A simple method for determination of fine resolution urban form patterns with distinct thermal properties using class-level landscape metrics

Abstract

Context Relationships between land surface temperature (LST) and spatial configuration of urban form described by landscape metrics so far have been investigated with coarse resolution LST imagery within artificially superimposed land divisions. Citywide micro-scale observations are needed to better inform urban design and help mitigate urban heat island effects in warming climates. Objectives The primary objective was to sub-divide an existing high-resolution land cover (LC) map into groups of patches with distinct spatial and thermal properties suitable for urban LST studies relevant to micro-scales. The secondary objective was to provide insights into the optimal analytical unit size to calculate class-level landscape metrics strongly correlated with LST at 2 m spatial resolution. Methods A two-tiered unsupervised k-means clustering analysis was deployed to derive spatially distinct groups of patches of each major LC class followed by further subdivisions into hottest, coldest and intermediary sub-classes, making use of high resolution class-level landscape metrics strongly correlated with LST. Results Aggregation class-level landscape metrics were consistently correlated with LST for green and grey LC classes and the optimal search window size for their calculations was 100 m for LST at 2 m resolution. ANOVA indicated that all Tier 1 and most of Tier 2 subdivisions were thermally and spatially different. Conclusions The two-tiered k-means clustering approach was successful at depicting subdivisions of major LC classes with distinct spatial configuration and thermal properties, especially at a broader Tier 1 level. Further research into spatial configuration of LC patches with similar spatial but different thermal properties is required.

Promoting sustainable landscape pattern for landscape sustainability

“…Identifying high-resolution urban landscape patterns with distinct spatial configuration and their thermal properties is not well explored in recent studies. Zawadzka. et al (2021) made a key contribution in this context based on a two-tiered unsupervised k-means clustering approach to subdividing major land cover classes according to the relationship between land surface temperature and urban form patterns at fine spatial resolution in three medium sized towns within the UK.…”

The importance of spatial configuration of neighbouring land cover for explanation of surface temperature of individual patches in urban landscapes

“…The data used in this study comprise several datasets derived and described in previous work: very high spatial resolution (2 m) LST maps downscaled from Landsat 8 TIR imagery for 6th June and 8th July 2013 using statistical methods (Zawadzka et al 2020a), a land cover map showing the distribution of five main land cover types at 2 m spatial resolution (Grafius et al 2016), Fig. 1, and a map of land cover subtypes obtained from a two-tiered K-means clustering analysis of selected landscape metrics and LST to yield land cover patches classified according to their spatial configuration and temperature (Zawadzka et al 2020b). Very high resolution LST data were downscaled from Landsat 8-derived thermal images using multivariate adaptive regression splines relating LST to spectral indices and distribution of impervious surfaces and water with overall accuracy measured by root mean square error (RMSE) ranging from 1.4 to Fig.…”

“…Analyses were carried out for areas within the 'Built-up Area Extent' boundary. Adapted from Zawadzka et al (2020b) 1.83 K depending on date and town. The two dates (6th of June and 8th of July 2013) were chosen as the only summer-time cloudless Landsat 8 images available for the three towns at the time of conducting the downscaling study.…”

“…In Tier 1, clustering of selected class-level landscape metrics (McGarigal and Marks 1995) related to spatial aggregation of land cover: COHE-SION, PLADJ and LSI yielded subdivisions of the main land cover classes with distinct spatial and thermal properties (Table S3 Supplementary Materials). The three landscape metrics were selected from a larger pool of various shape and aggregation metrics based on the strength of correlations with LST data within each land cover class (Zawadzka et al 2020b) and are rarely used in this type of studies. Thus formed land cover type patches were attributed with means of LST for June and subsequently for July, and further subdivided through Tier 2 k-means clustering to determine these land cover subtype patches that had the highest, intermediary, and lowest LST at both time steps.…”

“…LC land cover, cLHS conditioned latin hypercube sampling, SCDs spatial configuration descriptors that included spatial aggregation metrics (COHESION, LSI, PLADJ), distances to land cover classes, area, elevation, and feature heights. *refers to methodology steps described in detail in Zawadzka et al (2020b) classes that these SCDs referred to. Full list of SCDs considered in this study is given in Table S2 in Supplementary Materials).…”