Deep learning for intelligent irrigation decision-making: A review
Jingxin Yu, Qianhui Qu, Shuyi Peng, Xiaoming Wei, Ying Li, Congcong Sun
Abstract
Global agriculture faces the dual challenges of water scarcity and climate change, making efficient and precise irrigation management increasingly important. This review analyzes the role of deep learning (DL) technologies in intelligent irrigation decision-making: (1) DL technologies have shifted irrigation management from experience-based decisions to data-driven precision prediction. (2) Deep learning architectures demonstrate distinct advantages in different aspects of irrigation management, including spatial identification, soil water content prediction, long-term forecasting, and optimization of water use. (3) Hybrid DL models often demonstrate superior performance in practical applications. (4) Edge-cloud collaborative architectures are particularly effective, reducing communication volume and decreasing response times from minutes to seconds. Despite progress, intelligent irrigation using DL faces challenges related to data quality, model generalization ability, and computational resource limitations, as well as application barriers such as cost, acceptance, and regional adaptability. Future work should prioritize climate-adaptive models, extreme-weather response, and ultra-precise management in water-scarce regions, while evaluating federated, few-shot learning and large language models as enabling methods. • Deep Learning shifts irrigation from experience-based methods to data-driven, predictive decision-making. • Deep Reinforcement Learning optimizes schedules for significant water savings and yield increases. • Hybrid Deep Learning models integrating multiple architectures achieve the best practical performance. • Edge-cloud collaborative architectures enable real-time response by overcoming rural connectivity limitations. • Challenges include data quality and model generalization, requiring new climate-adaptive solutions.