Abstract
This study examines the dynamic mapping of impervious surface changes in optimising urban spatial structures and fostering sustainable development. A novel deep learning model and time-spectral-texture combination optimisation method were employed to identify pixel-based land-cover change trajectories. A piecewise linear regression model was also utilised to determine the time nodes of urban expansion. This methodology was applied to Panjin City, a resource-based city in China, to analyse temporal and spatial morphological changes related to urban expansion. The results reveal that the combination optimisation method achieved a trajectory classification accuracy of 93.10% and macro F1-score of 92.44%, with an urban expansion time identification accuracy of 84.24%. Panjin City's built-up area increased from 312.75 to 489.49 km² between 1990 and 2020, reflecting a growth rate of 56.51% and an average expansion speed of 5.89 km²/year. Furthermore, the spatial compactness of impervious surfaces declined, with urban expansion patterns shifting from leapfrog and edge expansion to infilling after 2016. These findings emphasise the need for strategic urban planning to enhance land-use efficiency and promote sustainable development, offering valuable insights for urban expansion mapping in other cities.