Introduction

Digital elevation models (DEMs) are among the most important resources used to map and extract hydrological processes and to delineate land components, which are related to natural resource management issues (Reddy et al. 2018; Quiroz Londoño et al. 2016; Mashimbye et al. 2014; Vaze et al. 2010). A DEM is a quantitative model of a portion of surface elevation or another form of a digital terrain model (DTM) in the event that the earth is bare. A DTM is a vector dataset composed of regularly spaced points and natural features, such as ridges and breaklines, also defined as a digital representation of variables relating to a topographic surface. The applications of a DTM or DEM include hydrologic modeling, drainage network management, landslide modeling, urban planning, and flood risk assessment (Reddy et al. 2018; Quiroz Londoño et al. 2016; Ouerghi et al. 2015; Chang et al. 2010; Vadon 2003; Kim and Kang 2001), geographic military applications, and topographic analysis. Moreover, a DEM is useful for understanding the environmental land properties and weathering processes that affect the soil.

There are several types of DEMs, and each one has a different application. For example, aerial photogrammetry is accurate and beneficial for extracting high-resolution DEMs that help to model more detailed earth surface characteristics and processes as well as landslide dynamics and geomorphology (Pulighe and Fava 2013). DEMs constructed from the Shuttle Radar Topography Mission (SRTM) are powerful resources for use in geological studies, geomorphology, monitoring water resources, and evaluation of natural hazards through integrating multi-source remote sensing images; this type of DEM generates a merged image with the advantages of different images (Yang et al. 2011; Kovalchuk et al. 2019). The Advanced Spaceborne Thermal Emission and Reflection Radiometer Global DEM (ASTER GDEM) 2 could be a good resource for topographical and geomorphological information in many scientific applications (Li et al. 2013; Fathy et al. 2019). While Google Earth (GE) data is considered to be a valuable resource for transportation and some engineering applications, it is inadequate for studies that must yield highly accurate results; GE data can also be used to conduct primary studies at a low cost (El-Ashmawy 2016; Wang et al. 2017).

DEMs are significant in many geographical and topographical fields. Surface hydrological modeling is one of the most widely used applications to study channel drainage networks, which can be extracted using DEMs (Reddy et al. 2018; Sahu et al. 2017; Reddy et al. 2004a, b). Furthermore, slope, aspect, topography, hill shade, and water streams can also be derived by a surface elevation model. Therefore, many researchers have been using DEMs to understand and interpret environmental, geographical, and geological issues as free availability of DEM data has increased (Reddy et al. 2018; Mashimbye et al. 2014; Singh et al. 2014; Pareta and Pareta 2011; Van Niekerk 2010; Sameena et al. 2009). DEM-derived techniques have been developed to help interpretation methods to become more accurate and helpful in geomorphological studies, especially in topographic and hydrological parameter analyses. These analyses are affected by two factors: the spatial resolution of elevation and accuracy (Quiroz Londoño et al. 2016; Ludwig and Schneider 2006; del Rosario González-Moradas and Viveen 2020).

Aerial photography, topographical surveys, remote sensing data, and radar interferometry-optical stereo images are used to create DEMs through raster images, which include a value for each pixel. In addition, topographic maps can be used to create DEMs through a grid that is digitized and interpolated using contour lines (Quiroz Londoño et al. 2016; Schumann et al. 2008; Szypuła 2019). In general, different types of DEMs can help to find various errors in derived DEMs or in the findings of DEMs, meaning that the quality of input DEM data determines the performance of a DEM and, therefore, significantly affects its usability.

This study aimed to implement the most common, open, and reachable sources to generate DEMs with different levels of accuracy, such as satellite images, aerial photos, GE data, United States Geological Survey (USGS) data, and national or field survey data. Extracts of some geomorphological features, such as slope, aspect, hill shade, hydrological streams, and watershed networks, were used for all the generated DEMs. Furthermore, an accuracy assessment and an evaluation were conducted on all the generated DEMs. The main objective of this study was to evaluate the methods used to compare elevation data from the SRTM and the ASTER GDEM. An evaluation of the available DEMs and validation of SRTM and ASTER GDEM 2 data were conducted based on the analysis of geomorphological features related to stream networks, hill shade, and watersheds.

