Avalanche Detection GeoTIFF Generation

I have 200 geo-referenced avalanche samples already validated on Sentinel imagery. Within the next two days I need a basic, from-scratch machine-learning workflow that can locate additional avalanches in Sentinel data from 2016 through 2025 and export each detection as a GeoTIFF layer.

Scope
• Train a simple but effective model on the 200 labelled samples.
• Run the model across the specified years of Sentinel scenes (same Area of Interest and spectral bands I used for the samples).
• Save every positive detection as a single-band GeoTIFF with the original projection preserved.

Key points
• No pre-trained weights are available, so please build the model pipeline yourself.
• Accuracy is important, yet I understand this is an initial, low-budget pass: focus on a clean, reproducible codebase (Python, Rasterio, GDAL, NumPy, scikit-learn or a lightweight deep-learning library) and solid avalanche masks rather than exhaustive hyper-parameter tuning.
• Deliver within 48 hours.

Deliverables
1. Python code/notebook with clear instructions.
2. Trained model file.
3. GeoTIFFs of detected avalanches (2016-2025).

A short demo GIF or screenshot proving the detections load correctly in QGIS would be a plus.