Integrating machine learning and system dynamics for climate-resilient viticulture: a case study from Slovenia

Purpose This paper explores how the integration of machine learning (ML) and system dynamics (SD) can enhance understanding of climate change impacts on viticultural ecosystems, with a focus on vineyard biomass growth and wine quality under rising temperatures. Design/methodology/approach The study combines multivariate temperature trend analysis (general linear model, GLM), ML models (Random Forest, SVM, Decision Trees, Linear Regression) and a SD model based on stock-and-flow architecture. Data include long-term weather records and over 70,000 wine quality scores across Slovenian wine regions. Findings ML models identified the vineyard region as the strongest predictor of wine quality, outperforming individual climate variables. However, their “black-box” nature limits interpretability. The SD model addressed this gap by simulating nonlinear growth responses, feedback loops and carrying capacity constraints. Simulations revealed optimal biomass growth at ∼ 25 °C, with a significant decline under heat stress conditions (>35 °C), confirming threshold-based system behaviour. Originality/value This study introduces an interdisciplinary modelling approach that bridges prediction and explanation. It contributes a novel integration of ML and system dynamics in viticulture, offering both empirical insights and a transparent tool for long-term climate adaptation planning.