Benchmarking framework for decomposition models in solar irradiance estimation within a temperate climate

• Establishment of a transparent framework for Diffuse Horizontal Irradiance estimation. • Developed and applied 4 rigorous tests to assess the robustness of decomposition models. • Identified 5 decomposition models that demonstrate robustness within the United Kingdom. • Altering the clear-sky irradiance results in a rise from 5 to 20 robust decomposition models. Accurate solar irradiance data is essential for the design and operation of solar energy systems, especially in areas lacking direct Diffuse Horizontal Irradiance (DHI) measurements. To address this gap, researchers often rely on decomposition models to estimate DHI. However, with a wide variety of models available, choosing the most suitable one remains challenging. This paper introduces a structured benchmarking framework aimed at identifying reliable models, with a focus on temperate climates using the United Kingdom as a case study. The framework evaluates 104 models based on physical, mathematical, and statistical grounds, applying a strict Mean Absolute Error (MAE) threshold of 15%. The analysis begins with a temporal resolution test, comparing model performance at hourly and minute intervals across three locations: Camborne, Chilbolton, and Lerwick. Only a handful of models maintain accuracy across both timescales and multiple sites, with Paulescu consistently performing best (MAE 12.76%). Starke1 also showed strength in coastal areas. Among the evaluated models, five − namely Paulescu, Yang5, Suehrcke-McCormick, Starke3, and Yang4 – consistently achieved the required accuracy across all datasets. To broaden the scope of the evaluation, the study explored 150 unique model pairings by combining 10 different clear-sky global horizontal irradiance (GHI Clear ) models with 15 decomposition approaches. This expanded analysis led to a notable increase in the number of models meeting the robustness criteria, raising the count from 5 to 20. Notably, integrating McClear with Engerer2a achieved the best MAE of 12.77%, revealing potential for further optimization. This framework not only identifies strong models but also demonstrates how clear-sky inputs influence DHI estimation, offering valuable insights for solar resource assessment.