Application of unmanned aerial systems for high throughput phenotyping of large wheat breeding nurseries

Reviewing the essential roles of remote phenotyping, GWAS and explainable AI in practical marker-assisted selection for drought-tolerant winter wheat breeding

Marker-assisted selection (MAS) plays a crucial role in crop breeding improving the speed and precision of conventional breeding programmes by quickly and reliably identifying and selecting plants with desired traits. However, the efficacy of MAS depends on several prerequisites, with precise phenotyping being a key aspect of any plant breeding programme. Recent advancements in high-throughput remote phenotyping, facilitated by unmanned aerial vehicles coupled to machine learning, offer a non-destructive and efficient alternative to traditional, time-consuming, and labour-intensive methods. Furthermore, MAS relies on knowledge of marker-trait associations, commonly obtained through genome-wide association studies (GWAS), to understand complex traits such as drought tolerance, including yield components and phenology. However, GWAS has limitations that artificial intelligence (AI) has been shown to partially overcome. Additionally, AI and its explainable variants, which ensure transparency and interpretability, are increasingly being used as recognised problem-solving tools throughout the breeding process. Given these rapid technological advancements, this review provides an overview of state-of-the-art methods and processes underlying each MAS, from phenotyping, genotyping and association analyses to the integration of explainable AI along the entire workflow. In this context, we specifically address the challenges and importance of breeding winter wheat for greater drought tolerance with stable yields, as regional droughts during critical developmental stages pose a threat to winter wheat production. Finally, we explore the transition from scientific progress to practical implementation and discuss ways to bridge the gap between cutting-edge developments and breeders, expediting MAS-based winter wheat breeding for drought tolerance.

Water Stress Alters Physiological, Spectral, and Agronomic Indexes of Wheat Genotypes

Selecting drought-tolerant and more water-efficient wheat genotypes is a research priority, specifically in regions with irregular rainfall or areas where climate change is expected to result in reduced water availability. The objective of this work was to use high-throughput measurements with morphophysiological traits to characterize wheat genotypes in relation to water stress. Field experiments were conducted from May to September 2018 and 2019, using a sprinkler bar irrigation system to control water availability to eighteen wheat genotypes: BRS 254; BRS 264; CPAC 01019; CPAC 01047; CPAC 07258; CPAC 08318; CPAC 9110; BRS 394 (irrigated biotypes), and Aliança; BR 18_Terena; BRS 404; MGS Brilhante; PF 020037; PF 020062; PF 120337; PF 100368; PF 080492; and TBIO Sintonia (rainfed biotypes). The water regimes varied from 22 to 100% of the crop evapotranspiration replacement. Water stress negatively affected gas exchange, vegetation indices, and grain yield. High throughput variables TCARI, NDVI, OSAVI, SAVI, PRI, NDRE, and GNDVI had higher yield and morphophysiological measurement correlations. The drought resistance index indicated that genotypes Aliança, BRS 254, BRS 404, CPAC 01019, PF 020062, and PF 080492 were more drought tolerant.

The potential of UAV and very high-resolution satellite imagery for yellow and stem rust detection and phenotyping in Ethiopia

Very high (spatial and temporal) resolution satellite (VHRS) and high-resolution unmanned aerial vehicle (UAV) imagery provides the opportunity to develop new crop disease detection methods at early growth stages with utility for early warning systems. The capability of multispectral UAV, SkySat and Pleiades imagery as a high throughput phenotyping (HTP) and rapid disease detection tool for wheat rusts is assessed. In a randomized trial with and without fungicide control, six bread wheat varieties with differing rust resistance were monitored using UAV and VHRS. In total, 18 spectral features served as predictors for stem and yellow rust disease progression and associated yield loss. Several spectral features demonstrated strong predictive power for the detection of combined wheat rust diseases and the estimation of varieties' response to disease stress and grain yield. Visible spectral (VIS) bands (Green, Red) were more useful at booting, shifting to VIS-NIR (near-infrared) vegetation indices (e.g., NDVI, RVI) at heading. The top-performing spectral features for disease progression and grain yield were the Red band and UAV-derived RVI and NDVI. Our findings provide valuable insight into the upscaling capability of multispectral sensors for disease detection, demonstrating the possibility of upscaling disease detection from plot to regional scales at early growth stages.

The Importance of Using Realistic 3D Canopy Models to Calculate Light Interception in the Field

Quantifying canopy light interception provides insight into the effects of plant spacing, canopy structure, and leaf orientation on radiation distribution. This is essential for increasing crop yield and improving product quality. Canopy light interception can be quantified using 3-dimensional (3D) plant models and optical simulations. However, virtual 3D canopy models (VCMs) have often been used to quantify canopy light interception because realistic 3D canopy models (RCMs) are difficult to obtain in the field. This study aims to compare the differences in light interception between VCMs and RCM. A realistic 3D maize canopy model (RCM) was reconstructed over a large area of the field using an advanced unmanned aerial vehicle cross-circling oblique (CCO) route and the structure from motion-multi-view stereo method. Three types of VCMs (VCM-1, VCM-4, and VCM-8) were then created by replicating 1, 4, and 8 individual realistic plants constructed by CCO in the center of the corresponding RCM. The daily light interception per unit area (DLI), as computed for the 3 VCMs, exhibited marked deviation from the RCM, as evinced by the relative root mean square error (rRMSE) values of 20.22%, 17.38%, and 15.48%, respectively. Although this difference decreased as the number of plants used to replicate the virtual canopy increased, rRMSE of DLI for VCM-8 and RCM still reached 15.48%. It was also found that the difference in light interception between RCMs and VCMs was substantially smaller in the early stage (48 days after sowing [DAS]) than in the late stage (70 DAS). This study highlights the importance of using RCM when calculating light interception in the field, especially in the later growth stages of plants.

Reflectance Measurements from Aerial and Proximal Sensors Provide Similar Precision in Predicting the Rice Yield Response to Mid-Season N Applications

Accurately detecting nitrogen (N) deficiency and determining the need for additional N fertilizer is a key challenge to achieving precise N management in many crops, including rice (Oryza sativa L.). Many remotely sensed vegetation indices (VIs) have shown promise in this regard; however, it is not well-known if VIs measured from different sensors can be used interchangeably. The objective of this study was to quantitatively test and compare the ability of VIs measured from an aerial and proximal sensor to predict the crop yield response to top-dress N fertilizer in rice. Nitrogen fertilizer response trials were established across two years (six site-years) throughout the Sacramento Valley rice-growing region of California. At panicle initiation (PI), unmanned aircraft system (UAS) Normalized Difference Red-Edge Index (NDRE_UAS) and GreenSeeker (GS) Normalized Difference Vegetation Index (NDVI_GS) were measured and expressed as a sufficiency index (SI) (VI of N treatment divided by VI of adjacent N-enriched area). Following reflectance measurements, each plot was split into subplots with and without top-dress N fertilizer. All metrics evaluated in this study indicated that both NDRE_UAS and NDVI_GS performed similarly with respect to predicting the rice yield response to top-dress N at PI. Utilizing SI measurements prior to top-dress N fertilizer application resulted in a 113% and 69% increase (for NDRE_UAS and NDVI_GS, respectively) in the precision of the rice yield response differentiation compared to the effect of applying top-dress N without SI information considered. When the SI measured via NDRE_UAS and NDVI_GS at PI was ≤0.97 and 0.96, top-dress N applications resulted in a significant (p < 0.05) increase in crop yield of 0.19 and 0.21 Mg ha^-1, respectively. These results indicate that both aerial NDRE_UAS and proximal NDVI_GS have the potential to accurately predict the rice yield response to PI top-dress N fertilizer in this system and could serve as the basis for developing a decision support tool for farmers that could potentially inform better N management and improve N use efficiency.

Unsupervised Plot-Scale LAI Phenotyping via UAV-Based Imaging, Modelling, and Machine Learning

High-throughput phenotyping has become the frontier to accelerate breeding through linking genetics to crop growth estimation, which requires accurate estimation of leaf area index (LAI). This study developed a hybrid method to train the random forest regression (RFR) models with synthetic datasets generated by a radiative transfer model to estimate LAI from UAV-based multispectral images. The RFR models were evaluated on both (i) subsets from the synthetic datasets and (ii) observed data from two field experiments (i.e., Exp16, Exp19). Given the parameter ranges and soil reflectance are well calibrated in synthetic training data, RFR models can accurately predict LAI from canopy reflectance captured in field conditions, with systematic overestimation for LAI<2 due to background effect, which can be addressed by applying background correction on original reflectance map based on vegetation-background classification. Overall, RFR models achieved accurate LAI prediction from background-corrected reflectance for Exp16 (correlation coefficient (r) of 0.95, determination coefficient (R²) of 0.90~0.91, root mean squared error (RMSE) of 0.36~0.40 m² m⁻², relative root mean squared error (RRMSE) of 25~28%) and less accurate for Exp19 (r =0.80~0.83, R² = 0.63~0.69, RMSE of 0.84~0.86 m² m⁻², RRMSE of 30~31%). Additionally, RFR models correctly captured spatiotemporal variation of observed LAI as well as identified variations for different growing stages and treatments in terms of genotypes and management practices (i.e., planting density, irrigation, and fertilization) for two experiments. The developed hybrid method allows rapid, accurate, nondestructive phenotyping of the dynamics of LAI during vegetative growth to facilitate assessments of growth rate including in breeding program assessments.

