid system and to represent low and high burned area occurrence strata in Madagascar. Over each tile Sentinel-2 time series were defined based on a set of conditions for minimizing cloud cover and to guarantee series length and a minimum time lag between image pairs. Sentinel-2 image pairs were classified with a Random Forest (RF) algorithm to provide burned perimeters of depicting areas that burned between the two dates that were combined in a synthetic burned area reference dataset. This dataset represent for each unit burned and unburned polygons and masked areas. A detailed description of the dataset can be found in Fernández-García et al.(2023) ("Madagascar’s burned area from Sentinel-2 imagery (2016–2022): four times higher than from lower resolution sensors", in preparation). Description of the project: The reference fire perimeters for Madagascar were obtained in the framework of the project "Fire regimes and ecosystem services in African biodiversity hotspots: can fire policies favoring climate change mitigation, biodiversity and local communities converge?" financed by the Swiss Network for International Studies (SNIS). Among the goals of the project is the accurate characterization of burned area and fire regimes at Sentinel-2 spatial resolution (20m). The burned area database has been produced for the period 2016-2022 in Madagascar, southern Mozambique, Eswatini and Eastern South Africa (Fernández-García et al., 2023: doi.org). The MGBAS2 reference data presented here is the validation data for the mentioned burned area database. The characterization of burned area is the first step of the project that also aims to analyze the drivers of fire regimes and their consequences on carbon dynamics, biodiversity and local livelihoods.