examples

SpiPy Examples

This directory contains example notebooks demonstrating the usage of SpiPy for snow property inversion from satellite imagery.

Main Workflow Examples

1. Setup

• 01_define_regions.ipynb - Define regions of interest for processing

2. Data Preparation (Prerequisites for Inversion)

• 02_pansharpening.ipynb - Pansharpening to improve spatial resolution of Sentinel-2 bands
• 03_create_background_reflectance.ipynb - Generate R0 (background/snow-free) reflectance maps from snow-free periods
• 04_cloud_masking.ipynb - Cloud detection and masking for satellite imagery

3. Core Processing Pipeline

• 05_sentinel_snow_inversion.ipynb - Main workflow: Sentinel-2 snow property inversion using SPIRES
  • Loading preprocessed Sentinel-2 data
  • Applying SPIRES algorithm
  • Batch processing with Dask
  • Saving inversion results (fsca, grain size, dust concentration)

4. Postprocessing

• 06_postprocess_clouds.ipynb - Interpolate cloud gaps and fix sharpening artifacts in results
• 07_postprocess_trees.ipynb - Tree masking and inpainting using deep learning

5. Analysis and Visualization

• 08_create_animations.ipynb - Generate temporal animations of snow properties

Test Notebooks

These notebooks demonstrate specific functionality and can be used for testing:

• test_interpolator.ipynb - Test LUT interpolation functionality
• test_inversion.ipynb - Test basic inversion on single pixels/images
• test_spectrum_diff.ipynb - Test spectral difference calculations

Development Notebooks

The following notebooks were used during development and are kept for reference:

• legacy_speedy_invert.ipynb - Legacy MATLAB-style implementation comparison
• cobyla.ipynb - COBYLA optimizer testing
• invert2d.ipynb - 2D inversion experiments
• compress_nc.ipynb - NetCDF compression utilities

Data Files

• UCSB_fsca.mp4 - Example animation output showing fractional snow-covered area

Requirements

Most notebooks require:

• SpiPy package installed
• Access to Sentinel-2 data (zarr format)
• Lookup tables (LUT) for snow reflectance
• Sufficient memory for processing satellite imagery

Some notebooks have additional requirements:

• postprocess_trees.ipynb requires PyTorch and simple-lama-inpainting
• Processing notebooks benefit from Dask for parallel computation

Getting Started

Quick Start

For new users, we recommend going through the notebooks in this order:

1. test_interpolator.ipynb - Understand the core LUT interpolation
2. test_inversion.ipynb - See basic inversion examples on single pixels
3. 05_sentinel_snow_inversion.ipynb - Full workflow with batch processing

Complete Workflow

For processing your own Sentinel-2 data:

1. 01_define_regions.ipynb - Define your region of interest
2. 02_pansharpening.ipynb - Improve spatial resolution
3. 03_create_background_reflectance.ipynb - Generate R0 maps
4. 04_cloud_masking.ipynb - Apply cloud detection
5. 05_sentinel_snow_inversion.ipynb - Run SPIRES inversion
6. 06_postprocess_clouds.ipynb - Fill cloud gaps
7. 07_postprocess_trees.ipynb - (Optional) Remove tree artifacts
8. 08_create_animations.ipynb - Visualize temporal evolution

Notes

• Notebooks assume data is stored in /scratch/tristate/ or /data/sentinel2/ directories
• Most workflows are designed for UCSB region but can be adapted
• Large notebooks (>1MB) contain embedded outputs and visualizations