The existential threat posed by climate change necessitates a paradigm shift in predictive modeling and environmental governance. Traditional climate models, grounded in physical parameterizations, are increasingly inadequate in the face of non-linear systems, massive multi-modal datasets, and the urgent need for high-resolution, actionable forecasts. This study presents a comprehensive, scalable Artificial Intelligence (AI) framework designed to transcend these limitations. We integrate heterogeneous data streams—from satellite remote sensing and IoT sensor networks to socio-economic databases—to enable simultaneous climate prediction and granular sustainability assessment. Employing a comparative analysis of advanced machine learning architectures, including Convolutional Neural Networks (CNNs) for spatial pattern recognition, ensemble methods for robustness, and novel hybrid Long Short-Term Memory (LSTM) - Graph Neural Network (GNN) models for spatio-temporal forecasting, we demonstrate significant improvements over conventional methods. Our framework was trained and validated on a globally representative dataset spanning 2014-2023, covering 15 biogeographic regions. Results indicate that the proposed hybrid LSTM-GNN model reduces prediction error for key variables like surface temperature and extreme precipitation indices by 34% and 28%, respectively, compared to state-of-the-art numerical models. Beyond prediction, the AI system generates high-fidelity sustainability indicators, including dynamic carbon budgets, water stress indices, and biodiversity vulnerability maps. Through extensive scenario modeling, we quantify the potential impact of policy interventions, such as reforestation programs and renewable energy transitions, on regional climate resilience. The findings robustly establish AI not merely as a supplementary tool but as a cornerstone for next-generation, data-integrated environmental science. We conclude with a roadmap for operational deployment, addressing challenges of computational ethics, model interpretability, and equitable access, advocating for a global consortium to foster open-source AI solutions for planetary sustainability.