Unsupervised Plot-Scale LAI Phenotyping via UAV-Based Imaging, Modelling, and Machine Learning.

Area of Science:

• Agricultural Science
• Remote Sensing
• Plant Breeding

Background:

• Accurate Leaf Area Index (LAI) estimation is crucial for crop growth monitoring and breeding.
• High-throughput phenotyping accelerates crop improvement by linking genetics to growth estimation.

Purpose of the Study:

• To develop and evaluate a hybrid method for estimating LAI using random forest regression (RFR) models trained on synthetic data.
• To assess the accuracy of RFR models for LAI estimation from UAV-based multispectral images under field conditions.

Main Methods:

• A hybrid approach combining a radiative transfer model for synthetic data generation and RFR for LAI estimation.
• Utilized UAV-based multispectral imagery and evaluated models on both synthetic and observed field data.
• Implemented background correction to address systematic overestimation in low-LAI conditions.

Main Results:

• RFR models accurately predicted LAI from background-corrected reflectance, achieving high correlation (r=0.95) and determination (R^2=0.90-0.91) coefficients in Exp16.
• Models showed slightly lower accuracy in Exp19 (r=0.80-0.83, R^2=0.63-0.69) but still captured spatiotemporal LAI variations.
• The method successfully identified variations related to growing stages, genotypes, and management practices.

Conclusions:

• The developed hybrid method provides a rapid, accurate, and non-destructive approach for phenotyping LAI dynamics during vegetative growth.
• This technique facilitates improved growth rate assessments, particularly beneficial for crop breeding programs.
• Background correction is essential for improving LAI estimation accuracy in scenarios with significant background effects.