ScholarWorks@경상국립대학교

초록

This study proposes a science-based pesticide safety management framework by integrating nationwide monitoring, environmental risk assessment, and AI-based predictive modeling. Residual levels of 92 pesticides were monitored at 496 paddy soil sites and 100 sites each for streams and lakes during April, July, and October. In soil, 77 % of compounds were detected above 0.01 mg/kg; in water, 44 compounds exceeded 0.05 μg/L. Frequently detected pesticides exhibited long soil half-lives and moderate to high Koc values. Principal component and cluster analyses identified butachlor and chlorantraniliprole as key contributors to cumulative Risk Quotient (RQ) in soil and water, respectively. Ridge regression effectively corrected SFO degradation predictions in soil (R2 improved from −4.25 to 0.93), while Random Forest improved GENEEC2 accuracy in aquatic systems (R2 = 0.39–0.44). For butachlor, AI modeling predicted detection timing and risk reduction (RQ ≤ 1), with earlier mitigation under 50 % reduced application. For chlorantraniliprole, risk levels declined rapidly below the RQ threshold even at full application. These results demonstrate the utility of AI-enhanced models for forecasting pesticide behavior and informing risk-based management strategies.

키워드