“…The direct way to determine LCZ classes is to match the calculated values of LCZ parameters with their reference ranges. This method has been most widely used to distinguish LCZs based on the LCZ parameters provided by Stewart & Oke (2012) or custom parameters (Agathangelidis et al, 2019;Bartesaghi Koc et al, 2018, 2017Cai et al, 2019;Emmanuel & Loconsole, 2015;Jin et al, 2020;Leconte et al, 2017Leconte et al, , 2015Mandelmilch et al, 2020;Mitraka et al, 2015;Nassar et al, 2016;Ndetto & Matzarakis, 2015;Perera & Emmanuel, 2018;Shi et al, 2018;Thomas et al, 2014;Villadiego & Velay-Dabat, 2014;Wang et al, 2018b;Zheng et al, 2018). In addition, some studies have used other methods to determine LCZ types, such as the score assignment method (Lelovics et al, 2014;Unger et al, 2014), the decision-making algorithm (Chen et al, 2020b;Quan, 2019;Zhao et al, 2019a), the multi-dimensional linear interpolation method (Quan et al, 2017), the Naive Bayes algorithm (Hammerberg et al, 2018), the random forest algorithm (Hu et al, 2019), and the k-means method (Hidalgo et al, 2019;Kwok et al, 2019;Zhan et al, 2018).…”