Quebec Trees Dataset

This dataset was generated for and used in the preprint "Influence of Temperate Forest Autumn Leaf Phenology on Segmentation of Tree Species from UAV Imagery Using Deep Learning". There can be found the detailed methodology.

Cloutier, M., Germain, M., & Laliberté, E. (2023). Influence of Temperate Forest Autumn Leaf Phenology on Segmentation of Tree Species from UAV Imagery Using Deep Learning (p. 2023.08.03.548604). bioRxiv. https://doi.org/10.1101/2023.08.03.548604

For rapid visualisation of the data:

Imagery and annotations (https://arcg.is/1L1DL00)

Point clouds

Abstract

Remote sensing of forests has become increasingly accessible with the use of unoccupied aerial vehicles (UAV), along with deep learning, allowing for repeated high-resolution imagery and the capturing of phenological changes at larger spatial and temporal scales. In temperate forests during autumn, leaf senescence occurs when leaves change colour and drop. However, few UAV-acquired datasets follow the same individual species throughout a growing season at the individual tree level, allowing for a multitude of applications when used with deep learning. Here, we acquired high-resolution UAV imagery over a temperate forest in Quebec, Canada on seven occasions between May and October 2021. We segmented and labeled 23,000 tree crowns from 14 different classes to train and validate a CNN for each imagery acquisition. The dataset includes high-resolution RGB orthomosaics for seven dates in 2021, as well as associated photogrammetric point clouds. The dataset should be useful to develop new algorithms for instance segmentation and species classification of trees from drone imagery.
 

Classes

Table 1. Main classes present in the dataset and total amount of annotations Label Common name Scientific name Family Annotations ABBA Balsam fir Abies balsamea Pinaceae 2895 ACPE Striped maple Acer pensylvanicum Sapindaceae 751 ACRU Red maple Acer rubrum Sapindaceae 5857 ACSA Sugar maple Acer saccharum Sapindaceae 1014 BEAL Yellow birch Betula alleghaniensis Betulaceae 290 BEPA Paper birch Betula papyrifera Betulaceae 5894 FAGR American beach Fagus grandifolia Fagaceae 222 LALA Tamarack Larix laricina Pinaceae 185 Picea Spruce Picea spp. Pinaceae 1022 PIST White pine Pinus strobus Pinaceae 569 Populus Aspen Populus spp. Salicaceae 1114 THOC Eastern white cedar Thuja occidentalis Cupressaceae 1510 TSCA Eastern hemlock Tsuga canadensis Pinaceae 59 Mort Dead tree - - 878 Total       22,260

The genus level classes, Picea spp. and Populus spp., include trees annotated at the species level (PIGL: Picea glauca, PIMA: Picea mariana, PIRU: Picea rubens, POGR: Populus grandidentata, POTR: Populus tremuloides). These classes were merged due to the difficulty in identifying the species and the similarities between the species.

Not included in this table are approximately 700 additional trees that were segmented and labelled and included in broader categories or in categories with too few individuals.

Included in the dataset

The data is organized by acquisition date (YYYY-MM-DD). There are seven acquisition dates and the study site is divided into three zones.

The data included for each of the dates and zones are:

• RGB imagery in Cloud-Optimized GeoTIFF (COG)
• Point cloud in Cloud-Optimized Point Cloud (COPC, .laz files)

The vector layers included are:

• Individual tree level annotations in GeoPackage (GPKG), one for each zone
• Polygons delimiting the inference data used in the publication

A copy of the vector data is in each compressed file for each date.

Metadata files are also included for all the data in a separate folder.