Detection of Low Elevation Outliers in TanDEM-X DEMs With Histogram and Adaptive TIN

Abstract

Low elevation outliers (LEOs) in TanDEM-X (TDX) digital elevation models (DEMs) have a great effect on the accuracy of digital terrain models (DTMs) generated by filtering DEM data. A novel method, combining local elevation histograms and adaptive triangulated irregular network (HATIN), was developed to detect LEOs in TDX DEMs. The performance of HATIN was evaluated by comparing LEOs identified by median, morphological, HATIN, and iterative inverse distance weighted (IIDW) interpolation methods with LEOs defined by differences between TDX DEMs and light detection and ranging (LiDAR) DTMs. The HATIN method detected almost all large clustered LEOs in the downtown area of Miami, USA with many high-rise buildings, and reduced the root-mean-square error (RMSE) of the resulting DTM from 10.37 to 1.11 m. In the same area, the IIDW method missed some large LEOs, producing a DTM with an RMSE slightly larger than that of the HATIN method. In contrast, the median and morphological methods made more omission and commission errors, failing particularly in the detection of many large LEOs, and producing larger RMSEs (2.43–2.91 m). In low-rise residential areas with isolated LEOs, the HATIN method also identified all large LEOs, producing the fewest commission errors of all methods. However, differences among DTM RMSEs for all four methods were minor. Therefore, it is recommended that, before TDX DEMs are filtered to generate DTMs for urban areas, isolated LEOs should be removed using the computationally simple median or morphological methods, whereas clustered LEOs should be removed by the complex and computationally intensive HATIN or IIDW methods.