BackgroundThough downward longwave radiation (DLR) models curb the paucity of data, they are mostly location dependent. Therefore, there is a need to evaluate their relevance given the increasing use of machine learning techniques. In this study, cloudless DLR estimates from regression models and soft computing models of neural networks (NN), support vector regression (SVR) and adaptive neuro-fuzzy inference system (ANFIS) were compared. Clear days from September 1992 to August 1994 and July 1995 to March 1998 in Ilorin (8.50 °N, 4.55 °E), Nigeria were considered, while the predictors for the models were water vapour pressure, e and air temperature, T.ResultsA new regression model in relation to the Boltzmann constant, σ: left(1.014left(frac{1.0times {10}^{30}times e}{T^{13}}right)+0.699right)sigma {T}^4 , was better than other regression models and applicable at another location. Between 1 and 8, the sixth degree was the best polynomial kernel function in SVR models’ estimations of cloudless DLR. Though the new regression model was comparable to expert systems, ANFIS was still the best model due to its consistent high correlations and lowest estimation errors.ConclusionsExperience-based computational procedures that combine enough logics with neural networks respond effectively to other data. Furthermore, the analytical relationship between water vapour pressure and air temperature in DLR’s mechanism should be redefined accordingly, while the sixth polynomial should be used as the default setting in SVR systems.