Improving estimation of days to maturity in field pea using RGB aerial imagery and machine learning

Abstract

Accurately estimating days to maturity (DTM) is essential for assessing local adaptation and yield potential in field pea (Pisum sativum L.) breeding programs. However, traditional manual scoring of DTM is labor-intensive and inefficient for large-scale, multi-environment trials. To address this challenge, we developed a high-throughput, low-cost phenotyping framework using uncrewed aerial systems (UASs) equipped with red-green-blue cameras, implemented within the North Dakota State University Pulse Crop Breeding Program. This study aimed to (1) compare aerial and manual phenotyping for DTM estimation, (2) identify the optimal assessment time point, and (3) detect significant loci associated with DTM in a panel of 300 genetically diverse pea accessions. Image-derived vegetation indices (VIs) collected 71 days after planting exhibited strong correlations with manually assessed DTM. Notably, vegetation indices demonstrated higher heritability (H2 = 0.91) compared to traditional DTM scores (H2 = 0.84). eXtreme Gradient Boosting models identified the visible atmospherically resistant index (31%), modified green-red vegetation index (17%), and redness index (13%) as the most predictive VIs. Genome-wide association mapping using these indices revealed three significant single nucleotide polymorphisms on chromosomes 3 and 5—variants not detected using traditional maturity data—highlighting the potential enhanced detection power of image-derived traits. This work demonstrates the utility of low-cost UAS platforms for scalable, non-destructive maturity estimation and illustrates their potential to uncover genetic components of economically important traits, offering new avenues for addressing missing heritability in legume breeding.