Reproduce “EWE: An Agentic Framework for Extreme Weather Analysis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhe Jiang, Jiong Wang, Xiaoyu Yue, Zijie Guo, Wenlong Zhang, Fenghua Ling, Wanli Ouyang, Lei Bai
EWE is an agent for diagnosing past extreme-weather events by planning an investigation, retrieving and analyzing weather data, producing maps, and revising its work after checks. It combines data access, computation, visualization, and interpretation in a repeating loop. In limited-sample, checklist-based LLM evaluations, its configuration scored 0.787, 0.680, and 0.679 across analysis scales; these scores do not establish causal truth.
Key insights
- Weather diagnosis is treated as a sequence of evidence-grounded tasks, rather than a single language-model response.
- Domain tools and auditing contribute meaningfully: removing tools particularly lowers thermodynamic-analysis scores, while removing auditors also reduces performance.
- Model strengths differ by analytical stage, and comparative grading reveals larger differences between systems in some categories than single-response grading.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.