Review of solar spectral irradiance modelling at ground level: Current methods and machine learning opportunities

The accurate acquisition of solar spectral irradiance is crucial for the efficient utilisation of solar energy and the strategic planning of solar energy harvesting systems. Due to the high costs associated with specialised measurement instruments, spectral irradiance modelling has become a popular alternative in various fields, including industrial electricity generation, agriculture, and building applications. This review provides a thorough evaluation of existing methods for modelling spectral irradiance at the Earth’s surface, while also exploring the potential of machine learning (ML) in this domain. The review begins by examining the atmospheric parameters (e.g., components of gas, aerosol and cloud) that influence spectral irradiance, summarising their wavelength-specific effects and accessibility via different physical or numerical models. A systematic comparison of various mainstream physical models reveals that using more relevant atmospheric parameters as input and enhancing the spatial–temporal resolutions could improve the accuracy of spectral irradiance prediction. However, their accuracy still depends on the accurate parameter acquisition by either measurement or modelling. In contrast, statistical parameterisation approaches can reduce dependence on complex input data, which enhances computational efficiency and maintains high accuracy under certain conditions, such as clear-sky conditions. In addition, machine learning is a promising method that can further enhance the accuracy and efficiency of spectral irradiance modelling. By leveraging large datasets and advanced algorithms, ML models can capture complex patterns and relationships in the data that are difficult to represent with traditional physical models. This enables more accurate predictions of spectral irradiance under a wide range of atmospheric conditions. However, compared to the ML models for total irradiance predictions, ML approaches for spectral modelling are still in the development stage. The challenge is to establish reliable wavelength-dependent correlations between atmospheric parameters and spectral irradiance outputs. Through this review, we aim to provide a comprehensive foundation for researchers and practitioners to better understand the current state of spectral irradiance modelling and to identify key areas for future development.