FloatChat - An Explainable Multimodal AI Platform for Predictive Oceanographic Data Intelligence with QC-Aware Natural Language Querying

Abstract

The Argo float program generates one of the most comprehensive global ocean observation datasets, yet its accessi-bility remains severely limited due to the technical complexity of data formats, quality-control procedures, and the programming expertise required for meaningful analysis. While existing conver-sational interfaces for oceanographic data provide basic natural language querying capabilities, they critically lack explainability, predictive intelligence, proactive anomaly detection, and rigor-ous quality-control awareness — significantly restricting their scientific utility. This paper presents FloatChat AI, an enhanced multimodal AI platform that addresses these limitations through six novel contributions: a QC-Guard Layer that automatically injects Argo-compliant quality-control constraints into generated SQL queries, an Explainable AI response module leveraging SHAP-based reasoning traces for non-expert interpretability, a Vision-Language Model integration enabling image-to-query multimodal input, a real-time anomaly detection microservice employing LSTM-Autoencoders for proactive ocean event alert-ing, a Temporal Fusion Transformer-based predictive forecasting engine for short-term oceanographic trend prediction, and a collaborative multi-user workspace supporting shared annotation and synchronized visualization. The system is evaluated through quantitative ML benchmarks including RMSE-based forecast accuracy, anomaly detection precision-recall metrics, and QC-injection error reduction rates, supplemented by a user compre-hension study comparing expert and non-expert interpretability outcomes. Experimental results demonstrate that FloatChat AI significantly advances the state of accessible, reliable, and scien-tifically rigorous oceanographic data intelligence.

Introduction

Although Argo data is highly valuable for climate research, it is difficult to use because it is stored in complex formats and requires advanced programming skills and domain expertise. Existing tools and conversational AI systems are limited by poor query accuracy, lack of quality-control awareness, no explainability, and absence of predictive or collaborative capabilities.

To address these issues, FloatChat AI integrates several advanced technologies into a unified system:

• QC-Guard layer to ensure data quality control during queries
• Explainable AI (SHAP) for transparent reasoning of results
• Vision-Language Models (VLMs) for multimodal input (text + images)
• LSTM-Autoencoder for real-time anomaly detection in ocean data
• Temporal Fusion Transformer (TFT) for forecasting oceanographic trends
• Collaborative workspace for multi-user scientific analysis

The literature review highlights progress in NLP, retrieval-augmented generation, explainable AI, anomaly detection, forecasting models, and multimodal AI, but shows that no existing system combines all these capabilities for oceanographic data exploration.

Conclusion

This paper presented FloatChat AI, a next-generation mul-timodal predictive intelligence platform for conversational exploration and analysis of Argo oceanographic datasets. Building upon the foundational conversational AI paradigm for ocean data access, FloatChat AI introduces six novel and independently validated contributions: a QC-Guard Layer achieving 100% Argo quality-control compliance in SQL query execution; a SHAP-based Explainable AI module im-proving non-expert comprehension by 82.6% relative to a no-explanation baseline; a Vision-Language Model integra-tion enabling multimodal image-to-query interaction with 91.4% geographic coordinate extraction accuracy; an LSTM-Autoencoder anomaly detection service achieving an F1-score

References

[1] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” Advances in Neural Information Processing Systems (NeurIPS), 2020. Available: https://arxiv.org/abs/2005.11401 [2] H. Touvron et al., “Llama: Open and Efficient Foundation Language Models,” Meta AI, 2023. Available: https://arxiv.org/abs/2302.13971 [3] A. Q. Jiang et al., “Mistral 7B,” Mistral AI, 2023. Available: https://arxiv.org/abs/2310.06825 [4] S. M. Lundberg and S. Lee, “A Unified Approach to Interpreting Model Predictions,” Advances in Neural Information Processing Systems (NeurIPS), 2017. Available: https://arxiv.org/abs/1705.07874 [5] B. Lim et al., “Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting,” International Journal of Forecasting, vol. 37, no. 4, pp. 1748–1764, 2021. Available: https://arxiv.org/abs/1912.09363 [6] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997. [7] H. Liu et al., “LLaVA: Visual Instruction Tuning,” Advances in Neural Information Processing Systems (NeurIPS), 2023. Available: https:// arxiv.org/abs/2304.08485 [8] “Argo Data Management,” International Argo Program / Argo Steering Team. Available: https://argo.ucsd.edu/data/ [9] “PostgreSQL: The World’s Most Advanced Open Source Relational Database,” The PostgreSQL Global Development Group. Available: https://www.postgresql.org/ [10] “PostGIS — Spatial and Geographic objects for PostgreSQL,” PostGIS Project. Available: https://postgis.net/ [11] “Ollama — Get up and running with large language models locally,” Ollama. Available: https://ollama.com [12] F. Liu et al., “A Survey on Large Language Models for Code Genera-tion,” arXiv, 2024. Available: https://arxiv.org/abs/2406.00515 [13] “Copernicus Marine Environment Monitoring Service (CMEMS),” European Union’s Copernicus Programme. Available: https://marine. copernicus.eu/

Copyright

Copyright © 2026 Harish J, Yeshwanthy S, Harish Kumar SV, Mrs. S. Chandrakala . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.