PolSAR Forest Height Estimation Enhancement With Polarimetric Rotation Domain Features and Multivariate Sensitivity Analysis
Forest height is a critical indicator of forest health and can directly influence the carbon storage capacity of ecosystems. Polarimetric features extracted from polarimetric synthetic aperture radar play a crucial role in forest height estimation. Typical polarimetric features, such as amplitude fe...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11108236/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Forest height is a critical indicator of forest health and can directly influence the carbon storage capacity of ecosystems. Polarimetric features extracted from polarimetric synthetic aperture radar play a crucial role in forest height estimation. Typical polarimetric features, such as amplitude features and polarimetric decomposition features, are susceptible to the influence of target scattering diversity, often leading to reduced interpretation performance. Advanced polarimetric rotation domain features effectively utilize the rich information embedded in target scattering diversity; however, there is a lack of research analyzing their sensitivity and application potential for forest height estimation. In addition, a univariate sensitivity metric is insufficient to comprehensively evaluate the contribution of polarimetric features to forest height estimation. In this study, we investigate the effectiveness of several typical polarimetric features and advanced polarimetric rotation domain features in forest height estimation. First, we propose a multivariate sensitivity analysis (MSA) method, which uses four metrics to comprehensively assess the sensitivity of all polarimetric features to forest height across different dimensions and to perform feature selection. Then, we propose a Bayesian-optimized ensemble learning algorithm to improve the accuracy of forest height estimation. Finally, various combinations of polarimetric features are used for modeling comparison. The results demonstrate the following: MSA can effectively select polarimetric features that contribute more significantly to forest height modeling; compared to typical polarimetric features, polarimetric rotation domain features exhibit higher sensitivity to forest height; and integrating polarimetric rotation domain features with typical polarimetric features achieves a complementary effect, further enhancing forest height estimation. |
|---|---|
| ISSN: | 1939-1404 2151-1535 |