Development of a Machine Learning Forecast Model for Global Horizontal Irradiation Adapted to Tibet Based on Visible All-Sky Imaging

Round 1

Reviewer 1 Report

Wu et al. compare the performance of random forest (RF) and long short-term memory (LSTM) in predicting  global horizontal irradiation. Ground measured GHI and cloud information are combined for prediction. The idea is interesting. The experiments are well designed and presented. However, some minor changes should be made before publication.

1) I would like to recommend authors reading more relevant papers from high quality journals, such as APEN, RSER, ECM, RENE, etc. Some papers that I have read are not cited by the authors. e.g., https://doi.org/10.1016/j.rser.2022.112680, https://doi.org/10.1016/j.apenergy.2019.113541, https://doi.org/10.1016/j.energy.2021.119887, http://dx.doi.org/10.1016/j.enconman.2016.04.009    

2) Please keep all texts the same font size

3) What does AI in the Abstract mean, please give its full name at the first appearance. Please check the manuscript thoroughly.

4) RF LSTM is not necessary in the Abstract

5) Please describe the time coverage of ground observations in Section 2.2 and 2.3.

Author Response

Please see the attachment

Author Response File: Author Response.doc

Reviewer 2 Report

In the paper "Development of a Machine Learning Forecast Model for global horizontal irradiation Adapted to Tibet based on Visible All Sky Imaging" by Lingxiao W. et all.,

tested a set of models derived from machine learning and artificial intelligence predictions adapted to Tibet and from ground-based observations of shortwave solar irradiance flux and from observed cloud cover obtained from all-sky imaging in the Yangbaying region. Key parameters were sensitively tested and optimized. The results show that the use of actual cloud cover as an input variable in a given model can significantly improve the prediction accuracy, and the RMSE of the prediction accuracy is reduced by more than 20% when the prediction horizon is 1 hour, compared with the model without the input cloud cover variable. This study is very important for global horizon radiation forecast models that are based on ground-level clouds, which substantially reduce the irradiance efficiency at shortwave solar radiation wavelengths. The Yangbajing research area has an altitude of 4 300 metres.

The work is based on real measurements of solar radiation and cloud cover from the ground in the study area. These data are compared with existing models based on a popular method of solar radiation prediction (machine learning) using artificial intelligence (AI), varying the number of time input data and cloud parameters. The complicated results are also graphically represented and discussed.    The obtained results show that the difference in prediction horizon and step size has different effects on different models, so the model should be selected according to the data resolution, prediction horizon and accuracy requirements.

Finally, the authors discuss the advantages of the models tested by them based on the input data to the models from machine learning and artificial intelligence. It is shown that the global horizon radiation, which depends on the amount and composition of clouds, can significantly improve the short-term forecast using their proposed method, which will undoubtedly contribute to achieving the " China's double-carbon strategy".

Author Response

Please see the attachment.

Author Response File: Author Response.doc