Global-scale Digital Elevation Models (DEMs) are essential for many studies, such as land surface hydrology modelling, flood inundation modelling, and terrain analysis. The Shuttle Radar Topography Mission (SRTM) DEM is the most widely used global-scale elevation data. However, the SRTM DEM contains height errors which come from different error sources. Long- and medium-wavelength (>500m) errors are mainly due to residual motion error of the interferometric mast. While short-wavelength errors (i.e. radar speckle) are caused by pixel-scale variations of surface brightness and slope. In addition to these random error components, the SRTM DEM contains systematic tree height bias because the radar beam cannot penetrate into forest canopy.
We have developed a global-scale method for removing height errors from the SRTM DEM, by combining the statistical and multi-satellite approach. First, we removed the medium-wavelength "striping error" using a 2D-Fourier-transform filter. Second, we have corrected the long-wavelength "absolute error" using ICESat Lasar altimerty "centroid" elevations. Third, we calculated the "tree height bias" by subtracting ICESat “lowest” elevation from SRTM elevation. We estimated tree height bias in pixels which do not have ICESat measurements, by using the Landsat tree density map and the global forest height map. Last, the short-wavelength "radar speckles" were removed by an adaptive smoothing filter.
By applying this 4-step method, the 90 percentile absolute bias was reduced from ~10m to ~3m, and remaining bias is mostly considered to be due to sub-pixel topography. We executed global flood simulations with the error-removed SRTM and the original SRTM, and found that the simulated inundated area showed much better agreement to observations when the error-removed DEM was used. The flood simulation results suggests that the height error removal is essential for better understanding of the global surface water dynamics and the global hydrological cycle.
各種検索
サポート
ヘルプ