CalPoly-Neon

Bucket path

gs://pyregence-tree-mortality/calpoly-neon/

Source data and documentation

• paper: www.mdpi.com/2072-4292/11/19/2326/htm
• repository: jonathanventura/canopy
• data: zenodo.org/record/3470250#.XZVW7kZKhPY

Preprocessing

To generate the training dataset here, I ran the following steps:

1. Ran their deep learning model for species classification (8 classes) and masked the ground using their airborne lidar data (canopy height < 2m).
2. Ran a three class (ground, live, dead) classification using SVM and Random Forest with the NEON pca-transformed spectroscopy data.
3. Created an ensemble model using the agreement of all three dead tree predictions (deep learning, RF, SVM).

You can find the high resolution file path at gs://pyregence-tree-mortality/calpoly-neon/seki_dead.tif.

There is also a directory, canopy/, which contains the scripts and raw datasets for the above. It's a bit of a mess, and not worth documenting in too much detail, but it contains the raw imaging spectroscopy data, a high resolution canopy height model, field datasets, the canopy package used by the original authors to run the deep learning models, and the intermediate SVM/RF models.