Crop Phenomics: Current Status and Perspectives

1Persistent food and feed supply needs, resources shortages, climate change and energy use are some of the challenges we face in our dependence on plants. Until 2050, crop production will have to double to meet the projected production demands of the global population. Demand for crop production is expected to grow 2.4% a year, but the average rate of increase in crop yield is only 1.3%. Moreover, production yields have stagnated in up to 40% of land under cereal production. Genetic improvements in crop performance remain the key role in improving crop productivity, but the current rate of improvement cannot meet the needs of sustainability and food security.

2Molecular breeding strategies pay more attention to selections based on genotypic information, but phenotypic data are still needed (; Araus et al., 2014). Phenotyping is necessary to improve the selection efficiency and reproducibility of results in transgenic studies (Table 1). Considering that molecular breeding populations can include a range from at least 200 to at most 10,000 lines, the ability to accurately and high-throughput characterize hundreds of lines at the same time is still challenging (; Araus et al., 2014). Apparently, compared with the vast genetic information, there is a phenotyping bottleneck hampering progress in understanding the genetic basis of complex traits. To break this bottleneck and improve the efficiency of molecular breeding, reliable, automatic, and high-throughput phenotypic technologies were urgently needed to provide breeding scientists with new insights in selecting new species to adapt to the resources shortages and global climate change.

3During the past decade, plant phenomics has evolved from an emerging niche to a thriving research field, which was defined as the gathering of multi-dimensional phenotypic data at multiple levels from cell level, organ level, plant level to population level. Crop phenotypes are extremely complicated because they are the result of interaction between genotypes (G) and a multitude of envirotypes (E). This interaction influences not only the growth and development process of crops measured by the structural traits on cellular, tissue, organ and plant level, but also the plant functioning measured by the physiological traits. These internal phenotypes in turn determine crop external phenotypes such as morphology, biomass and yield performance (Figure 1). Crop phenomics research integrates agronomy, life sciences, information science, math and engineering sciences, and combines high-performance computing and artificial intelligence technology to explore multifarious phenotypic information of crop growth in a complex environment, of which the ultimate goal is to construct an effective technical system able to phenotype crops in a high-throughput, multi-dimensional, big-data, intelligent and automatically measuring manner, and create a tool comprehensively integrating big data achieved from a multi-modality, multi-scale, phenotypic + environmental + genotypic condition, in order to develop new methods of mining genes associated with important agronomic traits, and propose new intelligent solutions for precision breeding.

4Here, we provide an overview of crop phenomics research, focusing on two parts, from phenotypic data collection through various sensors to phenomics analysis. We systematically introduced the crop phenotyping approaches from cellular, tissue, organ, and plant level to field level, discussing application and practical problems in research. Based on overwhelming amounts of phenotypic data, we then discussed the phenotyping extraction, phenotype information analysis and knowledge storage. We emphasize Phenomics is entering the big-data era with multi-domain, multi-level, and multi-scale characteristics. We highlight necessity and importance of building multi-scale, multi-dimensional and trans-regional crop phenotyping big database, researching E-trait depth analysis schemas and technical systems for precise identification of crop phenotypes, realizing functional-structural plant modeling based on phenomics, and developing bioinformatics technologies that integrate genomes and phenotypes.

5Currently, the plant phenotyping community seems some-what divided between high-throughput, low-resolution phenotyping and in-depth phenotyping at lower throughput and higher resolution. Phenotyping systems and tools applied in different scales are focused on different key characteristics –automated phenotyping platforms in controllable environment and high-throughput methodologies in field environment highlights high-throughput, while phenotyping covering the organ, tissue and cellular level emphasizes in-depth phenotyping and higher resolution. In this part, we systematically introduced the crop phenotyping approaches covering from cellular and tissue level to field level, and discussed the application and practical problems of which technologies in crop researches.

6Compared with the whole-plant phenotying technologies, phenotyping at higher spatial and temporal resolutions of tissue and cellar scales is more difficult. To reveal crop micro-phenotypes, the pre-treatment of plant or organ is usually destructive when sampled in a cumbersome and involves multi-step procedure, and the high resolution imaging of samples is also inefficient in view of the output in micron level. More challenging, automated image analysis techniques are urgently needed to quantify cell and tissue traits of crop from the larger high qualified image. In recent years, many emerging algorithms and tools have been proposed to handle with microscopic images from hand-cutting slices, micro-CT, fluorescent, laser, and paraffin sections imaging (Table 2).

