The field of meteorological satellite data processing is advancing rapidly, propelled by substantial developments in remote sensing technologies and the enhanced capabilities of modern satellites. The Fengyun satellite series, initiated by China in 1977, exemplifies this progress. Four generations of Fengyun satellites are operational, comprising two polar-orbiting satellites (Fengyun-1 and Fengyun-3) and two geostationary satellites (Fengyun-2 and Fengyun-4). These satellites demonstrate substantial technological advancements and offer comprehensive observational capabilities through sophisticated satellite networking.

Fengyun satellites have various optical remote sensing instruments that capture data across multiple spectral bands, ranging from ultraviolet to infrared. Instruments like the moderate resolution spectral imager-II on Fengyun-3D provide enhanced infrared detection capabilities with multiple channels, facilitating detailed surface cover classification, landform feature identification, and observing atmospheric, surface, and ocean characteristics. Consequently, these satellites deliver invaluable data for weather prediction, climate research, vegetation monitoring, land use classification, and atmospheric studies.

However, the exponential growth in data volume presents substantial challenges to traditional data processing methods. Increased number of satellite, enhanced sensor capabilities, and improved temporal and spatial resolution drive this data explosion. From the launch of Fengyun-1A in 1988 to Fengyun-3F in 2023, the series has generated a vast amount of historical and real-time data, necessitating the development of efficient and accurate analysis methods.

Artificial intelligence (AI) methods have become increasingly prominent in addressing the challenges of processing large-scale satellite datasets. Traditional data processing techniques typically involve complex workflows and rely heavily on expert knowledge, making them unsuitable for managing the vast amounts of data modern satellites generate. In contrast, AI methods utilize sophisticated algorithms and computational models for efficient and precise data analysis. Among the AI technologies, machine learning and deep learning techniques have shown immense potential in various satellite data processing tasks.

AI technology has demonstrated remarkable advantages in intelligent self-calibration, particularly in radiometric correction. Conventional radiometric correction methods often require intricate models and manual intervention. However, deep learning-based intelligent self-calibration methods can automatically learn the radiometric discrepancies between sensors and platforms. By leveraging extensive training data, these models can identify and correct radiometric biases in satellite sensors, resulting in consistent and reliable remote sensing data, as evidenced by the results shown in Table 1. This enhancement improves data quality and reduces dependency on manual operations, providing a solid foundation for subsequent remote sensing applications.

Traditional methods for cloud detection often rely on spectral features and threshold techniques, which frequently show limitations under complex cloud structures and surface conditions. Deep learning models, particularly those specifically trained to distinguish between cloud and non-cloud regions, as illustrated in Fig. 5, offer a precise interpretation of satellite imagery, substantially enhancing the cloud detection accuracy. This advancement is crucial for weather prediction, climate change research, and other cloud-related applications.

For cloud motion extrapolation, AI methods leverage recurrent neural networks and long short-term memory networks to predict future cloud movements based on historical data. Generative adversarial networks have also demonstrated strong performance in cloud motion studies, as shown in Fig. 6. Compared with traditional approaches, deep learning models more effectively capture the spatiotemporal patterns of cloud motion, improving the accuracy of cloud image predictions and offering reliable support for short-term weather predictions and severe convective weather warnings.

In precipitation inversion, the integration of physical and data-driven models, has driven substantial advancements in the field. Convolutional neural networks and vision transformer (ViT) excel at enhancing inversion accuracy, as shown in Fig. 8. They adeptly handle complex precipitation patterns and provide crucial data support for meteorological research and environmental monitoring. This integration improves the precision of precipitation distribution predictions.

AI models also show excellent potential in sea ice detection. By integrating multi-source data, deep learning models enhance the accuracy and reliability of sea ice detection, as illustrated in Table 4. These models can identify the presence of sea ice and estimate its thickness and coverage area, providing critical data support for climate research and marine environmental monitoring.

The advantages of AI methods include end-to-end processing, reduced reliance on expert knowledge, and enhanced generalization capabilities. Using vast historical datasets and advanced computational power, AI models autonomously learn latent patterns within the data, enabling accurate predictions and analyses.

Integrating AI technologies into satellite big data mining is ushering in a new era of efficient and accurate data processing. As AI methods continue to evolve, they will play an increasingly crucial role in satellite applications, enhancing the extraction of meaningful insights from the vast datasets. The future of satellite data processing lies in developing real-time, globally shared systems that fully leverage AI’s potential.

Despite these advancements, various challenges remain in the widespread adoption of AI in satellite remote sensing. Model interpretability, data quality, and computational demands must be addressed to ensure reliable and practical application of AI. Additionally, interdisciplinary collaboration among remote sensing experts, computer scientists, and domain specialists is essential for developing robust AI models tailored to specific satellite applications.

As AI technologies advance, they promise to revolutionize satellite data processing and enable more accurate and timely insights into our planet’s complex systems and phenomena.