Development of Sensor-based Phenotyping Methods for Ascochyta Blight Resistance Breeding of Chickpea

Abstract

Increasing pulse production can benefit global nutritional security and improve the sustainability of food production. The breeding of chickpea, the second most widely cultivated pulse, for resistance to Ascochyta blight can contribute to this goal, as this fungal disease limits yield in all major growing areas. Applying sensor-based phenotyping to disease resistance breeding programs allows for the rapid, precise, and nondestructive screening of large numbers of plants for improved disease resistance. In this way, the genetic gain can be increased. Previous studies have demonstrated the use of optical sensors for resistance breeding in other plant-pathosystems, but not yet for Ascochyta blight of chickpea. More broadly, the deployment of sensor-based phenotyping to uncontrolled conditions in the field, especially for measuring plant disease, remains a challenge. Therefore, the objectives of this thesis were (i) to develop sensor-based phenotyping methods for scoring Ascochyta blight disease severity in a chickpea breeding program and (ii) to identify within-scale and cross-scale functional resistance components. This was addressed by (a) reviewing the literature on sensor-based phenotyping of plant-pathogen interaction and (b) using an experimental approach in three different growth environments where chickpea and wild relatives were screened under disease pressure. The main outcomes from approach (a), the literature review, were the identification of potentially suitable sensor technology for Ascochyta blight of chickpea as RGB imaging, fluorescence imaging, hyperspectral imaging, and light detection and ranging, and the designation of potential measurable resistance components such as loss of healthy biomass, necrosis, chlorosis, and changes in plant metabolism. The experimental approach was conducted at (b1) single plant scale under controlled conditions in a glasshouse, (b2) at single pot scale in a disease nursery, and (b3) at plot scale in the field. In the glasshouse (b1), time course RGB imaging and extraction of growth rates in response to infection was a suitable and transferable method for predicting disease, achieving an R2 of 0.37 to 0.64 on unseen data. In the disease nursery (b2), also using RGB time course imaging derived growth rates, disease severity levels could be classified on unseen data with overall accuracies of 65 - 81 % (k = 0.43 - 0.59). For the purpose of early disease detection, infected and non-infected pots could be distinguished using lesion detection with 92.6 % accuracy 30 days after infection in one experiment, and 75.7 % accuracy 42 days after infection in a second experiment. In the field (b3), only single time points of data were acquired with a ground-based phenotyping platform that carried hyperspectral and lidar sensors. Using those single time point data, a general plant stress response expressed in the near infrared spectrum could rank genotypes according to their disease scores with r of 0.89. Across environments (b1) and (b2), models trained on spectral data for early disease detection and prediction of disease severity showed poor transferability to independent data. When hyperspectral data were used to predict disease severity in the glasshouse (b1) and field (b3), the degree of success was significantly impacted by the data preprocessing. These studies showed that time course RGB imaging to derive growth rate estimates and the measurement of general plant stress responses in the red edge and VNIR spectrum were suitable to assess the severity of Ascochyta blight disease for resistance breeding, and demonstrated the use of lesion detection for early detection of the disease. This thesis represents an important step toward supporting disease resistance breeding for Ascochyta blight of chickpea and for disease resistance breeding programs of other crops.