7Root anatomical traits have important effects on plant function, including acquisition of nutrients and water from the soil and transportation to the aboveground part. Over the last years, novel micro-image acquisition technologies and computer vision have been introduced to improve our understanding of anatomical structure and function of roots. In 2011, computer-aided calculation of wheat root vascular bundle phenotypic information extraction were introduced based on sequence images of paraffin sections. (Penn State College of Agricultural Sciences Roots Lab) have developed a high-throughput, high-resolution phenotyping platform, which combines laser optics and serial imaging with three-dimensional (3D) image reconstruction (RootSlice) and quantification to understand root anatomy by semi-auto RootScan. developed a fully automated RootAnalyzer software for root microscopic phenotypic information extraction to further improve image segmentation efficiency and ensure high accuracy. (Beijing Key Laboratory of Digital Plant) introduced X-ray micro-CT into 3D imaging of maize root tissues and developed an image processing scheme for the 3D segmentation of metaxylem vessels. Compared with traditional manual measurement of vascular bundles of maize roots, the proposed protocols significantly improved the efficiency and accuracy of the micron-scale phenotypic quantification.

8Different from the root system, crop stalks have a more complex microstructure. The complexity and diversity in microscopic image data poses greater challenges for developing suitable data analysis workflows in the detection and identification of microscopic phenotypes of stalk tissue. and presented semi- automated and automated analysis method for the stained microscopic images of stalk sections. and created an image analysis tool that could operate on images of hand-cut stalk transections to measure anatomical features in high throughput. Those tools have significantly improved the measurement efficiency of vascular bundle, but the anatomical traits corresponding to the rind and the detection accuracy remained a challenge. (Beijing Key Laboratory of Digital Plant) introduced micro-computed tomography (CT) technology for stalk imaging and developed the VesselParser 1.0 algorithm, which made it possible to automatically and accurately analyze phenotypic traits of vascular bundles within entire maize stalk cross-sections. So far, Beijing Key Laboratory of Digital Plant has built a novel method to improve the X-ray absorption contrast of maize tissue suitable for ordinary micro-CT scanning. Based on CT images, they introduced a set of image processing workflows for maize root, stalk and leaf to effectively extract microscopic phenotypes of vascular bundles.

9Phenotyping at tissue and cellar scales still requires complex procedures. The need to simplify the sample preparation process and explore advanced imaging techniques is crucial to accelerate microscopic phenotyping studies. Moreover, image processing is another major bottleneck in micro-phenotyping researches. Due to the specific crop organ and cell phenotypic characteristics of the huge differences, most micro-image analysis algorithms have been developed for specific biological assays currently.

10Most plant phenotyping platforms concentrated the high-throughput of individual plants. Hence the phenotypic accuracy of organs on the plants was always compromised. The most frequently used phenotyping index, such as leaf length, leaf area, and fruit volume, could be obtained in phenotyping platforms simply. The smartphone app platform was developed for field phenotyping by taking pictures and image analysis at organ level. It is very convenient to acquire leaf angles and leaf length using the app in a 2D view. The plants do not need to be destructively sampled indoors. 2D cameras are low cost and were integrated into most of the phenotyping platforms, which provided effective phenotypic solution for branch structured plants, especially for tracking the dynamic growth of organs on plants. However, 2D images lost another dimension data in 3D space, and some of the estimated morphological traits still need to be calibrated. Multi-view stereo (MVS) approach is another popular low-cost alternative for organ level phenotyping. 3D point clouds were reconstructed using multi-view images through structure from motion techniques, and then phenotypic traits were extracted through segmented organs of individual plants. This cost-effective 3D reconstruction method depicts an alternative to an expensive laser scanner in the studied scenarios with potential for automated procedures. There are significant differences between these organ traits within various cultivars, thus tiny errors could be ignored in large scale omics analysis.

