Agentic AI-Assisted Geostatistical and Stochastic Mapping of Heavy Metal Contamination in Debrecen Urban Soils, Hungary

Abstract

Implementing an agentic-AI-orchestrated Sequential Gaussian Simulation pipeline in Google Antigravity with Claude Opus 4.6 and Gemini 3.1 Pro, to reproduce the prior manual SGS results from the VVNL project annual reports and extended FEHÉR et al. (2025)'s Ordinary Kriging maps with probability-of-exceedance surfaces. The simulation showed that 81.2 % of Cd samples exceeded the 1 mg kg-1 threshold, 18.4 % of the study area had a Pb exceedance probability above 0.75, and the Cu maximum of 1,034.58 mg kg-1 in Köntösgát exceeded the intervention limit by 14 times. The primary contribution is agentic-AI execution with rerunnability on new conditioning data, the secondary is SGS reproduction and extension with explicit uncertainty, and the tertiary is the identification of subjective-decision points at which agentic AI cannot substitute for researchers’ judgement.

Using this rather than usual manual conventional way, helps to save time and to provide accurately calculated geostatistical values.