Surface solar radiation drives the water cycle and energy exchange on the earth's surface, and its diffuse component can promote carbon uptake in ecosystems by increasing the plant productivity. The accurate knowledge of their spatial distribution is of great importance to many studies and applications, such as the estimation of agricultural yield, carbon dynamics of terrestrial systems, site selection of solar power plants, as well as trends of regional climate changes. Therefore, we produce the hourly surface radiation datasets based on the hourly Multi-functional Transport Satellite (MTSAT) satellite imagery and the ground observations from the China Meteorology Administration (CMA) through deep learning techniques. The deep network is trained using training samples in 2014, and then utilized to generate the hourly radiation for other years. This dataset provides the gridded surface global and diffuse solar radiation in 2019 within 71.025°E - 141.025°E and 14.975°N - 59.975°N with an increment of 0.05°. Both the direct predicted hourly values and the integrated daily and monthly total values are available. The dataset should be useful for the analysis of the regional differences and temporal cycles of solar radiation in fine scales, and the impact of diffuse radiation on plant growth etc.

Data format: .h5 files. The files for hourly datasets are named as “RAD_yyyymmddhh.h5” where “yyyy”, “mm”, “dd”, and “hh” denote year, month, day and hour (UTC time).----File Structure: Hourly datasets:Variable name: global radiation; diffuse radiation Data type: int16 type Scaling factor: 0.01Unit: 0.01 MJ·m-2Notes: The values larger than 65536/2-1 are stored as negative values, thus their actual values are calculated by plus 65536 (refer to the Example of Visualization).----Source: Japan geostationary meteorological satellite MTSAT and observations from China Meteorological Administration (CMA) radiation sites.----Methods: A hybrid deep network, comprising of convolutional neural network and multi-layer perceptron, is designed to learn the mapping relationship between and image blocks from Japan geostationary meteorological satellite MTSAT and surface measurements at China Meteorological Administration (CMA) sites. The deep network is firstly trained using hourly GSR measurements at 98 CMA radiation sites in 2014, and the trained network with best performance is then used to map spatially continuous hourly GSR values taking geostationary satellite image blocks as inputs. After fine-tuned using hourly measurements of diffuse radiation at 16 Level-1 CMA radiation sites in 2014, the trained network is further utilized to map spatially continuous hourly diffuse radiation. Above complete trained networks are then used to produce the datasets in 2019.----Quality Control: Although tremendous efforts have been made to ensure the quality of in-situ measurements and the accuracy of the trained deep network, there still exist some deviations for our estimated datasets. The data provider disclaims any kind of liability for quality, performance, and fitness for a particular purpose arising out of the use.----Owner: Name: Hou Jiang, Ning LuE-mail: jh1225357498@gmail.com; ning.robin@gmail.com Address：No.11A, Datun Road, Chaoyang District, Beijing, 100101, P.R. China