190-9 Evaluating AI-Assisted Vibe Coding in Hydrologic Sciences

Abstract:

Advances in computing and the availability of large language models (LLMs) have created new opportunities for hydrologists to complete routine coding and data-analysis tasks more efficiently. However, many researchers still encounter challenges when writing code and understanding unfamiliar scripts. This research examines the potential of LLM-assisted vibe coding to support hydrologists in developing solutions, interpreting unfamiliar code, and more effectively utilizing computational tools. Participants were assigned to one of four leading LLM platforms, Copilot, ChatGPT, Claude, or Gemini, and tasked with completing a series of hydrologic coding exercises. Throughout the experiment, participants documented their prompts, exported generated code and outputs, and recorded observations regarding task completion and required user intervention. Results indicated that vibe coding can be an effective aid for computational tasks, although performance varies across platforms and is strongly influenced by prompt specificity and user interaction. This methodology showed great success in certain tasks such as hydrograph statistics, cross-correlation analyses, precipitation interpolation, and data pulls. However, more applied tasks such as developing unit hydrographs require more intervention and input from the user. As LLM capabilities continue to advance, vibe coding has the potential to become an increasingly valuable tool for updating legacy code, accelerating learning, and streamlining hydrologic research and workflows.

Geological Society of America Abstracts with Programs. Vol. 58, No. 5, 2026

doi: 10.1130/abs/2026AM-20333

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.