The Yield consortium uses satellite data and artificial intelligence to reliably predict agricultural yields. In collaboration with BASF Digital Farming, John Deere, and MunichRe DFKI develops predictive models for selected arable crops in the focus regions of Europe, South and North America. Later models will extend to other relevant crops and growing regions.
Environmental challenges in our world today are crystallizing on a global scale, in the form of food crises, wars, and the consequences of a changing climate. The agro industry needs transparency and assistance to form an active response to change and to successfully execute crisis management in the farming sector. DFKI launched the Yield Consortium project, funded through the “ESA InCubed Programme” to make an important contribution towards research. The consortium is part of the AI4EO Solution Factory, a collaborative research effort by ESA and DFKI, which seeks to find new, real world solutions for business and industry on the basis of satellite data.
The potential benefit of the Yield Consortium is found in the ability to proactively react to the impending changes in agricultural yields. Industry partners support the modelling process and share their complementary expertise in the areas of finance, agronomy, and harvesting systems. Satellite data is used at an early stage to reliably predict expected yields.
Yield predictions may one day in the future enable new insurance calculation concepts in terms of irrigation, fertilizers, crop protection, and profit planning that result in better loss estimates. These forecasts are not only beneficial for the industry and government policy makers, but also for the farmers themselves. They are useful in optimizing cultivation methods and improving crop protection.
In just a short time, the interdisciplinary team has managed to develop and successfully test an end-to-end model, which meets the requirements of the industry partners. Currently, yield forecasts can be determined in Germany for wheat and rapeseed crops and in South America, Argentina, and Uruguay for soybeans. The model achieves this on the basis of satellite data and also the integrated data on soil properties, plant growth stages, the weather information, and digital elevation models.
Yield forecasts are recorded at different points in time. For example, agricultural yields can be determined for harvest time and up to 120 days prior to the harvest. The closer the date to the harvest, the more exact the forecast will be. The answers to basic questions in the area of logistics and grain distribution are now more easily available: What location is best suited for individual crop? What kind of yields can we expect this season? Which herbicides/pesticides and how much fertilizer should be used? Yield maximization is significantly impacted by the answers to all of these questions.
"Our goal is to provide forecasts of agricultural yields as accurately as possible. This should help farmers in their day-to-day decisions. We would like to gradually expand the model to other countries, take more crops into account, and improve the existing models. A model will eventually be designed that can make global predictions for all important crops," said Dr. Marlon Nuske of the AI4EO Solution Factory.
The AI4EO Solution Factory was established in the Fall of 2021 through the “ESA InCubed Programme”. The aim of the AI factory is to apply artificial intelligence to exploit the potential of Earth Observation satellites in space and make it useful to industries on Earth. The possibilities of Earth observation are as limitless as the universe itself. Special solutions are worked out for each customer. At the same time, synergies in these different areas of application are captured to create a foundation on the programming side, parts of which can then be reused in a transfer ecosystem.
DFKI researcher Dr. Michaela Vollmer describes a practical application:"As it harvests a field, the combine continuously measures the yield in tons per hectare with such a high resolution that it can be assigned to a point of origin in the field. The measured harvest points or 'geo locations' allow us to see at the sub-field level, the point-by-point yield for the field. We use this high-resolution data to match the images from space with a 10 x 10 meter pixel resolution. This has the advantage of being useful for optimizing the training of our machine learning model. In turn, the farmer is provided with better information to apply individual measures on each field."
The team is interested in acquiring additional yield data to further develop the model."We have already begun successful collaborations with the agriculture-related companies Smartway and Manexa to supply their yield data. However, we are still looking for more cooperation partners who can provide high-quality yield maps," said Dr. Marlon Nuske.
"The success of the Yield Consortium will help us to attract additional industry partners from different subject areas for research, implementation, and practical model development. The combination of AI and EO in the project goes beyond basic research. The ongoing exchange with industry partners at DFKI facilitates knowledge transfer to different application areas via so-called TransferLabs. We look forward to collaborating with additional partners from other industrial sectors," said Prof Andreas Dengel, Managing Director, DFKI Kaiserslautern and head of the Smart Data & Knowledge Services research area.
© John Deere
© John Deere
Global food security and the agricultural sector are already facing global challenges. No other sector is so dependent on climatic factors and so directly affected by climate change: soil sealing, changing temperatures, weather events such as heavy rainfall, heat waves, and forest fires in the wake of global climate change are just some of the most serious phenomena. To be sure, extreme weather events have always existed. Yet, recent years have seen a rapid increase in the frequency of climate induced events. The continuous release of greenhouse gases is affecting the cultivation of corn, wheat, and rice. Traditional farming has relied on many years of experience to make determinations about which crop to plant and where. Continuing to implement this in the future will be difficult without having timely assistance through risk minimization and seasonal management that responds to shifting and unpredictable yields.
Present macrosocial challenges are becoming much more pronounced with regard to global food security. The agricultural sector is facing serious repercussions from a world food crisis triggered by the imminent shortfalls and threats from war zones of preventing grain exports. Sound data about agricultural yields can help to better assess these troubling scenarios. The AI models in the Yield Consortium project are designed to provide reliable yield forecasts and facilitate seasonal decision-making.
Dr. Marlon Nuske
Forschungsbereich Smarte Daten & Wissensdienste
Senior Researcher, DFKI Kaiserslautern
Tel.: +49 631 20575 7250
E-Mail: Marlon.Nuske@dfki.de
Dr. Michaela Vollmer
Forschungsbereich Smarte Daten & Wissensdienste
Senior Researcher, DFKI Kaiserslautern
Tel.: +49 631 20575 7611
E-Mail: Michaela.Vollmer@dfki.de