11However, besides these measured length, area, and volume phenotypic parameters, there are lots of obvious differences that can be observed by humans of plant organs, such as blade profiles, blade folds, vein curves, and leaf colors, of which are difficult to obtain the morphological data and quantitatively describe their differences. A series of mathematical approaches need to be developed to quantitatively describe these differences in order to discover more detailed phenotyping traits with high precision phenotyping data of organ level. Researchers use high resolution 3D scanners to acquire the morphological structure of plant organs. High resolution 3D scanners are relatively expensive while the acquired morphological data are more accurate than MVS reconstructed organs, especially for non-planner surface plants. 2D LiDAR scanners and depth sensors were also used, combined with turn table and translation devices to estimate phenotypic traits through 3D recovery of point clouds. Most 2D or 3D imaging could solve the 3D organ scale phenotyping data acquisition problems, except for tall plants for which it is difficult due to a limited field of vision. High resolution 3D point clouds of plant organs are quite complicated in order to extract more phenotypic traits, thus computer graphics algorithms, such as skeleton extraction, surface reconstruction, and feature-preserving remeshing, need to be applied to process the high-resolution morphological data for 3D phenotypic trait extraction. Because of the difficulty and complexity of detailed organ data acquisition and analysis, proposed a data acquisition standard and constructed a resource database of plant organs using measured in situ morphological data, to realize the integration and sharing of high quality data of plant organs.

12Automation and robotics, new sensors, and imaging technologies (hardware and software) have provided an opportunity for high-throughput plant phenotyping platforms (HTPPs) development. In the past 10 years, great improvements have been made in researching and developing HTPPs. Depending on the overall design, HTPPs in a growth chamber or greenhouse can generally be classified as sensor-to-plant or plant-to-sensor based on whether the plants occupy a fixed position during a measurement routine and an imaging setup moves to each of those positions or the plants are transported to an imaging station, respectively. Collectively, the techniques used in HTPPs in the growth chamber or greenhouse mainly include (Table 3):

13RGB imaging, which obtains the phenotypes of plant morphology, color, and texture.

14Chlorophyll fluorescence imaging, which obtains photosynthetic phenotypes.

15Hyperspectral imaging, which obtains phenotypes such as pigment composition, biochemical composition, nitrogen content, and moisture content.

16Thermal imaging, which obtains the plant surface temperature distribution, stomatal conductance and transpiration phenotype.

17Lidar, which obtains the three-dimensional structure phenotype of plants.

18Moreover, other advanced imaging techniques widely used in medicine, such as MRI, PET, and CT, have also been introduced into HTPPs in the growth chamber or greenhouse. The following table lists the major imaging techniques used in the HTPPs in the growth chamber or greenhouse.

19After integration of information technology, digital technology and platform equipment with plant phenotyping, several HTPPs in growth chamber or greenhouse have been listed in Table 4. Such high-throughput phenotyping platforms are characterized by automation, high-throughput, and high precision, which greatly improve plant data collection efficiency and accuracy, in order to improve the efficiency of crop breeding.

20However, most HTPPs still have high construction, operating, and maintenance costs, and most academic and research institutions do not have access to these techniques as a result. For example, the number of fully automatic and high-throughput phenotyping platforms is limited in China: one was developed by the Crop Phenotyping Center (CPC), http://plantphenomics.hzau.edu.cn/, and the others were introduced by LemnaTec, such as http://bri.caas.net.cn/, http://www.kenfeng.com/, and http://www.zealquest.com/. Because of this, there has been increased research into developing affordable, small-scale plant phenotyping platforms and technology. In fact, affordable phenotypic acquisition techniques or platforms are constantly being updated, such as the imaging system developed by that uses wireless-connected consumer digital cameras, and the low-cost Glyph phenotyping platform. Lowering the cost of these platforms could, therefore, significantly increase the scope of phenotypic research and advance the rapid expansion of phenotypic–genotypes analysis for complex traits.

21Field-based phenotyping (FBP) is a critical component of crop improvement through genetics, as it is the ultimate expression of the relative effects of genetic factors, environmental factors, and their interaction on critical production traits, such as yield potential and tolerance to abiotic/biotic stresses. Currently, the most commonly field-based phenotyping platforms (FBPPs) use ground wheeled, rigid motorized gantry or aerial vehicles, combined with a wide range of cameras, sensors and high-performance computing, to capture deep phenotyping data in time (throughout the crop cycle) and space (at the canopy level) in field environments (Table 5). It cannot be denied that the efficiency of ground wheeled phenotyping system is quite low if the plot area is too large. In 2016, FIELD SCANALYZERS, with rigid motorized gantry supporting a weather proof measuring platform that incorporates a wide range of cameras, sensors and illumination systems, were developed. The facility, equipped with high-resolution visible, chlorophyll fluorescence and thermal infrared camera, hyperspectral imager, and 3D laser scanner. Crops within a 10–20 m × 110–200 m area can be monitored, which realizes the continuous, automatic, and high-throughput detection of crop phenotyping detection in field. Meanwhile, the cable-suspended field phenotyping platform covering an area of ∼1 ha was also developed for rapid and non-destructive monitoring of crop traits. However, these large ground-based field phenotyping platforms also have high construction, operating, and maintenance costs, and it has to be located at certain sites limiting the scale at which it can be used. Furthermore, the ground-based FBPPs are not very suitable for large crops, such as maize, except in the early stages.