Utilization of Spectral Indices for High-Throughput Phenotyping

The conventional plant breeding evaluation of large sets of plant phenotypes with precision and speed is very challenging. Thus, consistent, automated, multifaceted, and high-throughput phenotyping (HTP) technologies are becoming increasingly significant as tools to aid conventional breeding programs to develop genetically improved crops. With rapid technological advancement, various vegetation indices (VIs) have been developed. These VI-based imaging approaches, linked with artificial intelligence and a variety of remote sensing applications, provide high-throughput evaluations, particularly in the field of precision agriculture. VIs can be used to analyze and predict different quantitative and qualitative aspects of vegetation. Here, we provide an overview of the various VIs used in agricultural research, focusing on those that are often employed for crop or vegetation evaluation, because that has a linear relationship to crop output, which is frequently utilized in crop chlorophyll, health, moisture, and production predictions. In addition, the following aspects are here described: the importance of VIs in crop research and precision agriculture, their utilization in HTP, recent photogrammetry technology, mapping, and geographic information system software integrated with unmanned aerial vehicles and its key features. Finally, we discuss the challenges and future perspectives of HTP technologies and propose approaches for the development of new tools to assess plants' agronomic traits and data-driven HTP resolutions for precision breeding.

Biomass Estimation of Agave durangensis Gentry Using High-Resolution Images in Nombre de Dios, Durango

The high demand for distilled agave products reduces wild populations. The use of geospatial technologies such as unmanned aerial vehicles (UAVs) offer enormous benefits in spatial and temporal resolution and lower costs than traditional direct field observation techniques for natural resource monitoring. The objective was to estimate the green biomass (Wt) of Agave durangensis Gentry using high-resolution images obtained by a UAV in Nombre de Dios, Durango. Random sampling was performed in the agave area. A Pearson correlation analysis was performed, followed by a regression analysis. The results showed that NDVI was the most correlated (r = 0.65). The regression analysis showed that the model obtained explains 59% (RMSE = 32.06 kg) of the total variability in the estimation of green biomass (Wt) of agave using images derived from the UAV. The best estimate was achieved with B1, B2, NDVI, GNDVI, EVI2, and SAVI as predictor variables. High-resolution images were shown to be a tool for estimating Wt of Agave durangensis Gentry.

Comparative Sensitivity of Vegetation Indices Measured via Proximal and Aerial Sensors for Assessing N Status and Predicting Grain Yield in Rice Cropping Systems

Reflectance-based vegetation indices can be valuable for assessing crop nitrogen (N) status and predicting grain yield. While proximal sensors have been widely studied in agriculture, there is increasing interest in utilizing aerial sensors. Given that few studies have compared aerial and proximal sensors, the objective of this study was to quantitatively compare the sensitivity of aerially sensed Normalized Difference Vegetation Index (NDVI) and Normalized Difference Red-Edge Index (NDRE) and proximally sensed NDVI for assessing total N uptake at panicle initiation (PI-NUP) and predicting grain yield in rice. Nitrogen response trials were established over a 3-year period (10 site-years) at various locations throughout the Sacramento Valley rice growing region of California. At PI, a multispectral unmanned aircraft system (UAS) was used to measure NDVIUAS and NDREUAS (average ground sampling distance: 3.7 cm pixel−1), and a proximal GreenSeeker (GS) sensor was used to record NDVIGS. To enable direct comparisons across the different indices on an equivalent numeric scale, each index was normalized by calculating the Sufficiency-Index (SI) relative to a non-N-limiting plot. Kernel density distributions indicated that NDVIUAS had a narrower range of values that were poorly differentiated compared to NDVIGS and NDREUAS. The critical PI-NUP where yields did not increase with higher PI-NUP averaged 109 kg N ha−1 (±4 kg N ha−1). The relationship between SI and PI-NUP for the NDVIUAS saturated lower than this critical PI-NUP (96 kg N ha−1), whereas NDVIGS and NDREUAS saturated at 111 and 130 kg N ha−1, respectively. This indicates that NDVIUAS was less suitable for making N management decisions at this crop stage than NDVIGS and NDREUAS. Linear mixed effects models were developed to evaluate how well each SI measured at PI was able to predict grain yield. The NDVIUAS was least sensitive to variation in yields as reflected by having the highest slope (2.4 Mg ha−1 per 0.1 SI). In contrast, the slopes for NDVIGS and NDREUAS were 0.9 and 1.1 Mg ha−1 per 0.1 SI, respectively, indicating greater sensitivity to yields. Altogether, these results indicate that the ability of vegetation indices to inform crop management decisions depends on the index and the measurement platform used. Both NDVIGS and NDREUAS produced measurements sensitive enough to inform N fertilizer management in this system, whereas NDVIUAS was more limited.

Phenomic data-facilitated rust and senescence prediction in maize using machine learning algorithms

Current methods in measuring maize (Zea mays L.) southern rust (Puccinia polyspora Underw.) and subsequent crop senescence require expert observation and are resource-intensive and prone to subjectivity. In this study, unoccupied aerial system (UAS) field-based high-throughput phenotyping (HTP) was employed to collect high-resolution aerial imagery of elite maize hybrids planted in the 2020 and 2021 growing seasons, with 13 UAS flights obtained from 2020 and 17 from 2021. In total, 36 vegetation indices (VIs) were extracted from mosaicked aerial images that served as temporal phenomic predictors for southern rust scored in the field and senescence as scored using UAS-acquired mosaic images. Temporal best linear unbiased predictors (TBLUPs) were calculated using a nested model that treated hybrid performance as nested within flights in terms of rust and senescence. All eight machine learning regressions tested (ridge, lasso, elastic net, random forest, support vector machine with radial and linear kernels, partial least squares, and k-nearest neighbors) outperformed a general linear model with both higher prediction accuracies (92-98%) and lower root mean squared error (RMSE) for rust and senescence scores (linear model RMSE ranged from 65.8 to 2396.5 across all traits, machine learning regressions RMSE ranged from 0.3 to 17.0). UAS-acquired VIs enabled the discovery of novel early quantitative phenotypic indicators of maize senescence and southern rust before being detectable by expert annotation and revealed positive correlations between grain filling time and yield (0.22 and 0.44 in 2020 and 2021), with practical implications for precision agricultural practices.

In-Field Wheat Reflectance: How to Reach the Organ Scale?

The reflectance of wheat crops provides information on their architecture or physiology. However, the methods currently used for close-range reflectance computation do not allow for the separation of the wheat canopy organs: the leaves and the ears. This study details a method to achieve high-throughput measurements of wheat reflectance at the organ scale. A nadir multispectral camera array and an incident light spectrometer were used to compute bi-directional reflectance factor (BRF) maps. Image thresholding and deep learning ear detection allowed for the segmentation of the ears and the leaves in the maps. The results showed that the BRF measured on reference targets was constant throughout the day but varied with the acquisition date. The wheat organ BRF was constant throughout the day in very cloudy conditions and with high sun altitudes but showed gradual variations in the morning under sunny or partially cloudy sky. As a consequence, measurements should be performed close to solar noon and the reference panel should be captured at the beginning and end of each field trip to correct the BRF. The method, with such precautions, was tested all throughout the wheat growing season on two varieties and various canopy architectures generated by a fertilization gradient. The method yielded consistent reflectance dynamics in all scenarios.

Designing Experiments for Physiological Phenomics

Phenomics has emerged as the technology of choice for understanding quantitative genetic variation in plant physiology and plant breeding. Phenomics has allowed for unmatched precision in exploring plant life cycles and physiological patterns. As new technologies are developed, it is still vital to follow best practices for designing and planning to be able to fully exploit any experimental results. Here we describe the basic - but sometimes overlooked - considerations of a phenomics experiment to help you maximize the value from the data collected: choosing population and location, accounting for sources of variation, establishing a timeline, and leveraging ground-truth measurements.

Statistical Methods for the Quantitative Genetic Analysis of High-Throughput Phenotyping Data

The advent of plant phenomics, coupled with the wealth of genotypic data generated by next-generation sequencing technologies, provides exciting new resources for investigations into and improvement of complex traits. However, these new technologies also bring new challenges in quantitative genetics, namely, a need for the development of robust frameworks that can accommodate these high-dimensional data. In this chapter, we describe methods for the statistical analysis of high-throughput phenotyping (HTP) data with the goal of enhancing the prediction accuracy of genomic selection (GS). Following the Introduction in Sec. 1, Sec. 2 discusses field-based HTP, including the use of unoccupied aerial vehicles and light detection and ranging, as well as how we can achieve increased genetic gain by utilizing image data derived from HTP. Section 3 considers extending commonly used GS models to integrate HTP data as covariates associated with the principal trait response, such as yield. Particular focus is placed on single-trait, multi-trait, and genotype by environment interaction models. One unique aspect of HTP data is that phenomics platforms often produce large-scale data with high spatial and temporal resolution for capturing dynamic growth, development, and stress responses. Section 4 discusses the utility of a random regression model for performing longitudinal modeling. The chapter concludes with a discussion of some standing issues.

