کاربرد مدل های رگرسیونی در پیش بینی کلاس خاک در بخشی از مناطق ایران مرکزی (منطقه زرند کرمان)

Abstract Soil digital survey as tool for soil spatial information provides pathways for producing of high resolution soil maps. Therefore, it should be developed strategic methods for making high resolution soil spatial information. Subsequently, this study was designed for prediction of soil classes by regression models in Zarand region of Kerman. Regression models includes of multinomial logistic regression and multiclass boosted regression tree were used for prediction of soil great groups by relating those with predictors such as remote sensing indices, terrain attributes and geomorphology map. A confusion matrix was used to calculate aspects of map accuracy. The geomorphology map at the fourth level (geomorphic surface) was a powerful predictor unlike the other levels (landscape, landform and lithology). Terrain attributes and finally remote sensing indices after geomorphic surface were imported as predictors in the prediction. The map purity over all soil great groups was above ۰.۶۰ in both calibration and validation locations. Poorer performance was observed for Calcigypsids and Haplocambids. Both methods provided good predictions for Haplosalids that shown by high values for users’ accuracy and producers’ reliability The results showed soils with better reliability are those highly influenced by topographic and geomorphic characteristics at least in this study area (e.g. Haplosalids, Haplogypsids and Torripsamments) and soils with very low reliability and accuracy of prediction are hardly influenced by the topographic and geomorphic characteristics (e.g. Haplocambids and Calcigypsids). Keywords: Digital soil mapping, Multinomial logistic regression, Boosted regression tree