Remote sensing for field pea yield estimation: A study of multi-scale data fusion approaches in phenomics.

Introduction: Remote sensing using unmanned aerial systems (UAS) are prevalent for phenomics and precision agricultural applications. The high-resolution data for these applications can provide useful spectral characteristics of crops associated with performance traits such as seed yield. With the recent availability of high-resolution satellite imagery, there has been growing interest in using this technology for plot-scale remote sensing applications, particularly those related to breeding programs. This study compared the features extracted from high-resolution satellite and UAS multispectral imagery (visible and near-infrared) to predict the seed yield from two diverse plot-scale field pea yield trials (advanced breeding and variety testing) using the random forest model. Methods: The multi-modal (spectral and textural features) and multi-scale (satellite and UAS) data fusion approaches were evaluated to improve seed yield prediction accuracy across trials and time points. These approaches included both image fusion, such as pan-sharpening of satellite imagery with UAS imagery using intensity-hue-saturation transformation and additive wavelet luminance proportional approaches, and feature fusion, which involved integrating extracted spectral features. In addition, we also compared the image fusion approach to high-definition satellite data with a resolution of 0.15 m/pixel. The effectiveness of each approach was evaluated with data at both individual and combined time points. Results and discussion: The major findings can be summarized as follows: (1) the inclusion of the texture features did not improve the model performance, (2) the performance of the model using spectral features from satellite imagery at its original resolution can provide similar results as UAS imagery, with variation depending on the field pea yield trial under study and the growth stage, (3) the model performance improved after applying multi-scale, multiple time point feature fusion, (4) the features extracted from the pan-sharpened satellite imagery using intensity-hue-saturation transformation (image fusion) showed better model performance than those with original satellite imagery or high definition imagery, and (5) the green normalized difference vegetation index and transformed triangular vegetation index were identified as key features contributing to high model performance across trials and time points. These findings demonstrate the potential of high-resolution satellite imagery and data fusion approaches for plot-scale phenomics applications.

Evaluation of Effective Class-Balancing Techniques for CNN-Based Assessment of Aphanomyces Root Rot Resistance in Pea (Pisum sativum L.).

Aphanomyces root rot (ARR) is a devastating disease that affects the production of pea. The plants are prone to infection at any growth stage, and there are no chemical or cultural controls. Thus, the development of resistant pea cultivars is important. Phenomics technologies to support the selection of resistant cultivars through phenotyping can be valuable. One such approach is to couple imaging technologies with deep learning algorithms that are considered efficient for the assessment of disease resistance across a large number of plant genotypes. In this study, the resistance to ARR was evaluated through a CNN-based assessment of pea root images. The proposed model, DeepARRNet, was designed to classify the pea root images into three classes based on ARR severity scores, namely, resistant, intermediate, and susceptible classes. The dataset consisted of 1581 pea root images with a skewed distribution. Hence, three effective data-balancing techniques were identified to solve the prevalent problem of unbalanced datasets. Random oversampling with image transformations, generative adversarial network (GAN)-based image synthesis, and loss function with class-weighted ratio were implemented during the training process. The result indicated that the classification F1-score was 0.92 ± 0.03 when GAN-synthesized images were added, 0.91 ± 0.04 for random resampling, and 0.88 ± 0.05 when class-weighted loss function was implemented, which was higher than when an unbalanced dataset without these techniques were used (0.83 ± 0.03). The systematic approaches evaluated in this study can be applied to other image-based phenotyping datasets, which can aid the development of deep-learning models with improved performance.

Biogenic VOCs Emission Profiles Associated with Plant-Pest Interaction for Phenotyping Applications.

Pest attacks on plants can substantially change plants' volatile organic compounds (VOCs) emission profiles. Comparison of VOC emission profiles between non-infected/non-infested and infected/infested plants, as well as resistant and susceptible plant cultivars, may provide cues for a deeper understanding of plant-pest interactions and associated resistance. Furthermore, the identification of biomarkers-specific biogenic VOCs-associated with the resistance can serve as a non-destructive and rapid tool for phenotyping applications. This research aims to compare the VOCs emission profiles under diverse conditions to identify constitutive (also referred to as green VOCs) and induced (resulting from biotic/abiotic stress) VOCs released in potatoes and wheat. In the first study, wild potato Solanum bulbocastanum (accession# 22; SB22) was inoculated with Meloidogyne chitwoodi race 1 (Mc1), and Mc1 pathotype Roza (SB22 is resistant to Mc1 and susceptible to pathotype Roza), and VOCs emission profiles were collected using gas chromatography-flame ionization detection (GC-FID) at different time points. Similarly, in the second study, the VOCs emission profiles of resistant ('Hollis') and susceptible ('Alturas') wheat cultivars infested with Hessian fly insects were evaluated using the GC-FID system. In both studies, in addition to variable plant responses (susceptibility to pests), control treatments (non-inoculated or non-infested) were used to compare the VOCs emission profiles resulting from differences in stress conditions. The common VOC peaks (constitutive VOCs) between control and infected/infested samples, and unique VOC peaks (induced VOCs) presented only in infected/infested samples were analyzed. In the potato-nematode study, the highest unique peak was found two days after inoculation (DAI) for SB22 inoculated with Mc1 (resistance response). The most common VOC peaks in SB22 inoculated with both Mc1 and Roza were found at 5 and 10 DAI. In the wheat-insect study, only the Hollis showed unique VOC peaks. Interestingly, both cultivars released the same common VOCs between control and infected samples, with only a difference in VOC average peak intensity at 22.4 min retention time where the average intensity was 4.3 times higher in the infested samples of Hollis than infested samples of Alturas. These studies demonstrate the potential of plant VOCs to serve as a rapid phenotyping tool to assess resistance levels in different crops.