Methodology

A large amount of resource data was used to achieve the aims of this study. The findings were compared via statistical analysis and visual comparisons, and these results were validated using reference data. The method utilized for the evaluation of the available digital elevation models (DEMs) for geomorphological features analysis comprises three steps as follows: automated delineation of geomorphological components (hill shade, water drainage geomorphology and watershed) from all types of DEMs using the ArcGIS package and comparison of results between all DEMs types by visual comparisons, linear regression analysis, t test factor, statistical analysis, and the assessment of the relative and absolute accuracy of the results. Scheme (1) summarizes the methods of this study.

Scheme 1
scheme 1

Study methodology

Full size image

Study area

The study area was located in the western part of the city of Al-Balaqa and in the northern part of Jordan, between 367,700 m and 375,200 m east and between 550,945 m and 552,750 m north according to the Jordan Transverse Mercator (JTM) Projection System (Fig. 1). The study area covered approximately 50 km2, was unpopulated, and mainly included hills. According to the reference DEM, the elevation of this area ranged from 205 m below the mean sea level near the King Abdullah Channel (KAC) to 990 m above the mean sea level (Fig. 1). Many springs are located in areas such as Ain Es-Sahen, Ain Sa’dah, and Ain Al-mir’ah; many valleys (wadis) are located in areas such as Es-Sahen, Al-Sa’da, Alnimr, Akrah, Aldaliya, and Al-Sauaf. These wadis flow from northeast to southwest (NE–SW) to reach the KAC.

Fig. 1
figure 1

Location of the study area

Full size image

Data use (DEM data)

Reference data: a triangulated irregular network (TIN) was used as a source of reference data to delineate, evaluate, and analyze DEMs derived from many types of available data (Shingare and Kale 2013; Vivoni et al. 2004). The TIN included vertices and nodes (points), edges (lines), facets (polygons), triangles, and topology. Therefore, the TIN helped to represent the earth’s surface through alternative methods (Chen et al. 2014; Kreveld and Silveira 2011; Zhou et al. 2011; Dæhlen et al. 2001; Vivoni et al. 2004). The TIN was converted to contour data (as reference DEM data) obtained from the Royal Jordanian Geographic Center (RJGC).


Digital aerial photography: aerial photography data was obtained from the RJGC at a high resolution (10 cm) and a scale of 1:25,000. Four images were acquired to cover the entire study area, and a DEM was extracted from digital aerial images using photogrammetry. PCI Geomatica software and the geographic information system (GIS) environment were used to estimate and delineate elevation data and hydrological parameters (Mashimbye et al. 2014; Chen et al. 2014). This type of data is useful in deriving geological structure parameters such as lineaments, fractures, and joints.


SRTM DEM: The National Aeronautics and Space Administration (NASA) provided global DEM data through the SRTM. These data were derived using X-band and C-band Interferometric Synthetic Aperture Radar (InSAR) (Quiroz Londoño et al. 2016; Gorokhovich and Voustianiouk 2006; Rabus et al. 2003). These data are available free of charge from the USGS. SRTM DEMs are highly accurate in that the data has a vertical error of less than 10 and 16 m (Quiroz Londoño et al. 2016; Mashimbye et al. 2014; Mulder et al. 2011; Van Niekerk 2008; Rodriguez et al. 2005; Farr 2000). SRTM elevation has three different resolutions: 30 m, 90 m, and 1 km. SRTM DEMs (30 m) are used in various applications due to their high resolution (Yang et al. 2011; Ludwig and Schneider 2006).


ASTER GDEM: The Japanese Ministry of Economy, Trade, and Industry (METI) and the United States NASA cooperated to develop an ASTER GDEM. The ASTER GDEM (second version) is supposed to have a relative vertical error of less than 17 m and was reported to have a relative circular geolocation error (horizontal error) of less than 71 m (Quiroz Londoño et al. 2016; Mashimbye et al. 2014; Mukherjee et al. 2013; Aster 2011).


