^{1}

^{2}

Accuracy and quality of DEM are of a great interest for a wide range of applications. In this study, quality of ASTER GDEM and SRTM DEMs were assessed in comparison with DGPS measurements. Impact of DEM resolution upon the accuracy of terrain representation and topographic attributes was also discussed. The study deduced that vertical error has a strong effect on error propagation and this highly obvious in higher elevations as the absolute standard error (SE) ranges between is ±0 - 2.5 and ±0 - 2.4 m for ASTER GDEM and SRTM respectively. This is reflected on slope and aspect as the vertical errors increase and uncertainty is relatively high in flat and low areas. Error propagation in low lands influenced drainage extraction and resulted in isolated and truncated water courses.

Digital Elevation Models (DEMs) provide a full 3d perspective-view of an elevation surface. DEMs are a corner-stone for a wide range of geoscience studies such as: hydrological modeling and flood simulation, civil engineering, soil and ecological studies, geomorphological mapping, geohazard assessment, etc.

DEMs are acquired from satellite remote sensing (optical or radar imaging systems), photogrammetry analysis, topographic maps by digitizing contour lines and spot heights, and site survey. Structure of the data is usually organized in a square digital elevation grid, triangular irregular network ( TIN ), and set of digital line graph contours or random points. Satellite remote sensing constellations in specific provide vast amounts of DEM data, ranging from global (~1 km e.g., USGS GTOPO30) to a very fine scale (~1 - 2 m e.g., GeoEye-1, WorldView-3). For each application, decisions are made on which elevation data to use driven by cost, resolution and accuracy [

DEM errors are generally categorized as either systematic, blunders or random [

Blunders errors are vertical errors which exceed the maximum absolute error permitted. They caused by misreading contours, transposing numeric values, erroneous correlations, or careless observations during data collection process. After removing systematic and blunders errors, the remaining errors are known as random errors. They result owing to accidental and unknown combinations of causes beyond the control of the observer [

These various errors often produce cell or groups of cell values that are artificially lower/higher in altitude than their surrounding cells [

Since DEM quality is influenced by several factors, e.g., sensor types, algorithm, terrain type, grid spacing and characteristics [

The main objective of this work is to assess the quality of ASTER GDEM (the Advanced Spaceborne Thermal Emission and Reflection Radiometer) and SRTM DEMs (Shuttle Radar Topography Mission) in comparison with in situ DGPS (Differential Global Positioning System) elevation data. This would reveal the impact of DEM resolution upon the accuracy of terrain representation and topographic attributes in a relatively low relief area. The results would be valuable to determine the appropriate DEMs for a particular application, improve topographic attributes and drainage extraction, and contribute to the previous works concern the assessment of ASTER GDEM and SRTM DEMs quality.

The study area lies on the eastern side of the River Nile between latitudes 26˚04'30"N to 26˚13'30"N and longitudes 32˚45'E to 33˚10'E (

~420.5 km^{2} and characterized by relatively low relief and marked by the mouth of Wadi (=dry valleys) El Serai (=mystery) that drains the north-east and south-east highlands (+600 m) and debouching into the Nile Valley (+72 m). This area has been selected because it is a promised area for land reclamation and agriculture development, so it is interspersed by many new villages and infrastructures although it is threatened by flash flood occurrence from the poorly ungauged Wadi El Serai. The area is covered by a great thick of sedimentary rocks (sand stone, shale and calcareous rocks), ranging in age from Cretaceous to Pliocene times. The sedimentary rocks are overlaid with loose and/or consolidated Quaternary deposits (alluvial, sand, gravels and Nile deposits). Land reclamation is based mainly on the shallow aquifers within the Quaternary alluvial sediments (10 - 90 m depth).

In this study, we used different data types and methods of analysis (

ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) GDEM version2 is a product that is generated from a pair of ASTER Level-1A images. This Level-1A input includes bands-3N (nadir) and −3B (after-viewing) from the visible near infra-red (NIR) telescope’s along-track stereo data that is acquired in the spectral range of 0.78 to 0.86 μm. The absolute vertical accuracy of the GDEM-2 mean error is −0.20 meters on average with accuracy of 17

meters at the 95% confidence level, which is a significant improvement compared to the GDEM v1 mean error of −3.69 meters [

The Shuttle Radar Topography Mission (SRTM) used the single-pass Interferometric Synthetic Aperture Radar (InSAR) technique to determine a nearly worldwide height model. The space shuttle carried in two radar antenna combinations

for C-band and X-band. The C-band used the scan-SAR mode with a swath width of 228 km; while the German-Italian X-band was limited to a swath width of 45 km. C-band was preferred for this study because it gives a continuous coverage nearly without gabs, while SRTM X-band has large gaps between the covered areas (1 arc second or ~30 m) and therefore it is not used in this study.

