Spatial prediction of soil qualitative variables using logistic regression and fuzzy techniques. Study region: Dobrovăţ basin (Central Moldavian Plateau)

Cristian-Valeriu Patriche,Dan Laurenţiu Stoica,Bogdan Roşca,Radu Gabriel Pârnău

doi:10.4316/georeview.2012.21.1.54

Abstract

The present study attempts to test the performance of two statistical approaches, namely the binary logistic regression and fuzzy techniques for spatial prediction of soil types. The study area is Dobrovaţ basin, located in NE Romania, within the Central Moldavian Plateau. The input parameters are the digital elevation model, slope, topographic wetness index, mean annual temperatures and precipitations. The logistic regression approach proved successful in estimating the spatial probabilities of Aluviosols, Chernozems, Preluvosols and Luvosols and generally failed in predicting the locations of Phaeozems. The fuzzy approach, implemented through SoLIM software, proved successful in predicting the occurrences of Aluviosols and Luvisols. On the whole, both methods managed to assign the same soil type as in the soil survey map for 55-56% of the basin. Though it is clear that the approaches need to be further improved, they do present, in the authors opinion, potential for the purpose of predicting soil qualitative variables

Highlights

Digital soil mapping refers to the application of mathematical and statistical methods for studying the spatial distribution of soils and their properties
Analysing the overall accuracy and the area under receiver operating characteristic (ROC) curve, as quality parameters for logistic regression models, it is found that Aluviosols and Chernozems are the best predicted soil types, followed by Preluvosols and Luvosols
The spatial probability of Aluviosols was computed from altitude, topographic wetness index and mean annual temperature

Summary

Introduction

Digital soil mapping refers to the application of mathematical and statistical methods for studying the spatial distribution of soils and their properties. A thorough synthesis of digital soil mapping techniques is provided by McBratney A.B. et al (2003). The application of such methods implies the use of a certain statistical software and / or a GIS software. The relief is characterized by the presence of cuesta flanks, structural plateaus at altitudes exceeding 300m and large floodplains. It has a temperate climate, with mean annual temperatures of 8.1-9.8oC and mean annual precipitations of 550-612 mm (Patriche C.V., 2005). The northern part of the region is covered by oak and beech forests, with a large extent of Luvisols, while the southern half is dominated by agricultural lands and Chernozems (Pirnău R.G., 2011)

Objectives

Results

Conclusion