Tutorial 3: Map uncertainty 101

Why no map is ever “the whole truth,” and how to use spatial evidence responsibly.

3.1 Where uncertainty comes from

Measurement error:
- Field measurements (e.g., soil samples, tree counts) have natural variability and possible errors.
- Satellite sensors have limitations in resolution, cloud cover, sensor noise, and retrieval algorithms.
Model uncertainty:
- Different modelling choices (type of model, predictors used, parameter settings) can lead to different predictions.
- Models may fit some environments better than others (e.g., data-rich vs. data-poor regions).
Data gaps and coverage:
- Some areas have more field data or higher-quality satellite observations than others.
- In data-scarce regions, models are forced to “guess” based on patterns learned elsewhere.
Temporal mismatch:
- Field data and satellite data may come from slightly different years or seasons.
- Land use may have changed since the data were collected.

Uncertainty can be represented in several ways:
- Error statistics:
  - A summary of how far predictions tend to be from observed values (e.g., root mean square error).
- Confidence intervals or ranges:
  - For each location, a range of likely values instead of a single number.
- Uncertainty classes or masks:
  - Simple categories (e.g., low / medium / high uncertainty) mapped across the region.
  - Masks showing areas where predictions should be treated with extra caution (e.g., outside data domain).
For users, the most practical representations are:
- Additional uncertainty layers alongside main indicators.
- Clear legends and descriptions explaining what the uncertainty values mean.

When uncertainty information is available:
- Look for map layers or views labelled as uncertainty, confidence, or data density.
- Compare these maps with the main indicator maps to see:
  - Where the indicator is both high (or low) and confident.
  - Where the indicator is uncertain, even if the values look extreme.
Questions to ask:
- Are the areas I am interested in mainly high-confidence or low-confidence zones?
- Does the level of uncertainty change between countries or regions?
- Are there patterns in uncertainty (e.g., highest in remote or under-sampled areas)?
Interpretation:
- High uncertainty does not mean the indicator is wrong; it means we should be more cautious.
- In high-uncertainty zones, it may be especially important to:
  - Use local knowledge.
  - Prioritise field verification.
  - Avoid making decisions based solely on the map.

Good practice when using uncertain maps:
- Combine multiple sources of evidence:
  - Maps, local knowledge, field observations, existing reports, and monitoring data.
- Communicate uncertainty openly:
  - When presenting maps, explain which areas and indicators are more or less reliable.
  - Use language such as “likely”, “approximate”, or “indicative” where appropriate.
- Avoid over-precision:
  - Do not treat pixel-level differences as exact.
  - Focus on patterns and gradients, not single-pixel values.
Example applications:
- Use high-confidence areas to prioritise interventions and demonstrate impact.
- Use uncertain areas to target further assessment, field surveys, or pilot projects.
The key message:
- Uncertainty is an essential part of honest, evidence-based decision-making.
- Maps remain valuable even when uncertain, as long as their limits are understood and respected.

Uncertainty can guide where to improve data and models:
- Highlight regions where additional field data would most improve predictions.
- Inform planning of new surveys, monitoring campaigns, or partnerships.
Over time:
- As more data and better models become available, uncertainty can be reduced.
- Tracking uncertainty across versions of a map shows progress in knowledge, not just changes on the ground.
For K4GGWA:
- Treating uncertainty transparently is part of building trust with partners and stakeholders.
- It supports a culture of learning, where evidence is continuously refined rather than treated as fixed.