Abstract The identification and analysis of urban roof materials using remote sensing technologies has attracted growing interest among Earth observation researchers in recent years. A georeferenced dataset has numerous applications in fields such as cadastral management, urban planning, and climate change. In this study, 907 spectral signatures were collected in June 2024 from seven roof materials and four types of urban land covers using Red Tide spectroradiometers (spectral range: 350 to 1000 nanometers) in the municipality of Puerto López, Meta, Colombia. Due to limited prior information, a stratified random sampling design was employed, with strata defined based on spectral variability obtained from an UltraCam Eagle M3 orthorectified image (50-centimeter spatial resolution). Furthermore, a subsampling strategy was also implemented to match the effective measurement area of the building roofs with the field of view of the spectroradiometer. Additionally, a feature selection analysis was performed using the Random Forest algorithm to identify the spectral bands most sensitive to material type, color, and conservation state, in relation to the spectral resolution of the orthorectified image. The results show that, within the spectral range of the instruments used, material color had the greatest influence on the spectral response. However, certain wavelengths in the field spectra exhibited greater sensitivity to material type and conservation state, although these did not align with the image’s central wavelength. Field-collected hyperspectral data is therefore a valuable resource for multispectral image classification, but its acquisition must be supported by accurate and rigorous fieldwork to ensure its effective application.
Valencia et al. (Fri,) studied this question.