Assessment of Zoning Plan Metadata using AI: Automating the Assessment of Land Parcels by Leveraging Large Language Models

Abstract

Zoning plans contain critical information about land use, building rights, and development potential, but analyzing these documents manually is time-consuming and often impractical at scale. This thesis investigates how automation and large language models (LLMs) can be used to extract meaningful data from Swedish zoning plan (detaljplan) metadata to support early-stage site selection and investment decisions. By combining spatial data analysis and economic indicators, the developed methods enables identification of underutilized parcels with high development potential. The study compares a regular expression keyword matching algorithm with Google’s LLM Gemini or interpreting regulations and extracting key parameters such as maximum building height and area utilization. Results show that the LLM achieves better overall performance and is consistently interpreting irregularly phrased regulations that traditional methods overlook. A tool has been created that integrates market data from recent real estate transactions to estimate local attractiveness, offering an economic dimension to the evaluation. Case studies from Järfälla, Halmstad, and Sundsvall demonstrate how the tool can improve the efficiency of zoning plan analysis, reduce manual workload, and provide a foundation for evaluating development sites.