Identification of volatile biomarkers for high-throughput sensing of soft rot and Pythium leak diseases in stored potatoes.

Soft rot and Pythium leak are postharvest storage diseases of potato tubers that can cause substantial crop losses in the US. This study focused on detecting volatile organic compounds (VOCs) associated with rot inoculated tubers during storage (up to 21 days) using headspace solid-phase microextraction (SPME) coupled to gas chromatography (GC) with mass spectrometry (MS) and flame ionization detector (FID) analysis. Russet Burbank and Ranger Russet tubers were inoculated with the rot pathogens. Static sampling with 50 min trapping time followed by GC-MS and GC-FID analysis identified 23 and 30 common VOCs from the pathogen inoculated tubers. Overall, n,n-dimethylmethylamine, acetone, 1-undecene, and styrene, occurred frequently and repeatability in inoculated samples based on GC-MS analysis, with the latter two found using GC-FID analysis as well. Identification of such biomarkers can be useful in developing high-throughput VOC sensing systems for early disease detection in potato storage facilities.

Dissecting the Genetic Architecture of Aphanomyces Root Rot Resistance in Lentil by QTL Mapping and Genome-Wide Association Study.

Lentil (Lens culinaris Medikus) is an important source of protein for people in developing countries. Aphanomyces root rot (ARR) has emerged as one of the most devastating diseases affecting lentil production. In this study, we applied two complementary quantitative trait loci (QTL) analysis approaches to unravel the genetic architecture underlying this complex trait. A recombinant inbred line (RIL) population and an association mapping population were genotyped using genotyping by sequencing (GBS) to discover novel single nucleotide polymorphisms (SNPs). QTL mapping identified 19 QTL associated with ARR resistance, while association mapping detected 38 QTL and highlighted accumulation of favorable haplotypes in most of the resistant accessions. Seven QTL clusters were discovered on six chromosomes, and 15 putative genes were identified within the QTL clusters. To validate QTL mapping and genome-wide association study (GWAS) results, expression analysis of five selected genes was conducted on partially resistant and susceptible accessions. Three of the genes were differentially expressed at early stages of infection, two of which may be associated with ARR resistance. Our findings provide valuable insight into the genetic control of ARR, and genetic and genomic resources developed here can be used to accelerate development of lentil cultivars with high levels of partial resistance to ARR.

Generalized Linear Model with Elastic Net Regularization and Convolutional Neural Network for Evaluating Aphanomyces Root Rot Severity in Lentil.

Phenomics technologies allow quantitative assessment of phenotypes across a larger number of plant genotypes compared to traditional phenotyping approaches. The utilization of such technologies has enabled the generation of multidimensional plant traits creating big datasets. However, to harness the power of phenomics technologies, more sophisticated data analysis methods are required. In this study, Aphanomyces root rot (ARR) resistance in 547 lentil accessions and lines was evaluated using Red-Green-Blue (RGB) images of roots. We created a dataset of 6,460 root images that were annotated by a plant breeder based on the disease severity. Two approaches, generalized linear model with elastic net regularization (EN) and convolutional neural network (CNN), were developed to classify disease resistance categories into three classes: resistant, partially resistant, and susceptible. The results indicated that the selected image features using EN models were able to classify three disease categories with an accuracy of up to 0.91 ± 0.004 (0.96 ± 0.005 resistant, 0.82 ± 0.009 partially resistant, and 0.92 ± 0.007 susceptible) compared to CNN with an accuracy of about 0.84 ± 0.009 (0.96 ± 0.008 resistant, 0.68 ± 0.026 partially resistant, and 0.83 ± 0.015 susceptible). The resistant class was accurately detected using both classification methods. However, partially resistant class was challenging to detect as the features (data) of the partially resistant class often overlapped with those of resistant and susceptible classes. Collectively, the findings provided insights on the use of phenomics techniques and machine learning approaches to provide quantitative measures of ARR resistance in lentil.

