Shelterbelts

This is an open-source Python package for mapping and categorising shelterbelts (windbreaks) across Australia using satellite imagery, in preparation for measuring their impacts on agricultural productivity at scale.

Key Features

1. Tree categorisation: classifies pixels as scattered trees, patch core, patch edge, or corridors based on nearby connectivity.

2. Cover categorisation: overlays ESA WorldCover 2021 land-cover classes (grassland, cropland, urban, water) onto the tree categories

3. Buffer categorisation: identify riparian and roadside tree buffers using the National Surface Hydrology Lines and National Roads datasets

4. Shelter categorisation: determine sheltered vs. unsheltered farmland based on tree density or wind direction, similar to Stewart et al. 2024

5. Patch & class metrics: compute patch and class landscape statistics similar to FragStats

6. Opportunities mapping: identify potential locations for additional tree plantings (currently a work-in-progress)

7. API integrations: download data from ANU BARRA-C2 (wind), WRI Canopy Height, and ESA WorldCover

8. Command-line interface: all index modules can be run in python scripts or directly from the terminal

9. Scalable: designed for national-scale processing on HPC systems (NCI Gadi)

10. Classifying trees from raw inputs: LAZ point clouds (via PDAL), Sentinel-2 imagery (with a bundled pre-trained neural network), or train your own model on labelled tiles

Example Output

The plot below shows the full categorisation pipeline: a binary tree-cover raster on the left, and the final shelter categories on the right (sheltered grassland and cropland labelled by the type of tree providing the shelter).

(Source code)

Visualise Results

You can explore results interactively in the Google Earth Engine App.

Github Repository

You can find installation instructions on the README of the GitHub repository.

Parameter Reference

The main parameters for categorising shelterbelts are:

Parameter	Default	Low	High	Description
min_patch_size	20	15	25	Minimum area (pixels) to classify as a patch rather than scattered trees
min_core_size	1000	100	10000	Minimum patch size (pixels) to classify as a core area
edge_size	3	2	5	Distance (pixels) defining the edge region around patch cores
buffer_width	4	3	5	Number of pixels away from a feature that still counts as within the buffer
distance_threshold	20	10	30	Distance from trees that counts as sheltered
density_threshold	5	3	10	Percentage tree cover within distance_threshold that counts as sheltered
wind_threshold	20	15	25	Wind speed threshold in km/h
wind_method	WINDWARD	MOST_COMMON	ANY	Method to determine primary wind direction
min_shelterbelt_length	20	15	25	Minimum skeleton length (pixels) to classify a cluster as linear
max_shelterbelt_width	6	5	7	Maximum skeleton width (pixels) to classify a cluster as linear

API Reference

• API Reference
  • WorldCover
  • BARRA Daily
  • Canopy Height
• Indices
  • Tree Categories
  • Cover Categories
  • Buffer Categories
  • Catchments
  • Patch Metrics
  • Shelter Categories
  • Class Metrics
  • Full Indices Pipeline
  • Opportunities
• Classifications
  • Binary Tree Rasters
  • Bounding Boxes
  • LiDAR
  • Sentinel Download
  • Training Pipeline
  • Prediction
  • Mosaicking