Forestry meets machine learning
When it comes to forestry, phenotyping – the act of quantifying and describing a forest’s physical, physiological and biochemical properties – presents a special variety of new challenges. Now, thanks to advances in remote sensing technology, accurate assessment of forest characteristics and growing conditions is possible. The next challenge is making the best use of this data.
Combining phenotype with genetics and forest management data creates very large, noisy and complex datasets. This year, Scion developed and implemented new methods based on machine learning to analyse these complex datasets and identify trends in growth patterns for different genotypes in different environments.
The significant challenges with forest phenotyping are caused by the size of the forested areas and the length of the forest rotation. This leads to a large amount of data over highly varied landscapes with differing terrain and localised climatic conditions. When you add technical challenges, including collecting data from trees that can be physically large and are often planted in remote locations, the complications mount quickly.
“The development of a successful phenotyping platform will ... enable us to identify greater opportunities to improve the productivity and value of our forest estate.” - Paul Adams, Forest Estate Manager, Rayonier Matariki Forests
Scion’s new forest phenotyping platform includes the development of new concepts of describing forest characteristics across large areas. The team has assembled stand management, genetic, soils, terrain, and climatic datasets (supplied by Timberlands) for the purpose of accurately describing the growing conditions experienced by crop trees. In the last year, New Zealand’s largest plantation forest has been characterised for forest phenotype and terrain characteristics and the techniques developed during this process have been extended to other large forests. This compilation of information results in extremely large datasets that can have more than a billion records.
Scion’s new machine learning techniques simplify this data, analysing the huge volume and turning it into applicable information that will be invaluable to forest growers, tree breeders, investors and other researchers. As an industry, forest growers will be able to make best use of our existing forests by maximising their profitability and enabling investors to have greater confidence in investing in regional infrastructure. This research could also significantly increase the speed with which new discoveries about tree growth are made and speed up the deployment of new genetic material.
This work begins to open up a world of potential for the forestry industry. Paul Adams, Forest Estate Manager at Rayonier Matariki Forests explains, “The development of a successful phenotyping platform will significantly improve our understanding of the linkages between genotype, environment and silviculture. This will enable us to identify greater opportunities to improve the productivity and value of our forest estate.”
The next steps for this research includes expanding the phenotyping concept to describe every tree in a forest.