Net Surface Shortwave Radiation Retrieval Using Random Forest Method With MODIS/AQUA Data

The net surface shortwave radiation (NSSR) at the Earth's surface drives evapotranspiration, photosynthesis, and other physical and biological processes. The primary objective of this study is to estimate NSSR in all sky conditions by using narrowband data of the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument onboard the AQUA satellite. The random forest (RF) machine learning method for retrieving NSSR was developed with MODerate resolution atmospheric TRANsmission model (MODTRAN 5) simulated data. The bias, root mean square error (RMSE), and R-2 for the training dataset of the model are 0.04 W m(-2), 2.03 W m(-2), and 1.00, respectively; for testing data, these values are 0.53 W m(-2), 5.50 W m(-2), and 1.00, respectively. Note that the proposed method is better than the traditional method (RMSE 7.29 W m(-2)) with MODTRAN data, and the sky conditions (clear and cloudy) do not need to be distinguished in the RF method. Seven in situ measurements of the Surface Radiation (SURFRAD) observation network were used to validate the estimated NSSR with MODIS/AQUA data using the proposed RF method, and the bias, RMSE, and R2 of the comparison are -8.4 W m(-2), 76.8 W m(-2), and 0.91, respectively. Approximately 70% of the absolute difference of all the samples is below 50 W m(-2). Considering its concise process and relatively improved accuracy, both in regard to model development and validation, it can be concluded that the retrieval of NSSR with RF will be an efficient and feasible method in the future.