This article aims to explore the potential application and challenges of large language model (LLM) in weather forecasting. Through analyzing LLM application in meteorology and related industries, including knowledge retrieval, foundation models, diagnostic analysis, tool calling and text generation, the article demonstrates that LLM has tremendous potentials in improving weather forecast accuracy and business intelligence. LLM serve as powerful assistive tools for forecasters by efficiently processing massive meteorological knowledge, integrating cross-domain multi-source information, and generating customized forecast products. The effectiveness of LLM in weather forecasting can be further enhanced by building high-quality meteorological corpora, optimizing benchmark testing frameworks, and incorporating external tools. While LLM brings new technological opportunities to the meteorological field, their widespread application still requires continuous exploration and improvement in corpus quality, model optimization, and human-machine collaboration. Moreover, LLM still has limitations in understanding the spatio-temporal dynamics of atmospheric motions and issues with bias and hallucination, which need to be addressed through data cleaning, debiasing, fine-tuning, and retrieval-augmented generation techniques.