R2RAG-Flood: A reasoning-reinforced training-free retrieval augmentation generation framework for flood damage nowcasting Publication type: ORIGINAL ARTICLE
Lipai Huang, Kai Yin*, Chia-Fu Liu, Ali Mostafavi · 2026
Read R2RAG-Flood: A reasoning-reinforced training-free retrieval augmentation generation framework for flood damage nowcasting Publication type: ORIGINAL ARTICLE by Lipai Huang, Kai Yin*, Chia-Fu Liu, Ali Mostafavi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-10312.txt.
Description
Flood damage nowcasting needs fast, clear predictions, but fine-tuning is costly. A 2026 paper led by Lipai Huang, with three co-authors, offers R2RAG-Flood, a training-free retrieval setup for flood damage nowcasting with reinforced thought. A thought-first knowledge base is built from labeled tabular records holding structured predictors, a compact text summary, and a model-made thought path.
During inference, the target prompt gains nearby neighbors and free-shots supporting case-based thought without fine-tuning. A two-stage routine first settles damage occurrence, then grades severity on a three-level damage-extent scale, with a careful downgrade check for shaky over-severe outputs. In a Harris County, Texas, test on Hurricane Harvey, the setup reaches about sixty-one to sixty-seven percent overall accuracy across seven backbones, with reasons for each prediction. Lighter versions beat the supervised baseline on severity per cost.
Catalog record
| CREATOR | Lipai Huang, Kai Yin*, Chia-Fu Liu, Ali Mostafavi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-10312