Neural Networks Estimations of Downward Longwave Radiation in Nigerian Environment
A Thesis Submitted to the School of Postgraduate Studies, University of Lagos
Data on downward longwave radiation (DLR) particularly at screen level are scarce due to problems such as pyrgeometer being delicate. Consequently, the instrument is rarely available in most weather stations and oftentimes durations of measurements are short. In Nigeria, ground measurements of DLR have been taken only at Ilorin (8.50oN, 4.55oE) from September 1992 - August 1994 and from June 1995 - April 1998. To ameliorate data scantiness, other means like mathematical modelling and remote sensing methods are employed. This study aims at developing clear skies empirical models through curve fittings and using neural networks to estimate DLR for ground and satellite data of Nigerian environments. The ground-measured DLR data and satellite-derived data for all the capital cities of the 36 states and the Federal Capital Territory in Nigeria from July 1983 to June 2005 were employed in this study. Additional, data of other parameters like temperature, water vapour pressure and rainfall were also employed. Both the ground data and satellite data were compared. The diurnal, monthly and yearly trends of the radiation across the country were investigated. Some existing DLR clear skies models were examined and new equations were developed for Ilorin. Lastly, estimations of DLR with artificial neural networks (ANNs) under different conditions were attained. The agreement between DLR ground measured data and satellite derived data is moderate and generally the radiation is low during the dry season. There are differing patterns from normal diurnal pattern of DLR particularly at the middle and end of the year. Two new separate clear skies equations are developed and a third one which involves the recalibration of the constants of Brunt (1932) model. It can be deducted from the new models that in this region, temperature T, in the power range of T5 to T6 and water vapour pressure term yield better estimation of DLR. ANNs gave an ideal fit for the radiation because in all cases of analyses of the estimations of DLR with ANNs, statistical significance was attained at 95% confidence level. Thus ANNs stand to provide the solution modelling tools even for situations of no DLR data and befitting equations. The extreme cases of diurnal patterns of the radiation around the June and December solstices need be further investigated.