We use satellite imagery in order to predict agricultural yield reliably and early in the season. This helps to use resources more efficiently and to make informed agronomic decisions.
Global agriculture is facing challenging times. Increasing amounts of impervious surfaces, unreliable and rapidly changing weather conditions as result of climate change and a growing population are posing an increasing number of challenges to the field. At the same time there is an increasing amount of legislation about the efficient use of resources.
Agriculture is becoming more and more demanding. In order to increase agricultural yields while using resources efficiently, it is important to be able to take decisions early on, for example the efficient distribution of fertilizers and pesticides.
Machine Learning can support farmers but is often limited by the availability of relevant data. Since it is often challenging to get access and harmonize the needed agricultural data, the Yield Consortium focusses on the use of Earth Observation data instead. We develop techniques for processing all relevant data sources, such as satellite data, yield maps, weather data, soil data and digital elevation maps. We then combine all available data sources as inputs to machine learning models to predict agricultural yields with sub-field level precision. These yield predictions can support agronomical decision making such as decisions about the time point of harvest, irrigation, fertilization and pesticide applications.
In addition, we aim for bringing digitalization and agriculture closer together. Many farmers have a multitude of relevant data sources. Their use is often limited because of missing harmonization. We help farmers to make profit from their data, such as agricultural yield maps.
Remote sensing is one of the core technologies of earth observation and has a multitude of possible applications in agriculture. They offer the potential to monitor large spatial regions with high temporal frequency.
The German Research Center for Artificial Intelligence is the world's largest independent research center for artificial intelligence. Our experts focus on the transfer of research results to industrial applications.
Domain expertise is key to success. We therefore collaborate closely with industrial partners who have a long-lasting experience in the domain of agriculture. Combining this with the AI expertise at DFKI and earth observation data allows us to develop new technology for industrial applications.
It is well known that indices derived from satellite imagery, such as the normalized difference vegetation index (NDVI), can be used to predict the biomass of crops. It is, however, a challenging task to predict the yield based on the amount of biomass. Crops with a large biomass can still result in low yield for example when suffering from nutrient deficiencies.
We use artificial intelligence to directly predict the yield based on the full spectral information of the satellite imagery. We therefore benefit from the additional multi-spectral information contained in satellite imagery invisible to humans. We employ pixel-wise yield predictions with a sub-field level resolution of 10x10m per pixel. Apart from the satellite imagery we fuse a range of additional data sources in the machine learning models in order to improve the prediction performance. Current models show promising results in predicting the sub-field level yield variations.
We are still looking for additional cooperation partners. Do you have high quality yield maps and are you interested in making a profit out of them?
The key to precise yield predictions with artificial intelligence is high-quality ground truth data for yield. To this end, we use yield maps from calibrated combine harvesters. Feel free to get in touch, if you would like to sell combine harvester yield maps and in return would like to benefit from preliminary results of the yield consortium.