UAV-Based Imaging Methods to Examine Herbicide Phytotoxicity in Lentil (Lens culinaris Medik.)

Abstract

Weed control is of great importance in the successful management of lentil due to its poor competitive ability and short stature. With a lack of effective herbicides and an increase in Group 2 herbicide resistant weeds, weed control is becoming more challenging to lentil producers. Field ratings to assess herbicide safety and phytotoxicity in crops is a tedious and error-prone process. The purpose of this research is to determine if phenotyping of lentil phytotoxicity and varietal tolerance is possible through UAV image-based methods. Field studies involving a two-factor randomized complete block design were conducted at two locations in Saskatchewan, Canada in the 2019 and 2020 growing seasons. The factors of lentil variety and herbicide rate were assessed, which included four commercial lentil varieties and two registered herbicide products-metribuzin and saflufenacil. Saflufenacil was applied in a dose-response style from 0 to 10x the recommended label rate in 2019 and 0 to 32x in the 2020 growing season. Metribuzin was applied from 0 to 10x the label rate in 2019 and from 0 to 20x in the 2020 growing season. Hyperspectral UAV imaging methods were utilized for the metribuzin study whereas multispectral UAV imagery methods were utilized for the saflufenacil study for multiple flights throughout the growing season. Increasing the dose of saflufenacil decreased ground-reference measures of above-ground lentil biomass and plant population density; as well as decreased canopy cover extracted from multispectral imagery which was represented by a threshold NDVI pixel count layer. The CDC Improve lentil variety was found to be more tolerant to saflufenacil application than CDC Impala through dose-response analysis of ground-reference data and pixel count data. Increasing the metribuzin dose decreased ground reference data measurements as well as influenced both visible and near infrared spectral reflectance from hyperspectral imagery. The spectra were noted to differ especially in the green peak, red-edge, and near-infrared regions of the electromagnetic spectrum between increasing doses of metribuzin. The CDC Greenstar lentil variety did not differ from CDC Maxim in tolerance levels following metribuzin application through dose-response analysis of both ground-reference and NDVI reflectance data. Both multispectral and hyperspectral imagery methods had strong correlations with ground reference data and detected differences (P<0.05) between the two varieties tested through dose-response models. The imagery methods of this thesis can aid in manual data collection when assessing varietal tolerance in lentils.