Deep learning for sustainable development across climate, energy, agriculture and urban systems

Abstract

Deep learning (DL) has emerged as a transformative paradigm for addressing the multidimensional and data-intensive challenges of sustainable development. The purpose of this review is to examine how DL contributes to four critical domains-climate action, sustainable energy, smart agriculture, and urban development-while identifying gaps that limit large-scale deployment. Methodologically, the study synthesizes peer-reviewed research published between 2000 and 2025, covering architectures such as convolutional neural networks, recurrent networks, transformers, graph neural networks, autoencoders, and multimodal frameworks. The findings reveal key trends including the rise of physics-informed models, the integration of deep reinforcement learning in energy and transport systems, and the increasing adoption of federated and edge AI for decentralized monitoring. At the same time, recurring challenges are identified: data scarcity, limited cross-regional generalizability, deficits in explainability, and ethical concerns surrounding fairness and accountability. The review concludes that addressing these issues requires hybrid physics–AI modeling, uncertainty-aware and participatory AI frameworks, and deployment-oriented research strategies. The key contributions and implications of this work are threefold: (i) the development of a cross-domain taxonomy mapping DL methods to sustainability tasks, (ii) benchmarking insights to guide model selection and evaluation, and (iii) a forward-looking research agenda to support researchers, practitioners, and policymakers. The originality of this review lies in its cross-sectoral synthesis, which extends beyond domain-specific surveys to highlight how DL can be responsibly scaled to advance the United Nations Sustainable Development Goals (SDGs).