1. Why the research question matters¶
A research question is the engine of your project. It guides what data you need, which method makes sense, how you evaluate outputs and how you explain results.
Without a clear question, a project can become a loose collection of tools: download data, run a model, make a map and then search for a story. With a clear question, the tools become part of a focused spatial analysis.
2. From topic to question¶
A topic is not yet a project. A project begins when you connect a question to data, method, output and evaluation.
| Level | Meaning | Example pattern |
|---|---|---|
| Topic | Broad area of interest | Urban heat, glacier change, tree crowns, flood damage |
| Problem | Why the topic matters | Heat exposure varies within cities and affects planning decisions |
| Research question | What your project tries to answer | Where are high land-surface-temperature patterns most visible in a selected city? |
| Analytical task | What your workflow needs to do | Acquire imagery, preprocess data, calculate or model an output, evaluate it |
| Expected output | What you produce | Map, figure, metric, comparison, prediction or interpretation |
3. What makes a good RQ?¶
A good SDS320 research question is usually:
spatial — it asks about location, distribution, distance, area, change, objects or spatial relationships,
feasible — it fits the available time, data and skills,
ethical — it avoids harm and handles sensitive topics responsibly,
data-driven — it can be addressed with accessible spatial data,
methodologically clear — it suggests what kind of analysis is needed,
narrow enough — it can be implemented and communicated in one semester,
open enough — it requires analysis, not only a yes/no answer.
The SDS110 project-planning material (lecture 4) used the FINER criteria: feasible, interesting, novel, ethical and relevant. For SDS320, add one practical test: can you imagine the main map, figure, metric or model output that would answer the question?
4. Use PICOT to sharpen the question¶
The PICOT framework can help you make a question more precise.
| PICOT element | Meaning for SDS320 |
|---|---|
| P — Population | Who or what is studied? This can be a place, object type, land-cover class, image set or spatial feature. |
| I — Intervention / Indicator | What is measured, detected, modelled or introduced? |
| C — Comparison | What is compared? For example before/after, area A/area B, class A/class B or method A/method B. |
| O — Outcome | What result is measured or mapped? |
| T — Time | Over what period or observation date? |
PICOT is not mandatory, but it helps reveal vague questions.
How does [indicator] differ between [comparison groups] for [population/features] in [area] during [time period]?5. Useful question patterns¶
How does [spatial phenomenon] vary across [area/time/classes]?Can [data source or method] identify [feature/change/pattern] in [study area]?To what extent does [variable] relate to [spatial outcome]?Where are [features/events/changes] most visible in [data]?How well can [method] map or predict [spatial feature] from [input data]?6. Weak vs improved examples¶
The examples below are illustrative.
| Weak question | Why it is weak | Improved question |
|---|---|---|
| How has biodiversity changed? | Too broad and not clearly spatial or measurable. | How has bird species richness changed across selected Swiss alpine valleys between 1990 and 2020? |
| Can I use deep learning on satellite images? | Starts with a method, not a spatial problem. | Can semantic segmentation of Sentinel-2 imagery identify broad land-cover classes in a selected study area? |
| Which regions have worse flood problems? | Ambiguous terms and unclear metric. | How did mapped flood extent differ between selected municipalities during a defined flood event? |
| Has this glacier changed? | Too close to yes/no. | By how much did glacier area change between two selected observation years, and where was retreat most visible? |
| Is the model good? | No task, metric or reference. | How accurately does the model classify selected land-cover classes compared with reference labels in the study area? |
7. Scope check¶
Before settling on a question, check:
Is the study area clearly defined?
Is the time period clear?
Can you access and inspect the data?
Does the method fit the data format and resolution?
Can you name the main output?
Can you evaluate or at least check the output?
Are there ethical or privacy risks?
Can you build a first prototype soon?
If several answers are unclear, narrow the question before adding technical complexity.
8. Decision points¶
Ask yourself:
Am I asking a spatial question or only using spatial data?
Does the question require analysis?
Can the question be answered with data I can access?
What would count as a useful result?
What would count as a limitation?
9. Mini task¶
Write three versions to improve your question:
Broad topic:
Draft question:
Improved SDS320 research question:Then complete:
My project asks ...
It uses ...
It compares or measures ...
It produces ...
It can be evaluated by ...10. Key takeaways¶
A good research question guides data, methods, workflow, evaluation and communication.
Start broad, then narrow.
FINER and PICOT can help make the question more precise.
Avoid questions that are too broad, too simple or method-first.
A feasible question connects to a realistic output and evaluation strategy.