Unmanned aerial system and satellite-based high resolution imagery for high-throughput phenotyping in dry bean

Abstract

Dry bean breeding programs are crucial to improve the productivity and resistance to biotic and abiotic stress. Phenotyping is a key process in breeding that refers to crop trait evaluation. In recent years, high-throughput plant phenotyping methods are being developed to increase the accuracy and efficiency for crop trait evaluations. In this study, aerial imagery at different resolutions were evaluated to phenotype crop performance and phenological traits using genotypes from two breeding panels, Durango Diversity Panel (DDP) and Andean Diversity Panel (ADP). The unmanned aerial system (UAS) based multispectral and thermal data were collected for two seasons at multiple time points (about 50, 60 and 75 days after planting/DAP in 2015; about 60 and 75 DAP in 2017). Four image-based features were extracted from multispectral images. Among different features, normalized difference vegetation index (NDVI) data were found to be consistently highly correlated with performance traits (above ground biomass, seed yield), especially during imaging at about 60–75 DAP (early pod development). Overall, correlations were higher using NDVI in ADP than DDP with biomass (r = −0.67 to −0.91 in ADP; r = −0.55 to −0.72 in DDP) and seed yield (r = 0.51 to 0.73 in ADP; r = 0.42 to 0.58 in DDP) at about 60 and 75 DAP. For thermal data, a temperature data normalization (utilizing common breeding plots in multiple thermal images) was implemented and the MEAN plot temperatures generally correlated significantly with biomass (r = 0.28–0.88). Finally, lower resolution satellite images (0.05–5 m/pixel) using UAS data was simulated and image resolution beyond 50 cm was found to reduce the relationship between image features (NDVI) and performance variables (biomass, seed yield). Four different high resolution satellite images: Pleiades-1A (0.5 m), SPOT 6 (1.5 m), Planet Scope (3.0 m), and Rapid Eye (5.0 m) were acquired to validate the findings from the UAS data. The results indicated sub-meter resolution satellite multispectral imagery showed promising application in field phenotyping, especially when the genotypic responses to stress is prominent. The correlation between NDVI extracted from Pleiades-1A images with seed yield (r = 0.52) and biomass (r = −0.55) were stronger in ADP; where the strength in relationship reduced with decreasing satellite image resolution. In future, we anticipate higher spatial and temporal resolution data achieved with low-orbiting satellites will increase applications for high-throughput crop phenotyping.