A Critical Review of Methodological Advances, Data Fusion Strategies, and Emerging Challenges

Abstract

The confluence of big data from Earth observation satellites, ground-based sensor networks, and climate models has created unprecedented opportunities for artificial intelligence to enhance climate change prediction. Yet the rapid proliferation of methodological approaches has outpaced systematic synthesis of best practices, data fusion strategies, and persistent limitations. This paper presents a critical systematic review of 142 peer-reviewed studies published between 2019 and 2025 that employ AI techniques for climate prediction using multi-source geospatial data. We synthesize advances across three analytical domains: temperature and extreme event forecasting, precipitation estimation, and long-term climate trend modeling. The review identifies a decisive shift from conventional machine learning methods toward hybrid architectures that combine deep learning with physical models, with convolutional neural networks and long short-term memory networks demonstrating particular efficacy for spatial and temporal pattern recognition respectively. However, we find that data heterogeneity remains the paramount methodological challenge, with inconsistent preprocessing, scale mismatches, and quality control procedures impeding model generalizability. Furthermore, a critical gap emerges: only 12% of reviewed studies explicitly address model interpretability or uncertainty quantification, limiting operational deployment for high-stakes decision-making. We propose a standardized framework for AI-driven climate prediction that prioritizes reproducible data fusion pipelines, benchmark datasets for model intercomparison, and integration of explainable AI techniques. The paper concludes that while AI has substantially advanced predictive capabilities, realizing its full potential requires shifting from performance maximization to systematic handling of data uncertainty, physical consistency, and actionable interpretability for climate resilience planning.