Global validation and hybrid calibration of CAMS and MERRA-2 PM2.5 reanalysis products based on OpenAQ platform

It is highly valuable to obtain high-quality PM2.5 concentration worldwide for continuous monitoring of global air pollution. Recently, global reanalysis products of PM2.5 have come into the view. However, most studies focus on the validation and calibration of a single product regionally, few studies expand to a global scale and integrate multiple products. With the help of global open-source data provided by the OpenAQ platform, we propose a hybrid calibration method aimed to improve the accuracy of CAMSRA and the MERRA-2 PM2.5 products. In the study, the accuracy of the two datasets are assessed on multi-time scales at first. Secondly, we try to use some machine learning models to correct the deviation of the original products alone and then further explore the possibility of the hybrid calibration. Global-scale validation results show that CAMSRA products are generally overestimated (daily R = 0.6), and MERRA-2 products are underestimated (daily R = 0.3), which supports our hybrid calibration method to an extent. Using the Extremely Randomized Tree (ERT) to implement the separate calibration scheme, two products show different degrees of accuracy improvement, to be specific, R increases by 0.19 and 0.43 for daily CAMSRA and MERRA-2 products, respectively. Compared with the separate calibration modeling, the hybrid method performs better, with R reaching up to 0.81. RMSE is only 14.94 mu g/m(3), which has a decrease of 60.99% and 64.42% to two abovementioned original products. The obtained daily PM2.5 maps have higher quality with no data gaps, which can be a promising data source of air pollution monitoring and health research. This dataset is published in GeoTIFF format at https://doi.org/10.5281/zenodo.5168102.