Fig 1.
Gender and academic level distribution of survey participants (N = 348).
(A) Pie chart showing gender composition with 45.1% female, 50.6% male, and 4.3% preferring not to disclose. (B) Distribution across academic levels, with master#39;s students comprising 81.9% of the sample.
Fig 2.
Large language model usage frequency and tool preferences.
(A) Usage frequency categories showing 49.1% daily users and 33.6% frequent users. (B) Platform preferences with DeepSeek (88.5%) and ChatGPT (73.9%) as most popular tools.
Fig 3.
Application scenarios and performance evaluation of LLMs in meteorological research.
(A) Horizontal bar chart of usage scenarios, with code writing as primary application (79.9%). (B) Radar chart of mean performance ratings (1-5 scale) across professional accuracy, efficiency, practicality, innovation, and ease of use. (C) Distribution of participant ratings across the five performance dimensions. Bars represent the percentage of responses on a 1–5 Likert scale.
Fig 4.
Challenges and research efficiency in LLMimplementation.
(A) Frequency of reported problems, highlighting issues with complex problem handling and answer generality. (B) Research efficiency assessment, showing that 30.7% of participants reported significant improvement in productivity.
Fig 5.
Improvement priorities and responsible usage practices of LLMsamong graduate students.
(A) Priority ranking of improvements, with professional knowledge accuracy (77.0%), inclusion of latest research developments (48.6%), and code generation quality (37.9%) identified as top needs.(B) Responsible usage practices, showing that most participants (88.2%) reported verifying output accuracy.
Fig 6.
Institutional perspectives and future outlook on LLM integration in academic research.
(A) Supervisor and institutional attitudes toward LLM usage, with the majority (63.2%) adopting a cautious but permissive stance. (B) Expert predictions on LLM's potential to replace human scientific writing within five years, with 40.5% believing replacement is unlikely in the short term but possible long-term.