Analysis of terrain attributes in different spatial resolutions for digital soil mapping application in southeastern Brazil

• Published in: Geoderma Regional Vol. 21; p. e00268
• Main Authors: Sena, Nathalie Cruz, Veloso, Gustavo Vieira, Fernandes-Filho, Elpídio Inácio, Francelino, Marcio Rocha, Schaefer, Carlos Ernesto G.R.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2020
• Subjects: Advanced Spaceborne Thermal Emission and Reflection Radiometer; Brazil; cluster analysis; Digital elevation model; digital elevation models; landscapes; lidar; Machine learning; Morphometry; Multivariate analysis; prediction; principal component analysis; soil; soil surveys; Brazil; Digital elevation model; Morphometry; Multivariate analysis; Machine learning
• ISSN: 2352-0094; 2352-0094
• DOI: 10.1016/j.geodrs.2020.e00268

Terrain attributes are used as auxiliary variables in spatial prediction of soil classes and properties, due to their important role in the pedogenetic process and the increasing availability of digital elevation models (DEMs) in different resolutions. This work analyzed the effect of the different spatial resolutions of the DEMs and attributes derived from terrain and their implications for application in DSM predictive models. We used three spatial resolutions from different DEMs: (1) LIDAR – 2 m; (2) ALOS PALSAR – 12.5 and 30 m; (3) SRTM – 30 m; and (4) ASTER GDEM – 30 m. Multivariate analyses were performed determined by the Pearson linear correlation coefficient (r), the K-means cluster analysis, and the principal component analysis (PCA). The prediction of soil classes was performed using the terrain attributes grouped as to sensitivity to resolution, for different spatial resolutions, applying the machine learning algorithms Random Forest and Support Vector Machine. The cluster analysis indicated that most attributes remained within the same group of resolution sensitivity with changes in the cell size of reference DEM. The attributes of the terrain grouped low sensitive to resolution derived from the SRTM DEM showed better precision and the main advantage was the low cost and facilitating computational processing for application in the DSM. •Most correlations between attributes were positive and strong when using SRTM as reference for DEM;•Despite the changes in spatial resolution of DEMs, most attributes remained with in the same group of sensitivity;•The performance of the prediction is related to the resolution of DEMs and the set of attributes of the terrain.