Global financial and climate infrastructures are increasingly characterized by complex interdependencies thatevolve dynamically over time. Capturing and predicting systemic risks within these interconnected systemsrequires advanced methods that can model both temporal sequences and relational structures. Traditional statisticalapproaches and single-modality machine learning frameworks often fall short in addressing the scale,heterogeneity, and volatility of such data. To overcome these limitations, this study proposes a blockchainorchestrated temporal graph forecasting framework that integrates hybrid recurrent neural network (RNN)–transformer architectures for robust predictive analysis. The approach leverages graph-based representations toencode relationships among entities such as financial institutions, energy networks, and climate systems, whiletemporal layers capture evolving dependencies across time. The fusion of RNNs with transformers ensures thatboth local sequential dynamics and long-range dependencies are effectively modeled. Blockchain infrastructureprovides a decentralized orchestration layer, ensuring data provenance, immutability, and trust in multistakeholder environments where systemic risks transcend institutional and national boundaries. By embeddingforecasting outputs into a blockchain-enabled ecosystem, predictions become verifiable, auditable, and resistantto tampering, which is particularly critical for high-stakes decision-making in global risk governance. Caseillustrations across financial contagion modeling and climate impact forecasting demonstrate the framework’spotential to enhance transparency, resilience, and accountability in managing systemic vulnerabilities. Thisconvergence of temporal graph learning and decentralized orchestration marks a significant step toward creatingpredictive infrastructures capable of guiding policy, regulation, and crisis response in domains where uncertaintyand interdependence define systemic stability