Towards a Robust Approach for Multitemporal Landcover Dataset: 3 Decades of Landcover Changes in Piauí, Brazil

Mapping the changes of land use and cover through the classification of satellite images is one of the essential sources to investigate and monitor the Earth´s surface. When performed in a multitemporal perspective, this approach requires specific procedures to match the images used. Considering that parkland/grassland savanna patches could increase over time due to different uses, this research aims to present a suitable method for processing Landsat-like images to investigate land cover dynamics. The study area covers seven municipalities in the Brazilian state of Piauí, Cerrado biome, which has been substantially affected by deforestation due to extensive agricultural projects in the past decades. The semi-automatic satellite imaging georegistration and the object-oriented classification of Landsat 5 and 8 satellites over the last 30 years in decadal periods are among the methodological procedures used. Findings demonstrate the semi-automatic image registration process as an effective method for the geometric correction of Landsat scenes, and the object-based classification procedures are appropriate for multitemporal studies allowing comparative metrics of landcover class changes by period. Regarding the remaining natural landcover within the study area, the results showed a substantial decrease of woodland savanna patches from 73% in 1986 to 43% in 2016, while agricultural fields increased from 4% to 25% in 30 years.