22In recent years, manned aircraft and unmanned aerial vehicle remote sensing platforms (UAV-RSPs) are becoming a high-throughput tool for crop phenotyping in the field environment, which meet the demands of spatial, spectral, and temporal resolutions (Table 5). For example, thermal sensors fitted to manned aircraft were used to measure canopy temperature. The sensors that UAV-RSPs carried typically included digital cameras, infrared thermal imagers, light detection and ranging (LIDAR), multispectral cameras, and hyperspectral sensors, which are applied to: canopy surface modeling and crop biomass estimation based on visible imaging; crop physiological status monitor, such as chlorophyll fluorescence and N levels, based on visible–near-infrared spectroscopy and high-resolution hyperspectral imaging; plant water status detection based on thermal imaging; and crop fine-scale geometric traits analysis based on LIDAR point clouds. There are definite advantages for UAV-RSPs, including portable, high monitoring efficiency, low-cost, and suitability for field environments. On the other hand, some limiting factors for UAV-RSPs also exist, including the lack of methods for fast and automatic data processing and modeling, the strict airspace regulations, and vulnerable to different weather conditions. Recently, combining ground-based platforms and aerial platforms for phenotyping offers flexibility. For example, the tractor-based proximal crop-sensing platform, combined with UAV-based platform, was used to target complex traits such as growth and RUE in sorghum.

23Phenomic experiments are not directly reproducible because of the multi-model of sensors (structure, morphology, color, and physiology information), multi-scales phenotypic data (from cellular to population level), and variability of environmental conditions. Crop phenotypic data collection is only the first step in Phenomic research. How to extract phenotypic traits from raw data? How to realize phenotype data standardization and storage? How to make cross-scale, cross-dimensionality meta-analyses? Finally, based on the phenotypic big data, how to realize the model-assisted phenotyping and phenotypic-genomic selection? Hereby this part focuses on hot topics in crop phenomics analysis raised above. We then suggest that research in this area is entering a new stage of development using artificial intelligence technology, such as deep learning methods, in which can help researchers transform large numbers of omic-data into knowledge.

24Image-based phenotyping, as an important and all-purpose technique, has been applied to measure and quantify vision-based and model-based traits of plant in the laboratory, greenhouse and field environments. So far, lots of image analysis methods and softwares have been developed to perform image-based plant phenotyping. From image analysis perspective, phenotypic traits of plant can be classified into 4 categories, i.e., quantity, geometry, color and texture, and are also classified into linear and non-linear features related to the pixel representation. Moreover, some valuable agronomic and physiological traits can be derived from image features. Generally, image techniques are specially designed according to specific crop varieties and phenotypic traits of interest, and always require with prior knowledge of research objects, as well as more or less man-machine interaction. Under highly controlled conditions, the classic image processing pipeline can provide acceptable phenotyping results, such as biomass, NDVI, chlorophyll responses, and compactness etc. However, the simple image processing pipeline is still very difficult to handle with non-linear, non-geometric phenotyping tasks. Facing species and environment diversities, fully automated and intelligent image analysis will remain a long-term challenge.

25Machine learning techniques, such as support vector machines, clustering algorithms, and neural networks, have been widely utilized to image-based phenotyping, which not only improve the robustness of image analysis tasks, also relieve tedious manual intervention. There is no doubt that machine learning techniques will have a prominent role in breaking through the bottlenecks of plant phenotyping. In a broad category of machine learning techniques, deep learning demonstrates impressive advantages in many image-based tasks, such as object detection and localization, semantic segmentation, image classification, and others. In essence, deep convolutional neural networks (CNNs) are well suited to many vision-based computer problems, e.g., recognition, classification, and instance detection and segmentation. Compared with the traditional image analysis methods, CNNs are simultaneously trained from end to end without image feature description and extraction procedures. As far as plant phenotyping, CNNs have been effectively applied to detect and diagnosis, classify fruits and flowers, and count leaf number. It is worth noting that those vision-based phenotyping tasks were driven by the massive captured and annotation plant images. From the view of machine vision perspective, deep learning has been a fundamental technique framework in image-based plant phenotyping.

