Integration of Multiscale Sensing Data for Phenomics Applications

Worasit Sangjan

doi:10.7273/000006323

Sensing technologies can be a powerful tool for phenotyping in breeding programs. Plant phenotypes can be assessed non-invasively and repeatedly across the whole population and throughout the plant development period utilizing advanced sensors and remote sensing platforms. In this study, multiscale sensing platforms—satellite, unmanned aerial vehicle (UAV), proximal sensing system, and Internet of Things (IoT) based sensing systems—equipped with sensors such as visible/RGB, multispectral, and hyperspectral systems were utilized for field-based phenomics applications. The applicability of a suitable sensing technology depends on the area of study, specific phenomics application, sensor specification, and data acquisition conditions. Three main phenomics applications were explored: (i) pasture crop health status evaluation, (ii) above-ground biomass quantity and quality evaluation in the field pea, and (iii) evaluating wheat yield potential in winter and spring wheat. The first study demonstrates the reliability of using a high-resolution satellite (ground sampling distance, GSD = 3 m) and UAV imagery for pasture management. The data from multiscale sensing data showed that the grazing density significantly affected pasture biomass (p < 0.05) only in 2019, and the vegetation index (VI) data from the two imagery types were highly correlated (r ≥ 0.78, p < 0.001, 2019). In the second study, the above-ground biomass (AGBM) and biomass quality (12 quality traits) were evaluated using UAV-based RGB and multispectral imaging, and hyperspectral sensing, respectively, in the winter pea breeding program (2019 and 2020 seasons). Three image processing approaches were evaluated for AGBM estimation, where the best results were acquired using the 3D point cloud model at 1.5 alpha shape technique showing high correlation with harvested fresh (r = 0.78–0.81, p < 0.001) and dry (r = 0.70–0.81, p < 0.001) AGBM. Similarly, the selected features from the normalized difference spectral indices and the ratio spectral indices extracted from hyperspectral data with the random forest model provided high predictive accuracy for all 12 biomass quality traits (0.81 < R2 < 0. 93; 0.05 < RMSE (%) < 1.80; 0.03 < MAE (%) < 1.32). In the wheat study, the vegetation indies were highly correlated between satellite (GSD = 0.31 m) and UAV data (0.42 ≤ r ≤ 0.99, p < 0.01) from winter and spring wheat breeding trials (2020 and 2021). The yield prediction using such VIs with the high-resolution satellite imagery (6.26 ≤ RMSE% ≤ 25.49; 5.11 ≤ MAE% ≤ 20.95; 0.17 ≤ r ≤0.78) and UAV imagery (5.53 ≤ RMSE% ≤ 17.20; 4.28 ≤ MAE% ≤ 14.20; 0.43 ≤ r ≤ 0.92) was also high. In addition to these two platforms, an intelligent and compact IoT-based sensor system was developed for independent and automated phenomics applications to measure and monitor plant responses in real-time. The sensor development, improvisation, and implementation encompassed three field seasons (2020, 2021, and 2022 seasons). The developed IoT-based sensor system could be successfully implemented to monitor multiple trials for timely crop management and increased resource efficiency. The system shows a high potential for supporting plant breeding programs for in-field phenotyping applications. All studies demonstrated promising results in monitoring and estimating crop performance and phenotypic traits using multiscale sensing systems.

Integration of Multiscale Sensing Data for Phenomics Applications

Files and links (1)

Abstract

Metrics

Details