The State of Hydroinformatics Prior to Generative AI: Establishing a Quantitative Baseline (2018–2023)

Carlos Erazo Ramirez , Kevin Song, Ibrahim Demir

The rapid advancement of web applications and machine learning has fundamentally reshaped hydrogeological sciences. While the recent emergence of Large Language Models (LLMs) dominates current discourse, understanding the trajectory of this shift requires quantifying the foundational digitization trends that preceded it. This study presents an automated web-mining framework designed to extract and classify technological keywords from the full text of scientific literature, overcoming the limitations of abstract-only bibliometrics. We applied this framework to 3,701 manuscripts published between 2018 and mid-2023 to establish a quantitative baseline for the hydrological digital landscape immediately prior to the Generative AI disruption. Utilizing the Elsevier Text Mining API and Latent Dirichlet Allocation, we identified a mature integration of predictive machine learning and web technologies. Our findings characterize the infrastructure phase of hydroinformatics, highlighting the critical consolidation of data-driven modeling that now serves as the backbone for emerging agentic AI workflows. Moreover, we aim to establish a reproducible technological baseline against which post-2023 LLM-driven transformations in hydrology can be quantitatively evaluated.