26A huge amount of complex data and the integration of a wide range of image, spectral and environmental data can be generated through by the above phenotypic technologies, usually up to GB or even petabytes, unstructured “Big Data.” Thus, the efficient storage, management and retrieval of phenotypic data have become the important issues to be considered. The current universally accepted principle of information standardization includes three aspects: (i) the ‘minimum information’ (MI) approach is recommended to define the content of the data set, (ii) ontology terms is applied for the unique and repeatable annotation of data, and in the form of data sharing and meta-analyses, (iii) and proper data formats, such as CSV, XML, RDF, MAGE-TAB, etc., are chosen for the construction of data sets. Up until now, a number of phenotyping resources have been built ranging from phenotypic data of one species to multi-data types. In 2010, PODD was developed for capturing, managing, annotating and distributing the data to support both Australian and international biological research communities; built a PHENOPSIS DB information system for Arabidopsis thaliana phenotypic data acquired by the PHENOPSIS phenotyping platform; Bisque is the first web based, cross-platform, developed into a repository to store, visualize, organize and analyze images in the cloud. In 2014, ClearedLeaves DB, an on open online database, was built to store, manage and access leaf images and phenotypic data; AraPheno1 was the first comprehensive, central repository of population-scale phenotypes (it integrated more than 250 publicly available phenotypes from six independent studies) for A. thaliana inbred lines; PhenoFront was a publicly available dataset of above-ground plant tissue to the LemnaTec Phenotyper platform; in 2016, the plant genomics and phenomics research data repository (PGP) were developed by the Leibniz institute of plant genetics and crop plant research and the German plant phenotyping network to comprehensively publish plant phenotypic and genotypic data. Obviously, from the perspective of database data standardization and storage, the storage scheme based on “cloud technology” is becoming the trend for the development of plant phenotype data storage. Cloud storage system can optimize the design of the system architecture, file structure, high-speed cache, etc., for the plant phenotype platform. At present, all kinds of phenotypic data collection platforms are still relatively independent, and have not been established at the level of regions, countries or continents. Through the advanced technology of artificial intelligence, establishing a typical crop phenotype database based on the multi-layer phenotypic information, for example GDB Human Genome Database, will of interest to a range of stakeholders.

27Plants are highly plastic to genotypes, environment and management via changing morphological traits and adjusting their physiological behavior. Complex interactions between genotypes, environment and managements at different scales determine the development of plants, but their separate contribution to the phenotype remains unclear. Dynamic models have been proven to be an efficient tool in dissection of abiotic and biotic effects on plant phenotypes. Functional–structural plant (FSP) models simulate plant growth and development in time and three-dimension (3D) space, and quantify complex interactions between architecture and physiological processes. The combination of FSP modeling and phenotyping have been used as a facile technique to address research questions in two ways.

28Firstly, FSP models offer a tool to dissect phenotype governed by a set of mechanisms. For example, used a FSP model to dissect net biodiversity effect into the effect induced by interspecific trait differences, and the effect induced by phenotypic plasticity by simulating whole-vegetation light capture for scenarios with and without phenotypic plasticity based on experimental plant trait data. The separate effect of each architectural trait (leaf angle, leaf curvature and internode length etc.) on dry mass production and light interception was quantified by simulating canopy growth using a dynamic FSP model for tomato. Radiation interception and radiation use efficiency were dissected into an environmental and a genetic term via conducting virtual multi-genotype canopies, in which the FSP model for maize was applied to calculate light interception for each plant. FSP modeling has proven to be highly effective for disentangling the relative contribution of each underlying process.

29Secondly, high-throughput phenotyping techniques facilitate the automate and precise calibration of FSP models. Image-based analysis was performed daily to reconstruct individual maize architecture, in order to calculate light interception using a FSP model and to estimate leaf area and the fresh plant weight of individual plants. Terrestrial LiDAR scanning was used to reconstruct complex tree canopy for predicting three-dimensional distribution of microclimate-related quantities in terms of net radiation, surface temperature and evapotranspiration. OPENSIMROOT integrated a root model that can simulate growth of a root system with 3D phenotyping techniques, such as magnetic resonance imaging (MRI) and X-ray computed tomography (CT). Phenotyping techniques not only provide an efficient method to evaluate the ability of the model to simulate plant architecture and geometry but also help researchers to understand functional responses based on images.

30Although genomic data has a major role in crop genetic improvements and breeding programs, with the advent of the era of omics, considerable gain can only be achieved by tightly coupling genomic discovery to plant phenomics. In recent years, phenomic researches that combine genomic data with data on quantitative variation in phenotypes have been initiated in many species, which rapidly decoded the function of a mass of unknown genes and improved understanding the G-P map.

