Evaluation of WRF-chem simulations of NO2 and CO from biomass burning over East Africa and its surrounding regions

In East Africa, biomass burning in the savanna region emits nitrogen dioxide (NO2), carbon monoxide (CO), and aerosols among other species. These emissions are dangerous air pollutants which pose a health risk to the population. They also affect the radiation budget. Currently, limited academic research has been done to study their spatial and temporal distribution over this region by means of numerical modeling. This study therefore used the Weather Research and Forecasting model coupled with chemistry (WRF-chem) to simulate, for the first time, the distribution of NO2 during the year 2012 and CO during the period June 2015 to May 2016 over this region. These periods had the highest atmospheric abundances of these species. The model’s performance was evaluated against satellite observations from the Ozone Monitoring Instrument (OMI) and the Measurement of Pollution in the Troposphere (MOPITT). Three evaluation metrics were used, these were, the normalized mean bias (NMB), the root mean square error (RMSE) and Pearson’s correlation coefficient (R). Further, an attempt was made to reduce the bias shown by WRF-chem by applying a deep convolutional autoencoder (WRF-DCA) algorithm and linear scaling (WRF-LS). The results showed that WRF-chem simulated the seasonality of the gases but made below adequate estimates of the gas abundances. It overestimated NO2 and underestimated CO throughout all the seasons. Overall, for NO2, WRF-chem had an average NMB of 3.51, RMSE of 2 × 1015 molecules/cm2 and R of 0.44 while for CO, it had an average NMB of − 0.063, RMSE of 0.65 × 1018 molecules/cm2 and R of 0.13. Furthermore, even though both WRF-DCA and WRF-LS successfully reduced the bias in WRF-chem’s NO2 estimates, WRF-DCA had a superior performance compared to WRF-LS. It reduced the NMB by an average of 3.2 (90.2%). Finally, this study has shown that deep learning has a strong ability to improve the estimates of numerical models, and this can be a cue to incorporate this approach along other stages of the numerical modeling process.