Introduction

This dataset is part of the plantAI project and contains images and field inventory data of tree seedlings in Swedish forestry.

The dataset includes:

• Field inventory data for individual seedlings
• Site and environmental properties
• Image metadata
• Images from the right camera lens
• Annotation of a subset of the right camera images

Images from the left camera lens are provided in a separate dataset. See the "Related works" section for links to associated datasets.

Dataset structure

The dataset is organized as follows:

• Images are provided as zip archives containing approximately 3000 images each, ordered by photo timestamp.
• seedling_observations.csv: 
  • The main data file containing seedling inventory data, image metadata, and site properties. Each row corresponds to one image.
  • All categorical variables encoded as numeric codes are supplemented by a corresponding "_label" column containing human-readable descriptions.
• columns.csv:
  • Describes all columns in seedling_observations.csv, including type and definitions.
• codebook.csv
  • Lists and describes all categorical codes used in the dataset.
• labelstudio_annotation_photo_angles_2-5.zip
  • Annotations of a subset of the images
  • Only side views are represented, that is photo_angle 2-5. 
  • Region of interest as bounding box
  • Seedling as oriented bounding box
  • Seedling pose keypoints

Column details

Detailed column descriptions are provided in columns.csv. This section highlights selected variables.

Column: `planting_spot`

The `planting_spot` variable describes the type of soil preperation done at the planting location, based on visual judgement, and the types are defined as:

Column: `site_preparation_method`

The `site_preparation_method` column describes the type of soil preparation done on the site as whole, and are defined as:

Soil types

The soil types are based on the Soil types 1:25 000–1:100 000 dated 2018-01-30, provided by SGU – Geological Survey of Sweden.

The GNSS position of each seedling was used to extract soil type information from two map layers:

1. Parent Material – Base Layer (JG2): Represents the dominant soil type at ~0.5 m depth. This includes areas of exposed bedrock or thin soil layers. All data points have a value in this layer.
2. Surface Layer – Thin or Discontinuous (JY1): Represents surface layers thinner than ~0.5 m, or discontinuous layers averaging 0.5–1 m. Common examples include thin peat or till on bedrock. If present, JY1 overlays JG2.

The following soil types appear in the dataset:

Label Studio annotation

The annotations are provided as exported from open source software Label Studio. 

Data quality and uncertainty

Several variables in the dataset are based on field observations and manual classification, including species, planting spot type, seedling vitality, and vegetation within the planting spot.

These variables are subject to observer interpretation and may include classification uncertainty. 

Position data is provided together with estimated horizontal and vertical errors in meters (as provided from the GNSS).