Continuous AI-based biodiversity assessments for Germany

KIBA-D marks the beginning of a new era in biodiversity research: for the first time in Germany, a system is being developed that uses artificial intelligence (AI) to automatically extract information on the nationwide state of biological diversity from digital text sources and evaluate it continuously – quickly, efficiently, and in real time.

The recently launched “Continuous AI-based Biodiversity Assessments for Germany” (KIBA-D) builds on the biodiversity fact check, which comprehensively documented the state of biodiversity in Germany for the first time. KIBA-D aims to use this data pool to train large language models (LLMs). The goal is to use AI to automatically update and expand this data set. To this end, in addition to scientific publications, government reports, statements, websites, and possibly even social media posts will be included, subject to strict source evaluation rules. These will be automatically found and evaluated by the AI and classified according to the structure of habitats, drivers, and measures developed by the Biodiversity Fact Check. The assessment is funded as accompanying research by the Central Coordination Office of the Research Initiative for the Conservation of Biodiversity (FEdA) of the Federal Ministry of Research, Technology, and Space.

“With KIBA-D, we are laying the foundation for an adaptive system that continuously evaluates biodiversity data nationwide and comprehensively, revealing trends at an early stage,” says coordinator Leonhard Hennig. “This enables us to obtain reliable information for politics, science, and society more quickly.” The assessment is being implemented by the German Research Center for Artificial Intelligence (DFKI), the German Center for Integrative Biodiversity Research Halle-Jena-Leipzig (iDiv), the Independent Institute for Environmental Issues (UfU), and the Öko-Institut e.V. and will be carried out over a period of 18 months.

Results are expected in early 2027. In the long term, the AI system being developed in KIBA-D will provide the basis for a comprehensive, continuous new biodiversity assessment in 2030. “Of course, we hope that our language models will one day also be used internationally, for example for IPBES reports,” says Hennig, setting ambitious goals.

FEdA spokesperson Volker Mosbrugger also emphasizes why it is so important that KIBA-D continuously expands the extensive data set from Faktencheck Artenvielfalt with the help of language models: “For the first time, we would have a system that automatically and continuously reveals biodiversity trends throughout Germany. The fact that the language model continuously extracts findings from scientific publications saves an enormous amount of time, gives us comprehensive insights into the trends and drivers of the biodiversity crisis, and helps us develop efficient and actionable recommendations for action.”

KIBA-D is funded with around one million euros as a module in the central coordination of the FEdA, which belongs to the Senckenberg Society for Nature Research. The assessment follows on from the renowned biodiversity fact check published in October 2024. More than 150 scientists from 75 institutions evaluated around 6,000 studies and compiled a unique overview of trends and causes of biodiversity loss in Germany. KIBA-D exemplifies a new, data-driven approach to protecting biological diversity – and how new technologies such as artificial intelligence can help us understand nature faster and better and preserve it in a more targeted manner.