MODELING CLIMATE CHANGE IN THE ASEAN REGION USING CO₂ TRADE NETWORK-BASED SPATIAL-TEMPORAL GRAPH NEURAL NETWORK (ST-GNN) AND IMPACT ANALYSIS ON INDONESIA

Abstract

Climate change in Southeast Asia is intensifying due to rapid economic growth, industrial expansion, and cross-border carbon transfers. This study employs a Spatial-Temporal Graph Neural Network (ST-GNN) to model CO₂ emissions embedded in ASEAN’s trade network, with Indonesia as the focal point due to its dominant role in regional emissions. Using Our World in Data (OWID) datasets (1990–2023), the ST-GNN framework captures interdependencies between trade-linked emissions and temperature change, outperforming traditional models with an RMSE of 0.011 when optimized (90% top features). Key findings reveal that Vietnam and Australia exert the strongest influence on Indonesia’s emission-driven temperature rise, while Singapore acts as a high-centrality hub in the carbon network. Permutation importance analysis identifies land-use change (CO₂_including_luc), energy consumption per capita, and coal-based emissions as the top predictors of warming trends. The temporal attention mechanism highlights critical periods, such as the 1998 financial crisis and post-2008 recovery, where economic shocks amplified emission impacts. Policy recommendations emphasize regional carbon accounting frameworks, deforestation control, coal phase-out strategies, and ASEAN-wide climate collaboration to mitigate transboundary emissions. This study demonstrates that ST-GNNs enhance climate modeling by quantifying spatial-temporal emission dynamics, offering actionable insights for decarbonizing trade-dependent economies.