The mission obtained elevation data on a nearly global scale of 90 m ground resolution for SRTM v2 and 30 m for SRTM v3. The elevation data representing the visible land surface (Digital Surface Model) including vegetation and buildings if exist. In this study, SRTM version2 image No. SRTM3N26E032V2 (

GPS (Global Positioning System) is a satellite-based system used to locate positions on the earth surface for 24 hours/day. Differential Global Positioning System (DGPS) in specific is an enhanced GPS system provides differential corrections to GPS receiver based on a static reference station at a known location. Ever since GPS is rapidly adapted for surveying as it provides positions in three dimensions (x, y, z). Data accuracy ranges from few millimeters to few meters (~15 m) depending on equipment and procedures applied to the process of data collection.

In this study, we used Sokkia GRX2 receiver that provides 226 and superior antenna quality with Real Time Kinematic (RTK) system (

^{1}Parts per million: 1 ppm corresponds to a 1 mm error over 1 km.

The base station re-broadcasts the phase of the carrier that it observes (measurements and coordinates) to the mobile receivers. The built-in software in a rover receiver combines and processes the GPS measurements collected at both the base station and the rover receivers to obtain the rover coordinates [^{1} horizontally and 15 mm + 2 ppm vertically. Location accuracy of approximately 50 × 50 meters regular grid was obtained with a vehicle tracker DGPS receiver for 81,454 points covering the study area based on RTK system, with a density of 193.7 points/km^{2}.

ASTER GDEM and SRTM tiles were obtained in a tiff format using geographic coordinate system (GCS) based on the Word Geodetic System 1984 (WGS84) horizontal datum and the EGM96 vertical datum. Tiles projection was then transformed into UTM zone 36 using WGS84 datum as default vertical datum to be in accord with the DGPS data and to compute the relative heights. Additional processing was required for noise reduction such as fill gabs to eliminate random sinks and peaks. Such random noises are a result from the technical procedures in the DEM production processes and highly correlated with coarser resolution DEMs. A sink is a cell surrounded by higher elevation values and therefore it has undefined drainage direction or internal drainage. A peak is a cell that is extremely higher than its neighbors. In this study, the arithmetic mean filter (low pass filter 3 × 3) was applied to all DEMs to smooth datasets and remove possible outliers [

Elevation histograms reveal nonsymmetrical or positive or right skewed distribution with long right tail towards higher elevations. In such case mean and median are not equal. Skewness ranges between 0.8 - 0.72 for ASTER GDEM and SRTM, respectively.

Kurtosis analysis reveals that elevation histograms also have more peaks means that a distribution also has fatter tails because of extreme outcomes compared to a normal distribution. Kurtosis ranges between 5.4 - 5.1 in the same order.

Plotting relative area versus relative height (hypsometric representation) manifests that relative low lands are the most dominant within the basin. They constitute 91.7% of the basin area, range in elevation between 80 - 320 m (a.s.l) and only 8.3% of the basin area ranges in elevation between 320 - 660 m (a.s.l). This confirms that the elevation data reveals positive skewed since it tends to concentrate towards the lower values reflecting old stage of denudation since the area below the hypsometric curve (hypsometric integral = HI) is close to zero and HI equals 0.36 [

Quantifying vertical accuracy was normally achieved by comparing the DGPS points to the nearest SRTM/ASTER GDEM Pixel value. For the purposes of model validation and statistical analysis, stratified randomly 100 DGPS elevation points were selected with their explicit x and y coordinates. Multi-value extraction tool in ArcGIS was used to extract cell values of the specified DGPS points from the ASTER GDEM and SRTM DEMs. Therefore, cell values were extracted for each input raster and a new field contains the cell values for each input raster was appended to the input point feature class.

