Deciphering Soil Fertility of Tobacco Planting Fields with Back Propagation Artificial Neural Networks in Southwest China

Assessing soil quality via the integrated soil fertility index (IFI) is essential for enhancing soil sustainability. In this context, the emergent application of machine learning provides a novel perspective, although its utilization in assessing soil quality is relatively limited compared to its widespread adoption in other fields. This study used the back propagation artificial neural networks (BP-ANN) to assess the IFI for tobacco planting in Dali Prefecture located at the southwest of China over the years 1982, 2012, and 2022. The BP-ANN model showed an accuracy exceeding 99.99%. Soil organic matter (SOM) and available phosphorus (AP) were found to be the primary contributors to the BP-ANN model, accounting for 37.59% and 29.56% of the variation, respectively. The IFIs positively correlated with those computed by Nemerow and other machine learning methods . The proportions of levels IV and V IFIs showed increasing trends, suggesting an excessive application of fertilizers. Soil properties showed varying changes as time progresses. Specifically, SOM, pH, and alkali-hydrolyzable nitrogen showed a downward trend, total nitrogen showed the maximum in 2012, while AP and available potassium significantly increased over time. Semi-variogram analysis further substantiated that the spatiotemporal variations in IFI values were attributed to a combination of random factor and inherent structural factor. These findings underscore the usefulness of BP-ANN in regulating soil quality by mainly controlling SOM and AP in Dali Prefecture. Consequently, precision agriculture, involving fertilizer application reduction, controlled release fertilizer usage, and organic farming practices, should be adopted for sustainable long-term tobacco cultivation.