Phenotyping for heat stress tolerance in wheat population using physiological traits, multispectral imagery, and machine learning approaches

Heat stress is a critical environmental factor that adversely affects crop productivity. With the increasing frequency and intensity of heat waves and extreme weather events, heat stress has become a challenge for wheat production, which is one of the most important cereal crops. To sustain wheat production under heat stress conditions, there is an urgent need to develop high-yielding, heat-tolerant wheat varieties. This requires characterizing the genetic and physiological mechanisms underlying heat tolerance, as well as developing efficient phenotyping methods to evaluate a large number of wheat genotypes under heat stress field conditions. In this study, we used 184 wheat genotypes that were sown at two times of sowing (TOS), i.e., optimal sowing as TOS1 and late sowing as TOS2, with higher temperatures faced by plants during heading and grain filling in TOS2. We used a combination of physiological traits, multispectral vegetative indices (VIs) derived from aerial imagery and machine learning approaches to effectively differentiate wheat genotypes for heat tolerance and susceptibility. The response of wheat genotypes to heat stress was delineated as being susceptible, moderate, and tolerant using the stress susceptibility index, percentage loss, and tolerance index. Different VIs varied significantly between the two TOS. The decline in VIs during anthesis and post-anthesis was minimal in heat tolerant genotypes compared to susceptible genotypes under TOS2. We classified the stress severity and yield using VIs with a machine learning approach. A model was created with a random forest classifier (RFC) trained to categorize genotypes based on the stress susceptibility index using Python libraries. The PCA was utilized to reduce dimensionality, and five principal components explaining 99 % of the variability were employed as input for developing the model. The RFC model achieved an accuracy of 64 % and excelled in recognizing crops under extreme stress, with a recall rate of 0.87 and an F1 score of 0.77 for the susceptible class. The model had high precision metrics, with values of 0.69, 0.42, and 0.80 for the susceptible, moderate, and tolerant classes, respectively. Our results suggest that multispectral-driven phenotypic traits can be used by breeders to select and develop wheat varieties tolerant to heat stress.