Then DEM validation was represented by using statistical measurements namely elevation error (Z_{dif}), mean error (ME), Standard deviation error (STD_{err}), Root mean square error (RMSE) (

Quantile-Quantile plots (Q-Q plots) based on the normal distribution are created for visual examination. The Q-Q plots of the reference data versus DEMs depict that the DEMs data are greatly undulated around the best fit linear relationship. That means DEMs data are mostly higher than the corresponding reference GPS data and to some extent other point samples are relatively lower than the reference data (^{2} = 0.74).

Effects of random errors in DEM for terrain analysis have been investigated using analytical and numerical error propagation techniques. Among the earliest works in DEM error propagation analysis, solutions for calculating standard

Statistical Method | Description | Equation No. |
---|---|---|

Elevation Error | Z d i f = Z D E M − Z G P S | Equation (1) |

Mean Error | M E = ∑ i = 1 n Z d i f ( i ) n | Equation (2) |

Standard Deviation | S T D e r r = ± ∑ i = 1 n ( Z d i f ( i ) − M E ) 2 n − 1 | Equation (3) |

Root mean square error | R M S E = ∑ i = 1 n ( Z d i f ( i ) ) 2 n | Equation (4) |

(After: [

DEMs | Z_{dif} | ME | STD | RMSE | ||
---|---|---|---|---|---|---|

Max. | Min. | Avg. | ||||

ASTER | 259.5 | −39.8 | 27 | 27.0 | 25 | 75.8 |

SRTM | 264.7 | −33.2 | 34.2 | 34.2 | 23 | 79.2 |

deviation of slope and aspect was represented by [

Standard error propagation (SE) of a partial derivative (∂Z) height(s) quantizes propagation errors of the partial derivative ∂Z_{DEM} from the remotely sensed DEMs with respect to the partial derivative ∂Z_{ref} from the reference elevation data, where:

∂ ( Z ) = ∂ ( Z D E M ) 2 + ∂ ( Z r e f ) 2 (5)

where (∂Z) represents the standard deviation of Z. SE assumes that the quantities a, b, etc. have uncorrelated and random errors and errors assumed to be spatially uniform. There is a very simple relationship between STD_{err} and SE which can be expressed as:

S E = S T D e r r n (6)

where n is number of sampled reference data (100). Linear geostatistical analysis of ASTER GDEM and SRTM revealed that both DEMs are highly correlated (r^{2} = 99.62) to each other (

Results reveal that the standard deviation of the vertical error has a strong effect on error propagation and this highly obvious in higher elevations. Error visualization shown in

The slope is derived from the first partial derivatives based on the average neighborhood approach (ANS). With the ANS method the slope is estimated by calculating the rate of change in elevation over the distance from the central cell to its eight neighbors using an average maximum approach [

directions (δZ/δY) and the horizontal direction (δZ/δX) from the central cell [

θ deg = atan ( Δ Z Δ X ) 2 + ( Δ Z Δ X ) 2 × 180 p i (7)

where ΔX and ΔY specify the cell dimensions. The lower the slope value the flatter the terrain; the higher the slope value the steeper the terrain.

DEMs | Slope | STD | RMSE | |
---|---|---|---|---|

Max. | Avg. | |||

ASTER | 58.5˚ | 6.1˚ | 6.2˚ | 7.56˚ |

SRTM | 60˚ | 4.9˚ | 5.66˚ | 7.22˚ |

DEMs | Max. | Mean | STD |
---|---|---|---|

ASTER | 359.9˚ | 182.7˚ | 102.6˚ |

SRTM | 359.9˚ | 187.5˚ | 97.2˚ |

that the overall RMSE of slope is 7.56˚ and 7.22˚ for ASTER GDEM and SRTM, respectively. Slope maps are presented in

Aspect identifies the slope direction or direction of the maximum rate of change elevation from each cell to its neighbors. The values of the output raster will be the compass direction of the aspect.

Magnitude of the DEM errors affects the results of topographic first partial derivatives (slope and aspect).

