CarbonMapWest Java biomass residue intelligence

Results

Rule-based suitability

ML predicted suitability

ML prediction error

U-Net suitability

U-Net confidence

Model Performance

U-Net

Macro-F1 93.3%

Mean IoU 87.6%. Best-performing segmentation model in the current experiment.

DeepLabV3Plus

Macro-F1 91.0%

Mean IoU 83.7%. Strong segmentation baseline, but lower than U-Net in the current experiment.

Confusion Matrices

U-Net confusion matrix

DeepLabV3Plus confusion matrix

Feature Importance

Feature importance classification

Feature importance regression