Equatorial All-Sky Downward Longwave Radiation Modelling

Abstract

Nsikan Ime Obot^*, Busola Olugbon, Ibifubara Humprey and Ridwanulahi Abidemi Akeem

Machine learning algorithms, such as random forests (RF), artificial neural networks (ANN), and support vector regression (SVR), are viable modelling tools because they can learn and replicate data patterns. However, linear regression models are relatively easy to implement. Downward longwave radiation (DLR) is rarely measured due to complications of its measuring instrument, notwithstanding the importance of the radiation in the atmosphere and the energy balance of the Earth’s surface. Besides linear regression, several machine learning modes, such as SVR and RF, were also used to model daily cloudless and all-sky DLR at Ilorin (8.53° N, 4.57° E), Nigeria. We further sought an appropriate ANN unit for estimating the radiation in this study. Predictors comprised the period, clearness index, air temperature, water vapour pressure, relative humidity, global solar radiation, and solar hour angle. We found that solar hour angle actively predicts all-sky DLR. The most vital variables used for an all-sky DLR linear regression model for this clime are water vapour pressure, relative humidity, and solar hour angle. Machine learning systems generalise better with vast data having well-correlated inputs. The results also reveal that several machine learning algorithms, like SVR with Pearson VII kernel function, can be used for modelling DLR.