- phenomaize@mrizp.rs
- Address: Slobodana Bajića 1,11185 Belgrade
METHODOLOGY
Methodology - UAV based maize phenotyping
High-throughput field phenotyping (HTFP) refers to the monitoring of crop development and measurement of plant traits in field conditions, using a wide range of sensors coupled with stationary platforms, ground vehicles, occupied or unoccupied aerial vehicles (UAVs, drones), or even satellite imaging. With UAV-based maize phenotyping, due to their ease of use, high spatial coverage, and affordable price, UAVs are the most popular remote sensing platforms in plant breeding.
PHENO_MaizE UAV-based maize phenotyping
Will use a multi-rotor wing UAV coupled with an RGB sensor to image 200 maize inbred lines and their test-crosses with one tester (400 genotypes in total) at multiple time points in three different environments in Serbia. Specifically, two locations in the first experimental year (2025) and one location in the second experimental year (2026) will be used.
UAS-based field imaging will involve flying the drone to collect single images that will be merged into a georeferenced image of the flying area, commonly known as an orthomosaic, or a 3D representation of the area surface like a 3D point cloud or digital surface model (DSM).
RGB-UAS imagery will be collected weekly from plant emergence to the start of flowering, twice per week during flowering, and once per week after flowering to physiological maturity, totaling 25-30 digital orthomosaics for each trial each year. Ground-truth measurements will be collected at one time point in each trial. Open-source UAS data extraction pipelines will be used for plot-level extraction of the following spectral features: vegetation indices, percentage of canopy cover, plant count in early growth stages, and plant height.
Utilization of UAV-based maize phenotyping
The project data will be utilized for the following purposes:
A. Automation of data collection: Assessment of digitally derived plant number per plot and plant height to replace manual measurements.
B. Trait prediction: Development of predictive models for traits of interest in maize breeding using image-derived variables and orthomosaics per se. Different predictive modeling techniques will be employed and compared for their performance in trait prediction. A common approach will be used where a portion of the data is sampled for a training set, which is then used to build a model.
The remaining data points represent a validation set, used to test how well the model fits. A testing pipeline will be implemented to predict known (tested) genotypes in known environments, known genotypes in unknown (untested) environments, unknown genotypes in known environments, and the most interesting scenario where unknown genotypes in unknown environments are predicted.
Additionally, model prediction accuracies will be examined in relation to the type of training set used: inbred lines per se, their test-crosses, and combined inbred lines and test-crosses training set.
C. Identification of informative image variables and flights: Selection of the most informative image-derived variables and flights to optimize the utilization of resources and enhance the effectiveness of data collection.
The most informative flights according to their predictive ability of agronomic performance will be selected. To increase the interpretability of the models for conventional breeding applications, the most informative variables will be extracted based on a stepwise regression approach and coefficient analysis.
Find out more about the Pheno_MaizE project
Project Objectives
The overall objective of the project is to explore the use of high-throughput field phenotyping (HTFP), specifically, Unoccupied Aerial Systems (UAS) derived data, in temperate hybrid maize breeding programs.