Google earth (GE) DEM: GE is a dataset of satellite imagery, aerial photography, and GIS that creates a 3D globe using DEM data collected by NASA (Sharma and Gupta 2014). GE is effective in providing a collection of data for many applications, such as urban planning, disaster management, visualizing the earth’s surface characteristics, and mapping natural resources. Two thousand elevation points were extracted from GE satellite images using the TCX converter tool. A 10-m DEM was generated using the kriging interpolation method with a GIS environment.

Data preparation

A 10-m reference DEM was generated from the reference contour using a 3D analysis tool in ArcGIS 10.4. A 10-m aerial photo DEM was also generated from two aerial photography stereo pairs using Geometrica Ortho-Engine 2015. For finer and easier comparisons, the 30-m SRTM DEM, the 30-m ASTER GDEM 2, and the elevation points were resampled to a 10-m resolution. This was achieved by converting the SRTM DEM and the ASTER GDEM 2 into points and by interpolating new elevation values for the three datasets using the kriging interpolation method in ArcGIS 10.4. All the DEMs were projected to the Universal Transverse Mercator system (UTM Zone 36 N) and clipped with the study area boundary. Table 1 presents univariate statistics, such as minimums (min), maximums (max), means, standard errors (SEs), standard deviations (SDs), kurtosis, and skewness, for different DEMs.

Table 1 Summary of general statistics for every DEM
Full size table

Finally, to show the differences in the derivation of drainage networks, length of stream, watershed, sub-basins, and outlets were calculated from the drainage network derived from input DEMs. Root mean square error (RMSE) was calculated as well to evaluate the accuracies of the DEMs in the study area. The degree of difference between the elevation values of the reference DEM and those of the other DEMs was determined, and an assessment of relative and absolute accuracy was used to assess the strength of the relationship between elevation data extracted from different DEMs.

Results and discussion

Linear regression analysis: the linear regression analysis showed a strong correlation between the elevation values of the reference DEMs and those of the DEMs derived from aerial photos, the ASTER GDEM 2, the SRTM, and GE (Table 2). This correlation was highly significant, with a probability value of P < 0.05. Based on the linear regression analysis of the DEM elevation values, it was found that the geomorphology parameter values derived from aerial photos are more similar to the reference DEM than the geomorphology parameter values derived from the SRTM, ASTER GDEM 2 and GE DEMs, respectively. In generally, DEMs derived from aerial photos showed the best fit with the reference DEM elevation values (R2 = 0.997). The SRTM, ASTER GDEM 2, and GE elevation values also proved to be very close to the reference DEM elevation values (R2 = 0.9813, R2 = 0.9809, and R2 = 0.9783, respectively). A t test analysis was used to evaluate the linear regression analysis, showing that the significance values of the differences between the reference DEM elevation values and the SRTM, ASTER GDEM 2, and GE elevation values were less than 0.05. This means that the differences between these values were statistically significant, whereas there was no statistically significant difference between the reference DEM elevation values and the aerial photo elevation values (significance value of approximately 0.89). These results confirmed that the aerial photo elevation values were best fitted to the reference DEM elevation values.

Table 2 Summary of linear regression analysis results for every DEM
Full size table

Visual comparisons: comparisons between hill shade, topography, and streams are considered to be the significant useful parameters for hydrological process and geomorphological feature analysis, which was used in the present study to evaluate the quality of the true values of hydrological behavior for each DEM. Both Table 3 and the red circles in Fig. 2 show the results of the visual comparisons. The aerial photo, SRTM, and ASTER DEMs showed hill shade in greater detail than the GE DEM.

