Vis-NIR Spectroscopy and Machine Learning for Estimating Soil Nutrient Quality Index and Chemical Properties in Remote Cacao-Growing Regions of Colombia Documento de conferencia uri icon

Abstracto

  • This study presents an innovative system integrating Vis-NIR spectroscopy and Machine Learning regressors to estimate soil chemical properties and a soil quality index in remote cacao-growing regions of Colombia. By incorporating expert evaluations, the developed soil nutrient quality index is practical and actionable, offering farmers a cost-effective and environmentally friendly alternative to traditional soil analysis. This study analyzes 138 soil samples from Nariño and 284 from Norte de Santander—two cacao-producing, hard-to-access regions in Colombia. The data were evaluated separately for each region and also in combination. Key chemical soil properties —such as pH, organic matter, calcium, magnesium, iron, cation exchange capacity, and the composite soil nutrient quality index—were predicted using regression models including Partial Least Squares Regression, Support Vector Regression, Least Absolute Shrinkage and Selection Operator, and Cubist. Model performance was optimized using cross-validation and hyperparameter tuning. Results highlight the effectiveness of Cubist model, with R2 values as high as 0.912 for calcium in Nariño and 0.831 in the combined dataset. The study also estimated the soil nutrient quality index (SNQI) with an R2 of 0.583. This approach demonstrates great potential for improving decision-making in sustainable cacao cultivation. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

fecha de publicación

  • enero 1, 2026

Número Estándar Internacional de Libros (ISBN) 13

  • 978-303223160-4

Última página

  • 74

Volumen

  • 16528 LNCS

Cuestión

Fecha / hora

  • enero 1, 2026