How s-kganz/forestlst works
Main System Components
a) Data Acquisition Layer
• External data sources:
• Aerial Detection Surveys (ADS)
• Earth Engine data
• VODCA data
• Daymet data
• MTBS (Monitoring Trends in Burn Severity)
• NLCD (National Land Cover Database)
• Terrain data
b) Data Processing Layer
• Two parallel pipelines:
• CA_MORT Pipeline (4km resolution)
• WEST_MORT Pipeline (1km resolution)
• Data cleaning and transformation steps
c) Model Development Layer
• Multiple model implementations:
• Neural Networks (PyTorch and TensorFlow)
• AutoML components
• GAM (Generalized Additive Models)
• Preisler CNN implementation
d) Analysis & Evaluation Layer
• Computational Time Analysis
• SHAP Analysis
• Model Assessment
• Training Benchmarks
Technology Stack Indicators
• Python ecosystem (main implementation)
• R (for GAM analysis)
• GDAL (for geospatial processing)
• Cloud Services:
• Google Cloud Storage
• Earth Engine
• Earth Access
Storage Components
Different storage formats:
• NetCDF files
• TFRecord files
• Zarr arrays
• CSV files
• Geodatabases
Specific Implementation Details
Create subsections showing:
• Data resolution differences (4km vs 1km)
• Processing pipeline steps
• Model architecture variations
• Utility functions and their relationships
Important Notes
Callouts for:
• Authentication requirements
• Environmental dependencies
• Resource requirements
• Processing sequence dependencies
Key Considerations:
• Modular nature of the utility functions
• Clear separation between data preparation and model training
• Different scales of operation between CA_MORT and WEST_MORT
• Feedback loops for model optimization and evaluation
• Integration points with external services and data sources
This structure will provide a comprehensive view of the system while maintaining clarity about the project's complex data processing and modeling workflows.