Model Validation

To validate the dynamic forest structure model, we iteratively withheld 1/10th of the training data in a k-fold crossvalidation exercise on 932 training locations distributed across the Columbia Basin region. Each fold was spatially independent, so the withheld fold did not overlap in space with the training folds. The table below shows the r2 adjusted, root mean square error (RMSE) for each variable, averaged across the 10 folds in the crossvalidation.

Variable	r2 Adjusted	RMSE (%)
Invasive Annual Grass Cover	0.508	9.05%
Perennial Grass Cover	0.606	16.96%
Sagebrush Cover	0.619	6.80%
Shrub Cover	0.677	8.87%