Surface solar radiation: AI satellite retrieval can outperform Heliosat and generalizes well to other climate zones

TL;DR

This study develops a first-of-its-kind machine-learning retrieval of instantaneous surface solar irradiance (SSI) from geostationary Meteosat SEVIRI data. The authors train a convolutional neural network to emulate the radiative-transfer-based Heliosat-SARAH-3 SSI and then fine-tune the model on diverse ground-station SSI measurements to achieve strong out-of-domain generalization across Europe and North Africa, particularly under cloudy conditions. Their results show that ground-station fine-tuning significantly reduces biases and RMSE relative to SARAH-3, with notable improvements in Alpine regions and in cloudy skies (CSI < 0.8). The work demonstrates the value of combining RTM emulation with data-driven refinement to produce fast, accurate, and broadly generalizable SSI retrievals suitable for solar resource assessments and near-term forecasting.

Abstract

Accurate estimates of surface solar irradiance (SSI) are essential for solar resource assessments and solar energy forecasts in grid integration and building control applications. SSI estimates for spatially extended regions can be retrieved from geostationary satellites such as Meteosat. Traditional SSI satellite retrievals like Heliosat rely on physical radiative transfer modelling. We introduce the first machine-learning-based satellite retrieval for instantaneous SSI and demonstrate its capability to provide accurate and generalizable SSI estimates across Europe. Our deep learning retrieval provides near real-time SSI estimates based on data-driven emulation of Heliosat and fine-tuning on pyranometer networks. By including SSI from ground stations, our SSI retrieval model can outperform Heliosat accuracy and generalize well to regions with other climates and surface albedos in cloudy conditions (clear-sky index < 0.8). We also show that the SSI retrieved from Heliosat exhibits large biases in mountain regions, and that training and fine-tuning our retrieval models on SSI data from ground stations strongly reduces these biases, outperforming Heliosat. Furthermore, we quantify the relative importance of the Meteosat channels and other predictor variables like solar zenith angle for the accuracy of our deep learning SSI retrieval model in different cloud conditions. We find that in cloudy conditions multiple near-infrared and infrared channels enhance the performance. Our results can facilitate the development of more accurate satellite retrieval models of surface solar irradiance.

Figures (13)

• Figure 1: Ground station measurements collected for this study. Bold-named stations are part of the BSRN network. Gray-bordered stations indicate sites within the DWD and MeteoSwiss training datasets. The domain shown corresponds to the domain of this study, 29°N–62°N and 9°W–28°E.
• Figure 2: Model comparison for the instantaneous SSI retrieval on 2022-08-20 at 15:30 UTC.
• Figure 3: Scatter plot of the instantaneous SSI ('SSI retrieval') compared to the SSI of the Heliosat SARAH-3 validation set ('SSI'). The dashed lines indicate the 45° line (green) and a linear regression fit (blue), respectively. The plot comprises 19.3 million SSI retrievals across Europe and North Africa from 2022.
• Figure 4: Instantaneous mean bias deviation (a) and instantaneous root-mean-square deviation (b) of instantaneous SSI estimates of the emulator model with regard to SARAH-3 SSI averaged pixel-wise during daytime periods of 2023 on the test set. Samples were binned in $0.25^\circ$x$0.25^\circ$ pixels to estimate regional MBDs and RMSDs.
• Figure 5: The instantaneous Mean Bias Error (MBE) on all observation ground station sets, averaged over all observations with a SZA $< 85^\circ$.