Simulation of representative nocturnal satellite imagery for urban areas with high spectral and high spatial resolution

Kurzfassung

Contrary to its daytime counterpart, nighttime VNIR satellite imagery are currently limited in both spectral resolution and spatial resolution. That does not mean, however, that the relevance of such a sensor is non-existent, with possible applications including the estimation of light pollution, energy consumption and socio-economic information, among others. In order to determine the optimal spectral bands, the required radiometric sampling and the spatial resolution, synthetic TOA spectral radiance values are simulated. These are computed through the combination of lamp spectra libraries, surface reflectance libraries, radiative transfer for the estimation of atmospheric effects, and typical luminance values based on well-established lighting standards. Various spectral band combinations are then evaluated for their ability to correctly estimate a number of important lighting quality parameters, as well as to discriminate between different lighting types. The tested lighting indicators include (1) LER, or the effiency to produce visible light; (2) spectral G index, which serves as an indicator for emissions in the blue part of the spectrum; and (3) CCT, which assesses the perceived colour of a light source. An optimal nanometre-level band selection is found for one panchromatic band and five additional multispectral bands. The selected multispectral bands are located in the blue, green, yellow, orange-red and near-infrared part of the spectrum, respectively, thereby offering a good spread over the full VNIR part of the spectrum. Since their choice is specifically adjusted to suit the spectra of artificial lights, however, spectral bands differ significantly from the typical daytime situation with the Sun as main illuminator, essentially emitting light equally across the spectrum. Whereas the main interest of daytime optical remote sensing is in surface reflectance, nighttime optical remote sensing focuses on the light sources. With respect to other nighttime sensor proposals and existing sensors, the recommended spectral bands reduce the estimation error of LER with 73% relatively, the G index error with 86% and the CCT error with 68%. Similarly, the classification performance of lighting types improves with about 10%. Based on the generated TOA radiances, detection limits of 10^{-8} to 10^{-7} W m^{-2} sr^{-1} nm^{-1} and saturation values of 10^{-4} to 10^{-3} W m^{-2} sr^{-1} nm^{-1} are recommended for the selected spectral bands. Additionally, results indicate that 12 or more bits are required for information storage. Finally, some road lighting patterns are simulated using a physically-based rendering software, in order to generate representative imagery. These are then used to determine the required spatial resolution for individual lighting detection. It is found that a ground sampling distance of 10 m is required in most cases. Overall, this thesis shows that significant improvements can be made in terms of the sensor design for nighttime VNIR remote sensing, opening the door to a new world of applications.