As part of the second project phase of the FEdA Central Coordination, the project "Continuous AI-based Biodiversity Assessments for Germany (KIBA-D)" is researching the foundations for an AI-based system in which DFKI AI experts in the field of Large Language Models, together with biodiversity experts from the Faktencheck Artenvielfalt and the FEdA consortium, will use the Faktencheck Artenvielfalt corpus to train AI models that can extract relevant data on biodiversity trends, causal relationships, and assessments of the effectiveness of instruments and measures from digitally available text information. In additional steps, the corpus will be expanded significantly, and the quality of the AI-based results will be analyzed jointly by biodiversity and AI experts. For this purpose, a workflow for AI-supported extraction of biodiversity-relevant information from text sources, a crawling solution for locating and collecting additional sources, and LLM-based extraction models for the various analysis tasks will be developed.
The Research Initiative for the Conservation of Biodiversity (FEdA) is funded by the Federal Ministry of Research, Technology and Space. It aims to sustainably deepen research into biodiversity and provide new insights into the state of biodiversity in Germany and the causes, dynamics, and consequences of biodiversity changes. This knowledge can then be used to develop effective system solutions. The research initiative's unique approach is to develop proposed solutions not only from a scientific and ecological perspective, but also to consider social and economic perspectives. The Biodiversity Fact Check is an FEdA project for the comprehensive assessment and evaluation of biodiversity in Germany, with the goal of identifying and developing options for action for concrete national and subnational policy. For this assessment, more than 150 scientists from 75 institutions and associations evaluated the findings from over 6,000 publications and compiled them in a specially developed database. To identify long-term developments, they created and analyzed an unprecedented data set of around 15,000 trends from almost 6,200 time series.
The results of the KIBA-D project will be incorporated into the comprehensive literature database of the Faktencheck Artenvielfalt and are intended to significantly expand this dataset. The establishment of a system for the continuous extraction of biodiversity data from digital texts, developed during the project, is intended to lay the foundation for continuous automated analysis of the extracted data. The long-term goal is an accelerated, renewed, comprehensive assessment by 2030.
Partners
- Deutsche Zentrum für integrative Biodiversitätsforschung Halle-Jena-Leipzig- iDiv e.V.
- Öko-Institut e.V.
- Unabhängiges Institut für Umweltfragen