The local climate zone (LCZ) classification scheme is effective for climatic studies, and thus, timely and accurate LCZ mapping becomes critical for scientific climate research. Remote sensing images can efficiently capture the information of large-scale landscapes overhead, while street-level images can supplement the ground-level information, thus helping improve the LCZ mapping. Previous study has proven the usefulness of street-level images in enhancing LCZ mapping results; however, how they help to improve the results still remains unexplored. To unveil the underlying mechanism and fill the gap, in this study, the feature importance analysis is performed on classification experiments using different data sources to reveal the contributions of different components, while feature correlation analysis is adopted to find the relationship between street view images and key LCZ indicators. The results show that fusing street view images can help improve the classification performance considerably, especially for compact urban types such as compact highrise and compact midrise. In addition, the results further show that the building and sky information embedded in the street view images contribute the most. The feature correlation analysis further demonstrates their strong correlations with key LCZ indicators, which define the LCZ scheme. The findings of the study can help us better understand how street-level images can contribute to LCZ mapping and facilitate future urban climate studies.
Supplementary notes can be added here, including code, math, and images.