Fusing multimodal data of nature-economy-society for large-scale urban building height estimation

• DOI: 10.1016/j.jag.2024.103809
• Publication Date: 2024-04-05T04:38:06Z
• Authors: Shouhang Du; Hao Liu; Jianghe Xing; Shihong Du
• Keywords: Autoencoder; Footprint

The building height holds significant importance for comprehensively understanding urban morphology, enhancing planning, and fostering sustainable development. Although many methods using optical SAR images have been presented estimation, these fall short in capturing the influences of economic social attributes on height. In this study, we introduced a Nature-Economy-Society (NES) feature model to represent information, established multi-scale One-Dimensional (1-D) Convolutional Neural Network predicting heights, referred as NES-CNN. First, derived natural buildings from time-series Sentinel-1 Sentinel-2 multispectral images, well World Settlement Footprint (WSF) data Digital Elevation Model (DEM), nighttime light Gross Domestic Product (GDP) data, function Points Interest (POI) data. Second, an autoencoder is employed reduce dimensionality high-dimensional attribute features, minimizing redundancy. Finally, 1-D CNN explore correlations between multi-source heterogeneous NES features facilitating prediction experiments, applied proposed method estimate heights Beijing Shanghai at spatial resolution 10 m. results indicated that Beijing, RMSE, MAE, R values are 6.93 m, 4.41 0.84, respectively, while Shanghai, 7.57 5.38 0.80, respectively. addition information decreases RMSE by 6 % both compared with only attributes. comparison existing studies same mapping resolution, 39 51 Shanghai. innovative inspiring nature study lies its application large-scale estimation.