Naiad: Novel Agentic Intelligent Autonomous System for Inland Water Monitoring

TL;DR

NAIAD addresses the need for integrated inland water monitoring by combining LLM-driven reasoning, retrieval-augmented generation, and on-the-fly DAG-based workflow orchestration to unify EO data, in-situ sensors, and meteorological context. It translates natural-language queries into actionable reports by dynamically invoking tools for Sentinel-2 data retrieval, NDCI computation, chlorophyll estimation, and CyFi forecasting within a self-hosted, explainable framework. Evaluations on three Greek lakes show high correctness (~82.98%) and substantial relevancy (up to ~78.72% with Qwen2.5), with ablations highlighting open models like Qwen2.5 as favorable for actionable outputs. The approach offers a deployable, domain-tailored solution for environmental decision support with clear paths to tool expansion, geographic scaling, and potential multi-agent collaboration.

Abstract

Inland water monitoring is vital for safeguarding public health and ecosystems, enabling timely interventions to mitigate risks. Existing methods often address isolated sub-problems such as cyanobacteria, chlorophyll, or other quality indicators separately. NAIAD introduces an agentic AI assistant that leverages Large Language Models (LLMs) and external analytical tools to deliver a holistic solution for inland water monitoring using Earth Observation (EO) data. Designed for both experts and non-experts, NAIAD provides a single-prompt interface that translates natural-language queries into actionable insights. Through Retrieval-Augmented Generation (RAG), LLM reasoning, external tool orchestration, computational graph execution, and agentic reflection, it retrieves and synthesizes knowledge from curated sources to produce tailored reports. The system integrates diverse tools for weather data, Sentinel-2 imagery, remote-sensing index computation (e.g., NDCI), chlorophyll-a estimation, and established platforms such as CyFi. Performance is evaluated using correctness and relevancy metrics, achieving over 77% and 85% respectively on a dedicated benchmark covering multiple user-expertise levels. Preliminary results show strong adaptability and robustness across query types. An ablation study on LLM backbones further highlights Gemma 3 (27B) and Qwen 2.5 (14B) as offering the best balance between computational efficiency and reasoning performance.

Figures (4)

• Figure 1: Graphical abstract of the NAIAD agentic AI system for inland water monitoring. User queries are processed by a LLM, which collaborates with an autonomous agent to dynamically retrieve information from a vector database and orchestrate external analytical tools. The workflow enables the generation of comprehensive, user-adapted water quality assessments via a simple prompt interface.
• Figure 2: Overview of the NAIAD system architecture and workflow. After the user submits a query, NAIAD interprets this query using RAG to retrieve relevant domain knowledge and add context. The LLM then determines whether external tools are needed, selecting and orchestrating them by dynamically constructing a Directed Acyclic Graph (DAG) that encodes the analytical workflow. Throughout the process, reflection mechanisms ensure output relevance and accuracy. The final report synthesizes all findings and is tailored to the user’s expertise.
• Figure 3: ROIs of our study with NDCI calculated overlay, (a) Lake Lysimachia, (b) Lake Trichonida and (c) artificial Lake Mornos.
• Figure 4: Example user queries and corresponding LLM-generated responses from NAIAD, across three Greek lakes (Mornos, Trichonida, Lysimachia). The queries address topics such as chlorophyll-a trends, precipitation and temperature effects, cyanobacteria levels, and weather influences. Responses are generated by models Qwen-2.5 and Gemma-3, showcasing NAIAD's ability to retrieve, reason, and summarize scientific insights based on structured document sources and environmental data.