Downscaling of Land Surface Temperature Using Airborne High-Resolution Data: A Case Study on Aprilia, Italy

Abstract

A regression-based downscaling of land surface temperature was developed over the heterogeneous urban area of Aprilia, Central Italy, using high resolution (HR) airborne data. Airborne sensors provided thermal and visible–near infrared (VNIR) measurements at 2-m pixel size. Coarse resolution images at 40, 30, and 20 m, upscaled by aggregation from the native airborne data, were sharpened to the finer resolution of 2 m. The main core of the downscaling method is the use of the spectral mixture analysis (SMA) to derive fractional pixel composition as predictors of the regression scheme. The HR VNIR data allow choosing detailed land cover types in the application of SMA, such as bright/dark roofs, and the benefit of this detailed selection is proved. The estimation error of the custom technique improves of about 10%–15% with respect to a classical regression downscaling.