Objectives

Objectives, Hypothesis, and Approach

Crops are the cornerstone for sustainable food production. Crop production is under pressure for several reasons. First, the growing world population needs to be fed and demands more renewable resources from agricultural production. Second, arable land and water resources are under increasing stress and climate change imposes additional constraints. Third, society increasingly demands high-quality and healthy food. Forth, increasing crop production needs to be attained while reducing the environmental footprint, for example, by restricting the use of certain agrochemicals, soil degradation, and water pollution.

Increasing yield while reducing environmental impact is at the core of sustainable crop production. To realize this, we need:

1. More detailed and specific information about the current status of the crops and the environment;

2. Better knowledge and understanding towards improved models handling the complex relationships of crop growth and the ecosystem, including the ability to make forecasts, simulations, and optimizations;

3. New robot-driven approaches to crop/soil management, weed control, and plant protection that reduce the environmental impact and increase yield;

4. Knowledge about the technological and socio-economical prerequisites such that these approaches will get widely adopted in order to achieve large-scale impact.

This cluster aims to address these needs through a technology- and data-driven approach.

Our main hypothesis is that a major shift toward sustainable crop production will be achieved by (1) multi-scale monitoring and modeling of plants and the environment using autonomous robotics with automated intervention and big data analysis combined with machine learning; and (2) integrating the developed innovations by considering crop farming in a systemic manner.

The various in-field monitoring activities of the PhenoRob Cluster will generate large amounts of heterogeneous data on plants, crops, and the environment. These data will include geometric and semantic models of individual plants, multispectral images at different scales, distributions of grown plant species in and around the fields, nutrition and water contents in soil, weather information, and further soil-crop-atmospheric variables and fluxes in the environment.

The main scientific challenges in such an approach are:

Challenge 1. How to identify relevant relations between inputs (crop genetic resources, crop management, and soil and weather conditions, etc.) and outputs (yield, plant growth, phenotypic performance, and environmental impact, etc.) in complex and changing field environments, to enable reliable forecasts and predictions about the consequences of changes in input parameters on the output and to support the design of improved production systems.

Motivation of Challenge 1. Today, plant- and agro-sciences generate extensive amounts of data. However, achieving good predictions for breeding and management based on these data is difficult. In this challenge, PhenoRob will identify through which measurements relevant parameters can be obtained to improve our breeding capabilities, to design effective agronomic management, and to improve the predictive power of models for future climate and management scenarios. PhenoRob will analyze how interconnections between input and output can be identified, by using big data and machine learning approaches.

Challenge 2. How to obtain more relevant information about the current state/parameters of the crop farming system, and how to exploit this information for crop management and robotic in-field intervention at individual plant level.

Motivation of Challenge 2. Spatial and temporal data are essential for breeding, early stress detection, farm management, and predictive modeling to improve agro- and ecosystem performance. Current technologies do not allow stakeholders (e.g., farmers and breeders) to obtain data with sufficient accuracy and throughput that enables the identification of novel traits for breeding and targeted intervention in an automated way. This requires models with prediction and forecasting capabilities in order to automatically optimize the management suggested to be executed by the robots. PhenoRob will develop the relevant technologies and tools to achieve this. In addition to that, the ability to perceive and treat every single plant according to it needs allows for substantially reducing the amount of required agrochemicals and fertilizer.

Challenge 3. How to design the process of upscaling the innovations to ease technology transfer as well as compatibility to current crop farming and to obtain information on how the new technology will change crop farming systems, landscape, and markets.

Motivation of Challenge 3. The appropriate measurements, novel technological capabilities, or an improved understanding of the underlying processes alone is not sufficient to meaningfully implement these innovations into crop farming. Sustainability can only be achieved with large-scale adoption. Thus, PhenoRob will address the questions of how to translate novel technologies into agricultural practice and what the risks and benefits are that the different stakeholders will face.

Approach. The PhenoRob Cluster will address these challenges by combining recent progress in probabilistic robotics, modern geodesy, computer vision, machine learning and big data processing with distributed sensing, high-precision monitoring, plant and root phenotyping, soil and ecosystem research, and crop growth modeling. This will be achieved through the following four actions:

(a) Systematically monitor all essential parts of crop farming with sensor networks and mobile ground and aerial robots, providing spatiotemporally-aligned and detailed data about individual plants, soil, nutrient, and ecosystem parameters such as weather or biodiversity of the flora. This will enable more effective and sustainable control and management of input and output parameters of the whole system comprising soil, crop, and environment.

(b) Develop novel technologies that enable real-time in-field interventions, such as automated control of weeds and selective spraying and fertilization of individual plants in field stands. First, such automated systems must be developed that can perform interventions on its own. Second, a productive ecosystem model is needed that allow for simulating the effects of different management action to select the most promising actions. This will be an enabling approach to reduce the environmental footprint of crop farming by reducing the amount of applied chemicals, and by minimizing soil degradation and erosion.

(c) Apply and extend modern machine learning techniques to analyze the recorded data, to improve our understanding and the models of plant growth, nutrient, and water efficiencies in relation to inputs, and to identify complex correlations with outputs.

(d) Analyze the expected impacts of the novel approaches to management decisions at farm-level, and investigate the requirements for a broad technology uptake considering market interactions, and benefits that will result from upscaling the developed technologies.

In particular, the approach of addressing crop and agro-/ecosystem modeling in combination with big data analytics and large-scale machine learning is especially novel, innovative, and promising, but also carries high risk. It is an orthogonal approach compared with omits and molecular approaches to modeling. We believe that this cluster will establish the foundations for a new direction of data-driven, sustainable crop production.