Analytical hillshading is used to visualize topography as shaded relief. It helps us to immediately determine major topographic reliefs such as ravines, ridges, peaks or valley. Hillshade considers both local illumination angles and shadows. The illumination value for each raster cell is determined by its orientation to the light source. Analytical hillshade was estimated using 315˚ sun azimuth and 45˚ sun elevation. Results were calculated in radians based on slope and aspect (in radians), and then converted into degrees for better interpretation. The overall STD ranges between 17.7˚ - 15.2˚ for ASTER GDEM and SRTM DEMs. It is noticeable that the calculated hillshade map from SRTM is less influenced by error propagation than the ASTER GDEM one (

ArcGIS Hydro-model was used in this study to automatically extract the drainage

DEMs | Min. | Max. | Mean | STD |
---|---|---|---|---|

ASTER | 0.26˚ | 83˚ | 49.1˚ | 17.7˚ |

SRTM | 0.17˚ | 76˚ | 48.52˚ | 15.2˚ |

network and drainage basin. Drainage extraction was done by determining the directions that water will flow out of each cell to its steepest down-slope neighbor (flow direction) relying on the 8D method [

Investigation of delineated stream networks depicts that automatic drainage basin delineation is very sensitive to DEM uncertainty. This is reflected on many parameters such as flow direction (

Comparison of the spatial patterns of the two extracted drainage networks from ASTER GDEM and SRTM DEMs (^{2} and stream frequency is 6.06 - 5.64 stream/km^{2} for ASTER GDEM and SRTM, respectively.

In the lower part of the drainage basin where relative low lands are dominant, there is unrealistic stream junctions and great changes in stream courses. By zooming into the stream map we observed that some streams are truncated and isolated. Others reveal artificial breaks since they follow the gridded-structure of the DEMs [

DEMs | Min. | Max. | Mean | STD |
---|---|---|---|---|

ASTER | east | northeast | 23.5˚ | 32.48 |

SRTM | east | northeast | 22.5 | 31.9 |

Orders DEMs | 1 | 2 | 3 | 4 | 5 | 6 | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

N | L | N | L | N | L | N | L | N | L | N | L | |

ASTER | 1304 | 580.7 | 641 | 296.5 | 309 | 132.7 | 207 | 92.7 | 89 | 36.4 | 1 | 11.9 |

SRTM | 1218 | 554.6 | 558 | 269.8 | 312 | 141.4 | 207 | 98.9 | 77 | 34.4 | 1 | 13.8 |

Differences | 86 | 26.1 | 83 | 26.7 | −12 | −8.7 | 0 | −6.2 | 12 | 2 | 0 | −1.9 |

N = stream number; L = stream length [km].

In this study, we assessed the quality of ASTER GDEM (30 m) and SRTM (90 m) in comparison with in situ DGPS elevation data to reveal its influence on the first derivatives topographic attributes (slope, aspect), topographic representation (hillshading), and the unconstrained derivatives stream networks in a relatively low relief area. Quantifying vertical accuracy is normally achieved by selecting a sample of in situ reference measurements in comparison with the original DEMs data. The vertical accuracy of the two DEMs was calculated using various statistical measurements. The overall absolute vertical accuracy (RMSE) was 75.8 for ASTER GDEM and 79.2 for SRTM (

the normal distribution reveal that the DEMs data are greatly undulated around the best fit linear relationship (

Although, linear geostatistical analysis of ASTER GDEM and SRTM revealed that both DEMs are highly correlated (r^{2} = 99.62) to each other, there are slight variations between the two DEMs on the basis of topographic attributes. The absolute standard error (SE) of the DEMs ranges generally between is ±0 - 2.5 and ±0 - 2.4 m for ASTER GDEM and SRTM respectively (

However, the overall RMSE of slope is 7.56˚ for ASTER GDEM and 7.22˚ for SRTM DEM. STD of slope direction is 102.6 for ASTER GDEM and 97.2 for SRTM DEM. Hillshade map extracted from SRTM is less influenced by error propagation than the ASTER GDEM. In brief, it can be concluded that SRTM DEM is relatively closer, accurate and consistent to the reference elevation data in terms of absolute accuracy than ASTER GDEM. On basis of the topographic attributes, there are slight variations between the ASTER GDEM and SRTM DEM.

Moawad, M.B. and El Aziz, A.O.A. (2018) Assessment of Remotely Sensed Digital Elevation Models (DEMs) Compared with DGPS Elevation Data and Its Influence on Topographic Attributes. Advances in Remote Sensing, 7, 144-162. https://doi.org/10.4236/ars.2018.72010