Cognitive experiments on labelling uncertainty for LCZ mapping across heterogeneous megacities in China: Introducing multi-source geospatial data and LCZ subclasses

World Urban Database and Access Portal Tools (WUDAPT) provides a prevailing workflow for mapping Local Climate Zones (LCZs) based on a supervised machine learning method. Yet, the low accuracy and numerous iterations caused by labelling uncertainty in heterogeneous urban context were still obstacles for its application. In this study, the cognitive experiments on labelling uncertainty for LCZ mapping were conducted over Wuhan, China. The multi-source geospatial data and LCZ subclasses cognition were introduced for pre-recognition on training areas (TAs). Sixty-four participants were randomly allocated into four equal-sized teams, with each team exclusively assessing the mapping accuracy of a designated scenario under a restricted protocol of nine iterations. Validated through randomized point sampling, the scenario set for mapping LCZ with subclass recognition show optimal performance, that it can achieve the overall accuracy (OA) of 83 % at five iterations, and reach up to OA of 89 %. This study highlights the critical role of LCZ subclasses cognition in improving mapping accuracy, and proposed an extended workflow of WUDAPT specific for heterogeneous megacities in China. The novel workflow improves LCZ classification accuracy while minimizing iterations in heterogeneous urban context, thereby supporting LCZ-based studies through reliable descriptions on the climate-related urban forms in Chinese megacities.