Influences of noise on paddy chlorophyll mapping and associated uncertainty across heterogeneous landscapes using Sentinel-2 and hybrid models
Satellite-derived crop functional traits are increasingly pivotal for effective agricultural monitoring and informed decision-making, satisfying the demands of sustainable, smart, and precision agriculture. However, retrieving these traits leveraging multispectral derivatives is challenging due to i...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1574954125003607 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Satellite-derived crop functional traits are increasingly pivotal for effective agricultural monitoring and informed decision-making, satisfying the demands of sustainable, smart, and precision agriculture. However, retrieving these traits leveraging multispectral derivatives is challenging due to inherent variability and uncertainties, highlighting the need for comprehensive quantification of the ill-posedness problem arising from both input data and predicted methods. This study investigated impacts of noise on paddy rice chlorophyll mapping and the associated pixel-level uncertainty across heterogeneous agricultural landscapes using Sentinel-2 imagery and hybrid models. Specifically, we employed Gaussian Process Regression (GPR) and Variational Heteroscedastic Gaussian Process Regression (VHGPR) to train models utilizing simulated data generated by the physically-based radiative transfer model PROSAIL. Subsequently, noise levels of 0 %, 5 %, and 10 % were incorporated to assess their effects, and pixel-level uncertainty was estimated across different land covers. Our findings revealed the influence of noise on model performance, band relevance, and associated uncertainty. The index of agreement (IOA) varying from the lowest 0.82 at 10 % noise to the highest 0.92 in the absence of noise. With relative root mean square errors (RRMSE) below 5 %, all models provided high-fidelity mapping of paddy chlorophyll content (mean relative uncertainties below 22 %), compared to the lower confidence levels observed for non-vegetated objects (mean relative uncertainties exceeding 50 %). This research contributes to the growing need for uncertainty mapping associated with plant trait products. Furthermore, it highlights the importance of noise management in satellite applications for achieving more robust and realistic retrievals across heterogeneous landscapes. |
|---|---|
| ISSN: | 1574-9541 |