Change detection script

[tamara-glazer]

Adding a script to calculate volumentric change detection for a given threshold between two timestamps (e.g., 2020 -> 2025).

Example usage:
1.  Compare growth between two COGs:
    python compute-change-clusters.py
        --start-cog /path/or/url/to/2020q2.tif
        --end-cog /path/or/url/to/2025q2.tif
        --change-percentile 0.95
        --output-gpkg growth.gpkg
        --output-layer clusters
        --min-cluster-pixels 8 # optional

2.  Compare decline between two COGs:
    python compute-change-clusters.py
        --start-cog /path/or/url/to/2020q2.tif
        --end-cog /path/or/url/to/2025q2.tif
        --change-percentile -0.95
        --output-gpkg decline.gpkg
        --output-layer clusters
        --min-cluster-pixels 8 # optional

3.  Compare tifs stored in a blob container, limit analysis to bbox provided:
    Expected directory structure:
        predictions_root / timestamp / data_dir / tile_id.tif
    python compute-change-clusters.py
        --account-url https://<account>.blob.core.windows.net
        --container-name <container>
        --predictions-root predictions_root
        --data-dir data_dir
        --start-ts 2020q2
        --end-ts 2025q2
        --bbox -97.5 32.4 -96.3 33.2
        --change-percentile 0.95  # switch to negative for declines
        --output-gpkg growth.gpkg
        --output-layer clusters

[calebrob6]

High level comment -- did you double check to see if this works for decline clusters?
Other high level comment -- do you think a README section on change detection (or CHANGE_DETECTION.md) with an example screenshot output would be helpful? If so, we can definitely push this to later too.

[Copilot]

Pull Request Overview

This PR adds a new Python script to compute change detection clusters (growth or decline) between two temporal prediction rasters. The script calculates volumetric change (density × height) between two timestamps, applies a percentile-based threshold, vectorizes regions exceeding that threshold into polygons, and outputs results to a GeoPackage file.

Key changes:

• New script for computing growth/decline clusters between two timestamps
• Supports Azure Blob Storage access via SAS tokens for prediction tiles
• Implements percentile-based thresholding for identifying significant changes
• Vectorizes change clusters using 8-connected component labeling

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[tamara-glazer]

High level comment -- did you double check to see if this works for decline clusters? Other high level comment -- do you think a README section on change detection (or CHANGE_DETECTION.md) with an example screenshot output would be helpful? If so, we can definitely push this to later too.

Yes confirmed, and double checked that it works for "decline clusters"

[tamara-glazer]

@calebrob6 Requesting a re-review. I added updates to the README with an example, and tested on the pre-processed COGs for Bamako. Ready to merge