Improved TROPOMI SO2 columns using aCovariance-Based Retrieval Algorithm (COBRA)

For more than two years, the operational SO2 product from the TROPOspheric Monitoring Instrument (TROPOMI) onboard Sentinel-5 Precursor (S5P) platform has provided important information on volcanic and anthropogenic SO2 emissions, with an unprecedented level of details. However, in light of the validation results obtained so far, the TROPOMI SO2 product performs very good for large SO2 columns but large scale biases exist for low SO2 columns making the product less suitable for weak SO2 emissions. The biases up to 0.2 DU (Dobson Units) in Vertical column density (VCD) are related to imperfect correction of the Ring effect and are therefore not easy to solve in the operational Differential Optical Absorption Spectroscopy (DOAS) algorithm. Here, we propose a different approach called Covariance-Based Retrieval Algorithm (COBRA), a method adapted from IASI thermal infrared trace gas retrievals. Application of COBRA to TROPOMI SO2 VCD retrievals solves almost completely the bias problem and reduces the retrieval noise by a factor of 2. We compare our COBRA results to the widely used Principal Component Analysis (PCA) algorithm applied to TROPOMI and found that COBRA is at least at the same quality level than the PCA alogrithm. The new COBRA SO2 data set is compared with CAMS regional model output and validated with ground-based MAX-DOAS data at two sites: Basrah, Iraq, and Xianghe, China. We present long-term averaged SO2 maps from TROPOMI measurements with improved data over volcanic and anthropogenic SO2 emission regions that reveal new point and area sources. Boundary layer SO2 vertical columns of 1-2 x 1015 molec/cm² (0.05-0.1 DU) are retrieved robustly worldwide, illustrating the step improvement in sensitivity to weak SO2 emissions offered by the COBRA approach. We also show first SO2 emission results using COBRA, and demonstrate the added value of the proposed new data set.