A MODIFIED LINEAR REGRESSION MODEL FOR PREDICTING AEROSOL OPTICAL DEPTH (AOD) IN ILORIN-NIGERIA

Abstract

In general, aerosol measurements generated from the ground-based stations are some of the most trusted results; however, these measurements are limited in spatial density. The measurement is also affected by the presence of heavy cloud in the atmosphere which makes obtaining free cloud satellite image for atmospheric correction very difficult. The current study focuses on the development of a modified empirical regression models aiming to predict a local Aerosol Optical density (AOD) in the atmosphere that is cloud free using ground meteorological data in Ilorin-Nigeria. Ten years (January, 2007 to December 2017) of Aerosol robotic network (AERONET) AOD and National Oceanic Atmospheric Administration-National climate Data Centre (NOAA-NCDC) meteorological data were used to establish a quantitative relationship between AOD on one hand and visibility, relative humidity, sea level pressure, temperature and wind speed on the other hand. The highest correlated model was used to predict AOD values during overall (January-December), Harmattan (November-March) and summer (April-October). The analysis of 10 years overall AOD data was satisfactory, with coefficient of determination (R²) =0.80 and root mean square error (RMSE) =0.07 with relatively low value of weighted mean absolute percentage error (wMAPE) <4% which indicates relative high accuracy of the model. Similarly, the prediction accuracy of the AOD model was high during summer (R² = 0.85, RMSE 0.05 and wMAPE =2%) and best during Harmattan (R² = 0.90, RMSE 0.03 and wMAPE =0.9%). This result is consistent with the results obtained in the previous literatures showing that, in as much meteorological variables will be