A generalizability analysis of a data-driven method for the Urban Heat Island phenomenon assessment

Cities worldwide are experiencing increasing temperatures due to the urban heat island (UHI) phenomenon. UHI is caused by the replacement of natural land surfaces with man-made dark surfaces. Among others, it causes a number of public health problems associated with heat events, particularly in the construction sector, where construction workers are more likely to die from heat-related illnesses compared to other industries. To address the negative effects of this phenomenon, researchers around the world have proposed different alternatives for studying the effects of UHI. Among these methods, data-driven approaches are becoming increasingly popular. However, as with all data-driven models, there is always the question of the extent to which they are generalizable. To answer this question, this research work applies the data-driven UHI assessment framework previously proposed by the authors to the cities of Montreal, Canada, and Apeldoorn, the Netherlands, in five different scenarios. The results showed that while models have good prediction capabilities within the scope of the training data set, they do not demonstrate good generalizability on the testing data from a different context. Also, the results of this research highlighted that as cities continue to grow, there is an urgent need to standardize the understanding and assessment of the UHI at a pedestrian level. The intrinsic differences in how UHIs are assessed and tackled worldwide can create confusion about the phenomenon, and limits the applicability and generalizability of data-driven approaches.