Author(s)

ROBERTA COCCI GRIFONI, MARIA SIMONETTA BERNABEI, ROSALBA D’ONOFRIO, GRAZIANO ENZO MARCHESANI, MOHAMMADJAVAD KHODAPARAST

Abstract

In this study, a novel high-resolution Local Climate Zone (LCZ) classification tool, the Framework for Environmental Type Classification Hub (FETCH), is introduced. It utilises Google remote-sensing data and a 30 m × 30 m segmentation method. Through the proposed tool, climate vulnerability within typical historical urban centres was effectively analysed. Ascoli Piceno, a historic Italian town, was selected as the study area because of its complex urban morphology. Advanced computer vision and machine learning techniques were employed to process and classify the remote sensing data, ensuring the accuracy and detail of the generated LCZ map. The results indicate that the LCZ map accurately represents urban morphology, revealing distinct climate zones that closely correspond to known urban features and climate vulnerability. A tool for assessing classification errors using the Root Mean Square Error of Prediction (RMSEP) is also introduced, which ensures the reliability of LCZ classification for further analysis and decision-making. The practical relevance of the LCZ map is underscored by its potential application in urban policy, climate adaptation and environmental analysis. This study addresses the challenge of developing consistent and high-resolution LCZ maps on a global scale by leveraging high-resolution data and advanced computational methods. The findings have significant implications for urban climate research, providing a robust tool for understanding and mitigating the impacts of urban heat islands in architecturally complex and historically significant cityscapes. The methodology developed in this study can be adapted for use in other historical urban contexts to promote the development of sustainable and resilient cities in the face of climate change.

Keywords

Local Climate Zone (LCZ), high-resolution, urban heat island (UHI), urban morphology, historical urban centres, machine learning, climate vulnerability, root mean square error of prediction (RMSEP)