Automatic Microplot Localization Using UAV Images and a Hierarchical Image-Based Optimization Method

To develop new crop varieties and monitor plant growth, health, and traits, automated analysis of aerial crop images is an attractive alternative to time-consuming manual inspection. To perform per-microplot phenotypic analysis, localizing and detecting individual microplots in an orthomosaic image of a field are major steps. Our algorithm uses an automatic initialization of the known field layout over the orthomosaic images in roughly the right position. Since the orthomosaic images are stitched from a large number of smaller images, there can be distortion causing microplot rows not to be entirely straight and the automatic initialization to not correctly position every microplot. To overcome this, we have developed a three-level hierarchical optimization method. First, the initial bounding box position is optimized using an objective function that maximizes the level of vegetation inside the area. Then, columns of microplots are repositioned, constrained by their expected spacing. Finally, the position of microplots is adjusted individually using an objective function that simultaneously maximizes the area of the microplot overlapping vegetation, minimizes spacing variance between microplots, and maximizes each microplot's alignment relative to other microplots in the same row and column. The orthomosaics used in this study were obtained from multiple dates of canola and wheat breeding trials. The algorithm was able to detect 99.7% of microplots for canola and 99% for wheat. The automatically segmented microplots were compared to ground truth segmentations, resulting in an average DSC of 91.2% and 89.6% across all microplots and orthomosaics in the canola and wheat datasets.

Improved phenotyping procedure for evaluating resistance in rice against gall midge (Orseolia oryzae, Wood-Mason)

BACKGROUND

The rice gall midge (RGM, Orseolia oryzae, Wood-Mason), an important stem-feeding pest worldwide, has caused serious production losses over the past decades. Rice production practices indicate that the most reliable method for managing RGM is the deployment of cultivars that incorporate host resistance. However, the conventional phenotypic screening method of rice resistance to RGM suggested by the International Rice Research Institute (IRRI) has been used for approximately 30 years, and only 12 rice varieties/lines (including controls) can be evaluated in one tray. It is not suitable for high-throughput phenotyping of rice germplasm. Moreover, a suitable method to prepare samples for molecular biological studies of rice resistance against RGM is imperative with the rapid development of modern molecular techniques.

RESULTS

The proper density of seedlings/RGM was determined for four seeding arrangements. A high-throughput phenotyping method (HTPM) for 60 lines/varieties infested with 36 female RGM adults in one tray, as described by method 4-3 (seeded 60 lines/varieties), was developed and verified using mutant screening. Furthermore, one RGM resistance gene flanked by markers 12RM28346 and 12RM28739 on chromosome 12 was simultaneously detected using method 2-2 (seeded 30 lines/varieties in one tray) treated with 24 RGM and analyzed using conventional and simplified grading systems. Genetic analysis of the RGM resistance gene was confirmed using a method identical to that suggested by IRRI. Finally, one bucket with 24 seedlings treated with at least five female RGM adults was efficacious and could offer adequate samples for insect development observation or molecular biological studies.

CONCLUSION

A highly efficient and reliable procedure for evaluation of resistance in rice to RGM was developed and improved, and was verified through mutant screening, gene mapping, genetic analysis, and insect growth and development observations.

Genetic dissection of seasonal vegetation index dynamics in maize through aerial based high-throughput phenotyping

Plant phenotyping under field conditions plays an important role in agricultural research. Efficient and accurate high-throughput phenotyping strategies enable a better connection between genotype and phenotype. Unmanned aerial vehicle-based high-throughput phenotyping platforms (UAV-HTPPs) provide novel opportunities for large-scale proximal measurement of plant traits with high efficiency, high resolution, and low cost. The objective of this study was to use time series normalized difference vegetation index (NDVI) extracted from UAV-based multispectral imagery to characterize its pattern across development and conduct genetic dissection of NDVI in a large maize population. The time series NDVI data from the multispectral sensor were obtained at five time points across the growing season for 1,752 diverse maize accessions with a UAV-HTPP. Cluster analysis of the acquired measurements classified 1,752 maize accessions into two groups with distinct NDVI developmental trends. To capture the dynamics underlying these static observations, penalized-splines (P-splines) model was used to obtain genotype-specific curve parameters. Genome-wide association study (GWAS) using static NDVI values and curve parameters as phenotypic traits detected signals significantly associated with the traits. Additionally, GWAS using the projected NDVI values from the P-splines models revealed the dynamic change of genetic effects, indicating the role of gene-environment interplay in controlling NDVI across the growing season. Our results demonstrated the utility of ultra-high spatial resolution multispectral imagery, as that acquired using a UAV-based remote sensing, for genetic dissection of NDVI.

Improving Wheat Yield Prediction Using Secondary Traits and High-Density Phenotyping Under Heat-Stressed Environments

A primary selection target for wheat (Triticum aestivum) improvement is grain yield. However, the selection for yield is limited by the extent of field trials, fluctuating environments, and the time needed to obtain multiyear assessments. Secondary traits such as spectral reflectance and canopy temperature (CT), which can be rapidly measured many times throughout the growing season, are frequently correlated with grain yield and could be used for indirect selection in large populations particularly in earlier generations in the breeding cycle prior to replicated yield testing. While proximal sensing data collection is increasingly implemented with high-throughput platforms that provide powerful and affordable information, efficient and effective use of these data is challenging. The objective of this study was to monitor wheat growth and predict grain yield in wheat breeding trials using high-density proximal sensing measurements under extreme terminal heat stress that is common in Bangladesh. Over five growing seasons, we analyzed normalized difference vegetation index (NDVI) and CT measurements collected in elite breeding lines from the International Maize and Wheat Improvement Center at the Regional Agricultural Research Station, Jamalpur, Bangladesh. We explored several variable reduction and regularization techniques followed by using the combined secondary traits to predict grain yield. Across years, grain yield heritability ranged from 0.30 to 0.72, with variable secondary trait heritability (0.0-0.6), while the correlation between grain yield and secondary traits ranged from -0.5 to 0.5. The prediction accuracy was calculated by a cross-fold validation approach as the correlation between observed and predicted grain yield using univariate and multivariate models. We found that the multivariate models resulted in higher prediction accuracies for grain yield than the univariate models. Stepwise regression performed equal to, or better than, other models in predicting grain yield. When incorporating all secondary traits into the models, we obtained high prediction accuracies (0.58-0.68) across the five growing seasons. Our results show that the optimized phenotypic prediction models can leverage secondary traits to deliver accurate predictions of wheat grain yield, allowing breeding programs to make more robust and rapid selections.

Improving Accuracy of Herbage Yield Predictions in Perennial Ryegrass with UAV-Based Structural and Spectral Data Fusion and Machine Learning

High-throughput field phenotyping using close remote sensing platforms and sensors for non-destructive assessment of plant traits can support the objective evaluation of yield predictions of large breeding trials. The main objective of this study was to examine the potential of unmanned aerial vehicle (UAV)-based structural and spectral features and their combination in herbage yield predictions across diploid and tetraploid varieties and breeding populations of perennial ryegrass (Lolium perenne L.). Canopy structural (i.e., canopy height) and spectral (i.e., vegetation indices) information were derived from data gathered with two sensors: a consumer-grade RGB and a 10-band multispectral (MS) camera system, which were compared in the analysis. A total of 468 field plots comprising 115 diploid and 112 tetraploid varieties and populations were considered in this study. A modelling framework established to predict dry matter yield (DMY), was used to test three machine learning algorithms, including Partial Least Squares Regression (PLSR), Random Forest (RF), and Support Vector Machines (SVM). The results of the nested cross-validation revealed: (a) the fusion of structural and spectral features achieved better DMY estimates as compared to models fitted with structural or spectral data only, irrespective of the sensor, ploidy level or machine learning algorithm applied; (b) models built with MS-based predictor variables, despite their lower spatial resolution, slightly outperformed the RGB-based models, as lower mean relative root mean square error (rRMSE) values were delivered; and (c) on average, the RF technique reported the best model performances among tested algorithms, regardless of the dataset used. The approach introduced in this study can provide accurate yield estimates (up to an RMSE = 308 kg ha−1) and useful information for breeders and practical farm-scale applications.

Influence of Spatial Resolution for Vegetation Indices’ Extraction Using Visible Bands from Unmanned Aerial Vehicles’ Orthomosaics Datasets

The consolidation of unmanned aerial vehicle (UAV) photogrammetric techniques for campaigns with high and medium observation scales has triggered the development of new application areas. Most of these vehicles are equipped with common visible-band sensors capable of mapping areas of interest at various spatial resolutions. It is often necessary to identify vegetated areas for masking purposes during the postprocessing phase, excluding them for the digital elevation models (DEMs) generation or change detection purposes. However, vegetation can be extracted using sensors capable of capturing the near-infrared part of the spectrum, which cannot be recorded by visible (RGB) cameras. In this study, after reviewing different visible-band vegetation indices in various environments using different UAV technology, the influence of the spatial resolution of orthomosaics generated by photogrammetric processes in the vegetation extraction was examined. The triangular greenness index (TGI) index provided a high level of separability between vegetation and nonvegetation areas for all case studies in any spatial resolution. The efficiency of the indices remained fundamentally linked to the context of the scenario under investigation, and the correlation between spatial resolution and index incisiveness was found to be more complex than might be trivially assumed.

