This study explores the application of AI-driven scientific visualization in climate change communication and its impact on public perception. By developing a deep learning-based visualization system (ClimateVis-AI) that transforms complex climate datasets into interactive, context-aware visuals, we conducted a mixed-methods experiment with 1,200 participants across the U.S. Results show that ClimateVis-AI significantly improves public understanding of climate trends (p<0.01) and enhances pro-environmental behavioral intentions (β=0.38, p<0.001) compared to traditional static visuals. Ethical considerations regarding data accuracy and algorithmic bias in AI-generated visuals are also discussed.