Hyperspectral imaging has a limited ability to remotely sense the onset of beech bark disease

This dataset includes hyperspectral and beech bark disease assessment data from our study area, Mont-Saint-Bruno National Park in Saint-Bruno-de-Montarville, Quebec, Canada. Here, we tested whether airborne hyperspectral imagery - involving data from 344 wavelengths in the visible, near infrared (NIR), and shortwave infrared (SWIR) - can be used to assess the severity and progression of beech bark disease in southern Quebec, in the hope of developing new methods for effective remote sensing of this fungal infection that is widespread in eastern North America. Field data on disease severity were linked to airborne hyperspectral data using georeferenced red-green-blue (RGB) drone imagery to delineate beech crowns of interest. We also looked for a relationship between this same disease severity variable and hyperspectral data at leaf level (with canopy leaf samples). This dataset therefore comprises four sections: 1. Canopy-level hyperspectral data (n=126); 2. Crown geometries for the 126 beech trees with aerial hyperspectral imaging; 3. Leaf-level hyperspectral data on selected beech trees (n=37) and 4. Beech bark disease assessment data (n=160).

Methods
Data for this project were collected in three different contexts. First, airborne hyperspectral data from Mont-Saint-Bruno National Park in Saint-Bruno-de-Montarville, Quebec, Canada, were collected for the CABO project in September 2018 with a resolution of 2 meters per pixel. Secondly, we assessed the progression of beech bark disease in the same study area on 160 beech trees across a broad disease gradient, using bark and canopy condition indices as predictors of disease progression. In the end, only 126 of these were linked to aerial hyperspectral imagery, as the spatial error on the geometries of the remaining 34 beech crowns in the flight line was too high. Finally, we measured the reflectance of the fresh leaves of 37 beech trees using a portable hyperspectral sensor.

For hyperspectral data processing, we have combined several techniques for reducing atmospheric noise in spectral data, including masking of atmospheric absorption regions (350-400 nm and 920-957 nm) followed by reflectance interpolation in the masked regions, and application of a smoothing filter (Savitzky-Golay) to the raw data. Downloaded datasets have already been processed using the above methods. Complete CABO raw hyperspectral datasets for the whole park are available on request. For further details on data collection and processing, please refer to the following preprint: https://doi.org/10.1101/2024.09.20.614150.