INDIVIDUAL ROOF ALBEDO ESTIMATION BY SUPER-RESOLUTION OF OPEN SATELLITE IMAGES

Abstract

 The urban heat island becomes severer. One of the countermeasures is "cool roof", which enhances the roof albedo and reduce the solar energy absorption to the city. There are many studies investigating the effect of cool roof. However, due to the lack of roof albedo data in city scale, there are uncertainties about temperature reduction by cool roofs in the previous works. We constructed a super-resolution model “ExEEGAN” using deep learning and attempted to estimate the individual roof albedo by increasing the spatial resolution of open satellite data. Training was done with a commercial satellite data of World View-3 which resolution is 1.24 m and fine enough to caputure individual buildings and houses. ExEEGAN successfully created finer images of open satellite data of both Landsat-8 and Sentinel-2, and identified roof albedos more accurately than the ordinary image interpolation methods, or bicubic and nearest neighbor interporation methods. RMSE of estimated roof albedos of individual buildings estimated from Sentinel-2 0.0279 in ExEEGAN, 0.0451 in bicubic interporation, and 0.0458 in nearest neighbor interporation.