The Crop Classification model delivered by Pixxel Aurora provides reliable identification of crop types across agricultural landscapes using satellite multispectral data. The model’s performance is validated through two complementary strategies: supervised classification, which leverages user-provided information about known crops, and unsupervised clustering, which groups farms based on similarities in vegetation patterns without prior labels.

  • Supervised Model Accuracy : This approach generally achieves classification accuracies in the range of 70% to 80% when compared against labeled reference farms. It is especially effective at differentiating crops with distinct growth signatures—achieving accuracies exceeding 90% for certain crops such as maize and potato.
  • Unsupervised Model Accuracy : When no prior crop labels are available, the model groups farms by similarity in vegetation temporal profiles, yielding accuracies typically between 70% and 80%. While this approach provides valuable insights, it may occasionally merge phenologically similar crops or over-segment single crop types.
  • Aditional Considerations :
    • Classification performance can be impacted by external factors such as cloud cover and the availability of clear temporal data within a single crop season.
    • The model focuses on crops grown within a single growing season and may not effectively classify crops with longer or overlapping phenological cycles.
    • The supervised model depends on the availability and quality of user-labeled reference farms for best results.
In summary, the Crop Classification model provides reliable and actionable crop type information at regional scales, supporting decision-making in agriculture, food supply chain management, and risk assessment with acceptable and interpretable accuracy metrics