Author(s): Janell Prater; Meagan Tobias; Branko Kerkez

Linked Author(s):

Keywords: No Keywords

Abstract: Continuous water data via distributed sensor networks is critical for effective water management, but maintaining these networks is costly, especially due to power demands. Data transmission is one of the largest energy consumers, making reporting frequency a major determinant of sensor lifespan. High-frequency reporting is important during storm events, while lower-frequency reporting suffices under baseflow conditions. Adaptive reporting could balance these needs but requires awareness of current hydrologic conditions and near-term weather forecasts—possible to do manually, but not scalable. Artificial intelligence (AI) agents combine large language models (LLMs) with weather and water level data through the model context protocol (MCP), offering a way to automate these decisions. We tested an AI agent on ten sensors across a four-day period containing storm events, using three prompts of varying domain specificity. Agent accuracy improved from 60% with generic instructions to 70% with water-systems context, demonstrating that domain-informed prompts are essential for reliable automated sensor management.

DOI:

Year: 2026