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

The objective of this research is to evaluate the precision and accuracy of local soil class maps generated with four computer algorithms: minimum distance, parallelepiped, maximum likelihood, and artificial neural networks, using digital elevation models and spectral signatures of local soil classes as input data. The study was done in the states of Mexico, San Luis Potosi, and Veracruz. Statistical binomial proportion tests were done to compare the difference between maps' precision and accuracy. The conclusion was that the combination of reflectance and elevation improved the quality of soil class maps produced by CAC, due to the reflectance variation of local soil classes according to altitude, which helped to better identify them. The best precision was 84% and the best accuracy was 30%.