Introducing AQUAH: Automatic Quantification and Unified Agent in Hydrology
We are excited to present AQUAH, the first end-to-end language-based agent purpose-built for hydrologic modeling. AQUAH allows users to start with a simple, natural-language prompt, such as “simulate floods for the Little Bighorn basin from 2020 to 2022”, and from there, it autonomously handles the entire modeling workflow—from retrieving necessary terrain, weather, and gauge data, to configuring a hydrologic model, running simulations, and generating an analyst-ready PDF report.
What sets AQUAH apart is its integration of vision-enabled large language models. The system interprets maps and rasters on the fly, makes critical modeling decisions (like outlet selection and parameter initialization), and offers commentary about simulation uncertainty. In initial trials across a variety of U.S. watersheds, AQUAH has demonstrated its ability to deliver “cold-start” hydrologic simulations and produce transparent, physically plausible documentation without manual intervention.
Early feedback from hydrologists highlights AQUAH’s ability to lower the barrier between Earth observation data, physics-based tools, and practical decision-making. While further calibration and validation are needed for operational adoption, AQUAH illustrates the promise of LLM-centered, vision-grounded agents to streamline and transform complex environmental modeling tasks.
Read the full paper here: arXiv:2508.02936
—