Can pedotransfer functions based on environmental variables improve soil total nutrient mapping at a regional scale?

Abstract

Numerous pedotransfer functions (PTFs) have been developed to predict the soil properties of interest from other soil properties and, less commonly, from environmental variables. However, only a few PTFs have been developed to predict soil nutrients using environmental variables and to extrapolate them to characterize spatial soil variations at a regional scale. In this study, we attempted to develop PTFs for the total nitrogen (TN), total phosphorus (TP) and total potassium (TK) concentrations in three typical pedo-climatic areas of China (Fujian Province, Jiangsu Province and Qilian Mountains) with diverse climate, terrain and soil types. A series of linear PTFs were developed to quantify the effect of terrain and climate on the predictive relations between the soil nutrients and other measured soil properties and environmental variables. In addition, digital soil mapping (DSM) based on the random forest (RF) technique was performed to test the hypothesis that the best-fit PTFs could be extrapolated, based on soil maps and environmental variables, to describe regional soil variations in the soil nutrients. The root mean square errors (RMSEs) of the best-fit PTFs for TN, TP and TK ranged from 0.21 to 0.79 g kg−1, 0.20 to 0.58 g kg−1, and 3.68 to 5.00 g kg−1, respectively. Different RMSEs were produced by DSM, namely 0.37-1.89 g kg−1, 0.19−0.56 g kg−1 and 3.79-4.83 g kg−1 for TN, TP and TK, respectively. PTFs provided a sound basis for database compilation if the soil properties were highly correlated. However, the extrapolation of best-fit PTFs to regional scales yielded greater errors than those produced by DSM. The comparison results reveal the limitations of PTFs and suggest that their performance could be improved by using environmental covariates or by fitting data in areas with relatively homogeneous soil landscapes. The DSM techniques may provide satisfactory alternatives to predict soil data at both regional and plot scales.