Object-based image analysis to map local climate zones

Local climate zones (LCZs) are seen as a useful concept to investigate the relationship between land use types and urban heat island effects, which has been substantially researched for cities of the Global North. However, in the Global South, the usefulness of the concept may be questioned, as spatial patterns of urban structure types are typically highly heterogeneous (e.g., a mix of commercial, formal residential and slum/informal areas). With the increased and reasonable availability of (very) high-resolution imagery, however, the concept of LCZ may also prove meaningful to investigate land use dependent thermal patterns in Global South cities, which is the main research question investigated in this paper. We compared LCZs based on very-high-resolution imagery (using Pleiades and SPOT-6) with the standard method using Landsat imagery, and we analyzed their relationship with land surface temperature (Landsat TIR) in Bandung, Indonesia. Two different methods of image classification are employed to extract the LCZs, i.e., the random forest algorithm on Landsat imagery and the object-based image analysis (OBIA) on very-high-resolution images. The result shows that the OBIA provides higher accuracy, reaching 89% for the object-based analysis on Pleiades imagery compared to 69% for the pixel-based analysis of Landsat imagery. In addition, the results of our analysis show that not only the type of LCZs but also composition and configuration patterns (i.e., density and aggregation of LCZs) significantly affect land surface temperature. To conclude, VHR imagery and OBIA allow mapping the complex patterns of LCZs in cities of the Global South, in support of developing mitigation strategies for the urban heat island effect via urban and landscape planning.