Evaluation of RGB and Multispectral Unmanned Aerial Vehicle (UAV) Imagery for High-Throughput Phenotyping and Yield Prediction in Barley Breeding

With advances in plant genomics, plant phenotyping has become a new bottleneck in plant breeding and the need for reliable high-throughput plant phenotyping techniques has emerged. In the face of future climatic challenges, it does not seem appropriate to continue to solely select for grain yield and a few agronomically important traits. Therefore, new sensor-based high-throughput phenotyping has been increasingly used in plant breeding research, with the potential to provide non-destructive, objective and continuous plant characterization that reveals the formation of the final grain yield and provides insights into the physiology of the plant during the growth phase. In this context, we present the comparison of two sensor systems, Red-Green-Blue (RGB) and multispectral cameras, attached to unmanned aerial vehicles (UAV), and investigate their suitability for yield prediction using different modelling approaches in a segregating barley introgression population at three environments with weekly data collection during the entire vegetation period. In addition to vegetation indices, morphological traits such as canopy height, vegetation cover and growth dynamics traits were used for yield prediction. Repeatability analyses and genotype association studies of sensor-based traits were compared with reference values from ground-based phenotyping to test the use of conventional and new traits for barley breeding. The relative height estimation of the canopy by UAV achieved high precision (up to r = 0.93) and repeatability (up to R2 = 0.98). In addition, we found a great overlap of detected significant genotypes between the reference heights and sensor-based heights. The yield prediction accuracy of both sensor systems was at the same level and reached a maximum prediction accuracy of r2 = 0.82 with a continuous increase in precision throughout the entire vegetation period. Due to the lower costs and the consumer-friendly handling of image acquisition and processing, the RGB imagery seems to be more suitable for yield prediction in this study.

Coupling of machine learning methods to improve estimation of ground coverage from unmanned aerial vehicle (UAV) imagery for high-throughput phenotyping of crops

Ground coverage (GC) allows monitoring of crop growth and development and is normally estimated as the ratio of vegetation to the total pixels from nadir images captured by visible-spectrum (RGB) cameras. The accuracy of estimated GC can be significantly impacted by the effect of 'mixed pixels', which is related to the spatial resolution of the imagery as determined by flight altitude, camera resolution and crop characteristics (fine vs coarse textures). In this study, a two-step machine learning method was developed to improve the accuracy of GC of wheat (Triticum aestivum L.) estimated from coarse-resolution RGB images captured by an unmanned aerial vehicle (UAV) at higher altitudes. The classification tree-based per-pixel segmentation (PPS) method was first used to segment fine-resolution reference images into vegetation and background pixels. The reference and their segmented images were degraded to the target coarse spatial resolution. These degraded images were then used to generate a training dataset for a regression tree-based model to establish the sub-pixel classification (SPC) method. The newly proposed method (i.e. PPS-SPC) was evaluated with six synthetic and four real UAV image sets (SISs and RISs, respectively) with different spatial resolutions. Overall, the results demonstrated that the PPS-SPC method obtained higher accuracy of GC in both SISs and RISs comparing to PPS method, with root mean squared errors (RMSE) of less than 6% and relative RMSE (RRMSE) of less than 11% for SISs, and RMSE of less than 5% and RRMSE of less than 35% for RISs. The proposed PPS-SPC method can be potentially applied in plant breeding and precision agriculture to balance accuracy requirement and UAV flight height in the limited battery life and operation time.

UAS-Based Plant Phenotyping for Research and Breeding Applications

Unmanned aircraft system (UAS) is a particularly powerful tool for plant phenotyping, due to reasonable cost of procurement and deployment, ease and flexibility for control and operation, ability to reconfigure sensor payloads to diversify sensing, and the ability to seamlessly fit into a larger connected phenotyping network. These advantages have expanded the use of UAS-based plant phenotyping approach in research and breeding applications. This paper reviews the state of the art in the deployment, collection, curation, storage, and analysis of data from UAS-based phenotyping platforms. We discuss pressing technical challenges, identify future trends in UAS-based phenotyping that the plant research community should be aware of, and pinpoint key plant science and agronomic questions that can be resolved with the next generation of UAS-based imaging modalities and associated data analysis pipelines. This review provides a broad account of the state of the art in UAS-based phenotyping to reduce the barrier to entry to plant science practitioners interested in deploying this imaging modality for phenotyping in plant breeding and research areas.

High-resolution spectral information enables phenotyping of leaf epicuticular wax in wheat

BACKGROUND

Epicuticular wax (EW) is the first line of defense in plants for protection against biotic and abiotic factors in the environment. In wheat, EW is associated with resilience to heat and drought stress, however, the current limitations on phenotyping EW restrict the integration of this secondary trait into wheat breeding pipelines. In this study we evaluated the use of light reflectance as a proxy for EW load and developed an efficient indirect method for the selection of genotypes with high EW density.

RESULTS

Cuticular waxes affect the light that is reflected, absorbed and transmitted by plants. The narrow spectral regions statistically associated with EW overlap with bands linked to photosynthetic radiation (500 nm), carotenoid absorbance (400 nm) and water content (~ 900 nm) in plants. The narrow spectral indices developed predicted 65% (EWI-13) and 44% (EWI-1) of the variation in this trait utilizing single-leaf reflectance. However, the normalized difference indices EWI-4 and EWI-9 improved the phenotyping efficiency with canopy reflectance across all field experimental trials. Indirect selection for EW with EWI-4 and EWI-9 led to a selection efficiency of 70% compared to phenotyping with the chemical method. The regression model EWM-7 integrated eight narrow wavelengths and accurately predicted 71% of the variation in the EW load (mg·dm-2) with leaf reflectance, but under field conditions, a single-wavelength model consistently estimated EW with an average RMSE of 1.24 mg·dm-2 utilizing ground and aerial canopy reflectance.

CONCLUSIONS

Overall, the indices EWI-1, EWI-13 and the model EWM-7 are reliable tools for indirect selection for EW based on leaf reflectance, and the indices EWI-4, EWI-9 and the model EWM-1 are reliable for selection based on canopy reflectance. However, further research is needed to define how the background effects and geometry of the canopy impact the accuracy of these phenotyping methods.

Assessing the Effect of Drought on Winter Wheat Growth Using Unmanned Aerial System (UAS)-Based Phenotyping

Drought significantly limits wheat productivity across the temporal and spatial domains. Unmanned Aerial Systems (UAS) has become an indispensable tool to collect refined spatial and high temporal resolution imagery data. A 2-year field study was conducted in 2018 and 2019 to determine the temporal effects of drought on canopy growth of winter wheat. Weekly UAS data were collected using red, green, and blue (RGB) and multispectral (MS) sensors over a yield trial consisting of 22 winter wheat cultivars in both irrigated and dryland environments. Raw-images were processed to compute canopy features such as canopy cover (CC) and canopy height (CH), and vegetation indices (VIs) such as Normalized Difference Vegetation Index (NDVI), Excess Green Index (ExG), and Normalized Difference Red-edge Index (NDRE). The drought was more severe in 2018 than in 2019 and the effects of growth differences across years and irrigation levels were visible in the UAS measurements. CC, CH, and VIs, measured during grain filling, were positively correlated with grain yield (r = 0.4–0.7, p < 0.05) in the dryland in both years. Yield was positively correlated with VIs in 2018 (r = 0.45–0.55, p < 0.05) in the irrigated environment, but the correlations were non-significant in 2019 (r = 0.1 to −0.4), except for CH. The study shows that high-throughput UAS data can be used to monitor the drought effects on wheat growth and productivity across the temporal and spatial domains.

Performance of the Two-Source Energy Balance (TSEB) Model as a Tool for Monitoring the Response of Durum Wheat to Drought by High-Throughput Field Phenotyping

The current lack of efficient methods for high throughput field phenotyping is a constraint on the goal of increasing durum wheat yields. This study illustrates a comprehensive methodology for phenotyping this crop's water use through the use of the two-source energy balance (TSEB) model employing very high resolution imagery. An unmanned aerial vehicle (UAV) equipped with multispectral and thermal cameras was used to phenotype 19 durum wheat cultivars grown under three contrasting irrigation treatments matching crop evapotranspiration levels (ETc): 100%ETc treatment meeting all crop water requirements (450 mm), 50%ETc treatment meeting half of them (285 mm), and a rainfed treatment (122 mm). Yield reductions of 18.3 and 48.0% were recorded in the 50%ETc and rainfed treatments, respectively, in comparison with the 100%ETc treatment. UAV flights were carried out during jointing (April 4th), anthesis (April 30th), and grain-filling (May 22nd). Remotely-sensed data were used to estimate: (1) plant height from a digital surface model (H, R ² = 0.95, RMSE = 0.18m), (2) leaf area index from multispectral vegetation indices (LAI, R ² = 0.78, RMSE = 0.63), and (3) actual evapotranspiration (ETa) and transpiration (T) through the TSEB model (R ² = 0.50, RMSE = 0.24 mm/h). Compared with ground measurements, the four traits estimated at grain-filling provided a good prediction of days from sowing to heading (DH, r = 0.58-0.86), to anthesis (DA, r = 0.59-0.85) and to maturity (r = 0.67-0.95), grain-filling duration (GFD, r = 0.54-0.74), plant height (r = 0.62-0.69), number of grains per spike (NGS, r = 0.41-0.64), and thousand kernel weight (TKW, r = 0.37-0.42). The best trait to estimate yield, DH, DA, and GFD was ETa at anthesis or during grain filling. Better forecasts for yield-related traits were recorded in the irrigated treatments than in the rainfed one. These results show a promising perspective in the use of energy balance models for the phenotyping of large numbers of durum wheat genotypes under Mediterranean conditions.

