Abstract
We present interesting application of artificial intelligence for investigating effect of the COVID-19 lockdown on 3-dimensional temperature variation across Nigeria (2°-15° E, 4°-14° N), in equatorial Africa. Artificial neural networks were trained to learn time-series temperature variation patterns using radio occultation measurements of atmospheric temperature from the Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC). Data used for training, validation and testing of the neural networks covered period prior to the lockdown. There was also an investigation into the viability of solar activity indicator (represented by the sunspot number) as an input for the process. The results indicated that including the sunspot number as an input for the training did not improve the network prediction accuracy. The trained network was then used to predict values for the lockdown period. Since the network was trained using pre-lockdown dataset, predictions from the network are regarded as expected temperatures, should there have been no lockdown. By comparing with the actual COSMIC measurements during the lockdown period, effects of the lockdown on atmospheric temperatures were deduced. In overall, the mean altitudinal temperatures rose by about 1.1 °C above expected values during the lockdown. An altitudinal breakdown, at 1 km resolution, reveals that the values were typically below 0.5 °C at most of the altitudes, but exceeded 1 °C at 28 and 29 km altitudes. The temperatures were also observed to drop below expected values at altitudes of 0-2 km, and 17-20 km.