In-Field Detection and Quantification of Septoria Tritici Blotch in Diverse Wheat Germplasm Using Spectral–Temporal Features

Abstract

Hyperspectral remote sensing holds the potential to detect and quantify crop diseases in a rapid and non-invasive manner. Such tools could greatly benefit resistance breeding, but their adoption is hampered by i) a lack of specificity to disease-related effects and ii) insufficient robustness to variation in reflectance caused by genotypic diversity and varying environmental conditions, which are fundamental elements of resistance breeding. We hypothesized that relying exclusively on temporal changes in canopy reflectance during pathogenesis may allow to specifically detect and quantify crop diseases while minimizing the confounding effects of genotype and environment. To test this hypothesis, we collected time-resolved canopy hyperspectral reflectance data for 18 diverse genotypes on infected and disease-free plots and engineered spectral–temporal features representing this hypothesis. Our results confirm the lack of specificity and robustness of disease assessments based on reflectance spectra at individual time points. We show that changes in spectral reflectance over time are indicative of the presence and severity of Septoria tritici blotch (STB) infections. Furthermore, the proposed time-integrated approach facilitated the delineation of disease from physiological senescence, which is pivotal for efficient selection of STB-resistant material under field conditions. A validation of models based on spectral–temporal features on a diverse panel of 330 wheat genotypes offered evidence for the robustness of the proposed method. This study demonstrates the potential of time-resolved canopy reflectance measurements for robust assessments of foliar diseases in the context of resistance breeding.

Predicting yellow rust in wheat breeding trials by proximal phenotyping and machine learning

“…For example, a SVI has been previously proposed to measure yellow rust in mid-late stage wheat [ 10 ] which relies on spectral data obtained at 860 nm, which is outside of the spectral range at which the spectroradiometer used in this study operates. Further improvement could come in form of replacing or combining the visual scores, used as ground truth data in this study, with more objective digital methods to estimate disease infection, such as destructive sampling and image analysis of diseased leaves, as was done in a similar study of field-based prediction of STB [ 13 ]. In that study, the authors reported a prediction accuracy of r = 0.53 when validating their proposed model in an independent test set.…”

“…Following the correlation analysis, the data at each timepoint was split into an 80/20 training and test data set, where the training data was used to perform recursive feature elimination and the test set was reserved for the validation of the final model trained based on the results of the feature elimination. Thus, a minimal set of predictors was established by performing supervised feature selection by recursive feature elimination with random forest (RF) regression, by adapting a protocol and associated code by [ 13 ]. The feature elimination was performed separately for each time point and proceeded by stepwise elimination of predictors from 119 predictors down to one, repeated 30 times.…”

“…Proximal phenotyping of both populations involved collecting data on Photosystem II quantum-yield using an active light fluorometer, spectral reflectance data (350–1150 nm) using a spectroradiometer, and finally leaf temperature data with an infrared thermometer. Random forest models in combination with recursive feature selection were successfully used for selecting and evaluating predictors for both chlorosis and necrosis [ 13 ] developed a promising time-resolved spectral method for quantifying STB severity in winter wheat under field conditions. The authors argued that spectral features estimated at single time points lack specificity to disease effects across time and are not robust to changes in reflectance caused by nuisance factors and environmental conditions.…”

Omics for the Improvement of Abiotic, Biotic, and Agronomic Traits in Major Cereal Crops: Applications, Challenges, and Prospects

“…During the last two decades, genomics has revolutionized plant breeding mainly due to a reduction in genotyping costs, which results in the adoption of new technologies, such as linkage mapping, genome-wide association studies, genome-wide selection, and rapid generation advance [189]. Accurate genetic mapping and genome-wide selection require precise phenotyping of the plants.…”

Detection of crown rot in wheat utilising near-infrared spectroscopy: towards remote and robotic sensing

“…Current reflectance-based crop disease models are not sufficiently robust to account for these environmental variations. This problem has been reviewed in Anderegg et al 24 and examined in detail in previous literature. 25,26…”