AI-Assisted Initial Planning of CCUS Networks in Colombia: A Geospatial-Generative Framework

Abstract

This paper proposes a comprehensive methodological framework for early-stage planning of Carbon Capture, Utilization, and Storage (CCUS) networks in Colombia. The approach integrates geospatial clustering, machine learning forecasting, and generative artificial intelligence (AI) to support the detection of emission hotspots, the estimation of CO₂ flows, and the conceptual simulation of logistics configurations. Based on historical emissions series (2000-2022), spatial infrastructure layers, and sectoral indicators, highpriority clusters are delineated using geospatial intelligence techniques such as kernel density estimation
and DBSCAN. To project CO₂ capture volumes under different macroeconomic and regulatory scenarios, Long Short-Term Memory (LSTM) and XGBoost models are employed, incorporating SHAP values to enhance interpretability. Generative AI (GPT-4o) is integrated to recreate context-sensitive supply chain designs, producing conceptual configurations that guide stakeholder engagement and territorial planning. Taken together, the methodology establishes a decision-support architecture that links spatial analytics and artificial reasoning to inform CCUS deployment in fragmented and data-scarce environments. The study contributes to the international debate on low-carbon transitions by offering a scalable and adaptable planning toolkit, grounded in Colombia’s institutional and territorial realities.