Deep Generative and Hybrid Models for Advancing Environmental Sustainability in Smart Cities

This chapter sets out to provide a multi-dimensional review of how generative artificial intelligence (GenAI), particularly deep generative model architectures, is being applied to support climate mitigation and adaptation in the context of sustainable smart cities. The tabulated analysis shows that generative adversarial networks (GANs) and variational autoencoders (VAEs) are the most widely applied generative models across climate mitigation and adaptation domains, primarily supporting tasks such as simulation, forecasting, augmentation, and imputation. In contrast, diffusion models (DMs) and transformers remain underutilized, particularly in real-world implementations, revealing significant gaps and untapped potential for broader deployment in climate-related contexts. Notably, hybrid architectures—combining generative, discriminative, deterministic, probabilistic, and/or heuristic components—are gaining traction across both mitigation and adaptation applications. The thematic synthesis reveals that GenAI plays an increasingly transformative role across six key climate domains by enabling real-time forecasting, intelligent decision-making, and scalable simulations that support climate-resilient urban systems. While its integration into energy, transportation, waste, water, disaster, and biodiversity management enhances operational efficiency and environmental responsiveness, it also introduces challenges related to data quality, algorithmic transparency, computational demands, and governance. By conceptualizing these insights through a multilayered framework, the study offers theoretical and practical contributions to the evolving discourse on GenAI-enabled climate action.