Advanced Imaging for Quantitative Evaluation of Aphanomyces Root Rot Resistance in Lentil.

Aphanomyces root rot (ARR) is a soil-borne disease that results in severe yield losses in lentil. The development of resistant cultivars is one of the key strategies to control this pathogen. However, the evaluation of disease severity is limited to visual scores that can be subjective. This study utilized image-based phenotyping approaches to evaluate Aphanomyces euteiches resistance in lentil genotypes in greenhouse (351 genotypes from lentil single plant/LSP derived collection and 191 genotypes from recombinant inbred lines/RIL using digital Red-Green-Blue/RGB and hyperspectral imaging) and field (173 RIL genotypes using unmanned aerial system-based multispectral imaging) conditions. Moderate to strong correlations were observed between RGB, multispectral, and hyperspectral derived features extracted from lentil shoots/roots and visual scores. In general, root features extracted from RGB imaging were found to be strongly associated with disease severity. With only three root traits, elastic net regression model was able to predict disease severity across and within multiple datasets (R 2 = 0.45-0.73 and RMSE = 0.66-1.00). The selected features could represent visual disease scores. Moreover, we developed twelve normalized difference spectral indices (NDSIs) that were significantly correlated with disease scores: two NDSIs for lentil shoot section - computed from wavelengths of 1170, 1160, 1270, and 1280 nm (0.12 ≤ |r| ≤ 0.24, P < 0.05) and ten NDSIs for lentil root sections - computed from wavelengths in the range of 630-670, 700-840, and 1320-1530 nm (0.10 ≤ |r| ≤ 0.50, P < 0.05). Root-derived NDSIs were more accurate in predicting disease scores with an R 2 of 0.54 (RMSE = 0.86), especially when the model was trained and tested on LSP accessions, compared to R 2 of 0.25 (RMSE = 1.64) when LSP and RIL genotypes were used as train and test datasets, respectively. Importantly, NDSIs - computed from wavelengths of 700, 710, 730, and 790 nm - had strong positive correlations with disease scores (0.35 ≤r ≤ 0.50, P < 0.0001), which was confirmed in field phenotyping with similar correlations using vegetation index with red edge wavelength (normalized difference red edge, 0.36 ≤ |r| ≤ 0.57, P < 0.0001). The adopted image-based phenotyping approaches can help plant breeders to objectively quantify ARR resistance and reduce the subjectivity in selecting potential genotypes.

High-Throughput Phenotyping of Fire Blight Disease Symptoms Using Sensing Techniques in Apple.

Washington State produces about 70% of total fresh market apples in the United States. One of the primary goals of apple breeding programs is the development of new cultivars resistant to devastating diseases such as fire blight. The overall objective of this study was to investigate high-throughput phenotyping techniques to evaluate fire blight disease symptoms in apple trees. In this regard, normalized stomatal conductance data acquired using a portable photosynthetic system, image data collected using RGB and multispectral cameras, and visible-near infrared spectral reflectance acquired using a hyperspectral sensing system, were independently evaluated to estimate the progression of fire blight infection in young apple trees. Sensors with ranging complexity - from simple RGB to multispectral imaging to hyperspectral system - were evaluated to select the most accurate technique for the assessment of fire blight disease symptoms. The proximal multispectral images and visible-near infrared spectral reflectance data were collected in two field seasons (2016, 2017); while, proximal side-view RGB images and multispectral images using unmanned aerial systems were collected in 2017. The normalized stomatal conductance data was correlated with disease severity rating (r = 0.51, P < 0.05). The features extracted from RGB images (e.g., maximum length of senesced leaves, area of senesced leaves, ratio between senesced and healthy leaf area) and multispectral images (e.g., vegetation indices) also demonstrated potential in evaluation of disease rating (|r| > 0.35, P < 0.05). The average classification accuracy achieved using visible-near infrared spectral reflectance data during the classification of susceptible from symptomless groups ranged between 71 and 93% using partial least square regression and quadratic support vector machine. In addition, fire blight disease ratings were compared with normalized difference spectral indices (NDSIs) that were generated from visible-near infrared reflectance spectra. The selected spectral bands in the range 710-2,340 nm used for computing NDSIs showed consistently higher correlation with disease severity rating than data acquired from RGB and multispectral imaging sensors across multiple seasons. In summary, these specific spectral bands can be used for evaluating fire blight disease severity in apple breeding programs and potentially as early fire blight disease detection tool to assist in production systems.