Abstract Detail



Biodiversity Informatics & Herbarium Digitization

Valdiviezo , Milton I [1], Dowell, Jordan [2], Mason, Chase [3].

Comparing Leaf Reflectance Analysis Prediction Models Based on Dried Whole Leaf Tissue against Dried Ground Tissue.

In recent years, the analysis of hyperspectral leaf reflectance has become increasingly applied as an alternative method of determining chemical and physiological traits in plants. Common applications include the derivation of photosynthetic traits, the analysis of physiochemical properties in leaves and the determination of nutrient content. This study focuses on the determination of nutrient content. Though such analyses may be conducted through the use of well-established conventional techniques, these are often costly, time-consuming, and thus ineffective for large-scale phenotyping efforts. Conversely, leaf reflectance analysis offers rapid, non-destructive data collection and negligible per-sample costs; an ideal method for high throughput phenotyping. Recent work has primarily been concerned with the analysis of single species composed of a few genotypes, however, the applicability of hyperspectral reflectance for analysis of nutrient contents and other physiological traits across diverse genera of plants remains understudied. Furthermore, as this modern method is relatively new and continually undergoing refinement, the optimal protocol for taking spectral measurements in leaves has yet to be determined. This study contributes to the refinement process by attempting to determine whether dried whole leaf tissue or dried ground leaf tissue is best for this kind of analysis. In this study, we explored the predictive capabilities of models based on these different tissue types within genus Cornus. We hypothesized that predictive models based on dried ground leaf tissue samples would be more accurate than those based on dried whole leaf samples as homogenized tissue is not subjected to the morphological and physiological differences among leaves that are apparent in whole leaf samples. We used a complimentary pair of spectrophotometers (FLAME-CHEM-UV-VIS Spectrophotometer 200-850nm and NIRQuest 512 Near-infrared Spectrophotometer 900-1700nm) to collect hyperspectral specular reflectance data from a wide cross-section of species of the genus Cornus sampled from Harvard University’s Arnold Arboretum. The samples were first sent to the Louisiana State University Soil Test and Plant Analysis Lab to obtain the ground-truthed nutrient content data via ICP-MS. We then used this dataset in conjunction with our reflectance dataset to build two sets of predictive models (one for dried whole leaf samples and one for dried ground tissue samples) using a support vector machine. Finally, we compared results obtained from both models to determine which tissue type offers greater predictive power, providing insights on whether morphological and physiological differences in dried ground tissue deliver a confounding signal that hinders the method’s predictive power.


1 - University of Central Florida, Biology, 1306 Dorado Drive, Apt A, Kissimmee, Fl, 34741, USA
2 - 14540 Lake Price Drive, Orlando, FL, 32826, United States
3 - University of Central Florida, Department of Biology, 4110 Libra Drive, Orlando, FL, 32816, US

Keywords:
foliar chemistry
Cornus
SVM
functional traits
reflectance spectroscopy
remote sensing
nutrient content estimation
machine learning
Hypersepctral reflectance
spectroscopy.

Presentation Type: Poster This poster will be presented at 5:30 pm. The Poster Session runs from 5:30 pm to 7:00 pm. Posters with odd poster numbers are presented at 5:30 pm, and posters with even poster numbers are presented at 6:15 pm.
Number: PBI007
Abstract ID:766
Candidate for Awards:Physiological Section Best poster presentation,Physiological Section Physiological Section Li-COR Prize


Copyright © 2000-2019, Botanical Society of America. All rights reserved