Dataset of Fourier-transform near-infrared (FT-NIR) spectral data for prediction of DTPA-extractable copper, iron, magnesium, and zinc in agricultural soil
Description
This dataset presents the Fourier-transform near-infrared (FT-NIR) spectroscopic data obtained from agricultural topsoil using a Thermo Scientific Antaris II spectrometer, along with the reference measurements and the calibration and validation of prediction models for DTPA-extractable copper (Cu), iron (Fe), magnesium (Mg), and zinc (Zn). We collected diffuse-reflectance spectral data from 731 topsoil samples taken at a depth of 0–20 cm from 38 cities (districts and prefectures) across 16 provinces of China, comprising 180, 180, 180, and 191 paired samples for Cu, Fe, Mg, and Zn, respectively. The acquisition process started by air-drying the soil and gently crushing and sieving it, and the ground fraction (<0.25 mm) was used for spectroscopic and chemical analysis. Spectra were recorded over 4000–10,000 cm⁻¹ (1000–2500 nm) at 8 cm⁻¹ resolution, yielding 1,557 variables per sample, and reference contents were determined by DTPA–TEA extraction followed by atomic absorption spectrometry. Sixteen preprocessing variants, including raw spectra, Savitzky–Golay smoothing, multiplicative scatter correction (MSC), standard normal variate (SNV), the first and second derivatives, and their combinations, were used with partial least squares (PLS) regression to create the prediction models, together with competitive adaptive reweighted sampling (CARS) for wavelength selection. Eighty percent of each sample set was divided into calibration and the remaining twenty percent was validation. Our results demonstrate the effectiveness of these models, with validation R² of 0.83 for Cu, 0.90 for Fe, 0.92 for Mg, and 0.89 for Zn. The FT-NIR dataset spanning 4000–10,000 cm⁻¹ provides a ready-to-use benchmark for developing portable/handheld soil-nutrient sensors, with potential for further accuracy improvements through wavelength selection and model refinement. The dataset includes the raw spectral library with paired DTPA-extractable contents, the complete PLS and CARS modelling results, descriptive statistics, and the figures.
Files
Steps to reproduce
Researchers can improve model accuracy by applying advanced preprocessing methods and both linear and nonlinear modeling techniques to this NIR spectral data and wet chemistry results or add this data to others models because utilizing NIR spectroscopy to predict soil properties necessitated a comprehensive and diverse dataset of soil samples to develop a accuracy predictive model.
Institutions
- Central South University of Forestry and TechnologyHunan, Changsha