Sensitivity of the C-band SRTM DEM vertical accuracy to terrain characteristics and spatial resolution
Résumé
This work reports the results of a careful regional analysis of the SRTM DEM (Shuttle Radar Topography Mission – Digital Elevation Model) vertical accuracy as a function of both topography and Land-Use/Land Cover (LULC). Absolute vertical errors appear LULC-dependent, with some values greater than the stated accuracy of the SRTM dataset, mostly over forested areas. The results show that the structure of the errors is well modeled by a cosine power n of the local incidence angle (loc). SRTM quality is further assessed using slope and topographical similarity indexes. The results show a lower relative accuracy on slope with a R2 = 0.5 and a moderate agreement (Kappa ≃ 0.4) between SRTM- and IGN-derived slopes. The application of a simple cosine squared correction on the 90 m SRTM dataset shows only a slight improvement of the relative accuracy despite a 7 m decrease of the mean absolute elevation error. The accuracy is strongly improved (R2 = 0.93 and Kappa = 0.75) for data resampled at a 150 m to 500 m horizontal resolution. These results support the idea that for regional application purposes the topographic correction as well as the spatial resampling of the SRTMdataset are needed.