Abstract
The Brazilian Savannas have been under increasing anthropic pressure for many years, and land-use/land-cover changes (LULCC) have been largely neglected. Remote sensing provides useful tools to detect changes, but previous studies have not attempted to separate the effects of phenology from deforestation, clearing or fires to improve the accuracy of change detection without a dense time series. The scientific questions addressed in this study were: how well can we differentiate seasonal changes from deforestation processes combining the spatial and spectral information of bi-temporal (normalized difference vegetation index) NDVI images? Which feature best contribute to increase the separability on classification assessment? We applied an object-based remote sensing method that is able to separate seasonal changes due to phenology effects from LULCC by combining spectral and the spatial context using traditional spectral features and semivariogram indices, exploring the full capability of NDVI image difference to train random forest (RF) algorithm. We found that the spatial variability of NDVI values is not affect by vegetation seasonality and, therefore, the combination of spectral features and semivariogram indices provided high global accuracy (97.73%) to separate seasonal changes and deforestation or fires. From the total of 13 features, 6 provided the best combination to increase the separability on classification assessment (4 spatial and 2 spectral features). How to accurately extract LULCC while disregarding the ones caused by phenological differences in Brazilian seasonal biomes undergoing rapid land-cover changes can be achieved by adding semivariogram indices in combination with spectral features as input data to train RF algorithm.
Highlights
The Brazilian Savanna biome is among the most endangered ecoregions on Earth due to high rates of deforestation and few formally protected areas (Hoekstra et al 2005), consisting of a mosaic of land cover types, undergoing a strong seasonality in climate
The semivariograms reached the sill within the calculated distance (900m), indicating that their spatial extents were sufficiently large to encompass the entire spatial variability of NDVI derived from Operational Land Imager (OLI)/Landsat-8 images
We found two distinct patterns: the shape and the overall variability of the data remained constant during the analysed period in areas with seasonal changes due to phenological effects between 2015 and 2016 (Figure 4ab); and the shape and sill increased during the analysed period in areas undergoing deforestation or fires (Figure 4cd)
Summary
The Brazilian Savanna biome is among the most endangered ecoregions on Earth due to high rates of deforestation and few formally protected areas (Hoekstra et al 2005), consisting of a mosaic of land cover types, undergoing a strong seasonality in climate. In these regions, a significant challenge in remote-sensing change detection is accurately extracting land-use and land-cover changes (LULCC) while disregarding those associated with phenological differences (Acerbi Junior et al 2015; Silveira et al 2018a; Silveira et al 2018b). They found challenges in the implementation of the method, that is a pixel-based approach, that is sensitive to registration errors and computational expensive (Zhu 2017)
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
