Application of Principal Component Analysis and Probabilistic Neural Networks in Ferralsols Recovery Evaluation Through Planting of Mabea Fistulifera and Eucalyptus Urograndis

ABSTRACT This study presents an innovative assessment model for analyzing the evolution of degraded soils subjected to different reclamation strategies. The proposal combines statistical and artificial intelligence tools to jointly integrate multiple physical and chemical soil properties, allowing for a more synthetic view of the processes. The model uses principal component analysis to synthesize information on the most relevant variables and subsequently applies probabilistic neural networks to identify the most likely values of the principal components when a given soil reclamation treatment is applied. Once the optimal ranges for a successfully reclaimed soil have been defined, the developed numerical methods are applied, defining an area considered optimal in the principal component diagram. The most appropriate treatment is considered to be the one most likely to occupy the optimal area. The methodology was applied to a dystrophic Red Oxisol degraded by the construction of the Ilha Solteira Hydroelectric Plant in Brazil, where a long‐term experiment with two tree species ( Eucalyptus urograndis and Mabea fistulifera ) and different doses of organic and mineral fertilization was established in 2010. The results show that the combination of M. fistulifera with the application of 20 Mg·ha −1 of compost significantly improves organic matter, porosity, and cation exchange capacity in the surface soil horizons, generating more favorable conditions for plant growth in the long term. Beyond the specific results, this multivariate model represents a useful tool for evaluating the effectiveness of long‐term soil restoration programs, providing objective criteria that can guide decision‐making in projects for ecological recovery and sustainable management of degraded lands.