AI-driven framework for high-precision seamless sunshine duration estimation using Himawari-8 satellite and ground observations

Abstract

Accurate sunshine duration (SD) estimation is vital for large-scale, dynamic applications, such as agricultural planning, urban development, and solar energy optimization. Ground-based observations, with their sparse stations and limited coverage, often fall short of meeting these requirements. In contrast, satellite-based methods, particularly those utilizing high spatial resolution data, have become increasingly important in addressing these limitations. To address the need for enhanced precision and coverage, this study developed an AI-driven framework using high-resolution Himawari-8 L1 Gridded data for the first time. By integrating multiple machine learning models with a stacked generalization algorithm, the framework enables seamless and highly accurate SD estimation. The proposed method demonstrates significant potential for wide-spread application. Tests conducted in Jiangxi Province China have confirmed its reliability. The model achieved a high R² of 0.969, with RMSE and MAE of 0.752 and 0.519 h, respectively. We further analyzed the distribution of SD across different land-use types. The results indicate that croplands and forests are predominantly located in areas with longer SD, reflecting the adaptive choices made by human activities and ecosystems in response to sunlight conditions. This finding suggests that SD can serve as a critical reference for the planning of croplands and forestlands. Moreover, it provides robust data support and scientific guidance for future regional land-use optimization, sustainable agricultural development, and renewable energy planning.