Abstract:
A quantitative non- experimental correlation study aimed at investigating the relationship between ozone and the solar activity indices for the period 1985- 2011 was carried out. Dataset for solar activity indices were collected from the National Geophysical Data Centre through their website while dataset for total column ozone over Nairobi (1.17º S; 36.46º E), Kisumu (0.03º S; 34.45º E) and Mombasa (4.02º S; 39.43º E) from combined multiple satellite based instruments prepared by Bodeker scientific were collected through their website; both of which are publicly accessible and available. Analysis of trend and variability was done for the variables for each city. Correlation analysis was performed to analyze the relationship between time series data. A spectral analysis was carried out to characterize the frequency dependence of daily variation. Total column ozone and solar activity indices data were fitted on to multiple linear regression and the artificial neural network models. Statistical indicators were employed to verify the quality and reliability of the fitted models. A decreasing trend of 0.031% year-1 was observed at Nairobi, 0.026% year-1at Kisumu and 0.031% year-1 at Mombasa over the study period. Lowest total ozone values appeared in warm dry season. Short rainy season recorded the highest total ozone. Total ozone annual variability ranged between 1% and 6%, while the seasonal variability ranged between 2% and 4%. Solar activity indices and total ozone were positively correlated and significant. Solar activity had both immediate and delayed impact on total ozone. An increase in total ozone of about 1 – 4 % was attributed to solar activity indices. A negative forcing of Mg II index was observed in all the cities. Three types of periodicities 6 months, 12 months and 30 months were identified. Only Mg II index significantly affected total column ozone in all the cities. Neural networks in the form of multilayer perceptron feed forward with backpropagation algorithm with a sigmoid activation function with supervised learning method in the form of 3-15-1 were obtained in all the cities. Neural networks showed slightly better skills in predicting total column ozone. Therefore the temporal variability of total column ozone in all the three cities was declining. Not all solar activity indices had significant contributions in the variation of stratospheric ozone.
