Key points are not available for this paper at this time.
Sustaining soil quality under intensive land use systems to feed the growing population is a major challenge in the developing world. A prerequisite for effective soil quality analysis is the selection of a minimum data set (MDS) which effectively represents a given production system. As soil is a complex and dynamic system, there is no universal consensus on the ideal set of soil indicators for assessing soil quality. To address this challenge, this study evaluated multiple machine learning-based feature selection techniques in conjunction with conventional principal component analysis (PCA) for selection of the MDS, with the hypothesis that non-linear machine learning-based feature selection algorithms could better capture the complexity of soil quality dynamics. Weights for the selected indicators were computed using both PCA and entropy methods, aiming to develop a novel framework for soil quality analysis in relation to crop yield. Integrated nutrient management practices had a significant influence on soil physical, chemical and biological properties, all of which are essential for maintaining soil health and sustaining crop productivity. MDS selection and weight assignment through PCA provided strong correlations with crop yield (R 2 = 0.91, p < 0.01 for rice; R 2 = 0.93 p < 0.001 for wheat), comparable to the performance of MDS selection using PCA with entropy-based weighting (R 2 = 0.88, p < 0.01 for rice; R 2 = 0.90, p < 0.01, for wheat). When the goal is to maximize crop yields, the use of MDS selected through PCA and PCA-based weights for soil quality index (SQI) calculation appears to be more effective for a better representation of the production system. The proposed framework holds promise for application across diverse cropping systems and management practices, offering valuable insights for enhancing sustainable crop production.
Ghosh et al. (Mon,) studied this question.