Large-Scale Phenotyping Inferences: From Trees to Forest through Machine Learning
2019-05-15T04:30:41Z (GMT) by
The genomics revolution has provided rapid gains in crop productivity by shortening the breeding cycle of many commercial species. Genomic data is most useful when carefully linked to crop phenotypic expression and environmental conditions (Dungey, et al., 2018). Within forests, individual tree phenotyping is exceptionally challenging due to their considerable size, the long breeding cycle, and the high variability of growing conditions (Pont, 2016). Advances in remote sensing and the emergence of sophisticated methods for large data analytics provide a means for describing and analysing phenotypic and environmental variation at a forest scale.
This study outlines the development and implementation of a phenotyping system that provides spatial estimates of stand productivity across a large plantation forest. Using a machine learning method forest productivity was modelled from an extensive set of 18 million observations of 93 variables describing climate, forest management, genetics and terrain, extracted from environmental surfaces, management records and LiDAR data (Watt, et al., 2013). The most important determinants of productivity were the genetic information and seasonal air temperatures, followed by variables describing the silvicultural treatment. The phenotyping method developed here can be used to identify superior and inferior genotypes and estimate a productivity index for each site, which will improve tree breeding and increase overall productivity across the forest.
Dungey, H. S., Dash, J. P., Pont, D., Clinton, P. W., Watt, M. S., & Telfer, E. J. (2018). Phenotyping Whole Forests Will Help to Track Genetic Performance. Trends in plant science.
Pont, D. (2016). Assessment of individual trees using aerial laser scanning in New Zealand radiata pine forests.
Watt, P., & Watt, M. S. (2013). Development of a national model of Pinus radiata stand volume from LiDAR metrics for New Zealand. International journal of remote sensing, 34(16), 5892-5904.
ABOUT THE AUTHOR
Maxime Bombrun is a data scientist in the Data Analytics team at Scion. He received a PhD in image processing and geology from the University of Clermont-Ferrand, France, in 2015, and completed a two-year post-doctoral fellowship in biomedicine at the University of Uppsala, Sweden. His research interests include image processing, statistical learning and their application in data science. He has published more than 15 papers, mostly in the field of image processing and data analysis.