Enhancing soil property retrieval in an Ali-Perudic Argosols–dominated region through multi-angle hyperspectral feature integration
5 affiliations
- 1State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, Chinese Academy of Sciences, Nanjing 210008, China; College of Geography and Environment, Shandong Normal University, Jinan 250358, China
- 2State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, Chinese Academy of Sciences, Nanjing 210008, China; Corresponding authors.
- 3College of Geography and Environment, Shandong Normal University, Jinan 250358, China; Corresponding authors.
- 4College of Geography and Environment, Shandong Normal University, Jinan 250358, China
- 5State Key Laboratory of Soil and Sustainable Agriculture, Institute of Soil Science, Chinese Academy of Sciences, Nanjing 210008, China
Abstract
Soil bidirectional reflectance distribution function (BRDF) is strongly influenced by soil physicochemical properties, yet most hyperspectral retrieval studies rely on single-view reflectance, leaving the value of multi-angle information insufficiently explored. In this study, bidirectional reflectance factors (BRFs) were measured at 48 viewing angles for 156 soil samples collected from a region dominated by Ali-Perudic Argosols. An optimal multi-angle retrieval framework was developed for different soil properties by systematically comparing spectral preprocessing methods, feature extraction, and modeling algorithms. The effects of viewing angle on retrieval accuracy and feature wavelength distribution were further analyzed, and the applicability of simulated multi-angle BRFs generated using SOILSPECT model was further assessed. Results show that the optimal framework for single-view reflectance combines Savitzky–Golay (SG) smoothing, a hybrid random frog–competitive adaptive reweighted sampling (RFROG–CARS) feature selection strategy, and nonlinear support vector machine (SVM) regression. Retrieval accuracy varied substantially with viewing geometry, and no universal optimal viewing angle was identified across soil properties. For multi-angle BRF data, a two-dimensional convolutional neural network (2D-CNN) feature extraction approach outperformed the RFROG–CARS wavelength selection method. Specifically, R 2 (RMSE) improved from 0.64 (8.8 g/kg) to 0.73 (7.7 g/kg) for SOM content, from 0.53 (6.9 %) to 0.61 (6.3 %) for sand content, and from 0.64 (4.7 %) to 0.69 (4.4 %) for silt content, while remaining comparable for cation exchange capacity (0.77/2.4 cmol/kg to 0.76/2.4 cmol/kg) and clay content (0.80/4.3 % to 0.79/4.3 %). Although the retrieval accuracy based on multi-angle BRFs remained slightly lower than that obtained from the optimal viewing geometry, it consistently outperformed most individual viewing angles. The selected multi-angle hyperspectral features were distributed across nearly all viewing angles, indicating complementary spectral–geometric information. Although retrieval accuracy based on simulated BRFs was lower than that based on measured BRFs, the results were broadly comparable. These findings support the use of multi-angle information in soil spectroscopy and remote sensing applications.
Related
- How Much Hyperspectral Information Does Chlorophyll Retrieval Really Need?
- Radiative Transfer Modelling of Ka-band Microwave Interactions with Snow-Covered Freshwater Ice: Implications for SWOT KaRIn Property Retrieval
- FireDataForge: A Unified Framework for Multi-Source Wildfire Data Retrieval and Integration
- Enhancing Tropical Cyclone Prediction Through AI Weather Model Ensemble Forecasts and Multi‐Source Data Integration
- Hyperspectral field and satellite radiometric matchups in Mediterranean oligotrophic waters
BibTeX
@article{zhou2026enhancing,
title = {{Enhancing soil property retrieval in an Ali-Perudic Argosols–dominated region through multi-angle hyperspectral feature integration}},
author = {Yufei Zhou and Lixia Ma and Xiaofei Wang and Lin Gao and Xianzhang Pan and Ziran Yuan and Changkun Wang and Dongsheng Yu},
journal = {International Journal of Applied Earth Observations and Geoinformation},
year = {2026},
month = oct,
doi = {10.1016/j.jag.2026.105646},
url = {https://doi.org/10.1016/j.jag.2026.105646},
note = {CC BY},
}
RIS
TY - JOUR
TI - Enhancing soil property retrieval in an Ali-Perudic Argosols–dominated region through multi-angle hyperspectral feature integration
AU - Yufei Zhou
AU - Lixia Ma
AU - Xiaofei Wang
AU - Lin Gao
AU - Xianzhang Pan
AU - Ziran Yuan
AU - Changkun Wang
AU - Dongsheng Yu
PY - 2026
DA - 2026/10/30
JO - International Journal of Applied Earth Observations and Geoinformation
DO - 10.1016/j.jag.2026.105646
UR - https://doi.org/10.1016/j.jag.2026.105646
AB - Soil bidirectional reflectance distribution function (BRDF) is strongly influenced by soil physicochemical properties, yet most hyperspectral retrieval studies rely on single-view reflectance, leaving the value of multi-angle information insufficiently explored. In this study, bidirectional reflectance factors (BRFs) were measured at 48 viewing angles for 156 soil samples collected from a region dominated by Ali-Perudic Argosols. An optimal multi-angle retrieval framework was developed for different soil properties by systematically comparing spectral preprocessing methods, feature extraction, and modeling algorithms. The effects of viewing angle on retrieval accuracy and feature wavelength distribution were further analyzed, and the applicability of simulated multi-angle BRFs generated using SOILSPECT model was further assessed. Results show that the optimal framework for single-view reflectance combines Savitzky–Golay (SG) smoothing, a hybrid random frog–competitive adaptive reweighted sampling (RFROG–CARS) feature selection strategy, and nonlinear support vector machine (SVM) regression. Retrieval accuracy varied substantially with viewing geometry, and no universal optimal viewing angle was identified across soil properties. For multi-angle BRF data, a two-dimensional convolutional neural network (2D-CNN) feature extraction approach outperformed the RFROG–CARS wavelength selection method. Specifically, R 2 (RMSE) improved from 0.64 (8.8 g/kg) to 0.73 (7.7 g/kg) for SOM content, from 0.53 (6.9 %) to 0.61 (6.3 %) for sand content, and from 0.64 (4.7 %) to 0.69 (4.4 %) for silt content, while remaining comparable for cation exchange capacity (0.77/2.4 cmol/kg to 0.76/2.4 cmol/kg) and clay content (0.80/4.3 % to 0.79/4.3 %). Although the retrieval accuracy based on multi-angle BRFs remained slightly lower than that obtained from the optimal viewing geometry, it consistently outperformed most individual viewing angles. The selected multi-angle hyperspectral features were distributed across nearly all viewing angles, indicating complementary spectral–geometric information. Although retrieval accuracy based on simulated BRFs was lower than that based on measured BRFs, the results were broadly comparable. These findings support the use of multi-angle information in soil spectroscopy and remote sensing applications.
SN - 1569-8432
SN - 1872-826X
ER -