Mapping urban Surface Materials Using Imaging Spectroscopy Data

Abstract

Urban environment and its processes directly affect human life. Detailed and up-to-date urban surface material maps are of great importance to modelers studying meteorology, climatology and ecology, as well as to authorities seeking to understand the urban growth dynamics and spatial evolution. However, mapping urban surface materials is challenging due to the complex spatial patterns. An established source of up-to-date information is remote sensing, as demonstrated by the widespread usage of SAR, LiDAR and optical data. Data from imaging spectrometers can identify detailed spectral features of surface materials through the fine and continuous sampling of the electromagnetic spectrum, which cannot be achieved with the same accuracy using multispectral or RGB images. To date, numerous studies in urban surface material mapping have been using data from airborne imaging spectrometers with high spatial resolution, demonstrating the potential and providing good results. Compared to these sensors, spaceborne imaging spectrometers have regional or global coverage, high repeatability, and avoid expensive, time-consuming, and labor-intensive flight campaigns. However, the spatial resolution of current spaceborne imaging spectroscopy data (also known as hyperspectral data) is about 30 m, resulting in a mixed pixel problem that is challenging to handle with conventional mapping approaches. The main objective of this study is to perform urban surface material mapping with imaging spectroscopy data at different spatial scales, simultaneously explore the information content of these data to detect the chemical and physical properties of surface materials, and take the mixed-pixel problem into account. Specifically, this thesis aims to (1) map solar photovoltaic modules using airborne imaging spectroscopy data based on their spectral features; (2) investigate the sampling robustness of urban material gradients; (3) analyze the area transferability of urban material gradients. To this end, we detected solar photovoltaics with an overall accuracy of about 80% to 90% by creating and combining spectral indices. This dissertation proved that the developed approach is suitable for accurate photovoltaic detection. We also demonstrated that the concept of urban surface material gradients is robust in sampling and transferable between similar urban areas. With these results, urban material gradients can be a generic technique for urban mapping with spaceborne imaging spectroscopy data. The methods developed invi the three parts of this dissertation improve the usefulness of imaging spectroscopy data for urban material detection from a classical method to the new concept of urban gradients, from airborne to spaceborne data, from pure pixel detection to solving the mixed pixel problem. By introducing and enhancing the gradient concept in urban mapping, the mixed pixel problem can be tackled, which is a promising approach for the analysis of imaging spectroscopy data from ongoing and upcoming spaceborne sensors. Overall, this thesis provides promising urban surface material mapping results by proposing a physical feature based approach as well as confirming and laying the foundation of the generic gradient concept in urban material studies. Further work can build on these results and could open a new field for the application of spaceborne imaging spectroscopy data.