Shelterbelts: A Python library for characterising farmland tree groups and their influence on nearby pasture and cropland

Christopher Gregory Bradley , John T Burley , Nicolas Pucino, Justin Borevitz

Shelterbelts is an open source Python library for mapping and categorising groups of trees in agricultural landscapes, and the corresponding shelter provided to nearby pixels based on the wind direction or density of tree cover.

Starting from a binary tree raster, pixels are labelled into ecologically meaningful categories based on their structure and location including: scattered trees, forest core and edge, riparian and road buffers, other linear (windbreaks/tree-rows) and non-linear patches. The library also provides functions to fetch and transform raster datasets for wind speed and direction, elevation and topographic attributes, along with national road and hydroline vector datasets. The library can also be used to process LiDAR through crown delineation, or fetch and transform global canopy height and land cover data, and even to train a new annual model to predict tree cover from satellite imagery, to create the necessary input binary raster for the pipeline.

The software was designed and tested on landscapes in Australia, but can be applied anywhere. It can be run locally in jupyter notebooks or at scale in high-throughput environments such as the National Computing Infrastructure (NCI).

The github repository is available here.
https://github.com/ChristopherBradley/shelterbelts

Australia-wide map of shelterbelt classifications can be viewed at this Earth Engine App. https://christopher-bradley-phd.projects.earthengine.app/view/shelterbelts