Mapeamento digital do solo com o uso de regressões logísticas múltiplas em região de Ortigueira, Paraná

Abstract

The present study aimed at the digital mapping of soil classes with the use of logistic regressions in several sample sizes, evaluating their potential as accurate mapping for soil classes. The study was conducted in a region of Ortigueira, Paraná, Brazil from a Digital Elevation Model, and its descendants, with a resolution of 4 m² and georeferenced samples of classified soil pro- file. We selected variable buffers of 75, 100 and 125 meters points that form databases for logistic regressions. The criteria used were the significance of field variables, the adjustment of logistic regressions to the sampled data, individual class accuracy, general accuracy and Kappa Cohen coefficient. The variables of greater extension explained the behavior of the regressions by altitude and declivity. The determination coefficients (R²) of Nagelkerke were 81.9 to 100% for CXa3, 42.5 to 51.1% for CXa1 and 16.5 to 19.7 for LVd1. State buffers resulted in values ranging from 65.2 to 69.6% for general accuracy and 63.2 to 67.3% for Kappa coefficients, in particular the sample buffers at points of 75 and 100 meters obtained the results of group results 69, 6 and 67.3% for the general set and Kappa coefficient respectively, but it was verified that a definition for the buffer for the mapping is related to the position of the soil samples classified in relation as terrain variables