A simple, cost-effective high-throughput image analysis pipeline improves genomic prediction accuracy for days to maturity in wheat

BACKGROUND

High-throughput phenotyping and genomic selection accelerate genetic gain in breeding programs by advances in phenotyping and genotyping methods. This study developed a simple, cost-effective high-throughput image analysis pipeline to quantify digital images taken in a panel of 286 Iran bread wheat accessions under terminal drought stress and well-watered conditions. The color proportion of green to yellow (tolerance ratio) and the color proportion of yellow to green (stress ratio) was assessed for each canopy using the pipeline. The estimated tolerance and stress ratios were used as covariates in the genomic prediction models to evaluate the effect of change in canopy color on the improvement of the genomic prediction accuracy of different agronomic traits in wheat.

RESULTS

The reliability of the high-throughput image analysis pipeline was proved by three to four times of improvement in the accuracy of genomic predictions for days to maturity with the use of tolerance and stress ratios as covariates in the univariate genomic selection models. The higher prediction accuracies were attained for days to maturity when both tolerance and stress ratios were used as fixed effects in the univariate models. The results of this study indicated that the Bayesian ridge regression and ridge regression-best linear unbiased prediction methods were superior to other genomic prediction methods which were used in this study under terminal drought stress and well-watered conditions, respectively.

CONCLUSIONS

This study provided a robust, quick, and cost-effective machine learning-enabled image-phenotyping pipeline to improve the genomic prediction accuracy for days to maturity in wheat. The results encouraged the integration of phenomics and genomics in breeding programs.

Improved Accuracy of High-Throughput Phenotyping From Unmanned Aerial Systems by Extracting Traits Directly From Orthorectified Images

The development of high-throughput genotyping and phenotyping has provided access to many tools to accelerate plant breeding programs. Unmanned Aerial Systems (UAS)-based remote sensing is being broadly implemented for field-based high-throughput phenotyping due to its low cost and the capacity to rapidly cover large breeding populations. The Structure-from-Motion photogrammetry processes aerial images taken from multiple perspectives over a field to an orthomosaic photo of a complete field experiment, allowing spectral or morphological trait extraction from the canopy surface for each individual field plot. However, some phenotypic information observable in each raw aerial image seems to be lost to the orthomosaic photo, probably due to photogrammetry processes such as pixel merging and blending. To formally assess this, we introduced a set of image processing methods to extract phenotypes from orthorectified raw aerial images and compared them to the negative control of extracting the same traits from processed orthomosaic images. We predict that standard measures of accuracy in terms of the broad-sense heritability of the remote sensing spectral traits will be higher using the orthorectified photos than with the orthomosaic image. Using three case studies, we therefore compared the broad-sense heritability of phenotypes in wheat breeding nurseries including, (1) canopy temperature from thermal imaging, (2) canopy normalized difference vegetation index (NDVI), and (3) early-stage ground cover from multispectral imaging. We evaluated heritability estimates of these phenotypes extracted from multiple orthorectified aerial images via four statistical models and compared the results with heritability estimates of these phenotypes extracted from a single orthomosaic image. Our results indicate that extracting traits directly from multiple orthorectified aerial images yielded increased estimates of heritability for all three phenotypes through proper modeling, compared to estimation using traits extracted from the orthomosaic image. In summary, the image processing methods demonstrated in this study have the potential to improve the quality of the plant trait extracted from high-throughput imaging. This, in turn, can enable breeders to utilize phenomics technologies more effectively for improved selection.

Generating the Baseline in the Early Detection of Bud Rot and Red Ring Disease in Oil Palms by Geospatial Technologies

Oil palm cultivation in Ecuador is important for the agricultural sector. As a result of it, the country generates sources of employment in some of the most vulnerable zones; it contributes 0.89% of the gross domestic product and 4.35% of the agricultural gross domestic product. In 2017, a value of USD $252 million was generated by exports, and palm contributed 4.53% of the agricultural gross domestic product (GDP). It is estimated that 125,000 hectares of palm were lost in the Republic of Ecuador due to Red Ring Disease (RRD) and specifically Bud Rot (BR). The current study aimed to generate an early detection of BR and RRD in oil palm. Image acquisition has been performed using Remotely Piloted Aircraft System (RPAS) with Red, Green, and Blue (RGB) cannons, and multispectral cameras, in study areas with and without the presence of the given disease. Hereby, we proposed two phases. In phase A, a drone flight has been conducted for processing and georeferencing. This allowed to obtain an orthomosaic that serves as input for obtaining several vegetation indices of the healthy crop. The data and products obtained from this phase served as a baseline to perform comparisons with plantations affected by BR and RRD and to differentiate the palm varieties that are used by palm growers. In phase B, the same process has been applied three times with an interval of 15 days in an affected plot, in order to identify the symptoms and the progress of them. A validation for the diseases detection has been performed in the field, by taking Global Positioning System (GPS) points of the palms that presented symptoms of BR and RRD, through direct observation by field experts. The inputs obtained in each monitoring allowed to analyze the spatial behavior of the diseases. The values of the vegetation indices obtained from Phase A and B aimed to establish the differences between healthy and diseased palms, with the purpose of generating the baseline of early responses of BR and RRD conditions. However, the best vegetation index to detect the BR was the Visible Atmospherically Resistant Index (VARI).

Fusion of Spectral and Structural Information from Aerial Images for Improved Biomass Estimation

Efficient, precise and timely measurement of plant traits is important in the assessment of a breeding population. Estimating crop biomass in breeding trials using high-throughput technologies is difficult, as reproductive and senescence stages do not relate to reflectance spectra, and multiple growth stages occur concurrently in diverse genotypes. Additionally, vegetation indices (VIs) saturate at high canopy coverage, and vertical growth profiles are difficult to capture using VIs. A novel approach was implemented involving a fusion of complementary spectral and structural information, to calculate intermediate metrics such as crop height model (CHM), crop coverage (CC) and crop volume (CV), which were finally used to calculate dry (DW) and fresh (FW) weight of above-ground biomass in wheat. The intermediate metrics, CHM (R2 = 0.81, SEE = 4.19 cm) and CC (OA = 99.2%, Κ = 0.98) were found to be accurate against equivalent ground truth measurements. The metrics CV and CV×VIs were used to develop an effective and accurate linear regression model relationship with DW (R2 = 0.96 and SEE = 69.2 g/m2) and FW (R2 = 0.89 and SEE = 333.54 g/m2). The implemented approach outperformed commonly used VIs for estimation of biomass at all growth stages in wheat. The achieved results strongly support the applicability of the proposed approach for high-throughput phenotyping of germplasm in wheat and other crop species.

PhenoCams for Field Phenotyping: Using Very High Temporal Resolution Digital Repeated Photography to Investigate Interactions of Growth, Phenology, and Harvest Traits

Understanding the interaction of plant growth with environmental conditions is crucial to increase the resilience of current cropping systems to a changing climate. Here, we investigate PhenoCams as a high-throughput approach for field phenotyping experiments to assess growth dynamics of many different genotypes simultaneously in high temporal (daily) resolution. First, we develop a method that extracts a daily phenological signal that is normalized for the different viewing geometries of the pixels within the images. Second, we investigate the extraction of the in season traits of early vigor, leaf area index (LAI), and senescence dynamic from images of a soybean (Glycine max) field phenotyping experiment and show that it is possible to rate early vigor, senescence dynamics, and track the LAI development between LAI 1 and 4.5. Third, we identify the start of green up, green peak, senescence peak, and end of senescence in the phenological signal. Fourth, we extract the timing of these points and show how this information can be used to assess the impact of phenology on harvest traits (yield, thousand kernel weight, and oil content). The results demonstrate that PhenoCams can track growth dynamics and fill the gap of high temporal monitoring in field phenotyping experiments.

Machine learning for high-throughput field phenotyping and image processing provides insight into the association of above and below-ground traits in cassava (Manihot esculenta Crantz)

BACKGROUND

Rapid non-destructive measurements to predict cassava root yield over the full growing season through large numbers of germplasm and multiple environments is a huge challenge in Cassava breeding programs. As opposed to waiting until the harvest season, multispectral imagery using unmanned aerial vehicles (UAV) are capable of measuring the canopy metrics and vegetation indices (VIs) traits at different time points of the growth cycle. This resourceful time series aerial image processing with appropriate analytical framework is very important for the automatic extraction of phenotypic features from the image data. Many studies have demonstrated the usefulness of advanced remote sensing technologies coupled with machine learning (ML) approaches for accurate prediction of valuable crop traits. Until now, Cassava has received little to no attention in aerial image-based phenotyping and ML model testing.

