Mapbiomas: Methodology to produce 30 years of annual maps for forest dynamics

Authors:
Marcos Rosa

Publication Date:
2019

Abstract/Summary:
The MapBiomas project began in July 2015 with the purpose of generating annual land use and land cover maps for all Brazilian territory, using digital image processing of Landsat imagery from 1985 to the current day (Souza & Azevedo, 2017). Automatic classification was performed using the Random Forest machine learning algorithm implemented in Google Earth Engine. Large-scale tropical forest dynamic analysis in Brazil Atlantic Forest Biome with 131.000 Mha was performed using the GEE platform, comparing pairs of land use maps in consecutive years (transition maps) from 1986 to 2016. Forest Restoration was divided in two different classes: Persistent Restoration includes areas of Pasture or “Mosaic of Agriculture or Pasture” mapped three consecutive years as forest and then persisted as Forest until 2016 and Non Persistent Restoration includes areas that were mapped as Forest and were deforested afterwards. Forest dynamics show a reduction of deforestation in old growth forests while restored forest increased forest cover over the last decade.

Resource Type:
Audio/Video, Conference Presentation, SER2019

Source:
Society for Ecological Restoration