The spatial configuration of local climate zones reveals effects on wayfinding in human walking

from origins, with higher values indicating higher urban accessibility [37]. LConn measures the local spatial connectivity with higher values indicating more connected local spaces [29]. Higher intelligibility values indicate the more consistent spatial configurations between the locals and the entire city. When the intelligibility value approaches 1, local places and cities or regions are accessible and well-connected synchronically for spatial co-developments. 2.4 Measuring spatial configurations of local climate zones 2.4.1 LCZ classification for characterizing regional spaces.. LCZ classification is used to illustrate urban forms including land-cover types and structures. The spatial characteristic of each LCZ type has been illustrated in [19, 38]. [39] formalized a commonly used method that includes three steps: 1) preparing the Landsat data; 2) identifying training areas for LCZ classi- fication in the SAGA software package [40], and 3) accuracy assessment and improvement. Compared to this conventional but complicated approach, a LCZ generator as a web applica- tion has been developed [38] to simplify the process, which provides not only an online plat- form to illustrate the urban forms of regions of interest as the LCZ maps but also an automated accuracy evaluation. Using this LCZ generator, the only thing we need to do in the LCZ classification is to define the valid training areas in the region. Training areas are the regions greater than 1 km2 that are defined using Google Earth™ to exemplify each LCZ type

provided by [19]. Between 5 to 15 training samples are required according to the classification guidelines, and more training areas are preferred for improving the LCZ classification accu- racy. Here, we defined 20 training areas for each LCZ type, serving as the inputs of the LCZ generator (https://lcz-generator.rub.de). The spatially explicit LCZ types in the GBA were pre- sented in Fig 3. Its overall accuracy has been higher than most LCZ classifications in Chinese cities [41] and can also pass the quality control using international standards [42]. 2.4.2 Landscape metric as a proxy for quantitative LCZs configurations. Following the LCZ classification to characterize regional spaces, landscape metrics of each LCZ type are cal- culated to indicate urban spatial configurations, such as where a particular LCZ type is more clustered. The landscape metrics quantifications were performed using FRAGSTATS tool which is simple to use and does not require any mathematical knowledge [43]. Two levels of landscape metrics were provided in FRAGSTATS—class-level and patch-level. Patch-level metrics quantify the spatial configurations of individual patches that are the continuous areas with the homogeneous LCZ type by assigning a unique numeric value to each patch [44, 45]. Class-level metrics suggest the overall spatial configurations of LCZ types by averaging over all patch-level metrics of the corresponding LCZ type [46]. Twelve class-level and two patch-level metrics were selected, and their details were given in Table 1. The potential qualitative effects

of changing landscape metrics values on wayfinding were also described in Table 1. 2.5 Determinations of strategies and locations for spatial reconfigurations 2.5.1 Correlation analysis. In this study, correlation analysis is used to demonstrate the associations between LCZs configurations and wayfinding. A correlation coefficient approach- ing +1 indicates that landscape metrics of LCZ types (proxies for LCZs configurations) and intelligibility (a proxy for wayfinding) are correlated in a more positive way, which, otherwise, are correlated in a more negative manner when the correlation coefficient approaches to -1. As a result, the correlation analysis results can be used to suggest solutions to spatial reconfigura- tions for wayfinding improvement. 2.5.2 Hot spot analysis. The hot spot analysis helps to examine whether local spatial asso- ciations, shown as spatial clusters, exist in the urban forms presented by LCZ configurations. We used Getis-Ord Gi* [47] in the ArcGIS Pro software to conduct hot spot analysis using PLOS ONE The spatial configuration for wayfinding PLOS ONE | https://doi.org/10.1371/journal.pone.0289780 September 8, 2023 6 / 18