Temporal field phenomics of transgenic maize events subjected to drought stress: Cross‐validation scenarios and machine learning models

Abstract

Abstract Global climate change has driven breeding programs to develop abiotic stress‐resilient plant varieties. Traditionally, assessing drought resilience involves labor‐intensive and time‐consuming processes. This study used an unmanned aerial system (UAS) to predict key phenotyping traits in maize ( Zea mays L.) and monitor plant response to drought during the crop cycle. We grew transgenic maize hybrids in two trials, one irrigated and another subjected to drought stress, and used a drone equipped with red–green–blue (RGB) and multispectral sensors to capture images of the plots over time. Machine learning models and various prediction scenarios revealed significant correlations between vegetation indices over time. Interestingly, the RGB sensor outperformed the multispectral sensor in trait prediction. Prediction accuracy across scenarios with untested genotypes and environments ranged from 0.40 to 0.70 for grain yield, 0.43 to 0.69 for days to anthesis, 0.51 to 0.67 for days to silking, and 0.35 to 0.57 for plant height. Ridge and random forest models consistently delivered the most accurate predictions across traits and environments. The vegetation indices normalized green–red difference index, VARI, and RCC also effectively predicted and captured the plant response to drought. This study highlights the value of UAS phenotyping as a practical tool for assessing abiotic stress due to its straightforward implementation.

Code and data availability

The paper's data availability statement says all codes and datasets (phenomic prediction scripts, folder 'Phenomic prediction', and described datasets) are publicly available at the authors' GCCRC publications page and on Dryad (doi:10.5061/dryad.0zpc8677b).