Frontiers in Artificial Intelligence (Jan 2025)
Phenology analysis for trait prediction using UAVs in a MAGIC rice population with different transplanting protocols
Abstract
Unmanned aerial vehicles (UAVs) are one of the most effective tools for crop monitoring in the field. Time-series RGB and multispectral data obtained with UAVs can be used for revealing changes of three-dimensional growth. We previously showed using a rice population with our regular cultivation protocol that canopy height (CH) parameters extracted from time-series RGB data are useful for predicting manually measured traits such as days to heading (DTH), culm length (CL), and aboveground dried weight (ADW). However, whether CH parameters are applicable to other rice populations and to different cultivation methods, and whether vegetation indices such as the chlorophyll index green (CIg) can function for phenotype prediction remain to be elucidated. Here we show that CH and CIg exhibit different patterns with different cultivation protocols, and each has its own character for the prediction of rice phenotypes. We analyzed CH and CIg time-series data with a modified logistic model and a double logistic model, respectively, to extract individual parameters for each. The CH parameters were useful for predicting DTH, CL, ADW and stem and leaf weight (SLW) in a newly developed rice population under both regular and delayed cultivation protocols. The CIg parameters were also effective for predicting DTH and SLW, and could also be used to predict panicle weight (PW). The predictive ability worsened when different cultivation protocols were used, but this deterioration was mitigated by a calibration procedure using data from parental cultivars. These results indicate that the prediction of DTH, CL, ADW and SLW by CH parameters is robust to differences in rice populations and cultivation protocols, and that CIg parameters are an indispensable complement to the CH parameters for the predicting PW.
Keywords
