Researchers have developed a new method for mapping tree species in Denmark by comparing traditional spectral-temporal features with embeddings from geospatial foundation models like TESSERA and AlphaEarth. The study found that a multilayer perceptron classifier using spectral-temporal features achieved the highest performance, with a macro F1 score of 0.843 for pure stands and 0.653 for mixed stands. However, TESSERA embeddings showed a significant advantage when training data was limited, outperforming spectral-temporal features with less than 25% of available plots. The best-performing model was then used to create the first high-resolution national tree species map of Denmark, offering a valuable resource for ecological research and land management. AI
IMPACT Demonstrates the potential of foundation models for specialized geospatial tasks, offering advantages in data-scarce scenarios.
RANK_REASON Academic paper detailing a new methodology for geospatial analysis using foundation models. [lever_c_demoted from research: ic=1 ai=1.0]
- Alkiviadis Koukos
- AlphaEarth
- Denmark
- multilayer perceptron
- National Forest Census
- random forest
- Sentinel-1
- Sentinel-2
- TESSERA
- XGBoost
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
