Global 30-m annual median vegetation height maps (2000-2022) based on ICESat-2 data and Machine Learning.

Area of Science:

• Ecology and Remote Sensing
• Geospatial Analysis
• Environmental Monitoring

Background:

• Accurate vegetation height measurement is vital for ecosystem assessment, carbon storage, and biodiversity studies.
• Existing global height models primarily focus on forests, neglecting crucial data for herbaceous and shrub ecosystems.
• A gap exists in global vegetation structure data for non-forest environments.

Purpose of the Study:

• To develop the first global estimate of median vegetation height for all ecosystems, including short vegetation.
• To provide annual vegetation height data at a 30 m resolution from 2000-2022.
• To enhance the detail and heterogeneity of vegetation height data in global models.

Main Methods:

• Utilized ICESat-2 satellite Lidar and Landsat cloud-free composites, alongside other Earth Observation raster data.
• Developed ensemble Gradient Boosted Tree (GBT) models using 32 million ICESat-2 segments across 10 independent draws.
• Estimated 90% prediction intervals to quantify model uncertainty.

Main Results:

• Achieved a root mean square error (RMSE) of 2.35 m, R² of 0.515, and D² regression score of 0.62 on the testing set.
• Generated annual global median vegetation height maps at 30 m resolution for the period 2000-2022.
• Demonstrated increased detail and heterogeneity in short vegetation ecosystems compared to existing products.

Conclusions:

• The developed model successfully provides a comprehensive global estimate of median vegetation height, filling a critical data gap.
• The approach enhances the characterization of vegetation structure in non-forest ecosystems.
• Output maps, reference samples, and trained models are publicly available under a CC-BY license to support further research.