Neural network models for predicting urban albedo of urban surfaces with different reflection directional properties

Abstract

The urban albedo is regarded as an important indicator for mitigating the urban heat island (UHI) phenomenon. The goal of this paper is to create neural network models that forecast the urban albedo of various reflection directional characteristic urban surfaces by simply inputting the sun’s position (altitude and azimuth) and solar radiation. In this study, two urban models with diffuse highly reflective (DHR) and retro-reflective (RR) urban coatings are created, and their urban albedo for two sunny days is calculated using standard (ASTM E1918A). Using the measured urban albedo for one sunny day, two predictive neural network models, Gaussian process (NNGP) and hyperbolic tangent function (NNTanH) are developed (August 1, 2021). The two developed neural network models are used to forecast the urban albedo for another sunny day (July 19, 2021). In the case of DHR urban coatings, the NNTanH model is more accurate with higher R-squared (R2) and lower root mean squared error (RMSE) than the NNGP model, and there is no significant difference between the two neural network models. However, it is demonstrated that the NNGP model is more accurate than the NNTanH model in predicting urban albedo in the case of RR urban coatings, with higher R2 and lower RMSE.