Building Urban Digital Elevation Model Using ICESat-2 Data

Cities gather over half of the world’s population,and the environment and development of cities are closely intertwined with human survival and well-being.In the urban environment,terrain and buildings are the two most important components of urban topography.Digital Elevation Model(DEM)and Building Height(BH)are digital representations of urban terrain and building elevation information,respectively.They play an irreplaceable role in expressing the three-dimensional form of cities and reproducing the real urban environment.Therefore,the production of DEM and BH is of great significance for understanding urban three-dimensional morphology,assessing urban environments,improving urban planning and design,and promoting sustainable urban development..Airborne LiDAR is typically an important data source for creating DEM and extracting BH.However,the high cost of data collection and the flight conditions of acquisition devices being subject to weather and relevant laws and regulations limit its application scope.In this context,spaceborne LiDAR,with its advantages of wide monitoring range,fast update speed,and low collection cost,provides a new available data source for establishing digital elevation models of cities.The latest generation of spaceborne LiDAR,ICESat-2/ATLAS,is one of the most advanced spaceborne LiDAR systems currently available.However,on the one hand,due to the scientific objectives of ICESat-2/ATLAS not being concentrated in urban areas,there are few studies applying ICESat-2/ATLAS data to urban areas.On the other hand,existing research has demonstrated the capability of ICESat-2/ATLAS to acquire high-precision height data within its orbit coverage,laying a foundation for the use of ICESat-2/ATLAS data in urban DEM modeling and BH extraction research.Based on this,taking New York City in the United States,Helsinki in Finland,and Shanghai in China as the research areas,the study explores methods for DEM modeling and BH extraction using ICESat-2/ATLAS data,providing data and technical support for research on urban sustainable development.The main research content and achievements of this study are as follows:(1)The accuracy of height measurement in urban areas using ICESat-2/ATLAS ATL03 data products was evaluated,demonstrating the ability of ATL03 data to acquire high-precision urban elevation information on a large scale.Firstly,the urban area was divided into three urban environments: high-rise intensive area,non-high-rise intensive area,and forest area,based on land cover type data.Secondly,ATL03 data were processed to obtain ground and non-ground ATL03 signal photons with medium/high confidence through denoising,classification,georeferencing,filtering,and other processing steps.Finally,the height measurement accuracy of ground and non-ground ATL03 signal photons in the three urban environments was evaluated.The research results showed that in the evaluation of urban ground elevation accuracy,the average absolute error of ATL03 ground photon height was 0.82 m,with average absolute errors of 2.54 m,0.74 m,and 0.39 m in the high-rise intensive area,non-high-rise intensive area,and forest area,respectively.In the evaluation of urban surface cover(such as buildings,vegetation,etc.)elevation accuracy,the average absolute error of ATL03 non-ground photon height was 1.39 m,with average absolute errors of 1.18 m,1.40 m,and 2.08 m in the high-rise intensive area,nonhigh-rise intensive area,and forest area,respectively.(2)DEM modeling based on ICESat-2/ATLAS ATL03 ground photon data is achieved using a spatial interpolation knowledge-constrained generative adversarial network(SIKGAN).SIKGAN utilizes ATL03 ground photon elevation data as the foundational input and takes into account the spatial relationships among the ground photons.By incorporating spatial distance data and inverse distance weighted interpolation,SIKGAN enables the generation of high-resolution(10 m)DEM.In comparison to traditional spatial interpolation methods such as inverse distance weighted interpolation and kriging,as well as deep learning methods like PIX2 PIX,SIKGAN demonstrates superior performance in terms of elevation and slope accuracy.Statistical analysis conducted in the research area of New York City indicates that the overall accuracy of the DEM generated by SIKGAN is MAE = 2.02 m,with a slope accuracy of MAE = 2.27°.Additionally,when applied to the research area of Helsinki,Finland,SIKGAN produces DEM with elevation and slope accuracies of MAE = 7.17 m and MAE = 4.03°,respectively.These results indicate the potential of SIKGAN for generalizing DEM modeling across different research areas while maintaining high accuracy.(3)Building height(BH)extraction is accomplished using ICESat-2/ATLAS ATL03non-ground photon data in conjunction with multi-source remote sensing data.A novel BH extraction framework is proposed,which relies primarily on ICESat-2/ATLAS ATL03 nonground photon data and incorporates building footprint data and high-resolution optical imagery.The framework utilizes deep learning techniques and an improved shadow-based height estimation method.It consists of five steps: 1)processing building footprint data,2)extracting building shadows,3)extracting building height annotations,4)estimating building heights,and 5)validating accuracy.The research results demonstrate considerable precision in BH extraction.In the research area of New York City,an average absolute error of 1.44 m is achieved among 60,046 extracted building heights.Furthermore,the proposed method is applied to the research area of Shanghai,where an average absolute error of 4.13 m is obtained for the heights of 15,875 extracted buildings.These findings highlight the transferability and potential of the proposed method for extracting BH over large-scale areas.