Unmanned aerial vehicle-based prediction of cold tolerance indicators in sugarcane (Saccharum spp. hybrids) varieties

Abstract

Louisiana is one of only two remaining sugarcane producing states in the U.S., and the industry is faced with labor shortage. Integration of predictive models incorporating markers offers a non-destructive tool for precision breeding. Chemical markers allow a direct measurement of damage and tolerance for sugarcane against winter freeze, which is the primary abiotic stress in Louisiana representing the northernmost sugarcane growing region worldwide. This study first utilized exploratory (cluster and principal component) analyses to show the effects of air temperature, but not genotype, on red, green, and blue reflectance data collected by unmanned aerial vehicle (UAV). Of tested algorithms (multiple linear regression (MLR), XGBoost, partial least squares, and artificial neural network), best fit models were obtained by MLR for yield (theoretical recoverable sugar, Cane Pol, Cane Brix, fiber, and moisture content), primary product (sucrose), and freeze damage indicators (fructose and glucose hydrolysis products of sucrose). Parts per million-level cold tolerance indicator (tyrosine-like fluorophore) and additional secondary products (polyphenols and trans-aconitic acid) in juice were modeled after concentrations were normalized to the canopy coverage, as the UAV sensor is detecting the canopy pixels. Built models could be used in freeze damage assessment as well as marker-assisted tolerant variety development, without the constraint of waiting for the abiotic stress to happen.