Ensuring soil health (SH) is essential for ecosystem sustainability and food security, making it a cornerstone of the one health concept. However, current SH assessment frameworks face critical engineering limitations: high analytical costs, extensive laboratory infrastructure requirements, and complexity that restricts implementation in resource-limited settings. SH is often assessed in agricultural contexts through dynamic soil health indicators (SHI) that integrate key SH indicators. For instance, the soil management assessment framework has been used worldwide to determine SHI. However, its reliance on soil science expertise limits its application, highlighting the need for a more autonomous and computationally efficient approach. We developed an artificial intelligence (AI) approach utilizing an artificial neural network to predict the SH index (SHI) based on a set of indicators selected through a genetic algorithm. This engineering solution addresses the multiobjective optimization challenge of maximizing predictive accuracy while minimizing analytical complexity and cost. The study used a database containing 33 soil indicators (physical, chemical, and biological) from 432 samples under different land uses in Southern Brazil. This ANN-based approach demonstrated superior performance in predicting SHI compared to 11 other machine learning regression algorithms. Additionally, the GA-based indicator selection outperformed the correlation-based filter method, SelectKBest, achieving up to 82% higher R2 for small subsets and an average of 33% improvement overall, thereby demonstrating more efficient and informative feature selection. The minimum set of 13 indicators achieved the highest accuracy in predicting SHI, as indicated by a high coefficient of determination (R2 = 0.86), a low mean absolute error (MAE = 0.024), and a low mean square error (MSE = 0.001), reducing analytical requirements by over 60%, and enabling cost-effective monitoring across agricultural systems. Therefore, the proposed AI engineering approach accurately predicts SH and emerges as a promising solution for SH assessments, facilitating adaptive management practices and supporting sustainable intensification across diverse agricultural contexts without requiring specialized expertise. This technology addresses critical barriers to precision agriculture adoption, particularly in developing regions, where most agricultural lands lack regular SH monitoring due to resource constraints.
Silva et al. (Sat,) studied this question.