Data fusion of visible near-infrared and mid-infrared spectroscopy for rapid estimation of soil aggregate stability indices

Abstract

Soil aggregate stability (AS) is a crucial soil physical property for sustainable agricultural and environmental land management practices. Yet, its conventional laboratory methods are fastidious and time consuming thereby restricting high-density sampling and large-scale estimation. Therefore, the potential of fusing visible near-infrared (vis-NIR) and mid-infrared (MIR) spectroscopy to enhance the prediction of three AS indices, namely, fast wetting (FW), slow wetting (SW) and mechanical breakdown (MB) on some Belgian Retisol, Cambisol and Luvisol topsoils was evaluated. Partial least squares regression (PLSR) was used to build calibration models for the three AS indices using the vis-NIR spectra, MIR spectra and the fusion of both spectra (SF). Another data fusion approach (i.e. model output averaging, MOA) included the use of four averaging algorithms. Results showed that MIR models outperformed the vis-NIR models for all three indices. The SF approach showed an improved prediction performance over both individual sensor techniques for the FW index only. MOA models outperformed the individual and SF models and yielded the best prediction accuracy for SW and MB indices. Data fusion modelling thus enhanced the accuracy of FW, SW and MB predictions, although the selection of the best data fusion approach is dependent on the nature of the dataset and the stability index to be assessed.