Predicting Soil Nutrient Classes Using Vis–NIR Spectroscopy to Support Sustainable Farming Decisions
Article Swipe
YOU?
·
· 2025
· Open Access
·
· DOI: https://doi.org/10.1002/ldr.70007
· OA: W4412136445
Efficient nutrient management is crucial for sustainable agriculture and ecosystem health. Traditional approaches to nutrient recommendations often rely on soil nutrient classes rather than specific numerical values. Visible–Near‐Infrared (Vis–NIR) spectroscopy offers a rapid, nondestructive method for predicting soil properties. This study explores the use of Vis–NIR spectroscopy to classify soil nutrient levels for improved agricultural decision‐making. A dataset of 216 soil samples, collected from diverse land uses in the Gummlapalli subwatershed, Karnataka, India, was analyzed for 11 soil properties, including pH, soil organic carbon (SOC), macronutrients (N, P 2 O 5 , K 2 O, S), and micronutrients (B, Cu, Fe, Mn, Zn). The Partial Least Squares Regression (PLSR) model, enhanced with Savitzky–Golay (SG) smoothing and Standard Normal Variate (SNV) transformation, showed variable prediction performance ( R 2 : 0.04 for K 2 O to 0.70 for pH). Two approaches for nutrient classification were evaluated: (1) indirect classification based on predicted soil properties and (2) direct classification using Partial Least Squares Discriminant Analysis (PLS‐DA). To address class imbalance during classification, the Synthetic Minority Over‐sampling Technique (SMOTE) was employed. Direct classification outperformed the indirect approach, achieving higher overall accuracy (OA) for key properties, including pH (0.72), SOC (0.61), P 2 O 5 (0.84), K 2 O (0.50), and Cu (0.96). These results underscore the reliability of direct classification for assessing soil nutrient classes. This study highlights the potential of Vis–NIR spectroscopy as a robust tool for soil nutrient classification, enabling precise fertilizer recommendations, supporting sustainable farming practices, and promoting ecosystem sustainability.
