Forest degradation in the Brazilian arc of deforestation: Land cover classification and biomass estimation using very high-resolution satellite images
Abstract
Tropical forests cover the last unexploited frontier lands that may be converted to agricultural production so they have been hotspots of land use change. Perhaps the hottest of these spots for the past 40 years is the Brazilian arc of deforestation in the southern and eastern Amazon. This region has been a focus of development with forests converted to agriculture, including industrial commodity crops, and pastures. Degradation of forests by extensive selective logging and fires has accompanied the advance of the frontier and has resulted in significant impacts on ecosystem services. While the deforestation and, to some extent, the agricultural use in the region are well quantified, the degradation of forests has been difficult to study. In order to better understand this land use and land cover change hotspot, we tested an approach to classify and map intact forests, logged and burned forests as well as agricultural and pastoral land cover. Our tests were performed for a portion of the Feliz Natal Municipality in Mato Grosso, Brazil that has been subject to intense development pressure since the 1980s. We used very-high-resolution images and divided them into 250 m x 250 m tiles. We calculated two-dimensional Fourier transforms of enhanced vegetation index (EVI) for each tile and used the resulting frequency decomposition (reflecting dominant tree crown dimensions) as predictors in boosted decision tree models to both classify land cover into five classes (agriculture, pasture, logged forest, burned forest, or intact forest) and estimate biomass. Using training data derived from field visits, visual interpretation of historical Landsat images, and airborne lidar data, we were able to classify five land cover classes (intact forests, logged forests, burned forests, agriculture, and pasture) with between 80% to 95% accuracy. Forest biomass ranged from near zero to over 100 Mg C ha-1 and was highly accurate based compared to the lidar data. Existence of historical very-high-resolution images and the likelihood for continued availability of commercial imagery far into the future makes this method attractive for economical long-term monitoring. The importance of this research is underscored by the recent surge in deforestation in the Brazilian Amazon and associated forest degradation and fires.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMB128...03K
- Keywords:
-
- 3305 Climate change and variability;
- ATMOSPHERIC PROCESSES;
- 3337 Global climate models;
- ATMOSPHERIC PROCESSES;
- 0414 Biogeochemical cycles;
- processes;
- and modeling;
- BIOGEOSCIENCES;
- 0426 Biosphere/atmosphere interactions;
- BIOGEOSCIENCES