本文回顾了大气资料同化的传统方法，并探讨了将资料同化分辨率从数十千米提高到数百米的科学策略。这些科学策略得益于计算机硬件和计算方法方面的最新技术发展。本文选取两个重点问题说明其中的机遇与挑战：① 如何在高分辨率资料同化中充分利用卫星观测到的热带气旋云和雨带结构；② 传统资料同化的哪些核心技术需要重新评估和改进。具体讨论包括：通过探索卫星观测亮温与热带气旋内部不可直接观测的相对涡度之间的联系，推动全天空亮温同化的创新发展；构建适应卫星轨道观测时刻的全球资料同化框架；发展先进的资料稀疏化和质量控制方法，避免丢失最需要的大气小尺度和大梯度结构信息。最后，讨论如何有效融合气象资料同化与深度学习，例如将人工智能物理参数化模型引入四维变分同化系统，将图像同化与人工智能相结合，直接利用图像数据进行预测，以及实现资料同化与气象深度学习的相互促进。

This article reviews traditional practices in atmospheric data assimilation and explores scientific strategies to assimilate data at resolutions spanning from tens of kilometers to even hundreds of meters. Such advances leverage the latest technological developments in computing hardware and methods. Two focal points are specifically chosen to illustrate the opportunities and the associated challenges: ① How to fully exploit satellite-observed cloud and rainband structures in tropical cyclones for high-resolution data assimilation; and ② which traditional data assimilation core techniques need re-evaluation and improvement. Specific topics include making an innovative advancement in all-sky brightness temperature assimilation by seeking the connection between satellite observed brightness temperature and unobservable relative vorticity within tropical cyclones; constructing a global data assimilation framework suitable for satellite-orbit-data observation times; developing advanced data thinning and quality control methods to avoid losing the most needed atmospheric small-scale and large gradient structural information. Finally, how to effectively integrate meteorological data assimilation with deep learning is discussed, such as incorporating artificial intelligence (AI) physical parameterization models into 4D-Var system; combining image assimilation with AI to make prediction from image data; making data assimilation and meteorological deep learning mutually beneficial.