Soil organic carbon prediction based on scale-specific relationships with environmental factors by discrete wavelet transform

Abstract

The spatial distribution of soil organic carbon (SOC) is vital to agricultural and environmental management, and its variation is influenced by environmental factors operating at different scales and intensities. The objective of this study was to explore the scale-dependent effects of environmental factors on SOC distribution and to predict SOC based on their scale-specific relationships using wavelet transform. The spatial data of SOC, slope, soil water content (SWC), soil bulk density (SBD), sand, silt, and Landsat 8 remote sensing reflectance (Rrs) were extracted at 330 m interval along three transects in the arable land of Taiyuan basin, China. The spatial series of SOC and environmental variables along each of three transects were separated into six detail components and one approximation representing different scales using wavelet decomposition. The specific scale of each detail component was identified by Hilbert transform. The SOC variances over the entire basin were mainly explained (54%–86%) by scales of 12.70 and 21.12 km. Compared with the relationships between SOC and environmental factors at sampling scale, their multiscale correlations were better at larger scales. The SOC estimation using wavelet reconstruction based on predicted SOC at all scale components outperformed its prediction using stepwise multiple linear regression (SMLR) based on the original sampling data. The major contributing scale to SOC prediction was 12.70 km over the entire basin. In the prediction of overall SOC, Rrs was the major predictor in the upstream and downstream portions, whereas soil texture was the major contributor in the midstream portion. In this study, the SOC prediction using wavelet transform based on scale-dependent relationships with environmental variables generated new insights in soil properties estimation, and wavelet transform has potential for determining the multiscale relationships of soil properties with influencing factors.