I need help turning recent Sentinel-2 scenes into an actionable forest-health product. The workflow is already defined: compute NDVI, SAVI, and EVI from the pre-processed imagery, feed those layers into a Random Forest model, and generate a binary classification—healthy versus stressed vegetation. Please deliver the resulting health map as both GeoTIFFs and matching Shapefiles so I can load them directly in ArcGIS or QGIS. A brief visual report that summarises the confusion matrix and highlights the main stress hotspots will round off the job nicely. Python, Rasterio, NumPy and scikit-learn sit at the core of this task, so experience with those libraries (or equivalent geospatial tools) is essential. If everything runs smoothly and the model meets reasonable accuracy benchmarks, the project is complete.