Land use and land cover (LULC) are intrinsically tied to ecological and social dynamics. Still, classifying LULC in ecotones, where landscapes are commonly heterogeneous and have a wide range of physiognomies, remains a challenge. Here we present a three-level hierarchical classification approach, using both Landsat and MODIS images, and both pixels and objects as units of information. We applied this multi-temporal and -spatial approach to classify land use in the Upper Xingu River Basin (∼170,000 km 2 ), located in the arc of deforestation of the Brazilian Amazon. The first level includes five classes and differentiates managed land from native vegetation with high overall accuracy (93%). The second level has 11 classes (overall accuracy = 86%) and separates main land uses and native vegetation domains. The third level has 16 classes (overall accuracy = 83%) and addresses productivity of both managed and natural systems. We find that this new method presented here is more efficient than existing regional and global land cover products. Applying this approach to assess land cover transitions in the basin from 1985 to 2015, we find that agricultural production increased, yet manifested itself differently in the northern (Amazon biome) and southern (Cerrado biome) portions of the basin. Analyzing land use change in different levels, we identify that agricultural intensification occurred mainly in the Amazon while the Cerrado has undergone an expansion in agricultural area. The method presented here can be adapted to other regions, improving efficiency and accuracy of classifying land cover in heterogeneous landscapes.
|Original language||English (US)|
|Journal||Remote Sensing Applications: Society and Environment|
|State||Published - Aug 2019|
Bibliographical noteFunding Information:
This work was supported by the Belmont Forum /funded Xingu project (“Integrating land use planning and water governance in Amazonia: towards improved freshwater security in the agricultural frontier of Mato Grosso”) via São Paulo Research Foundation (FAPESP) , grant number 2013/50180-5 ); Andrea Garcia is recipient of grant 2015/05103-8 and 2017/12787-6 from FAPESP; Rodnei Rizzo is recipient of grant 2013/20377-1 and 2017/12567-6 from FAPESP. Andrea Garcia, Vivian Vilela, and Rodnei Rizzo were also recipients of grants from CAPES, and Maria Victoria Ballester from CNPq. Paul West and James Gerber received support from the Belmont Forum/FACCE-JPI funded DEVIL project (Delivering Food Security from Limited Land) (NE/M021327/1) via NSF award no. 1540195 . We also thank Instituto Socioambiental (ISA) and Instituto de Pesquisa Ambiental da Amazônia (IPAM) for making data available, contributing to this manuscript.
© 2019 Elsevier B.V.
- Land use change