Data Preprocess
This section explain preprocessing guideline in this project for each dataset.
The data used for development consists of image data from 3 satellites. Sentinel-1, Sentinel-2 from ESA and GEDI from NASA. Sentinel-1 and Sentinel-2 serve as predictor variables and GEDI serves as a ground truth reference,
Sentinel 1
Input of the model : Model will inference AGB from pattern and relation in each band.
Using raw Band : Relation of the band related to AGB will extract from model in training process.
Use in training and inference process
Stack with sentinel 2
Temporal coverage for 6 month to keep seasonal information consistent.
Interferometric Wide Swath
Ascending : VV + VH total 2 layer
Median Composite
Thermal noise removal
Sentinel 2
Input of the model : Model will inference AGB from pattern and relation in each band.
Using raw Band : Relation of the band related to AGB will extract from model in training process.
Stack with sentinel 1
Use in training and inference process
For Training
Temporal coverage for 6 month to keep seasonal information consistent.
Filter cloud < 50 %
Median Composite
For Inference
- Use sentinel2 image with minimum cloud coverage
GEDI
- Target reference of the model. Use L4A layer to represent above ground biomass density .