Improving ICESat-2 photon classification and tree height estimation using Moran's I and machine learning
The Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) provides valuable data for vegetation mapping using photon counting lidar (PCL) technology. However, its ATL08 data product, designed for canopy height and terrain classification, exhibits classification inaccuracies due to algorithm limitati...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Science of Remote Sensing |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666017225000574 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) provides valuable data for vegetation mapping using photon counting lidar (PCL) technology. However, its ATL08 data product, designed for canopy height and terrain classification, exhibits classification inaccuracies due to algorithm limitations and noise contamination. This study aimed to address these challenges by leveraging local spatial autocorrelation, Moran's I, as a feature input in machine learning methods to enhance photon classification accuracy. Random Forest models were developed and compared, with one model incorporating Moran's I to capture spatial patterns. The study covered 12 diverse ecoregions across the United States, including conifer forests, broadleaf forests, and savannas. Canopy heights derived from different models were validated against Airborne Laser Scanning (ALS) data. The results demonstrated that the model incorporating Moran's I improved classification accuracy, with R2 values ranging from 0.30 to 0.76 across ecoregions. Top of Canopy (TOC) classifications in dense forests, such as those in South Carolina, achieved higher agreement with ALS data, whereas sparse environments like Louisiana savannas exhibited lower accuracy. This study highlights the importance of incorporating spatial autocorrelation measures in machine learning workflows to improve vegetation classification, which can be beneficial for more accurate ecological assessments using ICESat-2 data in diverse landscapes. |
|---|---|
| ISSN: | 2666-0172 |