Comparative Machine Learning Bias Correction of Satellite-Derived Solar Irradiance at Tropical Stations in Central Java, Indonesia

Eko Riyanto; Iwan Setiawan; Uky Yudatama

doi:10.63158/journalisi.v8i5.1778

Authors

Eko Riyanto Diponegoro University
Iwan Setiawan Diponegoro University
Uky Yudatama Muhammadiyah University of Magelang

DOI:

https://doi.org/10.63158/journalisi.v8i5.1778

Keywords:

Bias Correction, Global Horizontal Irradiance, NASA POWER, Site Adaptation, Tropical Climate

Abstract

Satellite-derived Global Horizontal Irradiance products carry systematic bias in tropical regions, so site-level correction is a prerequisite for their use where ground networks are sparse. This study compares bias-correction methods for NASA POWER daily irradiance in Central Java, Indonesia, and asks how much reported accuracy reflects algebraic dependence between predictors and the ground reference. Ten PUSDATARU stations were analysed over 2020–2024 (17,072 station-days; 8,459 from five exactly co-located stations). Random Forest, XGBoost, LightGBM, Elastic Net, and Quantile Mapping were trained chronologically (2020–2022/2023/2024) against persistence and climatology baselines, with preprocessing fitted on training data only, paired significance tests, leave-one-station-out validation, and SHAP attribution. Random Forest reduced RMSE from 3.398 to 2.674 MJ m⁻² day⁻¹ (21.3%), lowered mean bias from 0.988 to 0.221, and raised R² from 0.266 to 0.546; across ten stations XGBoost reduced error from 5.921 to 2.831. Differences among tree ensembles were not statistically robust, and the weighted ensemble gave no gain at the highest inference cost. Adding ground-observed sunshine as a predictor raised the coefficient of determination to 0.9994, identified as recovery of the Ångström–Prescott expression defining the reference (70.9% attribution). Near-perfect accuracies against sunshine-derived references therefore measure leakage, not skill; correction also failed to transfer spatially, motivating sub-degree geostationary products as the next step.

Downloads

Download data is not yet available.

References

[1] W. Tercha, S. A. Tadjer, F. Chekired, and L. Canale, “Machine Learning-Based Forecasting of Temperature and Solar Irradiance for Photovoltaic Systems,” Energies (Basel)., vol. 17, no. 5, p. 1124, 2024, doi: 10.3390/en17051124.

[2] S. Al-Dahidi, B. Rinchi, R. Dababseh, O. Ayadi, and M. Alrbai, “A geographic multi-scale machine learning framework for predicting solar irradiation on tilted surfaces,” Energy, vol. 313, Dec. 2024, doi: 10.1016/j.energy.2024.133767.

[3] M. U. Khan and M. A. Jama, “Evaluation and correction of solar irradiance in Somaliland using ground measurements and global reanalysis products,” Heliyon, vol. 10, no. 16, p. e35256, 2024, doi: 10.1016/j.heliyon.2024.e35256.

[4] Artificial Intelligence for Renewable Energy Systems. Elsevier, 2022. doi: 10.1016/C2020-0-03645-5.

[5] I. A. Aditya, P. M. Putra, R. A. Rajagukguk, and Y. Asri, “Estimating Irradiance through Satellite-Driven Deep Learning Models,” in Proceedings - International Seminar on Intelligent Technology and its Applications, ISITIA, Institute of Electrical and Electronics Engineers Inc., 2024, pp. 48–52. doi: 10.1109/ISITIA63062.2024.10667941.

[6] T. Nguyen-Duy, T. Ngo-Duc, M. Choisy, I. P. Fernandez, and S. Sparrow, “Performance and added value of a high-resolution (2 km) rainfall product based on WRF-downscaled ERA5 for Ho Chi Minh City, Vietnam,” Theor. Appl. Climatol., vol. 156, no. 12, Dec. 2025, doi: 10.1007/s00704-025-05894-1.

[7] D. Puspita et al., “Hybrid ANN WRF solar radiation forecasting in data limited tropical region,” Renewable and Sustainable Energy Transition, vol. 8, Dec. 2025, doi: 10.1016/j.rset.2025.100124.

[8] N. S. Sitohang, I. Budi, K. G. H. Mangunkusumo, F. Albar, and G. R. Afdholy, “Assessing Satellite-Based GHI Accuracy via Statistical and LSTM Approaches: Nusa Penida PVPP Case Study,” in Proc. ICT-PEP, IEEE, 2025, pp. 43–48. doi: 10.1109/ICT-PEP67281.2025.11232049.

