Accurate monitoring and operation of solar power systems require high-resolution solar radiation measurements and precise separation models. This study aims to improve the accuracy of classic diffuse fraction-clearness index piecewise separation models by applying data-driven classifications of sky conditions. This is achieved through a novel outlier-insensitive clustering algorithm and shape prescriptive modeling, applied to 1-, 10-, 30-, and 60-min ground measurements from 4 different locations in the MENA region. This study shows that classifications of sky conditions are not uniform among the selected locations even though all stations fall in the arid desert climate category. This highlights the importance of extracting the sky conditions from measurements rather than using available classifications in the literature. The selection of the number of clusters has to undergo optimization. The number of clusters is also a function of the time resolution. One of the selected locations shows four optimal clusters for 1-min data and six clusters for 60-min data. All developed piecewise separation models show high accuracy and stability with the mean bias errors approaching zero values and the mean absolute errors ranging between 8.7 and 11.8%. The models also outperform existing ones and have good generalization capabilities under the same climate classification. (C) 2021 Elsevier Ltd. All rights reserved.