Skip to main content

Erosion Classification

This guide walks you through the 9-step process for performing erosion classification in TytonAI using supervised machine learning. The workflow includes everything from imagery upload to generating detailed erosion metrics.

Erosion Overview​

The Erosion Classification workflow in TytonAI enables users to detect and quantify areas of erosion using high-resolution imagery and AI-powered analysis. The classification process combines polygon-based training data, point-based accuracy data, and elevation inputs like DSM or GSM bands to produce high-quality erosion features. These can be then analysed to calculate metrics such as erosion volume, depth, and spatial extent. This workflow is ideal for environmental monitoring, rehabilitation projects, land condition assessments, and erosion risk tracking across mining, agriculture, and conservation sites.

1. Imagery​

  • Upload your RGB imagery
  • Upload a DSM (Digital Surface Model) layer — this is required for Erosion Classification and is used to generate the Normalized DSM band used by the erosion model.
  • If available, include an Alpha band to improve boundary accuracy (optional).

2. Study Areas​

  • Use the Study Area Tool to draw one or more Study Area polygons.
  • These define the boundaries for where your training and classification will occur.

3. Class List​

  • Create two classes:
    • Erosion
    • Not Erosion
  • These classes will be used for both your training areas and assessment points.
  • These are required for erosion classification. Additional vegetation classes can be added but are not used in this model.

4. Accuracy Points​

Accuracy points are your ground truth reference. They are essential for validating the model’s predictions.

  • Use the Accuracy Points Tool to place labeled points across the study area.
  • Assign each point a class (e.g., Erosion or Not Erosion).
  • These serve as your ground truth — the source of truth for what’s actually present in the landscape.
tip
  • Accuracy points can be placed inside or outside training areas, depending on what’s being evaluated (e.g. you may place vegetation accuracy points inside a training area if you’re only training erosion and not erosion).
  • Don't place erosion and not erosion accuracy points inside erosion/non erosion training areas.

5. Training Data​

  • Create a Training Area using the Training Data tool.
  • Use the Training Data Tool to draw training areas around known erosion and not erosion features.
  • Use the Polygon Tool to manually draw polygons around visible erosion features.
  • Use the Merge Tool to combine adjacent polygons of the same erosion class.

Layer Correction​

For large areas of Not Erosion between erosion features:

  1. Open the Layer Correction Tool
  2. Add a rule:
    • Where: Master ≠ Erosion
    • Burn-in: Not Erosion
  3. This auto-labels all unclassified areas as Not Erosion to balance training data and save time.
tip
  • When using the polygon tool, Hold Ctrl + Drag to freehand draw polygons for more complex shapes.

6. Train Model​

  • Go to the Train Model tab.
  • Select the Erosion Mega Model.
  • Review and select your Assessment Points.

⚠️ Warning: The Erosion Mega Model requires a Normalized DSM band.

If the required band is missing:

  • The Assessment Points selector will be grayed out.
  • Hover over the red warning icon to reveal the tooltip.
  • Click Generate Missing Bands and configure:
    • Input: your DSM
    • Algorithm: Normalized DSM

Once the band is successfully generated:

  • The selector will become active.
  • You can now choose the relevant Training Areas and Assessment Points.

Training Output Includes:​

  • F1 scores per epoch
  • Confusion matrix (actual vs predicted)
  • Class breakdown and input bands
tip

TytonAI will automatically recommend the best performing epoch based on F1 Score once training completes.

7. Classify​

  • Open the Classify tab and choose Classify Erosion.
  • Select your Study Area and your previously trained Model.
  • Confirm the classification cost (credits based on area and resolution).
  • Click Classify to run the model and generate a new classified raster layer.

8. Analysis​

Option 1: Using a DSM Band​

  • In the Explorer, right-click your erosion classification layer.
  • Select Calculate Erosion Metrics.
  • This will auto-select your vegetation classification and allow you to calculate depth using a Ground Surface Model (GSM).
  • If you deselect the vegetation classification, the interface will default to using the DSM band instead.
tip

This is ideal for bare areas where no vegetation classification is available.

Option 2: Using a GSM Band​

  • Navigate to the Analysis tab.
  • Select a vegetation classification and study area.
  • Enable:
    • Erosion Metrics
    • Optionally: Vegetation Cover
  • Click Generate Report to run the workflow.
  • Once your analysis workflow is commplte, it will show in the reports tab.
tip

This is ideal for areas where both vegetation and erosion are in the same location.

9. Reports​

  • Review the generated report in the Reports tab.
  • Available outputs:
    • Erosion Metrics: Average depth, volume, max width
    • Vegetation Cover (if enabled): Class % breakdown
  • Export options:
    • CSV table exports
    • PNG chart images
tip
  • Enable the Depth Map overlay from the Explorer under the erosion classifcation for a visual heatmap of erosion depth.

Tips for Best Results​

  • Ensure training polygons accurately reflect visible erosion features.
  • Distribute training areas evenly across your study area.
  • Use Layer Correction to fill in large, homogeneous areas.
  • Place accuracy points strategically to capture all class types.
  • Validate training quality using the F1 Score and confusion matrix