Modeling the atmospheric dispersion of SO2 from Mount Nyiragongo

Mount Nyiragongo, an active volcano, is the most dominant natural source of sulphur dioxide (SO2) in Africa. While a number of studies have employed atmospheric models to simulate the dispersion of SO2 from this mountain, prior to this study, no attempt has been made to use deep learning to bias correct the model's esti-mates. Here, the Weather Research and Forecasting model coupled with chemistry (WRF-Chem) was used to simulate massive SO2 plumes degassed from this mountain between September 2014 and August 2015. Satellite observations by the Ozone Monitoring Instrument (OMI) showed that the SO2 spread to over 500 km from the volcano site. A deep convolutional autoencoder algorithm (WRF-DCA) was then applied to reduce the bias that WRF-Chem showed against the OMI observations. Finally, the correction performance of WRF-DCA was compared with a conventional bias correction method, linear scaling (WRF-LS). The performance of WRF-Chem, WRF-DCA, and WRF-LS was analyzed using three metrics, that is, the normalized mean bias (NMB), the root mean square error (RMSE), and Pearson's correlation coefficient (R). The results showed that WRF-Chem overestimated SO2 at locations near the volcano site and underestimated SO2 at locations further away from the volcano site. It generated an overall average NMB of -0.61 against the OMI observations. Respectively, WRF-DCA and WRF-LS reduced this bias by an average of 0.25 (40.9%) and 0.21 (34.4%). Furthermore, although both methods also reduced the RMSE and improved the correlation, WRF-DCA consistently performed better than WRF-LS. This study demonstrates the advantage that deep learning can provide in estimating volcanic SO2 emissions.