| Property | Value |
|---|---|
| Framework | PyTorch |
| License | Apache-2.0 |
| Paper | Research Paper |
| Downloads | 1,040 |
What is granite-geospatial-canopyheight?
The granite-geospatial-canopyheight is an advanced AI model designed to predict canopy heights from satellite imagery. It utilizes the Harmonized Landsat and Sentinel-2 (HLS) L30 optical satellite imagery and has been fine-tuned using training data from the Global Ecosystem Dynamics Investigation (GEDI) L2A across 15 different global biomes.
Implementation Details
The model is built on a Swin-B transformer backbone, chosen for its superior characteristics including smaller patch size for higher resolution, efficient windowed attention, and hierarchical merging capabilities. It employs SimMIM self-supervised learning for pretraining and uses a UPerNet adapted for pixel-wise regression during fine-tuning.
- Built using Terratorch framework
- Implements windowed attention for computational efficiency
- Uses UPerNet with Pixel Shuffle layers for upscaling
- Trained on NASA's HLS L30 and GEDI L4A datasets
Core Capabilities
- Accurate prediction of tree and vegetation heights
- Carbon cycle quantification
- Crop yield estimation
- Forest timber production monitoring
- Carbon sequestration measurement
Frequently Asked Questions
Q: What makes this model unique?
This model uniquely combines Swin-B transformer architecture with UPerNet for regression, trained across 15 global biomes, making it highly versatile for different geographical regions. Its ability to process satellite imagery for canopy height prediction makes it valuable for environmental monitoring and forest management.
Q: What are the recommended use cases?
The model is ideal for environmental scientists, forest managers, and agricultural researchers who need to monitor vegetation height, assess carbon sequestration, estimate crop yields, or study forest dynamics using satellite imagery.