Mapping clear-sky surface solar ultraviolet radiation in China at 1 km spatial resolution using Machine Learning technique and Google Earth Engine

Abstract

Ultraviolet (UV) radiation is an important fundamental data for solar energy utilization, climate change, human health, photochemical reaction studies, etc. However, it is still a problem to get UV radiation estimations with high spatial resolution rapidly. This study attempted to develop a Machine Learning (ML) model to estimate clear-sky UV radiation with high accuracy and high spatial resolution (1 km) in China. Based on Moderate-resolution Imaging Spectro-radiometer (MODIS) data and ERA5 reanalysis data obtained from Google Earth Engine (GEE), we established input dataset composed of different variables and developed 29 ML models to estimate clear-sky UV radiation using 37 Chinese Ecosystem Research Network (CERN) stations measurements for model training and validation. The results showed that compared with other ML models the Deep Neural Networks (DNN) model had a high and stable performance with a determination coefficient (R2) of 0.904, a Root Mean Square Error (RMSE) of 3.100 Wm-2, a Mean Absolute Error (MAE) of 2.274 Wm-2 for 10-fold cross-validation. To realize fast estimation of online clear-sky UV radiation, the DNN model was deployed to Google Cloud Platform. The online estimation results showed that northern China had more UV radiation than southern China, and eastern China had less radiation than western China. This study would provide a useful reference for the study of solar energy resources, human health, and ecological system studies.