Exploiting soil spectroscopy in the VNIR-SWIR range to estimate soil organic carbon stock: what role can sampling density and spacing play?
Journal of Agricultural Engineering · 23.06.2026 · 57 (4)
DOI: 10.4081/jae.2026.1995 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Agricultural Engineering, online erschienen am 23.06.2026. Verfasst von Caterina Mazzitelli und 5 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Soil organic carbon stock (SOCS) is a key indicator of soil fertility, ecosystem functioning, and climate change mitigation, yet its direct measurement remains labor-intensive, time-consuming, and costly. Visible, near-infrared, and short-wave infrared (VNIR-SWIR) soil reflectance spectroscopy offers a rapid and cost-effective alternative for SOCS estimation, although its predictive performance may strongly depend on sampling design and landscape heterogeneity. This study investigated the extent to which sampling density, spatial scale and spacing affect the accuracy of spectroscopy-based SOCS prediction. An ensemble modeling framework integrating partial least squares regression (PLSR), random forest (RF), and artificial neural networks (ANN), combined with five spectral pre-processing techniques, was applied to two contrasting datasets collected in southern Italy. The first dataset represented the entire heterogeneous region of Campania (CAM), comprising 2,957 topsoil samples collected on a sparse irregular grid with spacing ranging from 1 to 4 km. The second dataset corresponded to a local experimental field (MFC2, 8 ha), where 135 topsoil samples were collected on a dense regular grid with 25 m spacing. Model performance was evaluated using independent validation datasets through the coefficient of determination (R²), root mean square error (RMSE), and residual predictive deviation (RPD). Marked differences emerged between spatial scales. A fair predictive performance was achieved at MFC2 (best model: R² = 0.65; RPD = 1.71), whereas all models performed poorly at the regional CAM scale (best model: R² = 0.24; RPD = 1.14). Variogram analysis showed greater unresolved spatial variability (nugget effect) in the CAM dataset, indicating that the sparse regional sampling design failed to adequately capture fine-scale SOCS heterogeneity. To further investigate the role of sampling density, a denser subset of 130 CAM soil samples, with approximately 1 km spacing, was analyzed. Although model performance improved moderately (best R² = 0.35), predictive accuracy remained unsatisfactory (RPD
Zitieren
APA
Mazzitelli, C., Romano, N., Ben-Dor, E., Jonge, L. W. D., Hermansen, C., & Nasta, P. (2026). Exploiting soil spectroscopy in the VNIR-SWIR range to estimate soil organic carbon stock: what role can sampling density and spacing play?. Journal of Agricultural Engineering, 57(4). https://doi.org/10.4081/jae.2026.1995
Vancouver
Mazzitelli C, Romano N, Ben-Dor E, Jonge LWD, Hermansen C, Nasta P. Exploiting soil spectroscopy in the VNIR-SWIR range to estimate soil organic carbon stock: what role can sampling density and spacing play?. Journal of Agricultural Engineering. 2026;57(4). doi:10.4081/jae.2026.1995
BibTeX
@article{Mazzitelli2026,
author = {Mazzitelli, Caterina and Romano, Nunzio and Ben-Dor, Eyal and Jonge, Lis Wollesen de and Hermansen, Cecilie and Nasta, Paolo},
title = {Exploiting soil spectroscopy in the VNIR-SWIR range to estimate soil organic carbon stock: what role can sampling density and spacing play?},
journal = {Journal of Agricultural Engineering},
year = {2026},
volume = {57},
number = {4},
doi = {10.4081/jae.2026.1995}
}
EUVIMED