Small-sample-data augmentation and transfer strategies for forest cover change monitoring
The Qilian Mountains serves as a critical ecological barrier in northwest China, where the forest coverage strongly connected with the regional ecosystem stability, water conservation as well as climate change. However, a high resolution and accuracy of forest coverage data is still missing for this...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Ecological Indicators |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1470160X25008003 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The Qilian Mountains serves as a critical ecological barrier in northwest China, where the forest coverage strongly connected with the regional ecosystem stability, water conservation as well as climate change. However, a high resolution and accuracy of forest coverage data is still missing for this region due to the scarce of sample data, which leads to the large discrepancy of existing forest coverage products. Here, we are aiming to develop a forest mapping approach, which will adopt the augmentation and transfer strategy of small sample to relief the sample scarcity problem. By integrating small-sample data from field observations and expert visual interpretation, as well as utilizing Dempster-Shafer (D-S) evidence theory to assimilate multi-source land cover datasets, a high-quality large-sample label database for the Qilian Mountains was constructed. The large sample data were used to develop the 30-meter annual forest cover dataset (AFD_QLM) from 1986 to 2023, using a locally adaptive machine learning algorithm at 1°×1° grid cells. This approach offers an effective technical framework for forest mapping in regions with scarce sample data and complex terrain. We found that: (1) The AFD_QLM dataset achieved a significantly higher overall accuracy (OA = 0.951) than existing forest cover products, and captured forest dynamics and spatiotemporal patterns in the Qilian Mountains; (2) AFD_QLM excelled in monitoring fine forest details in differentiating between forest and shrubland; and (3) The forest area within the study region showed a stable growth trend, primarily driven by afforestation activities along the southeastern forest edges. This study highlights the usefulness of small-sample data augmentation and transfer strategy in addressing the issue of sample scarce. The forest data, generated in this study, could serve as a robust data foundation for ecological protection and sustainable management in the Qilian Mountains. |
|---|---|
| ISSN: | 1470-160X |