Atmospheric Temperature Prediction across Nigeria using Artificial Neural Network

Abstract

Atmospheric temperature is one of the dominating atmospheric parameters that impact on the propagation of radio waves through the troposphere. Adequate knowledge of the atmospheric temperature of an environment is therefore essential for radio wave propagation planning. In this study, thirty-four (34)-year (1981-2014) atmospheric temperature data of 10 selected weather stations across the climatic zones of Nigeria, obtained from the Nigerian Meteorological Agency (NIMET) through the data bank of the West African Science Service Centre on Climate Change and Adaptive Land Use (WASCAL) of the Federal University of Technology Minna, Nigeria was used in Artificial Neural Network (ANN) for the prediction of mean monthly atmospheric temperature. The ANN architecture comprised of 2 inputs (the climatic zones and the corresponding month for the mean monthly atmospheric temperature), 1 hidden layer and 1 output (atmospheric temperature). Levenberg-Marquardt algorithm was used with 9 different pairs of activation functions formed from 3 activation functions (logsig, purelin and tansig). The number of neurons in the hidden layer was varied from 33-39 with an increasing steps of 2 (33, 35, 37 and 39). The network architecture of 2-37-1 (2 inputs, 37 neurons in the hidden layer and 1 output), with tansig/tansig pair of activation functions had the least mean square error value of 2.2280, and was used for the prediction process. The computed correlation values for measured and predicted atmospheric temperature ranged from 0.9733 to 0.8787, depicting strong positive correlation and good accuracy of the developed model. Comparisons of the measured and the ANN predicted atmospheric temperature across selected stations in the climatic zones of Nigeria, showed that the developed model can effectively predict mean monthly atmospheric temperature, using month and climatic zone as input parameters.