Use Landsat 8 data to test dimensionality reduction, clustering, and eventually clustering algorithms.

Note that source files are not currently housed here due to size. They can be found in the project's OneDrive folder: https://1drv.ms/u/s!ApknW2-syaCYg4VIVh-HrXRhUB2RbA?e=i5MV9G.

Done during Metis:

• Wrangle geospatial data using GDAL and GRASS.
• PCA to reduce channels 1 through 7 to three PCA components.
• Visualize false color maps in a plotly Dash app, including a PCA component map.
• Categorize pixels in GRASS.
• Gridsearch over hyperparameters in DBSCAN and Mean Shift algorithms to categorize pixels & compare results.
• Segment pixels in GRASS.
• Segment pixels in DBSCAN and Mean Shift and compare results.

Future:

• Collect training data on lava flow ages.
• Conduct regression on pixels or segments to estimate ages & compare to test data.
• Expect a major problem to be explaining the difference between windward (wet) and leeward (dry) weathering results.
• Explore whether a neural network with appropriate connection graph can address the windward / leeward problem and improve regression.