31Many agronomic traits are complex and controlled by many genes, each with a small effect. Identifying the molecular basis of such complex traits requires genotyping and phenotyping of suitable mapping populations, enabling quantitative trait locus (QTL) mapping and genome-wide association studies (GWAS), which have been widely carried out in crop plants. associated phenotypic traits of small grain cereals with genome information to dissect the genetic architecture of biomass accumulation. In 2014, based on 13 traditional agronomic traits and 2 newly defined traits of rice, identified 141 associated loci by GWAS. In 2015, Combining GWAS with 29 leaf traits at three growth stages using high-throughput leaf scoring (HLS), 73 new loci with leaf size, 123 of leaf color, and 177 of leaf shape were detected. In 2017, large-scale quantitative trait locus (QTL) mapping was performed, combined with 106 agronomic traits of maize inbred line from seedling to tasselling stage, and a total of 988 QTLs were identified. Also, plant sizes of 252 diverse maize inbred lines were monitored at 11 different developmental time points, and 12 main-effect marker-trait associations were identified.

32Obviously, combining the high-throughput phenotyping technology and large-scale QTL or GWAS analysis not only greatly expanded our knowledge of the crop dynamic development process but also provided a novel tool for studies of crop genomics, gene characterization and breeding. We believe that with a complete system of genetic information, combined with crop high-throughput phenotyping technology, phenotypic-genomic analysis will revolutionize how we deal with complex traits and underpin a new era of crop improvement.

33Phenomics is entering the era of ‘Big Data,’ thus the crop science community need to combine artificial intelligence technology and collaborative research at the national and international levels, to build a new theory for analyzing crop phenotypic information, construct an effective technical system able to phenotype crops in a high-throughput, multi-dimensional, big-data, intelligent and automatically measuring manner, and create a tool comprehensively integrating big data achieved from a multi-modality, multi-scale, phenotypic + environmental + genotypic condition. There is no denying that there are many challenges that crop phenomics need to address in the next 5–10 years, such as:

34Phenomics is entering the big-data era with high-throughput, multi-dimensionality, and multi-scale. We emphasize various phenotyping approaches for crop morphology, structure, and physiological data with three multi- characteristics: multi-domain (phenomics, genomics etc.), multi-level (traditional small to medium scale up to large-scale omics), and multi-scale (crop morphology, structure, and physiological data from cell to whole-plant). The single and individual phenotypic information cannot satisfy the association analysis in the new era called ‘-omics,’ and the systematic and complete phenomics information will be the foundation of future research.

35In response to emerging challenges, new methods and techniques based on artificial intelligence shall be introduced to advance image-based phenotyping. An automated phenotyping system and platform result in lots of digital features, which need to prove their values throughout large sample statistics and relationship analysis with traditional agronomic traits. How to precisely and efficiently evaluate, understand and interpret these digital image-based features, and dig out valuable quantitative traits for functional genomes are key problems in the development and application of plant phenotyping.

36With the multi-domain, multi-level, and multi-scale phenotypic information, we urgently need to make use of the latest achievements of artificial intelligence in depth learning, data fusion, hybrid intelligence and swarm intelligence to develop big-data management producers for supporting data integration, interoperability, ontologies, shareability and globality.

37Modeling is a powerful tool to understand G × E × M interactions, identify key traits of interest for target environments. Nevertheless, several scientific and technical challenges need to be overcome. For example the validity and practicality of the models in terms of modeling processes and their interactions need further verification, and the interaction and feedbacks of multi-scale phenotypes between modeling processes also need to be solved. Only then will we be able to streamline and speed up the tortuous gene-to-phenotype journey through modeling to develop the required agricultural outputs and sustainable environments for everybody.

38Crop genotype (G) -phenotype (p) -envirotype (E) information comprehensive analysis and utilization. In short, as said “the future of plant phenotyping lies in synergism at the national and international levels.” We need to seek novel solutions to the grand challenges of multi-omics data, such as intelligent data-mining analytics, which offers a powerful tool to unravel the biological processes governing plant growth and development, and to advance plant breeding for much-needed climate-resilient and high-yielding crops.

39CZ, XG, and YZ conducted the literature survey and drafted the article. JD, WW, SG, JW, and JF provided the data and tables about the crop phenotyping researches by NERCITA. All authors read and approved the final manuscript.

40The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.