RESULTS

To accelerate image processing, an automated image-analysis framework called CIAT Pheno-i was developed to extract plot level vegetation indices/canopy metrics. Multiple linear regression models were constructed at different key growth stages of cassava, using ground-truth data and vegetation indices obtained from a multispectral sensor. Henceforth, the spectral indices/features were combined to develop models and predict cassava root yield using different Machine learning techniques. Our results showed that (1) Developed CIAT pheno-i image analysis framework was found to be easier and more rapid than manual methods. (2) The correlation analysis of four phenological stages of cassava revealed that elongation (EL) and late bulking (LBK) were the most useful stages to estimate above-ground biomass (AGB), below-ground biomass (BGB) and canopy height (CH). (3) The multi-temporal analysis revealed that cumulative image feature information of EL + early bulky (EBK) stages showed a higher significant correlation (r = 0.77) for Green Normalized Difference Vegetation indices (GNDVI) with BGB than individual time points. Canopy height measured on the ground correlated well with UAV (CHuav)-based measurements (r = 0.92) at late bulking (LBK) stage. Among different image features, normalized difference red edge index (NDRE) data were found to be consistently highly correlated (r = 0.65 to 0.84) with AGB at LBK stage. (4) Among the four ML algorithms used in this study, k-Nearest Neighbours (kNN), Random Forest (RF) and Support Vector Machine (SVM) showed the best performance for root yield prediction with the highest accuracy of R2 = 0.67, 0.66 and 0.64, respectively.

CONCLUSION

UAV platforms, time series image acquisition, automated image analytical framework (CIAT Pheno-i), and key vegetation indices (VIs) to estimate phenotyping traits and root yield described in this work have great potential for use as a selection tool in the modern cassava breeding programs around the world to accelerate germplasm and varietal selection. The image analysis software (CIAT Pheno-i) developed from this study can be widely applicable to any other crop to extract phenotypic information rapidly.

Evaluation of Cotton Emergence Using UAV-Based Narrow-Band Spectral Imagery with Customized Image Alignment and Stitching Algorithms

Crop stand count and uniformity are important measures for making proper field management decisions to improve crop production. Conventional methods for evaluating stand count based on visual observation are time consuming and labor intensive, making it difficult to adequately cover a large field. The overall goal of this study was to evaluate cotton emergence at two weeks after planting using unmanned aerial vehicle (UAV)-based high-resolution narrow-band spectral indices that were collected using a pushbroom hyperspectral imager flying at 50 m above ground. A customized image alignment and stitching algorithm was developed to process hyperspectral cubes efficiently and build panoramas for each narrow band. The normalized difference vegetation index (NDVI) was calculated to segment cotton seedlings from soil background. A Hough transform was used for crop row identification and weed removal. Individual seedlings were identified based on customized geometric features and used to calculate stand count. Results show that the developed alignment and stitching algorithm had an average alignment error of 2.8 pixels, which was much smaller than that of 181 pixels from the associated commercial software. The system was able to count the number of seedlings in seedling clusters with an accuracy of 84.1%. Mean absolute percentage error (MAPE) in estimation of crop density at the meter level was 9.0%. For seedling uniformity evaluation, the MAPE of seedling spacing was 9.1% and seedling spacing standard deviation was 6.8%. Results showed that UAV-based high-resolution narrow-band spectral images had the potential to evaluate cotton emergence.

Development and Validation of a Phenotyping Computational Workflow to Predict the Biomass Yield of a Large Perennial Ryegrass Breeding Field Trial

Increasing dry matter yield (DMY) is the most important objective in perennial ryegrass breeding programs. Current yield assessment methods like cutting are time-consuming and destructive, non-destructive measures such as scoring yield on single plants by visual inspection may be subjective. These assessments involve multiple measurements and selection procedures across seasons and years to evaluate biomass yield repeatedly. This contributes to the slow process of new cultivar development and commercialisation. This study developed and validated a computational phenotyping workflow for image acquisition, processing and analysis of spaced planted ryegrass and investigated sensor-based DMY yield estimation of individual plants through normalized difference vegetative index (NDVI) and ultrasonic plant height data extraction. The DMY of 48,000 individual plants representing 50 advanced breeding lines and commercial cultivars was accurately estimated at multiple harvests across the growing season. NDVI, plant height and predicted DMY obtained from aerial and ground-based sensors illustrated the variation within and between cultivars across different seasons. Combining NDVI and plant height of individual plants was a robust method to enable high-throughput phenotyping of biomass yield in ryegrass breeding. Similarly, the plot-level model indicated good to high-coefficients of determination (R ²) between the predicted and measured DMY across three seasons with R ² between 0.19 and 0.81 and root mean square errors (RMSE) values ranging from 0.09 to 0.21 kg/plot. The model was further validated using a combined regression of the three seasons harvests. This study further sets a foundation for the application of sensor technologies combined with genomic studies that lead to greater rates of genetic gain in perennial ryegrass biomass yield.

Functional QTL mapping and genomic prediction of canopy height in wheat measured using a robotic field phenotyping platform

Genetic studies increasingly rely on high-throughput phenotyping, but the resulting longitudinal data pose analytical challenges. We used canopy height data from an automated field phenotyping platform to compare several approaches to scanning for quantitative trait loci (QTLs) and performing genomic prediction in a wheat recombinant inbred line mapping population based on up to 26 sampled time points (TPs). We detected four persistent QTLs (i.e. expressed for most of the growing season), with both empirical and simulation analyses demonstrating superior statistical power of detecting such QTLs through functional mapping approaches compared with conventional individual TP analyses. In contrast, even very simple individual TP approaches (e.g. interval mapping) had superior detection power for transient QTLs (i.e. expressed during very short periods). Using spline-smoothed phenotypic data resulted in improved genomic predictive abilities (5-8% higher than individual TP prediction), while the effect of including significant QTLs in prediction models was relatively minor (<1-4% improvement). Finally, although QTL detection power and predictive ability generally increased with the number of TPs analysed, gains beyond five or 10 TPs chosen based on phenological information had little practical significance. These results will inform the development of an integrated, semi-automated analytical pipeline, which will be more broadly applicable to similar data sets in wheat and other crops.

Functional QTL mapping and genomic prediction of canopy height in wheat measured using a robotic field phenotyping platform

Genetic studies increasingly rely on high-throughput phenotyping, but the resulting longitudinal data pose analytical challenges. We used canopy height data from an automated field phenotyping platform to compare several approaches to scanning for quantitative trait loci (QTLs) and performing genomic prediction in a wheat recombinant inbred line mapping population based on up to 26 sampled time points (TPs). We detected four persistent QTLs (i.e. expressed for most of the growing season), with both empirical and simulation analyses demonstrating superior statistical power of detecting such QTLs through functional mapping approaches compared with conventional individual TP analyses. In contrast, even very simple individual TP approaches (e.g. interval mapping) had superior detection power for transient QTLs (i.e. expressed during very short periods). Using spline-smoothed phenotypic data resulted in improved genomic predictive abilities (5-8% higher than individual TP prediction), while the effect of including significant QTLs in prediction models was relatively minor (<1-4% improvement). Finally, although QTL detection power and predictive ability generally increased with the number of TPs analysed, gains beyond five or 10 TPs chosen based on phenological information had little practical significance. These results will inform the development of an integrated, semi-automated analytical pipeline, which will be more broadly applicable to similar data sets in wheat and other crops.

High-Throughput Prediction of Whole Season Green Area Index in Winter Wheat With an Airborne Multispectral Sensor

INTRODUCTION

In recent decades, the interest has grown to quantify the green area index as one of the key characteristics of crop canopies (e.g. for modelling transpiration, light interception, growth). The approach of estimating green area index based on multispectral reflection data from unmanned airborne vehicles with lightweight sensors might have the potential to deliver data with sufficient accuracy and high throughput during the whole season.

MATERIALS AND METHODS

We therefore examined the applicability of a recently launched drone-based multispectral system (Sequoia, Parrot) for the prediction of whole season green area index in winter wheat, with data from field trials in Northern Germany (2017, 2018 and 2019). The explanatory power of different modeling approaches to predict green area index based on multispectral data was tested: linear and non-linear regression models, multivariate techniques, and machine learning algorithms. Further, different predictors were implemented in these models: multispectral data as raw bands and as ratios. Additionally, a new approach for the evaluation of green area index predictions during senescence is introduced. It is shown that a robust calibration during growth phase is applicable during senescence as well.

RESULTS AND DISCUSSION

A linear model which includes all four wavebands provided by the sensor in three ratios (VIQUO) and a Support Vector Machine (SVM) algorithm allow a reliable and sufficiently accurate whole season prediction. The VIQUO-model is recommended as the best model, as it is precise but still relatively simple, thus easier to communicate and to apply than the SVM. The integrated values of predicted green area indices in an independent trial are highly correlated with their final biomass (R²: VIQUO = 0.84, SVM = 0.85) which represents the process of radiation interception, one of the determining factors of growths. This is an indicator for both, a robust model calibration and a high potential of the tested multispectral system for agricultural research and crop management.

