One deep tree is better than many small trees

Each tree must see the entire dataset

Combining multiple trees improves predictions

Trees must use all features at every split