Relationships in soil distribution as revealed by a global soil database

Abstract

The ISRIC WISE Global soil database was used to develop broad quantitative relationships between the environmental factors of climate, parent material and topography and a range of important soil properties (pH, sum of bases, organic carbon, clay content and others). Three different analytical approaches were used in the analysis, involving (i) multiple linear regressions, (ii) fitted decision trees and (iii) categorical analysis with median values. The strengths of the predictive relationships are generally only moderate at best, with R 2 values in the regression relationships typically in the range 0.2 to 0.4 and a wide spread in inter-quartile values in categorical median results. Testing with 100 random samples revealed only broadly moderate accuracy for the soil property predictions, with root mean square error down to 0.84 for pH predictions and down to 2.6 cmol c/kg for sum of bases. Despite their low strengths, the predictive relationships can provide useful first approximations of soil character under different environmental conditions and could be applied in broad quantitative soil modelling and mapping programs. They have the potential for widespread application as they should be universally applicable, are based on readily available data and do not require sophisticated quantitative modelling techniques. The relationships revealed in the study can assist in our understanding of soil formation and soil distribution. Most relationships are in accord with accepted pedological thinking and support the state factor model of soil formation, but some anomalies are observed and deserve further examination. The results reveal the dominant influence of climate and parent material in controlling the distribution of many soil properties, with the influence of topography being less evident, at least at the global scale.