Skip to content

SCynergy 2026 - GeoAI Workshop — Worksheet

📘 Notebook 1 — Understanding Data Acquisition

  1. What problem does STAC solve in geospatial workflows?
    → Think about how data is discovered and filtered.

  2. Why do we define multiple temporal phases (pre-event, event, post-event)?
    → What additional information do they provide?

  3. What factors determine the “best” satellite scene?
    → Consider differences between optical and radar data.

📗 Notebook 2 — Understanding Data Packaging

  1. Why must all datasets be aligned onto a common grid?
    → What would happen if they were not?

  2. What is reprojection, and why is it necessary?
    → Think about coordinate systems and pixel alignment.

  3. Why do we split large images into smaller chips?
    → Consider GPU limitations and model requirements.

  4. How does the workflow handle missing or invalid data?
    → Why is this important for AI models?

📕 Notebook 3 — Understanding Model Inference

  1. What does each dimension of the input tensor represent?
    → [Batch, Channels, Time, Height, Width]

  2. Why is DEM repeated across the time dimension?
    → Does the terrain change over time?

  3. What does the model output represent?
    → How would you interpret the prediction?

Reflection Questions

  • Which part of the pipeline was most surprising or new to you?
  • Where do you think errors or uncertainties could arise?
  • How could this workflow be adapted to another application (e.g., wildfire, agriculture)?