Crop Phenotyping

Abstract

Crop phenotyping aims to quantify traits like photosynthesis, development, architecture, biomass or quality of crops using a broad variety of sensors and analysis procedures. The course aims to provide the necessary basic knowledge in agronomy and plant breeding along with knowledge in image acquisition, computer vision, machine learning and crop modelling to improve crops and cropping systems.

Objective

The course aims to get you acquainted with different aspects of crop phenotyping. A major focus is to develop your competence to work in interdisciplinary teams to accomplish different aspects of the crop phenotyping workflow. We will focus on different aspects. Crop physiology Using in-field observations, you will get a basic understanding of the developmental and physiological processes leading to yield formation using wheat and soybeans as examples. Basics in sensors In our sensor hands-on practice, you will learn how to use different types of sensors including laser scanning, thermography, RGB imaging and chlorophyll fluorescence. Basics in modelling Using a range of data analysis scripts in python and R, you will get acquainted with AI-based feature extraction from images: You will acquire a basic understanding how to define training strategies for new computer-vision algorithms to quantify traits in complex crop canopies. 1) You will learn how to use community-based labelling software to collaboratively label targeted aspects (such as leaves) in images of complex canopies. 2) We will use this training data to improve AI-based models or train them from scratch. Using exercises in R, you will get acquainted with the downstream modelling tasks including correcting for field heterogeneity and dynamic modelling of measurements over time. Basics towards FAIR, standardized data To make data fair and big you will get to know crop ontology standards ( https://cropontology.org/ ) and we will discuss how to improve or adapt these standards to comply with new digital technologies. In your collaborative work you will get acquainted with the “Minimum Information About a Plant Phenotyping Experiment” ( https://www.miappe.org/ ) and how to use it to describe your experiment. You will learn how to organize your measurements and data using unique identifiers (UIDs) and relational data tables and to work collaboratively on a git repository. You will utilize methods to calibrate and validate new digital technologies by getting acquainted with the ”Breeders’ Equation” to calculate heritability and international standards to evaluate new methods, such as PP1/333.

Content

Arable crops such as wheat or soybean cover a large part of our planet. The efficiency with which they are produced has a major impact on food security and ecology. During their development, crops are exposed to a wide range of biotic and abiotic stresses. Adaptation of crops to a changing climate including extreme environmental conditions (e.g. cold and heat stress or drying soils) and to new diseases are important breeding aims. New cropping systems, such as mixed cropping or intercropping, introduce additional complexity for crop improvement: Breeders need to understand the interaction among plants of different varieties or species to make the right selection decisions. The crop phenotyping community offers tools to improve the selection process by helping to understand how crops achieve high yield and how yield components are affected by stresses. Their phenotyping toolbox consists of: i) sensor carriers (drones, ground-based robots, gantries and hand-held devices), ii) all kinds of active and passive sensors, and iii) an extensive model and data analysis framework. This framework is increasingly deploying 3D information from crop canopies combined with crop models and AI-driven image and data analysis workflows. Due to this complexity the community is divided into different sub disciplines with research specialists such as i) physiological breeders, who bring in a mechanistic understanding how crops function; ii) automation and sensor developers, who bring in new, plant-specific sensing solutions; iii) modellers, who make sure that sensor-derived data is summarized to predict the target traits (mostly yield and quality parameters); iv) data management specialists making sure that data is FAIR and that thousands of small experiments carried out across the globe can be harnessed into big data to understand the interaction between genotypes, the production environment and management practices (GxExM). All these specialists need to be able to collaboratively contribute to the overall data acquisition and data processing workflow. In Crop Phenotyping, we teach the main aspects of this workflow and form teams that collaboratively work on a common project. The course will take place on the field phenotyping platform FIP (kp.ethz.ch/FIP) of the ETH research station in Eschikon, which is part of the International Plant Phenotyping Network ( https://www.plant-phenotyping.org ), European Research Infrastructure for Plant Phenotyping ( https://emphasis.plant-phenotyping.eu/ ) and the DigiCrop network ( https://digicrop.net/ ). We planted a “variety garden” with the most important crops of Switzerland and an experiment including all varieties listed on the wheat variety list of Switzerland to work on. Both experiments will be used also for teaching and intense observations with the aim to explain how the different crops go through the season. This will include in-field observation of the response towards stresses and diseases. We will look at different types of sensors ranging from active sensing of chlorophyll fluorescence as proxy for photosynthesis over thermography to high-resolution RGB imaging including reconstruction of point clouds from multi-view images. Modelling steps will include the acquisition of image training data by labelling, the extraction of features using deep neural networks, spatial correction of field heterogeneity and dynamic modelling of growth and stress response. In a common project, we will split up into four groups mimicking the specializations of different crop phenotyping experts like working out the basics in crop physiology, utilizing imaging sensors in the field, extracting targeted features from the image and processing the data to compare it with ground truth. The results of this collaborative research will be presented at the final field day after the end of the semester. At this day we will be in the field with experts to learn additional phenotyping aspects and the quantification of diseases.

Resources