Comparative Study of Urban Thermal Landscape Identification Methods Based on Landsat Imagery

Abstract

The urban heat island (UHI) has become a significant environmental concern due to the rapid pace of urbanization and climate change. Urban thermal landscapes (UTLs), which reflect the thermodynamic manifestation of urban landscapes, are usually identified by reclassifying land surface temperature (LST) images. Different UTL identification methods can lead to inconsistent results and hinder cross-study comparability. This study compares four widely used UTL identification methods using a Landsat 9 LST image acquired in the summer of 2022 in Zhengzhou, China. The UTLs were classified into five levels, and the influence of spatial extent was examined by applying different multiples of the built-up area. Landscape metrics and land cover composition were then used to evaluate the outcomes. Results reveal that when the spatial extent expands from 1 to 4 times the built-up area, the segmentation thresholds for UTLs generally decrease across all methods. As a result, within the built-up area, the percentage of landscape (PLAND) of the high-temperature landscape (HL) increases by 24.19%, 21.45%, 27.76%, and 36.75% for the mean-standard deviation (MSD) method, natural breaks (Jenks) method, Hot-spot analysis method, and urban thermal field variance index (UTFVI) method, respectively. In contrast, the PLAND of the low-temperature landscape (LL) decreases by 11.83%, 11.70%, 14.44%, and 33.60%, respectively. The MSD and hot-spot methods produce similar results and effectively identify UHI core areas. However, the hot-spot method is more computationally intensive. The Jenks method excels in illustrating thermal gradient transitions. Nevertheless, with larger study areas, the segmentation threshold for LL is sensitive to spatial extent, leading to small patches being identified as LL. The UTFVI method yields overly aggregated classification results, thereby limiting its ability to represent mixed-function areas accurately. Overall, the MSD and Jenks methods are recommended. Moreover, a spatial extent approximately twice the size of the built-up area, rather than the administrative boundary, is suggested as the study area in UTL identification.