Digital soil mapping at pilot sites in the northwest coast of Egypt: A multinomial logistic regression approach

The study examines a digital soil mapping approach for the production of soil maps by using multinomial logistic regression on soil and terrain information at pilot sites in the Northwestern Coastal region of Egypt. The aim is to reproduce the original map and predict soil distribution in the adherent landscape. Reference soil maps produced by conventional methods at Omayed and Nagamish areas were used. Spectral and terrain parameters were calculated and logit models of the soil classes were developed. Predicted soil classes’ maps were produced. Software’s IDRISI/SAGA/SATISTCA/SPSS were used. The terrain and spectral parameters were found to be significantly influential and the selection of the land surfaces predictors was satisfactory. The McFadden pseudo R-squares ranged from 0.473 to 0.496. The most significant terrain parameters influencing the spatial distribution of the soil classes were elevation, valley depth, multiresolution ridgetop flatness index, multiresolution valley-bottom flatness index, and SAGA wetness index. However, the most influential spectral parameters are the first two principal components of the six Landsat Enhanced Thematic Mapper bands. The overall accuracy of the predicted soil maps ranged from 72% to 74% with a Kappa Index ranging from 0.62 to 0.64. The developed probability models were successfully used to predict the spatial distribution of the soil mapping units at pixel resolutions of 28.5 m × 28.5 m and 90 m × 90 m at adjacent unvisited areas at Matrouh and Alamin. The developed methodology could contribute to the allocation and to the soil digital mapping and management of new expansion sites in remote desert areas of Egypt.