Dataset for: GLI-Fusion: A Low-Cost Strategy Integrating Geographical Location Identifiers with Residual Models to Enhance Soil Nitrogen Prediction using Near-Infrared Spectroscopy

This dataset contains the soil visible-near infrared (Vis-NIR) spectral data, laboratory-measured soil total nitrogen (TN) contents, and corresponding Geographical Location Identifiers (GLI) used in the aforementioned study. The dataset is designed to evaluate the performance of the GLI-Fusion deep learning framework, which leverages zero-cost spatial auxiliary features (GLI) to improve the Vis-NIR prediction accuracy of soil TN across different spatial scales.

The repository includes data from two distinct spatial scales: a regional small-sample dataset (Nanhu, China) and a processed macro-scale dataset (LUCAS 2015, Europe).

Data Files Overview:

1. Nanhu.xlsx
Contents: It includes the sample ID, Geographical Location Identifier (GLI, representing the specific village of the sampling point), Vis-NIR spectral absorbance data (900–1700 nm, 227 bands), and the corresponding laboratory-measured soil total nitrogen (TN) content (g/kg).

2. Lucas2015.xlsx
This file contains a processed subset of the public European LUCAS 2015 topsoil dataset, adapted to validate the cross-regional generalization of the GLI-Fusion model.
Contents: To ensure comparability with the regional Nanhu dataset, the original high-resolution LUCAS spectra were linearly interpolated to match the specific 900–1700 nm wavelength range (retaining 227 bands). It also includes the corresponding physicochemical values, specifically soil TN content.

3. Lucas-GLI.xlsx
This file contains the spatial auxiliary features for the LUCAS dataset.
Contents: It provides the country-level Geographical Location Identifiers (GLI) for each sample in the LUCAS 2015 dataset.
Usage Note: The row order in this file strictly corresponds to the sample sequence in Lucas2015.xlsx. Users can easily merge this GLI column with the spectral data in Lucas2015.xlsx via row-wise concatenation for multi-modal machine learning tasks.