Spatial Prediction in Agricultural Landscapes of Soil Nutrients Using NIR Spectroscopy and Advanced Regression Models for Precision Agriculture in Rice Paddies, Tomato Fields, and Montado Ecosystems

Soil fertility is a critical factor for sustainable agricultural production, yet traditional soil analysis methods are often time-consuming and costly. As such, the aim of our research was to evaluate the potential of near-infrared (NIR) spectroscopy as a rapid, cost-effective, and non-destructive tool to assess soil fertility properties in agricultural soils. The soil samples were collected from the Quinta da Foz located in Benavente (Portugal) across agricultural regions including rice paddies, tomato fields, and Montado ecosystems. The sampling locations were guided by the Normalized Difference Vegetation Index (NDVI) to capture spatial variability, and soil analyses included near-infrared (NIR) spectral measurements and laboratory-based chemical determinations of soil fertility parameters (pH, electrical conductivity, total carbon and nitrogen, organic matter, macro- and micronutrients) as well as multiple soil carbon fractions. Two predictive modeling approaches (Random Forest (RF) and Partial Least Squares Regression (PLSR)) were developed to estimate soil chemical properties from spectral data. The RF models consistently outperformed PLSR, achieving high accuracy (R 2 = 0.85) for nutrients such as Mg, Fe, Ni, Ca, and Na and organic matter. A moderate predictive performance (R 2 between 0.70 and 0.80) was observed for different elements, namely, K and Mn. On the other hand, P, S, Zn, electrical conductivity, pH, total N, and various carbon fractions were poorly predicted. The spatial interpolation of predicted values enabled the generation of soil fertility maps that informed site-specific nutrient management. The results indicate that NIR spectroscopy combined with robust modeling offers a promising approach for rapid spatial assessment of selected soil nutrients, supporting precision agriculture and sustainable land management.