Solar Radiation Analysis for Predicting Climate Change Using Deep Learning Techniques

Introduction

The largest source of energy on Earth is solar radiation, and the quantity of solar radiation reaching the Earth's surface is influenced by the atmosphere, hydrosphere, and biosphere (Budyko, 1969; Islam et al., 2009). Solar radiation also plays an important influence in global climate, and even little variations in the Sun's energy output induce significant changes in the Earth's temperature (Beer et al., 2010; Siingh et al., 2011). The Sun's irradiance has the largest impact on Earth's high atmosphere, while the lower atmosphere protects the planet from the increasing heat. If the Sun is causing Earth's warming, one would anticipate the upper atmosphere to get more heated. Variations in solar radiation influence global temperatures, global mean sea level, and the occurrence of severe weather events (Bhargawa & Singh, 2019). These variables are accountable for climate change. Accurate measurements and assessments of solar radiation's temporal and geographical variability are therefore critical in research on solar energy, construction materials, and severe weather and climate events (Cline et al., 1998; Garland et al., 1990; Grant & Tuohimaa, 2004; Hoogenboom, 2000). The warming caused by growing quantities of man-made greenhouse gases is several times higher than any impacts caused by recent fluctuations in solar activity. Many approaches for predicting solar radiation have been developed, including theoretical parameter models, empirical models, artificial intelligence algorithms, and satellite retrieval data (Halabi et al., 2018; Li et al., 2008; Lu et al., 2011; Makade et al., 2019; Mellit, 2008; Wild, 2009). The A-P model, initially suggested by Angstrom (1924) and Prescott (1940), is commonly used to estimate solar radiation. The BCM model was developed by Bristow and Campbell (1984) by examining the connection between solar radiation and daily maximum and minimum temperatures. Yang et al. (2001) created a hybrid model (YHM) to improve the A-P model by investigating the influence of meteorological factors before evaluating the model's accuracy in Japan. Salazar (2011) contrasted the YHM with a climatological solar radiation model to determine the horizontal direct and diffuse components of solar radiation, resulting in the CYHM (corrected YHM). Gueymard (2003) investigated sun irradiance forecasts using 19 solar radiation models, indicating that detailed transmittance models outperform bulk models. Many academics have been motivated by the advancement of machine learning algorithms to create solar radiation prediction models (Azadeh et al., 2009; Chen et al., 2011; Jiang, 2009; Voyant et al., 2012). Fadare (2009) and Linares-Rodrguez et al. (2011) built solar radiation prediction models using artificial neural network (ANN) technology and tested their predictive capacity. Xue (2017) developed a solar radiation prediction model using a back-propagation algorithm and demonstrated that the forecast accuracy was dependent on the combination and configuration of the input parameters. Chen et al. (2011) built a solar radiation prediction model utilizing the support vector machine (SVM) approach and demonstrated that the SVM-based algorithm had varying predicted accuracy while employing various kernel functions. Both Olatomiwa et al. (2015) and Shamshirband et al. (2016) optimized the SVM algorithm and produced accurate predictions. Tree techniques, such as the random forest approach and the gradient boosting regression tree (GBRT), have been utilized to build solar radiation prediction models with promising results (Fan et al., 2018; Persson et al., 2017; Sun et al., 2016; Zeng et al., 2020). Some researchers have conducted comparative analyses of a range of machine learning algorithms in recent years (Meenal & Selvakumar, 2018; Pang et al., 2020; Shamshirband et al., 2016), and all of these studies reveal that the ANN algorithm does not provide excellent prediction outcomes but does suggest a path for algorithm development. Deep learning algorithms have been used in several research to estimate solar radiation. Shamshir Rahman et al. (2019), for example, analyze several forms of deep learning algorithms used in the area of solar, and the findings demonstrate that hybrid networks outperform single networks. Mishra et al. (2020) suggested and got excellent results using a short-term solar radiation prediction model based on WT-LSTM, demonstrating that deep learning technology has significant promise in solar radiation. Gao et al. (2020) propose a CEEMDAN-CNN-LSTM model for hourly multi-region solar irradiance estimation, and the results show that the model can achieve more accurate prediction performance than existing models.