Using Large Language Models for generating biodiversity reports
Master
Semester programme:Master of Applied IT
Client company:Nest Natuurinclusief
Timo Sevarts
Project description
This project studied if large language models can analyze geographic datasets to create valuable insights for ecological reports.
Context
This project is part of an ongoing Fontys research effort towards using technology for good causes and
is formulated in cooperation with Nest Natuurinclusief, a Rotterdam consultancy firm specializing in
urban ecology and impact of construction projects on that. The problem we focus on is the orientation
difficulty in the ever-growing information jungle on urban ecology, new biodiversity laws, (local)
governance rules etc. Parties like project developers, terrain owners and city residents would gladly
contribute to increasing urban nature if the needed information was readily available for them to make
efficient nature-inclusive decisions in their projects.
Results
This project has three main results:
1. A suitable architecture for generating biodiversity reports using LLMs;
2. A prototype using this architecture, that is able to generate nature analyses for municipalities;
3. A comparison between five LLMs (Gemma3, Qwen3, o3-mini, gpt-4o-mini and Gemini 2.5 Flash) to see what LLM fits best for generating biodiversity report analyses.