Quantifying the driving force of urban morphologies on canopy urban heat island: a machine learning approach with educational application

This study quantifies the nonlinear driving force of urban morphological factors on canopy urban heat island intensity (CUHII) in Anhui Province, integrating relocated meteorological station data, remote sensing imagery, and machine learning frameworks. CUHII values exhibit a range of 0.06°C–1.12°C, with the built-up largest patch index (LPIbt, importance score = 0.25) and built-up area ratio (ARbt, 0.18) emerging as dominant drivers. Cropland coverage (ARc, Pearson’s r = −0.59) demonstrates significant cooling effects on urban thermal environments. The random forest (RF) model outperforms support vector regression (SVR) model, achieving training/test R 2 values of 0.95/0.76 and RMSE of 0.04/0.08°C. This superiority highlights its capability to capture complex interactions between urban morphologies and local thermal environment. The research framework is innovatively adapted to a flipped classroom educational model: students not only replicate the machine learning workflow using the same dataset but also design comparative experiments to test how urban morphological indicators affect CUHI outputs, thereby deepening their understanding of both physical mechanisms of CUHI and the interpretability of machine learning modeling. This integration of cutting-edge climate research with hands-on educational practice bridges the gap between academic inquiry and practical skill development. The study provides a replicable methodological framework for urban climate research and its translation into educational applications.