Temporal Dynamics of Wheat Blast Epidemics and Disease Measurements Using Multispectral Imagery

Wheat blast is a devastating disease caused by the Triticum pathotype of Magnaporthe oryzae. M. oryzae Triticum is capable of infecting leaves and spikes of wheat. Although symptoms of wheat spike blast (W_SB) are quite distinct in the field, symptoms on leaves (W_LB) are rarely reported because they are usually inconspicuos. Two field experiments were conducted in Bolivia to characterize the change in W_LB and W_SB intensity over time and determine whether multispectral imagery can be used to accurately assess W_SB. Disease progress curves (DPCs) were plotted from W_LB and W_SB data, and regression models were fitted to describe the nature of W_SB epidemics. W_LB incidence and severity changed over time; however, the mean W_LB severity was inconspicuous before wheat began spike emergence. Overall, both Gompertz and logistic models helped to describe W_SB intensity DPCs fitting classic sigmoidal shape curves. Lin's concordance correlation coefficients were estimated to measure agreement between visual estimates and digital measurements of W_SB intensity and to estimate accuracy and precision. Our findings suggest that the change of wheat blast intensity in a susceptible host population over time does not follow a pattern of a monocyclic epidemic. We have also demonstrated that W_SB severity can be quantified using a digital approach based on nongreen pixels. Quantification was precise (0.96 < r> 0.83) and accurate (0.92 < ρ > 0.69) at moderately low to high visual W_SB severity levels. Additional sensor-based methods must be explored to determine their potential for detection of W_LB and W_SB at earlier stages.

Assessment of Water and Nitrogen Use Efficiencies Through UAV-Based Multispectral Phenotyping in Winter Wheat

Unmanned aerial vehicle (UAV) based remote sensing is a promising approach for non-destructive and high-throughput assessment of crop water and nitrogen (N) efficiencies. In this study, UAV was used to evaluate two field trials using four water (T0 = 0 mm, T1 = 80 mm, T2 = 120 mm, and T3 = 160 mm), and four N (T0 = 0, T1 = 120 kg ha^-1, T2 = 180 kg ha^-1, and T3 = 240 kg ha^-1) treatments, respectively, conducted on three wheat genotypes at two locations. Ground-based destructive data of water and N indictors such as biomass and N contents were also measured to validate the aerial surveillance results. Multispectral traits including red normalized difference vegetation index (RNDVI), green normalized difference vegetation index (GNDVI), normalized difference red-edge index (NDRE), red-edge chlorophyll index (RECI) and normalized green red difference index (NGRDI) were recorded using UAV as reliable replacement of destructive measurements by showing high r values up to 0.90. NGRDI was identified as the most efficient non-destructive indicator through strong prediction values ranged from R ² = 0.69 to 0.89 for water use efficiencies (WUE) calculated from biomass (WUE.BM), and R ² = 0.80 to 0.86 from grain yield (WUE.GY). RNDVI was better in predicting the phenotypic variations for N use efficiency calculated from nitrogen contents of plant samples (NUE.NC) with high R ² values ranging from 0.72 to 0.94, while NDRE was consistent in predicting both NUE.NC and NUE.GY by 0.73 to 0.84 with low root mean square errors. UAV-based remote sensing demonstrates that treatment T2 in both water 120 mm and N 180 kg ha^-1 supply trials was most appropriate dosages for optimum uptake of water and N with high GY. Among three cultivars, Zhongmai 895 was highly efficient in WUE and NUE across the water and N treatments. Conclusively, UAV can be used to predict time-series WUE and NUE across the season for selection of elite genotypes, and to monitor crop efficiency under varying N and water dosages.

Ground-Based LiDAR Improves Phenotypic Repeatability of Above-Ground Biomass and Crop Growth Rate in Wheat

Highly repeatable, nondestructive, and high-throughput measures of above-ground biomass (AGB) and crop growth rate (CGR) are important for wheat improvement programs. This study evaluates the repeatability of destructive AGB and CGR measurements in comparison to two previously described methods for the estimation of AGB from LiDAR: 3D voxel index (3DVI) and 3D profile index (3DPI). Across three field experiments, contrasting in available water supply and comprising up to 98 wheat genotypes varying for canopy architecture, several concurrent measurements of LiDAR and AGB were made from jointing to anthesis. Phenotypic correlations at discrete events between AGB and the LiDAR-derived biomass indices were significant, ranging from 0.31 (P < 0.05) to 0.86 (P < 0.0001), providing confidence in the LiDAR indices as effective surrogates for AGB. The repeatability of the LiDAR biomass indices at discrete events was at least similar to and often higher than AGB, particularly under water limitation. The correlations between calculated CGR for AGB and the LiDAR indices were moderate to high and varied between experiments. However, across all experiments, the repeatabilities of the CGR derived from the LiDAR indices were appreciably greater than those for AGB, except for the 3DPI in the water-limited environment. In our experiments, the repeatability of either LiDAR index was consistently higher than that of AGB, both at discrete time points and when CGR was calculated. These findings provide promising support for the reliable use of ground-based LiDAR, as a surrogate measure of AGB and CGR, for screening germplasm in research and wheat breeding.

An efficient RGB-UAV-based platform for field almond tree phenotyping: 3-D architecture and flowering traits

BACKGROUND

Almond is an emerging crop due to the health benefits of almond consumption including nutritional, anti-inflammatory, and hypocholesterolaemia properties. Traditional almond producers were concentrated in California, Australia, and Mediterranean countries. However, almond is currently present in more than 50 countries due to breeding programs have modernized almond orchards by developing new varieties with improved traits related to late flowering (to reduce the risk of damage caused by late frosts) and tree architecture. Almond tree architecture and flowering are acquired and evaluated through intensive field labour for breeders. Flowering detection has traditionally been a very challenging objective. To our knowledge, there is no published information about monitoring of the tree flowering dynamics of a crop at the field scale by using color information from photogrammetric 3D point clouds and OBIA. As an alternative, a procedure based on the generation of colored photogrammetric point clouds using a low cost (RGB) camera on-board an unmanned aerial vehicle (UAV), and an semi-automatic object based image analysis (OBIA) algorithm was created for monitoring the flower density and flowering period of every almond tree in the framework of two almond phenotypic trials with different planting dates.

RESULTS

Our method was useful for detecting the phenotypic variability of every almond variety by mapping and quantifying every tree height and volume as well as the flowering dynamics and flower density. There was a high level of agreement among the tree height, flower density, and blooming calendar derived from our procedure on both fields with the ones created from on-ground measured data. Some of the almond varieties showed a significant linear fit between its crown volume and their yield.

CONCLUSIONS

Our findings could help breeders and researchers to reduce the gap between phenomics and genomics by generating accurate almond tree information in an efficient, non-destructive, and inexpensive way. The method described is also useful for data mining to select the most promising accessions, making it possible to assess specific multi-criteria ranking varieties, which are one of the main tools for breeders.

High-Throughput Phenotyping of Indirect Traits for Early-Stage Selection in Sugarcane Breeding

One of the major limitations for sugarcane genetic improvement is the low heritability of yield in the early stages of breeding, mainly due to confounding inter-plot competition effects. In this study, we investigate an indirect selection index (Si), developed based on traits correlated to yield (indirect traits) that were measured using an unmanned aerial vehicle (UAV), to improve clonal assessment in early stages of sugarcane breeding. A single-row early-stage clonal assessment trial, involving 2134 progenies derived from 245 crosses, and a multi-row experiment representative of pure-stand conditions, with an unrelated population of 40 genotypes, were used in this study. Both experiments were screened at several stages using visual, multispectral, and thermal sensors mounted on a UAV for indirect traits, including canopy cover, canopy height, canopy temperature, and normalised difference vegetation index (NDVI). To construct the indirect selection index, phenotypic and genotypic variance-covariances were estimated in the single-row and multi-row experiment, respectively. Clonal selection from the indirect selection index was compared to single-row yield-based selection. Ground observations of stalk number and plant height at six months after planting made from a subset of 75 clones within the single-row experiment were highly correlated to canopy cover (rg = 0.72) and canopy height (rg = 0.69), respectively. The indirect traits had high heritability and strong genetic correlation with cane yield in both the single-row and multi-row experiments. Only 45% of the clones were common between the indirect selection index and single-row yield based selection, and the expected efficiency of correlated response to selection for pure-stand yield based on indirect traits (44%–73%) was higher than that based on single-row yield (45%). These results highlight the potential of high-throughput phenotyping of indirect traits combined in an indirect selection index for improving early-stage clonal selections in sugarcane breeding.

The Contribution of Functional Traits to the Breeding Progress of Central-European Winter Wheat Under Differing Crop Management Intensities

Wheat yields in many of the main producing European countries stagnate since about 20 years. Hence, it is of high interest, to analyze breeding progress in terms of yield and how associated traits changed. Therefore, a set of 42 cultivars (released between 1966 and 2012) was selected and yield as well as functional traits defined by the Monteith and Moss equation were evaluated under three levels of management intensity. The Monteith Moss equation thereby calculates grain yield as the product of incident photosynthetically active radiation, fraction of intercepted radiation, radiation use efficiency, and harvest index. The field trial was performed in a high yielding environment in Northern Germany in two seasons (2016-2017 and 2017-2018) with very contrasting rainfall rates. The three differing managements were: intensive (high N + pesticides), semi-intensive (high N - pesticides), and extensive (low N - pesticides). The results indicate that the stagnation of wheat yields in Central-Europe is not caused by a diminishing effect of breeding on yield potential. This equally applies to suboptimal growing conditions like extensified crop management and restricted water supply. Nearly all functional sub-traits showed a parallel progress but coefficients of determination of relationships between traits and year of variety release are decreasing along the hierarchy of yield formation. One exception is radiation interception which did not show a stable linear increase during breeding history. In recent years, biomass is getting more important in comparison to harvest index. Values of harvest index are slowly approaching theoretical maxima and correlations with grain yield are decreasing.

