Disturbance Attribution: Initial Results

First draft classification models for both binary classification and multi class classification

Binary Classification

Disturbance Classes

• Harvest
  
• Non Harvest

Metric	Estimator	Estimate
accuracy	binary	0.9425
roc_auc	binary	0.9573

Multi Class Classification

Classes

• Harvest
• Flooding
• Weather
• Fungal
• Insect
• Unknown Non Harvest Disturbances (largely ‘spectral decline’ from timesync, but also includes unknowns from DEC forest health flights)

Metric	Estimator	Estimate
accuracy	multiclass	0.8961
roc_auc	hand_till	0.7289

This classifier notably fails to predict any disturbances in both the ‘insect’ and ‘fungal’ classes. These were the two smallest classes in the data set and as such, this is not a surprising result. The ’unknown’category represents a significant portion of the non harvest disturbances and was left in for both models. It could likely be excluded from later version of the multiclass classifier.