Local polynomial regression outperforms standard solar radiation model in Mogadishu study

An empirical evaluation of daily global solar radiation models conducted for Mogadishu, Somalia, has demonstrated that Local Polynomial Regression achieves superior estimation accuracy compared to the conventional Angström–Prescott formulation. In the research published by Shuaib Mursal Ibrahim and Aşır Genç in Discover Applied Sciences, the non-parametric Local Polynomial Regression framework attained a coefficient of determination (R²) of 0.907, outperforming the coefficient of determination of 0.797 recorded for the Angström–Prescott model.

The accurate estimation of daily global solar radiation, designated as H in atmospheric and solar engineering literature, represents a critical requirement for solar energy system design, building energy simulation, and regional renewable energy planning. These predictive modeling tools are especially vital in data-scarce regions where comprehensive ground-based measurement facilities are scarce or entirely absent. To evaluate potential solar estimation options for the Mogadishu region, the research team analyzed four empirical formulations: the Angström–Prescott model, Local Polynomial Regression, quadratic spline regression, and cubic spline regression.

To evaluate each model thoroughly, the researchers assessed the four frameworks using both conventional statistical metrics and information-theoretic indicators. The traditional statistical evaluation relied on standard error metrics, namely the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). In addition to these conventional measures, the investigation incorporated entropy-based indicators, specifically evaluating the Shannon entropy of residuals and the Kullback–Leibler divergence to assess the degree to which each modeling approach retained the statistical characteristics of the underlying data.

The empirical findings demonstrated distinct performance advantages for the non-parametric regression technique over the traditional parametric baseline. The Local Polynomial Regression model achieved a root mean square error of 0.156 kWhm^-2d^-1, with a 95% confidence interval of [0.132, 0.178], while recording a residual Shannon entropy value of 1.679 with a 95% confidence interval of [1.577, 1.917]. By comparison, the Angström–Prescott formulation yielded a higher root mean square error of 0.231 kWhm^-2d^-1, with an associated 95% confidence interval of [0.215, 0.262], reflecting greater residual error in its daily radiation predictions.

Beyond standard error values, the entropy-based metrics demonstrated that non-parametric models preserved the statistical structure of observed radiation series, thereby minimizing information loss during estimation. Based on these findings, the validated Local Polynomial Regression estimator offers a deployable baseline for solar resource assessment in regions that lack ground-based meteorological infrastructure. At the same time, the authors noted that locally calibrated Angström–Prescott coefficients provide a parsimonious, site-adapted formula that remains practical for application whenever local sunshine duration data are available.

Subject of Research: Empirical daily global solar radiation modeling in Mogadishu, Somalia

Article Title: Entropy-based evaluation of empirical daily global solar radiation models for Mogadishu, Somalia

Article References: Ibrahim, S. M., & Genç, A. (2026). Entropy-based evaluation of empirical daily global solar radiation models for Mogadishu, Somalia. Discover Applied Sciences. https://doi.org/10.1007/s42452-026-09859-5

Image Credits: AI Generated

DOI: 10.1007/s42452-026-09859-5

Keywords: global solar radiation, daily solar radiation, solar radiation models, Mogadishu, Somalia, local polynomial regression, Angström–Prescott model, quadratic spline regression, cubic spline regression, Shannon entropy, Kullback–Leibler divergence, residual entropy, solar resource assessment, solar energy system design, building energy simulation, renewable energy planning, non-parametric models, sunshine duration