How docampo16/frayjorge works
Project type and purpose
• This is not a traditional multi-tier web app but a data-science workflow implemented entirely as Jupyter notebooks. The goal is to perform land-use and land-cover (LULC) classification for Fray Jorge National Park, using Sentinel-2 (s2) imagery and clustering/classification techniques.
File structure
• Top-level files are all notebooks (.ipynb), each covering a stage or variant of the LULC pipeline:
• s2workflow.ipynb – likely ingestion and preprocessing of Sentinel-2 data
• NSClustering_clean.ipynb / NonSupervisedClustering.ipynb – unsupervised clustering experiments
• LULC.ipynb / FJ_LULC.ipynb – supervised classification or final mapping
• No separate ‘src’ or ‘backend’ directories: the logic lives in notebooks.
• Standard .gitignore and README.md show this is a research/code-prototype repo.
Extract insights from the README
• Project name: FrayJorge.
• Purpose: Land Use and Land Classification for Fray Jorge National Park.
• No prescribed architecture is described, so we must infer a pipeline workflow.
Main components and their interactions
a. Data Ingestion
• External data source: Sentinel-2 satellite imagery (and any ancillary shapefiles or DEMs).
• Notebook: s2workflow.ipynb reads raw raster tiles.
b. Data Preprocessing
• Tasks: atmospheric correction, cloud masking, band stacking, cropping to park boundary.
• Output: clean, analysis-ready raster layers.
c. Feature Extraction
• Compute indices (NDVI, EVI, etc.), dimensionality reduction if any.
• Could happen in the same or a separate notebook.
d. Unsupervised Clustering
• Notebooks: NonSupervisedClustering.ipynb / NSClustering_clean.ipynb.
• Apply k-means, DBSCAN, etc., to segment land cover.
• Output: cluster maps.
e. Supervised Classification
• Notebooks: LULC.ipynb / FJ_LULC.ipynb.
• Train ML models (Random Forest, SVM) using ground-truth labels.
• Produce final classified map.
f. Validation & Visualization
• Generate accuracy metrics, confusion matrices, and final map overlays.
g. Reporting / Export
• Export GeoTIFFs, PNGs, or interactive maps for sharing.
Architectural patterns and design principles
• Linear data-pipeline pattern: each notebook represents a stage.
• Reproducibility: notebooks + fixed environment (requirements.txt or conda env).
• Modular: if scaled up later, you could turn notebooks into Python modules/services.
Technologies & libraries
• Jupyter Notebook (presentation & orchestration)
• Python geospatial stack: rasterio/GDAL, numpy/pandas, scikit-learn, possibly geopandas, matplotlib/folium
• Sentinel-2 data access via API (e.g., Copernicus Open Access Hub)
• Optional: conda or virtualenv for environment management
Diagrammatic conventions
• Keep it to a single A4/slide-size view but allow zoom-in on complex steps (e.g., a zoom-callout for clustering).