Comparative accuracy assessment of digital elevation models (SRTM, ALOS WORLD 3D, ASTER GDEM, MERIT DEM) on the example of forest and floodland zones of the National park “Nizhnyaya Kama”

Background. Digital elevation models are important data sources for solving a number of practical and scientific problems. All available global remote sensing data are digital surface models by definition as they represent an earth surface with all natural and man-made objects on it. At the moment, there is no globally consistent digital terrain model, which is free of building and trees canopy and represents relief itself. A significant contribution to the creation of such models was made by D. Yamazaki et al., presented “Multi-Error-Removed Improved-Terrain DEM” (MERIT DEM). The novelty of this terrain model is that it attempts to remove vegetation bias and other random and systematic errors of the other sources, such as SRTM and ALOS WORLD 3D. The current study was performed to assess the local accuracy of the MERIT DEM, as well as other four open DEMs in comparison to topographical survey data on the example of the forest and floodplain areas. Materials and methods. Accuracy assessment was performed on the example of the National Park “Nizhnyaya Kama”, located on the north-east part of the Republic of Tatarstan, Russia (Fig. 1). Four digital elevation models were assessed: SRTM, ASTER GDEM, ALOS WORLD 3D and MERIT DEM. Available data of topographic maps at a scale of 1:100,000 were used as the assessment basis. The topographic data was vectorized and interpolated using Multilevel B-Splines (MBS) (Fig. 2a). A test subset (2,000 points) was sampled from the vectorized topographic data, including contour lines points and single landmark points (Fig. 2b). The test subset was used to calculate accuracy metrics: absolute values of minimal and maximal elevation difference, mean difference, Pearson's correlation coefficient, and root-mean-square error (RMSE). In addition, vertical profiles were extracted from the elevation models. Results. The interpolated elevation model MBS was characterized by the highest accuracy as expected. The RMSE = 1.67, and near-zero mean error allowed to conclude adequate representation of the park's relief. The range of elevation values for the forest areas were 54.27-176.11 m with the mean of 109.39 (Fig 3a). The remote sensed digital elevation models had the lowest accuracies (Table 1). The elevation values were 6.08-15.88 m higher than the actual topographical data. The corrected model MERIT DEM had the lowest error values in comparison to the remote sensed models. Nevertheless, the mean error and RMSE of the MERIT DEM were higher than of the interpolated MBS model. In the floodplain areas the models were ordered in the following order by the increase of the mean elevation: ASTER GDEM (50.1 m) < SRTM (52.3 m) < MBS (55.1 m) < ALOS WORLD 3D (57.2 m) < MERIT DEM (57.4 m). The mean errors and root mean squared errors of the remote sensed DEMs were lower in comparison to the forest areas (Table 2). Conclusion. The results showed that the overall accuracy of open digital elevation models increase in the order of ASTER GDEM < ALOS WORLD 3D < SRTM < MERIT DEM. The vertical profiles showed high noisiness of the ASTER GDEM and ALOS WORLD 3D for the forest and floodplain areas of the park.

Keywords

Authors

References