WHEAT.AI aims to revolutionize the way we gain ecophysiological and agronomic knowledge by leveraging modern machine learning techniques and earth observation.
The main goals of this project are:
- Develop and evaluate a sophisticated machine learning model to predict growth by means of Leaf Area Index (LAI) development for winter wheat using data on weather, soil, and cultivation types (conventional and organic farming).
- Investigate the impact of soil characteristics, weather conditions, and cultivation types on crop growth and yield, thereby enhancing our understanding of the risk and resilience factors in winter wheat cultivation.
- Compare and validate the WHEAT.AI model against mechanistic models and field experiment results, ensuring its reliability and applicability to real-world agricultural practices.
We utilizing transformer-based deep neural networks for precise growth modeling and yield prediction, moving beyond traditional time-series and function-fitting models by leveraging a unique, comprehensive, multi-year, country-wide dataset , which includes LAI observations, environmental variables, and detailed information on cultivation types over more than eight years across the entire agricultural landscape of Switzerland. By integrating ecophysiological knowledge into earth observation approaches, we bridge the gap between earth observation and crop physiology .
In conclusion, we believe that the synthesis of this truly interdisciplinary research will be groundbreaking. Just as modern machine learning has revolutionized applications such as protein folding and weather forecasting, we envision how systems like WHEAT.AI may revolutionize how we investigate ecophysiological processes of plant growth at the landscape scale and identify risks for crop production. This will be evaluated by addressing exciting research questions. Moreover, the proposed research will be pivotal for earth observation applications, demonstrating a new way for gap-filling and denoising time series data of plant traits beyond typical satellite data-driven approaches.
Team
Partners
Michele Volpi, Swiss Data Science Center
Funding
We are thankful for the funding by the Swiss National Science Foundation (SNSF)
