The average precision (mAP) on the test set for detection of Plumes is 61.6%. Compared to the other classes, the model is best at detecting plumes with the highest mean AP, mean Recall and mean Precision. The mean Precision across all classes is relatively low, with mean Precision for plumes around 14.0%, whilst all other classes the mean Precision is less than 1%. The low mAP rates for the other classes is a result of the incomplete tagging. This was more severe with the sources of contamination because fewer examples were captured for each of those classes, both proportionally to the number of contaminants in the dataset and relative to the total plume count.