Article Dans Une Revue Atmosphere Année : 2025

Deep Learning and Spatial Context for Global Horizontal Irradiance Estimation: Addressing Independent Pixel Approximation Limitations with Satellite Imagery

Résumé

Accurate estimation of Global Horizontal Irradiance (GHI) is essential for solar energy applications, climate modeling, and various geophysical processes.Traditional satellite-based methods rely on the Independent Pixel Approximation (IPA), which treats each pixel as radiatively isolated from its neighbors, neglecting 3D cloud effects and horizontal photon transport. These limitations could be amplified by the higher spatial, temporal, and spectral resolutions of third-generation geostationary satellites. In this study, we evaluate deep learning models that explicitly incorporate spatial context from GOES-16 multispectral satellite imagery to improve satellite-based GHI estimation and address IPA limitations.We compare two architectures—a Fully Connected Network (FCN) and a convolutional-based model—against a state-of-the-art physical retrieval method (PSM3), using in-situ GHI measurements from 31 U.S. stations.Our results show that deep learning models leveraging spatial context outperform PSM3 across most metrics, especially under cloudy and partially clear conditions, yielding improved performance, stability, and reduced bias. The best-performing model achieves a 26.5% lower RMSE and a 21% lower MAE compared to PSM3 on a year-long test set. However, deep learning models still struggle to consistently outperform PSM3 in some scenarios in terms of bias, particularly under clear-sky conditions or on some specific test stations.Qualitative analysis highlights specific weakness modes of PSM3, particularly when it misclassifies cloudy scenes as clear-sky, where deep learning models correctly capture cloud-induced variability.We discuss the implications of these findings and potential directions for model improvements. This work underscores the potential of spatial-context-aware deep learning models to overcome IPA limitations for the next generation of satellite-based GHI retrieval methods, and improve GHI retrieval in heterogeneous atmospheric conditions.

Dates et versions

hal-05217874 , version 1 (21-08-2025)

Identifiants

Citer

Vadim Becquet, Philippe Blanc, Yves-Marie Saint-Drenan, Yehia Eissa. Deep Learning and Spatial Context for Global Horizontal Irradiance Estimation: Addressing Independent Pixel Approximation Limitations with Satellite Imagery. Atmosphere, 2025, 16 (8), pp.949. ⟨10.5194/ems2025-294⟩. ⟨hal-05217874⟩
224 Consultations
0 Téléchargements

Altmetric

Partager

  • More