Abstract
The purpose of this study is to compare the different performances of three commonly used models (i.e., linear regression, random forest regression, and deep neural network (DNN)) to predict cloud fraction (CF) using ground-based shortwave solar radiation measurements and analyze the importance of the input features. The CF data are obtained from the Surface Radiation Budget (SURFRAD) and the Atmospheric Radiation Measurement (ARM) and the irradiance data from the USDA UV-B Monitoring and Research Program. The study finds that CF of opaque and total clouds can be best predicted using both Random Forest Regression and DNN with the validation R2 up-to 0.959 and MSE as low as 0.00688.
Recommended Citation
Lerma, Jamison; Blackaby, Emily K.; Chen, Maosi; Madronich, Sasha; and Gao, Wei
(2023)
"Predicting Cloud Fraction with Instantaneous Direct and Diffuse Shortwave Solar Irradiance,"
The Geographical Bulletin: Vol. 64:
Iss.
2, Article 9.
Available at:
https://digitalcommons.kennesaw.edu/thegeographicalbulletin/vol64/iss2/9