Scholarlink

<p>Weather profoundly influences agriculture, transportation, energy, health, and industrial operations. Traditional meteorological forecasting, though well established, faces persistent limitations in data coverage, computational demand, and human subjectivity. The rapid expansion of ground-based sensors, satellite and radar observations, and multimodal climate datasets together with advances in artificial intelligence (AI), machine learning (ML), deep learning (DL) now enables more precise, scalable weather detection, classification, and forecasting. This review consolidates recent progress and introduces a three-dimensional taxonomy that organizes AI-based meteorological approaches by data sources (ground, satellite, radar, microwave, hyperspectral, SAR, reanalysis), tasks (detection, classification, nowcasting, forecasting, extreme-event prediction), and methodologies (classical ML, CNNs, hybrid CNN-LSTMs, transformers, and graph neural networks). A comprehensive synthesis compares state-of-the-art models in terms of accuracy, computational efficiency, and deployment readiness, supported by an expanded dataset summary covering major benchmarks such as MODIS, CloudSEN12, WeatherBench, BharatBench, and AMOS. The review further evaluates trade-offs between model complexity, accuracy, and energy consumption to guide practical deployment. Despite substantial progress, open challenges remain most notably data imbalance, non-standard evaluation protocols, real-time limitations in IoT and edge environments, and limited explainability and cross-regional generalization. Promising research directions include multi-label weather classification, Explainable AI (XAI) for interpretability, lightweight architectures for resource-constrained platforms, and federated learning for cross-domain adaptation. By bridging meteorological science and intelligent computation, this review provides a structured foundation for developing next-generation, AI-driven weather systems that are accurate, interpretable, and globally scalable.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-Driven Meteorology for Weather Detection, Classification, Nowcasting and Forecasting (1991–2025): A Comprehensive Review of Deep Learning and Machine Learning Approaches

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Yogesh H. Bhosale

摘要

Weather profoundly influences agriculture, transportation, energy, health, and industrial operations. Traditional meteorological forecasting, though well established, faces persistent limitations in data coverage, computational demand, and human subjectivity. The rapid expansion of ground-based sensors, satellite and radar observations, and multimodal climate datasets together with advances in artificial intelligence (AI), machine learning (ML), deep learning (DL) now enables more precise, scalable weather detection, classification, and forecasting. This review consolidates recent progress and introduces a three-dimensional taxonomy that organizes AI-based meteorological approaches by data sources (ground, satellite, radar, microwave, hyperspectral, SAR, reanalysis), tasks (detection, classification, nowcasting, forecasting, extreme-event prediction), and methodologies (classical ML, CNNs, hybrid CNN-LSTMs, transformers, and graph neural networks). A comprehensive synthesis compares state-of-the-art models in terms of accuracy, computational efficiency, and deployment readiness, supported by an expanded dataset summary covering major benchmarks such as MODIS, CloudSEN12, WeatherBench, BharatBench, and AMOS. The review further evaluates trade-offs between model complexity, accuracy, and energy consumption to guide practical deployment. Despite substantial progress, open challenges remain most notably data imbalance, non-standard evaluation protocols, real-time limitations in IoT and edge environments, and limited explainability and cross-regional generalization. Promising research directions include multi-label weather classification, Explainable AI (XAI) for interpretability, lightweight architectures for resource-constrained platforms, and federated learning for cross-domain adaptation. By bridging meteorological science and intelligent computation, this review provides a structured foundation for developing next-generation, AI-driven weather systems that are accurate, interpretable, and globally scalable.

Graphical Abstract