Soil Mapping with a Limited Number of Samples by Coupling EMI and NIR Spectroscopy in Hazelnut Tree Orchard
Author(s)
Date Issued
2024
Type
conferenceObject
Start Page
604
End Page
604
Abstract
Precision agriculture relies on high-detail soil maps to optimize resource use. Traditional soil proximal
mapping methods, such as EMI sensing, require a certain number of soil sampling and laboratory analyses
to predict maps of soil characteristics. VIS-NIR and NIR diffuse reflectance spectroscopy offer rapid and low-
cost alternatives, allowing for increased datapoints and better map prediction accuracy.
The aim of this work is to test and optimize a methodology for high-detail soil mapping in a hazelnut grove
of approximately 4 ha, located in Corte Migliorina farm (Southern Tuscany, Italy), using both EMI proximal
sensor and NIR handheld spectrometer. The maps of ECa obtained by EMI provided the pattern of soil
spatial variability. Only 5 topsoil samples (Ap horizon, 0-30 cm) were collected for laboratory analysis,
following the maximum variability of ECa and elevation. In addition, other 40 topsoil samples (0-30 cm)
were collected by a regular grid and used for NIR spectroscopy to increase the number of datapoints within
the study field.
The spectrometer used for this work was the Neospectra Scanner (Si-Ware Systems, Menlo Park, USA), a low-
cost NIR spectrometer (1350-2500 nm) based on MEMS (micro-electromechanical systems) technology.
Partial Least Square Regression (PLSR) using a national spectral library, augmented by the 5 local samples
analyzed, was used to predict clay, sand, organic carbon (SOC), total nitrogen (TN), and cation exchange
capacity (CEC). The 40 datapoints with predicted soil variables were used for spatial interpolation, using ECa
map, elevation, and DEM derivatives as covariates. Two methods of interpolation were tested: Universal
Kriging (UK) and Regression Kriging (RK). The errors of predictive maps were calculated by 5 additional
points analyzed by conventional laboratory analysis. RK and UK showed similar accuracy, with lower
prediction errors for SOC and clay (R2>0.8) and slightly lower for TN (R2>0.5). Low accuracy was calculated
for sand and CEC mapping.
File(s)![Thumbnail Image]()
Name
IUSS2024_ABSTRACT BOOK (2).pdf
Size
3.51 MB
Format
Adobe PDF
Checksum (MD5)
95e61fe6faaf67fa2fed299aaf18b156
Conference(s)
Centennial Celebration and Congress of the International Union of Soil Sciences