Recommendation systems are widely applied across various domains, particularly in providing users with personalized content and service recommendations. However, traditional recommendation systems face challenges in meteorological services when handling complex atmospheric data and real-time dynamic user needs. Accurate cloud top height (CTH) retrieval is crucial for enhancing meteorological services, especially in delivering precise weather forecasts and warnings. This paper proposes a self-attention-based CTH retrieval model integrated into an intelligent meteorological service recommendation (IMSR) system. Leveraging self-attention mechanisms, the model captures long-range dependencies and extracts key global features from satellite-based meteorological data, improving the accuracy of CTH retrieval even in data-sparse environments. Additionally, the system utilizes real-time CTH data to provide personalized weather recommendations for agriculture, transportation, and tourism users. Experimental results demonstrate that the proposed model outperforms existing methods in both CTH retrieval accuracy and recommendation effectiveness, significantly enhancing the timeliness and relevance of meteorological services. This approach integrates advanced deep learning techniques with practical applications in weather services, offering potential for cross-domain applications.