Skip to main content

4. Accuracy Points

Accuracy assessment points are essential for validating your vegetation classification. These points serve as "ground truth" data that help train and evaluate the model's performance.

Overview​

The Accuracy Data functionality in Tyton AI enables users to generate and manage accuracy assessment points that serve as critical ground truth references for vegetation classification. These points play a dual role in both training and validating classification models, ensuring that outputs are scientifically robust and spatially accurate.

By manually placing assessment points in clearly identifiable vegetation types, users can improve model performance, quantify classification accuracy and identify areas needing further review. This process supports both initial ("first pass") classifications and ongoing refinement, helping the model adapt to local vegetation patterns.

Creating well-distributed, high-confidence assessment points across various vegetation classes and environmental conditions lays a strong foundation for automated classification and enhances the ecological reliability of the analysis.

Creating Assessment Points​

  1. From the project view, locate the Assessment Data section in the left panel.
  2. Select the Assessment Point tool (or press 5 on your keyboard).
  3. Choose the correct vegetation class from the dropdown menu.
  4. Click on the map to place assessment points:
    • Place points in areas where you are confident of the vegetation type.
    • Aim for at least 50 points per class for robust validation.
    • Distribute points across different vegetation types and areas.

Managing Assessment Points​

Viewing Assessment Data​

  • Assessment points appear in the project explorer under 'Assessment Data'.
  • Points are organized by class with counts shown in parentheses.
  • Each point stores:
    • Geographic coordinates.
    • Assigned vegetation class.
    • Additional metadata.

Editing Point Classifications​

  1. Use the Selection tool (press 1 on your keyboard).
  2. Hold Shift and drag a box to select multiple points.
  3. In the Feature Properties panel:
    • Click the Classification field.
    • Choose the correct class from the dropdown.
    • All selected points will be updated.

Supporting First Pass Classification​

Accuracy assessment points are crucial for your initial classification efforts:

Why First Pass Matters​

  • Creates a foundation for automated classification.
  • Helps the model understand local vegetation patterns.
  • Establishes baseline accuracy metrics.
  • Identifies areas needing more detailed assessment.

Preparing for First Pass​

  1. Create accuracy points before running classifications:
    • Focus on areas you plan to classify first.
    • Include representative samples of each vegetation type.
    • Place points in areas with clear, distinct vegetation classes.
  2. Ensure good coverage:
    • Place points across the full extent of your study area.
    • Include different environmental conditions (slopes, aspects, soil types).
    • Mark boundaries between different vegetation types.

Using Points for Training​

  • Points serve dual purposes:
    • Training data for the classification model.
    • Validation data for accuracy assessment.
  • More points generally lead to better classification results.
  • Well-distributed points help the model understand spatial patterns.

Best Practices​

  • Create assessment points in areas representative of each vegetation class.
  • Include transition zones between different vegetation types.
  • Mark points where you have high confidence in the vegetation type.
  • Consider seasonal variations when placing points.
  • Regularly review and update point classifications as needed.
Accuracy Assessment Strategy

For reliable model validation:

  • Include 50 points minimum per vegetation class
  • Distribute points across the full study area
  • Add extra points for rare or important vegetation types
  • Place points in both pure and mixed vegetation areas
Field Validation

When possible:

  • Use GPS to mark actual field locations
  • Take reference photos of assessment points
  • Note seasonal conditions and date of observation
  • Record dominant species and vegetation structure
  • Document any disturbances or unusual conditions