ID: 2151
Presenting Author: Saeed Givehchi
Session: 663 - Climate risks assessment in the age of climate misinformation
Status: pending
This study develops an AI-based framework that detects misinformation in climate risk data using NLP, anomaly detection, and graph analytics, improving data credibility and supporting reliable climate
Misinformation within climate risk data can seriously distort assessments and weaken adaptation strategies. This study addresses the problem by introducing an artificial intelligence (AI) framework designed to enhance the credibility of climate-related datasets. The framework integrates multiple machine learning methods to detect and reduce unreliable or misleading information before it influences analysis outcomes.
Natural Language Processing (NLP) models based on BERT are applied to evaluate the credibility of textual materials such as research reports and social media content, identifying inconsistencies or unsupported claims. For numerical datasets, Isolation Forest algorithms are employed to flag outliers and possible data manipulation. Additionally, graph-based analysis helps reveal how inaccurate information circulates across data-sharing networks.
A case study on coastal flood risk demonstrated that the system detected more than 80% of known misinformation cases, significantly improving the reliability and transparency of the resulting risk model. Beyond its technical contribution, the approach reduced manual data checking and allowed experts to focus on interpretation and stakeholder communication. The findings highlight the potential of explainable AI tools to complement human expertise, creating a stronger evidence base for climate resilience planning and supporting more informed and trustworthy climate actions.
Associate Professor, Faculty of Environment, University of Tehran, Tehran, Iran