Workflow design - SDS320 Spatial Data Analytics

Planning the steps from raw data to interpretable results

1. Why workflow design matters¶

A workflow turns your research question into an implementable project. It shows how raw data become results, figures and interpretation.

Without a workflow, it is easy to jump between notebooks, tools and ideas without knowing what depends on what. With a workflow, you can implement the project step by step, debug problems more easily, explain your decisions and improve Reproducibility.

2. What is a spatial analytics workflow?¶

A spatial analytics workflow connects:

Question
→ Data
→ Preprocessing
→ Method
→ Output
→ Evaluation
→ Interpretation
→ Communication

In practice, this might become:

Define study area
→ download imagery and vector context data
→ clip and reproject data
→ create tiles or features
→ run model or analysis
→ compare output with reference data
→ create map and figure
→ interpret pattern and limitations

The workflow should show both technical steps and project logic. It should answer not only “What code do I run?” but also “Why does this step belong in the project?”

3. Workflow diagram¶

A simple diagram can make your project easier to discuss during feedback sessions.

Adapt the diagram to your project. The goal is clarity, not decorative complexity.

4. Design your workflow backwards¶

Start from the result you want to communicate.

Ask:

What is the main figure, map, metric or result?
What output is needed to create it?
Which method can produce that output?
Which input data are required?
Which preprocessing steps are needed?
Which checks are needed before I trust the output?

This prevents workflows that collect many data and methods without a clear final purpose.

5. Minimum viable workflow¶

A minimum viable workflow might be:

One study area
→ one input dataset
→ one preprocessing step
→ one method
→ one output map
→ one evaluation or plausibility check
→ one caption explaining the result

Once this works, you can decide whether to expand.

6. Decision points¶

Decision	Example question
Data choice	Is the spatial or temporal resolution suitable?
Study area	Can the full area be processed, or should it be smaller?
Method	Is a complex model needed, or is a simpler baseline enough?
Evaluation	Do I have reference data or another way to assess output quality?
Output	Which figure or map best supports the research question?

7. Flags & checks¶

Use this table when your workflow becomes difficult to explain, rerun or connect to the research question.

Red flag	First check
You cannot explain your project in 6–8 steps	Write the workflow as plain text before coding more.
Your method is clear but your question is not	Return to Research question.
You have many datasets but no output plan	Design backwards from the final figure, map, metric or result.
Preprocessing is hidden or scattered	Make the sequence explicit and save important intermediate outputs.
Everything happens in one large notebook	Split exploration, preprocessing, analysis and results into clearer notebooks or scripts.
You have no evaluation step	Decide how you will judge whether the output is useful or trustworthy.
The workflow feels too complex to finish	Define a minimum viable workflow with one dataset, one method, one output and one check.

8. Mini task¶

Write your project workflow in 6–8 steps.

1. ...
2. ...
3. ...
4. ...
5. ...
6. ...

Then add:

Main output:
Evaluation check:
Most fragile step:
Fallback if this step fails:

9. Key takeaways¶

A workflow connects your question, data, method, output and interpretation.
Designing backwards keeps the project focused.
A minimum viable workflow is better than an ambitious workflow that never runs.
Workflow diagrams support feedback and reproducibility.
Every project should include an evaluation or plausibility-check step.