Easy MPE: Extraction of Quality Microplot Images for UAV-Based High-Throughput Field Phenotyping

Microplot extraction (PE) is a necessary image processing step in unmanned aerial vehicle- (UAV-) based research on breeding fields. At present, it is manually using ArcGIS, QGIS, or other GIS-based software, but achieving the desired accuracy is time-consuming. We therefore developed an intuitive, easy-to-use semiautomatic program for MPE called Easy MPE to enable researchers and others to access reliable plot data UAV images of whole fields under variable field conditions. The program uses four major steps: (1) binary segmentation, (2) microplot extraction, (3) production of ∗.shp files to enable further file manipulation, and (4) projection of individual microplots generated from the orthomosaic back onto the raw aerial UAV images to preserve the image quality. Crop rows were successfully identified in all trial fields. The performance of the proposed method was evaluated by calculating the intersection-over-union (IOU) ratio between microplots determined manually and by Easy MPE: the average IOU (±SD) of all trials was 91% (±3).

Combining Crop Growth Modeling and Statistical Genetic Modeling to Evaluate Phenotyping Strategies

Genomic prediction of complex traits, say yield, benefits from including information on correlated component traits. Statistical criteria to decide which yield components to consider in the prediction model include the heritability of the component traits and their genetic correlation with yield. Not all component traits are easy to measure. Therefore, it may be attractive to include proxies to yield components, where these proxies are measured in (high-throughput) phenotyping platforms during the growing season. Using the Agricultural Production Systems Simulator (APSIM)-wheat cropping systems model, we simulated phenotypes for a wheat diversity panel segregating for a set of physiological parameters regulating phenology, biomass partitioning, and the ability to capture environmental resources. The distribution of the additive quantitative trait locus effects regulating the APSIM physiological parameters approximated the same distribution of quantitative trait locus effects on real phenotypic data for yield and heading date. We use the crop growth model APSIM-wheat to simulate phenotypes in three Australian environments with contrasting water deficit patterns. The APSIM output contained the dynamics of biomass and canopy cover, plus yield at the end of the growing season. Each water deficit pattern triggered different adaptive mechanisms and the impact of component traits differed between drought scenarios. We evaluated multiple phenotyping schedules by adding plot and measurement error to the dynamics of biomass and canopy cover. We used these trait dynamics to fit parametric models and P-splines to extract parameters with a larger heritability than the phenotypes at individual time points. We used those parameters in multi-trait prediction models for final yield. The combined use of crop growth models and multi-trait genomic prediction models provides a procedure to assess the efficiency of phenotyping strategies and compare methods to model trait dynamics. It also allows us to quantify the impact of yield components on yield prediction accuracy even in different environment types. In scenarios with mild or no water stress, yield prediction accuracy benefitted from including biomass and green canopy cover parameters. The advantage of the multi-trait model was smaller for the early-drought scenario, due to the reduced correlation between the secondary and the target trait. Therefore, multi-trait genomic prediction models for yield require scenario-specific correlated traits.

High-throughput phenotyping with deep learning gives insight into the genetic architecture of flowering time in wheat

BACKGROUND

Measurement of plant traits with precision and speed on large populations has emerged as a critical bottleneck in connecting genotype to phenotype in genetics and breeding. This bottleneck limits advancements in understanding plant genomes and the development of improved, high-yielding crop varieties.

RESULTS

Here we demonstrate the application of deep learning on proximal imaging from a mobile field vehicle to directly estimate plant morphology and developmental stages in wheat under field conditions. We developed and trained a convolutional neural network with image datasets labeled from expert visual scores and used this "breeder-trained" network to classify wheat morphology and developmental stages. For both morphological (awned) and phenological (flowering time) traits, we demonstrate high heritability and very high accuracy against the "ground-truth" values from visual scoring. Using the traits predicted by the network, we tested genotype-to-phenotype association using the deep learning phenotypes and uncovered novel epistatic interactions for flowering time. Enabled by the time-series high-throughput phenotyping, we describe a new phenotype as the rate of flowering and show heritable genetic control for this trait.

CONCLUSIONS

We demonstrated a field-based high-throughput phenotyping approach using deep learning that can directly measure morphological and developmental phenotypes in genetic populations from field-based imaging. The deep learning approach presented here gives a conceptual advancement in high-throughput plant phenotyping because it can potentially estimate any trait in any plant species for which the combination of breeder scores and high-resolution images can be obtained, capturing the expert knowledge from breeders, geneticists, pathologists, and physiologists to train the networks.

High-throughput phenotyping with deep learning gives insight into the genetic architecture of flowering time in wheat

BACKGROUND

Measurement of plant traits with precision and speed on large populations has emerged as a critical bottleneck in connecting genotype to phenotype in genetics and breeding. This bottleneck limits advancements in understanding plant genomes and the development of improved, high-yielding crop varieties.

RESULTS

Here we demonstrate the application of deep learning on proximal imaging from a mobile field vehicle to directly estimate plant morphology and developmental stages in wheat under field conditions. We developed and trained a convolutional neural network with image datasets labeled from expert visual scores and used this "breeder-trained" network to classify wheat morphology and developmental stages. For both morphological (awned) and phenological (flowering time) traits, we demonstrate high heritability and very high accuracy against the "ground-truth" values from visual scoring. Using the traits predicted by the network, we tested genotype-to-phenotype association using the deep learning phenotypes and uncovered novel epistatic interactions for flowering time. Enabled by the time-series high-throughput phenotyping, we describe a new phenotype as the rate of flowering and show heritable genetic control for this trait.

CONCLUSIONS

We demonstrated a field-based high-throughput phenotyping approach using deep learning that can directly measure morphological and developmental phenotypes in genetic populations from field-based imaging. The deep learning approach presented here gives a conceptual advancement in high-throughput plant phenotyping because it can potentially estimate any trait in any plant species for which the combination of breeder scores and high-resolution images can be obtained, capturing the expert knowledge from breeders, geneticists, pathologists, and physiologists to train the networks.

Development and Validation of a Model to Combine NDVI and Plant Height for High-Throughput Phenotyping of Herbage Yield in a Perennial Ryegrass Breeding Program

Sensor-based phenotyping technologies may offer a non-destructive, high-throughput and efficient assessment of herbage yield (HY) to replace current inefficient phenotyping methods. This paper assesses the feasibility of combining normalised difference vegetative index (NDVI) from multispectral imaging and ultrasonic sonar estimates of plant height to estimate HY of single plants in a large perennial ryegrass breeding program. For sensor calibration, fresh HY (FHY) and dry HY (DHY) were acquired destructively, and plant height was measured at four dates each in 2017 and 2018 from a selected subset of 480 plants. Global multiple linear regression models based on K-fold and random split cross-validation methods were used to evaluate the relationship between observed vs. predicted HY. The coefficient of determination (R2) = 0.67–0.68 and a root mean square error (RMSE) between 5.43–7.60 g was obtained for the validation of predicted vs. observed DHY. The mean absolute error (MAE) and mean percentage error (MPE) ranged between 3.59–5.44 g and 22–28%, respectively. For the FHY, R2 values ranged from 0.63 to 0.70, with an RMSE between 23.50 and 33 g, MAE between 15.11 and 24.34 g and MPE between ~22% and 31%. Combining NDVI and plant height is a robust method to enable high-throughput phenotyping of herbage yield in perennial ryegrass breeding programs.

Predicting Zea mays Flowering Time, Yield, and Kernel Dimensions by Analyzing Aerial Images

Image analysis methods for measuring crop phenotypes may replace traditional measurements if they more efficiently and reliably capture similar or superior information. This study used a recreational-grade unmanned aerial vehicle carrying a spectrally-modified consumer-grade camera to collect images in which each pixel value is a vegetation index based on the normalized difference between the blue and near infrared wavelength bands (BNDVI). The subjects of the study were Zea mays hybrids with good yield potential grown in 4-row plots. Flights were conducted at least once per week during three successive growing seasons in south-central Wisconsin. Average BNDVI for each plot (genotype) rose steadily through June, peaked in July, and then declined as plants matured. BNDVI histograms changed shape over the season as the canopy concealed soil, became more uniformly green, then senesced. Principal Components Analysis (PCA) captured the change in histogram shape. PC1 represented canopy closure. PC2 represented the mean of the BNDVI distribution. PC3 represented the spread of the distribution. Correlation analysis showed that flowering time correlated with PC2 and PC3 best (r ≈ 0.5) a few days before the event (day in which 50% of the plants exhibited tassels). Three ears were picked from each plot to quantify kernel dimensions by image analysis before each plot was mechanically harvested to determine grain weight per plot. Correlations between this measurement of yield and PC2 were low in June but exceeded 0.4 within 10 days after flowering. Kernel length correlated similarly with PC2. The correlation between PC2 and kernel thickness displayed a similar but inverted time course. These results indicate that greater mid-season BNDVI values correlate positively with yield comprised of tall, thin kernels. Partial least squares regression performed on the BNDVI time courses predicted flowering time (r = 0.54-0.79) and yield (r = 0.4-0.69). This three-year experiment demonstrated that readily available hardware and software can create a phenotyping platform capable of predicting maize flowering time, yield, and kernel dimensions to a useful degree.