University of Natural Resources and Life Sciences, Vienna (BOKU) - Research portal
Gewählte Doctoral Thesis:
Jiangsan Zhao
(2017):
Root phenotypic characterization and classification of pan-European legume cultivars.
Doctoral Thesis - Institut für Waldökologie (IFE),
BOKU-Universität für Bodenkultur,
pp 139.
UB BOKU
obvsg
FullText
- Abstract:
- Roots play an important role in crop yield and have become an increasingly widespread research topic. Although non-destructive methods are generally high-throughput and popular, destructive methods are still indispensable. One of the problems limiting the usability of destructive harvesting methods is the quality of data generated: low replicate number & non-Gaussianity – both problematic for traditional statistical analysis methods. Machine learning is generally more sensitive in detecting differences compared with traditional statistical methods. A significant improvement in the number of high-accuracy pairwise classifications of pea cultivars confirmed the usefulness of root traits selection –avoiding the “curse of high dimensionality”. The higher frequency of important root traits from the top soil layer generally highlighted their importance in differentiating between pea cultivars. Besides the advantages of machine learning compared with traditional statistical analysis, supervised ones (SML) take the information from both the label of the data and data itself into consideration. This is superior to unsupervised ones (UML), which only consider the data information. UML found clusters with a mixture of faba bean cultivars from both northern and southern Europe origins, which failed to explore the pattern of root adaptation to northern and southern environments. Based on SML, however, we found the clear separation of northern/southern Europe origin and the underlying root traits, which was also later confirmed by a permutation test and McNemar’s test. Faba bean cultivars from southern Europe were likely more drought tolerant due to the lengthier laterals along the taproot in the top 10 cm. Even though the non-destructive study of seedling roots is inherently high-throughput, the gap of transferability between non-destructive-high and generally destructive-low throughput (closer to reality) is immanent. I used models to simulate root growth of pea seedlings grown in agar towards mature pea plants with linear and exponential functions with different constraints; the best agreement of modeled root length and its distribution with real mature plants was the linear function considering both flowering time and scaling function on branching frequency of mature plants as constraints.
- Betreuer: Rewald Boris
- 1. Berater: Bodner Gernot
- 2. Berater: Sykacek Peter