Code and data for: Depth-stratified modelling improves hyperspectral prediction of soil organic carbon content

This Zenodo repository contains the data and analysis code accompanying the manuscript submitted to SOIL (Copernicus Publications): Depth-stratified modeling improves hyperspectral prediction of soil organic carbon content: a comparison with integrated modeling on the Loess Plateau, China. This repository includes: (i) Measured soil organic carbon (SOC) contents for 388 soil samples collected from 97 sampling sites in Changzhi City, Shanxi Province, China, across four soil depth layers: 0–20, 20–40, 40–60, and 60–80 cm; (ii) Corresponding visible–near-infrared (Vis–NIR) reflectance spectra (350–2500 nm, 1 nm resolution) measured using an ASD FieldSpec 4 spectrometer; (iii) Complete Python source code for the full preprocessing and modeling pipeline, including Savitzky–Golay filtering, Kennard–Stone splitting, Pearson correlation-based sensitive band selection, PLSR, GRID‑SVR, and PSO‑SVR. See README.md for installation and replication instructions.