Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Turning this lesson into progress on your SDS320 project

1. Why this page matters

Every idea in this lesson, GeoAI’s definition, why spatial is special, the domains it is used in, the tools, the seven core tasks, and the traditional-to-foundation-model trajectory, exists to help you answer one question for your own project: what should your model actually predict, and with what kind of approach? This page does not introduce new theory. It asks you to commit, provisionally, to a first direction.


2. Project checklist


3. Decision points

Domain: Pick something you are genuinely curious about, not just the first example from this lesson. A domain you find personally interesting is easier to stay motivated on through a full semester.

Task type: Use the three-question check from Core GeoAI tasks: whole-image or located, boxes or pixels, category or number. If two task types both seem plausible, note both and let L02 – Data acquisition help you decide based on what data is actually available.

Paradigm: A reasonable default: if you expect to hand-label fewer than roughly 100 examples, lean toward a fine-tuned foundation model or a traditional method; if you expect several hundred or more, a CNN trained more specifically for your task becomes realistic within a semester.

Scope: A study area you can meaningfully cover in one semester, rather than “the whole country,” is a reasonable choice at this early stage. You can always expand later if time allows.


4. Common pitfalls


5. Mini deliverable

Write a short Project Idea Sketch (150–250 words) containing:

  1. Your candidate domain and a one-sentence research question.

  2. Your tentative core task type, and a one-sentence justification.

  3. The one or two spatial-data properties (from this lesson) most likely to shape your data choices.

  4. A first guess at data sources, even if unconfirmed, that you will check in L02.

  5. A first guess at paradigm (traditional method, CNN, or fine-tuned foundation model) and why.

Save this sketch somewhere you will return to. You are not committing to it permanently, but it gives L02 something concrete to test against real data availability.


6. Reflection questions