A high-throughput UAV millimetric imaging pipeline for plant density phenotyping reveals high heritability in wheat-pea sole crops and intercrops.

Abstract

Within-field diversification, notably through cereal legume intercropping, is a major lever of the agroecological transition. Breeding cultivars adapted to intercropping requires major changes in breeding practices, owing to strong genotype-by-stand-type interactions and the high spatial heterogeneity of low-input field trials. Part of this heterogeneity stems from variation in plant density, a first yield component that strongly influences resource competition and is critical to monitor in intercrops. Given the many plots in such platforms, high-throughput phenotyping of plant density is essential. Over two seasons, we designed a large field trial involving 200 wheat cultivars observed as sole crop as well as intercropped with two pea tester cultivars, in a pseudo-randomized complete block design. Plant density was manually counted on 40 cultivars, alongwith UAV flights at both high and low altitudes. We propose a method to predict plant density from millimeter-resolution RGB images, using linear regression on log-transformed fractional vegetation cover and additional colorimetric and textural indices. We predict total density of wheat or pea in sole crops, or the combined pea+wheat total density in intercrops with relative RMSE of 19.2% and 27.4% depending on season. Thanks to the pipeline, total plant density was predicted across the entire trial, comprising more than 1,000 microplots (200 varieties); this density showed high heritability (0.75-0.85). We provide guidelines regarding the composition of the training set for such intercrop trials. High-throughput phenotyping approaches such as the one proposed here can be instrumental in mobilizing genetic diversity for agroecology.