A double-layer ensemble framework for rubber plantation mapping using multi-source data in the google earth engine: a case study of the southwestern border region of China
The remote sensing mapping of rubber plantations still faces significant challenges in terms of classification accuracy and the availability of remote sensing data, particularly in areas with high landscape fragmentation and tropical cloud and rainfall interference. To address these issues, we propo...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-08-01
|
| Series: | International Journal of Digital Earth |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/17538947.2025.2520472 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The remote sensing mapping of rubber plantations still faces significant challenges in terms of classification accuracy and the availability of remote sensing data, particularly in areas with high landscape fragmentation and tropical cloud and rainfall interference. To address these issues, we propose a double-layer ensemble framework for rubber plantation extraction using multi-source data in the Google Earth Engine. The first layer is the Phenological Four-Temporal Ensemble Model (PFT-EM), which leverages the capability of machine learning to incorporate full-season image inputs, mitigating the limitations posed by data quality during specific phenological periods. This layer utilizes five machine learning algorithms, namely Random Forest, Maximum Entropy Model, Gradient Tree Boosting, Support Vector Machine, and Classification and Regression Tree, to construct the corresponding PFT-EMs. The second layer is the Stacking Ensemble Model, which integrates information from various types of PFT-EMs to enhance the overall accuracy and generalization capability. The proposed framework was validated in the southwestern border region of China, achieving a mean F1-score of 0.932 and a mean overall accuracy of 91.25%. The results demonstrate that this framework effectively mitigates the limitations of image availability during specific periods and enables the accurate extraction of rubber plantations across large areas. |
|---|---|
| ISSN: | 1753-8947 1753-8955 |