Mapping of urban forest canopy height through diffusion-enhanced deep learning by using sentinel-2 and urban geospatial data

Urban canopy height is a key indicator for assessing urban ecosystem services and ecological functions. However, the fragmentation and heterogeneity of urban forests pose challenges to the deep learning methods that perform well in large-scale forest canopy height mapping. To address this, we propose a novel and scalable Multimodal Urban Forest Canopy Height (MUFCH) estimation model that integrates diffusion mechanism and geospatial auxiliary data. Experimental results show that the MUFCH estimation model achieves a mean absolute error (MAE) of 2.019 m and a root mean square error (RMSE) of 3.090 m, representing improvements of 26.39% and 19.28%, respectively, compared to the baseline model. This enables high-precision canopy mapping with open-access data, providing a cost-effective solution for assessing critical forest indicators.