Towards Transferable Soil Property Estimation from Multi-Source Spectral Data

Soil properties such as soil organic carbon (SOC) and clay content are key indicators of ecosystem functioning, land degradation, and environmental change. Advances in spaceborne hyperspectral remote sensing enable new possibilities for large-scale soil property monitoring. However, differences in sensor characteristics, acquisition conditions, and surface heterogeneity continue to limit the transferability of retrieval models across regions and observation systems. This study investigates the role of spectral preprocessing in improving the transferability of soil property estimation using multi-source spectral data. We evaluate continuum removal (CR) and first-derivative (FD) transformations to improve the interpretability and alignment of diagnostic soil absorption features in laboratory and satellite reflectance spectra. Using different spectral datasets, we assess the impact of preprocessing on feature comparability, predictive performance, and robustness under varying data distributions. We further examine how spectral heterogeneity and distribution shifts influence model generalization. Our results demonstrate that robust preprocessing improves the comparability of spectral features and strengthens model transferability. These findings highlight the importance of sensor-independent preprocessing strategies for reliable and scalable soil property mapping using multi-source remote sensing data in environmental monitoring applications.