[9] J. Zhu, X. Shi, and S. Zhang, “Machine Learning-Based Grammar Error Detection Method in English Composition,” Sci. Program., vol. 2021, 2021, doi: 10.1155/2021/4213791.

[10] D. Selvamuthu, V. Kumar, and A. Mishra, “Indian stock market prediction using artificial neural networks on tick data,” Financial Innovation, vol. 5, no. 1, 2019, doi: 10.1186/s40854-019-0131-7.

[11] L. Breiman, “Random Forests,” Mach. Learn., vol. 45, no. 1, pp. 5–32, 2001, doi: 10.1023/A:1010933404324.

[12] M. S. Alam, F. S. Al-Ismail, M. S. Hossain, and S. M. Rahman, “Ensemble Machine-Learning Models for Accurate Prediction of Solar Irradiation in Bangladesh,” Processes, vol. 11, no. 3, p. 908, 2023, doi: 10.3390/pr11030908.

[13] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ACM, 2016, pp. 785–794. doi: 10.1145/2939672.2939785.

[14] G. Ke et al., “LightGBM: A Highly Efficient Gradient Boosting Decision Tree,” in Advances in Neural Information Processing Systems 30, 2017, pp. 3146–3154.

[15] R. Ledesma and G. Salazar, “Machine Learning for Site Adaptation of Satellite-Derived Solar Irradiance in Northwestern Argentina,” 2025. [Online]. Available: https://datalsasaf.lsasvcs.ipma.pt/

[16] H. Zou and T. Hastie, “Regularization and variable selection via the elastic net,” J. R. Stat. Soc. Series B Stat. Methodol., vol. 67, no. 2, pp. 301–320, 2005, doi: 10.1111/j.1467-9868.2005.00503.x.

[17] L. Yezeguelian, C. Vernay, T. Carrière, and P. Blanc, “Characterizing the convergence and robustness of the kernel density mapping method for site-adaptation of global horizontal irradiation in Western Europe,” HAL, 2021. [Online]. Available: https://hal.science/hal-03337523

[18] M. D. Bailey, D. Nychka, M. Sengupta, and S. Bandyopadhyay, “Adapting Quantile Mapping to Bias Correct Solar Radiation Data,” arXiv, 2024. doi: 10.48550/arXiv.2405.20352.

[19] L. Gudmundsson, J. B. Bremnes, J. E. Haugen, and T. Engen Skaugen, “Technical Note: Downscaling RCM precipitation to the station scale using quantile mapping – a comparison of methods,” May 15, 2012. doi: 10.5194/hessd-9-6185-2012.

[20] S. Zainali, D. Yang, T. Landelius, and P. E. Campana, “Site adaptation with machine learning for a Northern Europe gridded global solar irradiance product,” Energy and AI, vol. 15, Jan. 2024, doi: 10.1016/j.egyai.2023.100331.

[21] F. A. Suwadana et al., “Solar Irradiance Estimation in Tropical Regions Using Recurrent Neural Networks and WRF Models,” Energies (Basel)., vol. 18, no. 4, p. 925, 2025, doi: 10.3390/en18040925.

[22] P. M. P. Garniwa et al., “Comparison of Semi-Empirical Models in Estimating Global Horizontal Irradiance for South Korea and Indonesia,” Resources, vol. 14, no. 11, p. 170, 2025, doi: 10.3390/resources14110170.

[23] S. Chen, C. Li, Y. Xie, and M. Li, “Global and direct solar irradiance estimation using deep learning and selected spectral satellite images,” Appl. Energy, vol. 352, Dec. 2023, doi: 10.1016/j.apenergy.2023.121979.

[24] A. Damiani et al., “Spatially generalizable bias correction of satellite solar radiation for regional climate assessment — a case study in Japan,” International Journal of Applied Earth Observation and Geoinformation, vol. 145, p. 104947, 2025, doi: 10.1016/j.jag.2025.104947.

[25] H. Shi et al., “First estimation of high-resolution solar photovoltaic resource maps over China with Fengyun-4A satellite and machine learning,” Renewable and Sustainable Energy Reviews, vol. 184, p. 113549, 2023, doi: 10.1016/j.rser.2023.113549.