Capturing long-tailed individual tree diversity using an airborne multi-temporal hierarchical model

This dataset contains the hyperspectral images and train-test split from the article. 

The corresponding comet ML experiment is:

CometML

To protect ongoing scientific activities in the area, the geospatial position of the crops have been omitted. Please contact ForestGEO for data requests. 

https://forestgeo.si.edu/

For the corresponding git repo please see: 

Contents:

train.csv: The image crops used for model training. The image_path corresponds to the relative path to the image file in the directory.

test.csv: The image crops used for model evaluation.

*.tif images: A 369 band hyperspectral image using the NEON surface reflectance data cropped by the tree crown.

The original HSI data: https://data.neonscience.org/data-products/DP3.30006.001. The location of the crown was predicted using the RGB data product and the deepforest model: https://deepforest.readthedocs.io/. 

The taxonID abbreviations follow NEON's taxonomy: https://data.neonscience.org/taxonomic-lists

OSBS.shp: Shapefile with ensemble predictions for the full Ordway Swisher Biological Station. The ensemblaTa column is the predicted taxonID for each crown. The ens_score is the confidence prediction for that crown. The CHM_height is the extracted raster value from NEON's ecosystem structure canopy height model.