Evaluation of four digital classifiers for automated cartography of local soil classes based on reflectance and elevation in Mexico

Localization of local soil classes in an arid region of Mexico, using satellite imagery

“…The size of sampling was calculated with the following: six classes (Bl, Ba, Pr, N, A or none), so c = 6. In considering the findings by Cruz-Cárdenas et al (2010) in an arid region of Mexico, the precision value (b i ) was 30%, the confidence level (α) of 95% and the largest expected percentage of any category (Bl, Ba, Pr, N, A or none) was 50%. Using these data resulted in a sample size of 28 pixels for each class of soil in digital maps.…”

“…Demarcation of the training sites: The digitized map of LSC was superimposed over the study area. In each class of mapped soil, a polygon was digitized which constituted its training site (Cruz-Cárdenas et al 2010), considering within each site the georeference of the pedological pit.…”

“…Comparative studies between field survey maps of soils and maps of soil classes developed by means of computer-assisted cartography have demonstrated the effectiveness of the latter (Barrera-Bassols et al 2006). Cruz-Cárdenas et al (2010) found that the combination of satellite image reflectance and a digital elevation model (DEM) gives maps of soil classes in arid, temperate and tropical regions, more than 30% accuracy and precision, the best classifier being the MAXLIKE (maximum likelihood). In analysing this and considering advances in techniques of soil surveying and the use of satellite images, it is probable that spectral properties of topsoil of soil classes in arid areas will allow the establishment of their distribution, first at a local level and then at a regional level.…”

“…Payton et al (2003) mentioned that remote sensors and images classification are useful tools for compiling maps. In arid regions, satellite images have been used to map soils with sodium and gypsum (Neild et al 2007), to differentiate soil orders (Ríos-Martínez and Curtis 2002), to reconstruct lacustrine scenarios (Gaber et al 2009) and even for the mapping of local soil classes (LSCs; Cruz-Cárdenas et al 2010). However, these studies require exhaustive and systematic sampling as well as geostatistical methods for the compilation of soil maps.…”

Historical Delineation of Landscape Units Using Physical Geographic Characteristics and Land Use/Cover Change

“…The image for the 1983 scene consists of four bands and the remaining scenes had seven bands, each recording the intensity of the energy reflected or emitted by the objects on the surface at different portions of the electromagnetic spectrum [28]. We analyzed the spectral bands using the maximum likelihood classifier that has shown acceptable results for the generation of thematic maps [29][30][31][32]. The classifier uses the vector of means and the covariance matrix from pixels of the training areas of each class to assign the pixels of the image to each defined classes according to their probability of belonging to one category or another [12,15,33,34].…”