![]()
The New South Wales Riverina townships of Griffith, Leeton, Narrandera and Yenda cover an area of over 6,200 km2 of land, much of which is used for fruit and nut production. These are perennial tree and vine crops, which live for many years without needing to be replanted. The mapping of these intensive cropping areas aids natural resource planning and analysis, and supplements response plans for any biosecurity threats or natural disasters in the area.
We sat down with James Brinkhoff, Senior Research Fellow and part of the team at the Applied Agricultural Remote Sensing Centre at the University of New England. As part of the #GEEImpact campaign to uncover how people are using Google Earth Engine (GEE) to make a difference, James shared how his team is using GEE to produce a land cover map of part of the Riverina region, with a specific focus on perennial crops such as those grown on fruit trees and vines. The work was done in collaboration with Justin Vardanega of Riverina Local Land Services.
James, what was the specific challenge that you wanted to tackle in this project?
Current land cover maps in Australia have limited spatial resolution, and often don't discriminate between classes at the commodity level, for example: separating citrus from hazelnut from olive orchards.
We wanted to investigate techniques to provide a medium resolution (10-metre pixel size) map of perennial crops, separating between nine fruit and vineyard crops that are common in the Riverina region.
How has the challenge impacted your community?
These maps will be useful in delivering more rapid responses to biosecurity threats, providing targeted agronomic information to growers relevant to their crops and give a better understanding of the different crops over the large Riverina region. In the future, we anticipate that the maps will provide information to help predict per-crop production.
How has the challenge impacted you or your organisation?
Through the project, we have seen increased development in the skillset of our group at the Applied Agricultural Remote Sensing Centre.
We developed and used the following skills in developing the land cover map of the Riverina area:
- Application development for gathering geospatial crop data
- Processing large volumes of satellite imagery into tidy time-series datasets
- Using these data sources to train machine learning algorithms with a focus on per-crop phenology
- Utilising object-based image analysis
As a team, we can now deliver industry-relevant information in the form of maps and statistics, including scientifically robust analysis of product accuracy.
Tell us more about your project and how Google Earth Engine has helped you achieve your project-related goals?
GEE was vital to this project and we were able to make use of the large quantities of remote sensing products available on the platform. For this project in particular, we used:
- Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 multispectral images aggregated into a monthly time-series
- In-built algorithms such as Simple Non-Iterative Clustering (SNIC) to generate field boundaries
- Machine learning classifiers such as Random Forest and Support Vector Machine to automatically identify crops.
We also used GEE to deploy a web app to gather crop-type training data from regional crop experts. The end product displayed the final crop-type map.
For more information, some of the work undertaken during this project was published: https://www.mdpi.com/2072-4292/12/1/96
What is 'the win' for you and your community?
We have demonstrated the potential of mapping many different perennial crop types and assessed the accuracy of the 10-meter resolution using freely available satellite imagery.
The techniques we used were computationally feasible on the Google Earth Engine Platform and could be scaled to cover other areas and additional crop types.
Do you have a #GEEImpact story or project that you’d like to share?
Use the hashtag #GEEImpact on Twitter, Facebook or LinkedIn, or get in touch with our team!
Back To News Stories