Abstract
Abstract
Background and aims
Root distribution over the soil profile is important for crop resource uptake. Using machine learning (ML), this study investigated whether measured square root of planar root length density (Sqrt_pRLD) at different soil depths were related to uptake of isotope tracer (15N) and drought stress indicator (13C) in wheat, to reveal root function.
Methods
In the RadiMax semi-field root-screening facility 95 winter wheat genotypes were phenotyped for root growth in 2018 and 120 genotypes in 2019. Using the minirhizotron technique, root images were acquired across a depth range from 80 to 250 cm in May, June, and July and RL was extracted using a convolutional neural network. We developed ML models to explore whether the Sqrt_pRLD estimates at different soil depths were predictive of the uptake of deep soil nitrogen - using deep placement of 15N tracer as well as natural abundance of 13C isotope. We analyzed the correlations to tracer levels to both a parametrized root depth estimation and an ML approach. We further analyzed the genotypic effects on root function using mediation analysis.
Results
Both parametrized and ML models demonstrated clear correlations between Sqrt_pRLD distribution and resource uptake. Further, both models demonstrated that deep roots at approx. 150 to 170 cm depth were most important for explaining the plant content of 15N and 13C isotopes. The correlations were higher in 2018.
Conclusions
The results demonstrated that, parametrized models and ML-based analysis provided complementary insight into the importance of deep rooting for water and nitrogen uptake.
Funder
Royal Library, Copenhagen University Library
Publisher
Springer Science and Business Media LLC
Subject
Plant Science,Soil Science
Reference38 articles.
1. Avati A (2019) Bias-variance analysis: theory and practice, Stanford University. Available at. http://cs229.stanford.edu/summer2019. Accessed 21 Jun 2023
2. Awika HO, Mishra AK, Gill H, Awika HO, Mishra AK, Gill H, DiPiazza J, Avila CA, Joshi V (2021) Selection of nitrogen responsive root architectural traits in spinach using machine learning and genetic correlations. Sci Rep 11:9536. https://doi.org/10.1038/s41598-021-87870-z
3. Baker RE, Pena J-M, Jayamohan J, Jérusalem A (2018) Mechanistic models versus machine learning, a fight worth fighting for the biological community? Biol Lett 14:20170660. https://doi.org/10.1098/rsbl.2017.0660
4. Breiman L (2001) Random forests. Mach Learn 45:5–32. https://doi.org/10.1023/A:1010933404324
5. Breiman L (1996) Bagging predictors. Mach Learn 24:123–40. https://doi.org/10.1007/BF00058655