Table 3 Statistics for the extracted streams of all DEMs
Full size table
Fig. 2
figure 2

Hill shade maps of the five DEMs: a reference (contour), b aerial photo, c SRTM, d ASTER GDEM 2, and e GE (Google Earth)

Full size image

The standard deviation of the extracted stream values of the aerial photo DEM was very close to that of the reference DEM (Table 3). Moreover, the results concerning extracted streams for the reference DEM were very close to those for the aerial photo DEM. The differences in the extracted streams for different DEMs resulted from factors such as the surface area pattern and pixel size of each DEM, as well as technique effects.

A second visual technique superimposed topographic profiles, such as the north-to-south profile (2-km and 10-m sampling distances) from each of the DEMs over 2-m and 10-m sampling distances (Fig. 3). This technique showed the relative relationships between the different DEMs. In this respect, the aerial photo DEM was closest to the reference DEM, followed by the SRTM, ASTER GDEM 2, and GE DEMs, respectively.

Fig. 3
figure 3

Superimposed topographic profiles from the north-to-south profile

Full size image

Figure 4a–e shows five types of DEM data. The aerial photo and SRTM DEMs appeared to be similar in shape and were distinctively different from the land watercourse system generated from the ASTER GDEM 2 and GE DEMs (Tables 4, 5). Closer visual comparison revealed that the aerial photo DEM very effectively identified water drainage geomorphology (i.e., water streams, outlets, and water basins) (Fig. 4a–e). The water drainage geomorphology derived from the SRTM and ASTER GDEM 2 DEMs appeared to mostly coincide with the reference data, whereas that derived from the GE DEM appeared to be somewhat different from the reference data. However, the geomorphological components of the aerial photo and SRTM DEMs were more accurate than those of the other DEMs. Compared to the SRTM DEM, the aerial photo DEM yielded more detailed water drainage systems and incorporated more land surface features (e.g., canals and streams) in certain areas. These differences were due to the way these DEMs were derived. The aerial photo DEM was created from stereo imagery, whereas the SRTM DEM was derived using InSAR with a vertical error of less than 10 and 16 m. Consequently, the aerial photo DEM was more detailed than the SRTM DEM. Although the ASTER GDEM 2 land components appeared to be similar in shape to those of the SRTM DEM, they were less accurate. Based on the data sources, the vertical errors of the SRTM DEM ranged from 10–16 m, while those of the ASTER GDEM 2 reached 71 m. Figure 4a–e shows that the geomorphological components of the GE DEM did not coincide with some significant water channels and streams, because the GE DEM was created using large-scale elevation data fused with the SRTM DEM. These results were consistent with those of previous studies by Gichamo et al. (2012), Shafique et al. (2011), and Mashimbye et al. (2014), who indicated that factors such as image acquisition angle, terrain complexity, and quality of the image affect the quality and accuracy of DEMs. Furthermore, DEMs derived from contours are affected by steep areas and exact features that do not appear in flat terrain (Vaze et al. 2010; Wise 2007; Xie et al. 2003; Mashimbye et al. 2014).

Fig. 4
figure 4

Stream network maps of the five DEMs: a reference (contour), b aerial photo, c SRTM, d ASTER GDEM 2, and e GE (Google earth)

Full size image
Table 4 Statistics for the extracted water drainage geomorphology of all DEMs
Full size table
Table 5 Statistics for the extracted watershed data of all DEMs
Full size table

Hydro-processing results related to the extracted watershed and drainage information from different DEMs are shown in Table 5. These results revealed variable differences between the extracted watershed data of the reference DEM and those of the other DEMs.

Visual comparisons of drainage networks extracted from different DEMs showed that they had the same resolution. Specifically, the streams of the aerial photo DEM were similar to those of the reference DEM, particularly with respect to sub-basins and stream length (km) (Fig. 4; Table 4). Furthermore, the streams of the SRTM and ASTER GDEM 2 DEMs were somewhat close to those of the reference DEM. The drainage network of the GE DEM was not any more accurate than that of the reference DEM. Analysis of the extracted stream networks of all the DEMs indicated that the DEMs were able to extract the main stream threshold values. However, the drainage network derived from the aerial photo DEM was the most accurate, particularly in open areas.

Elevation differences between the reference DEM and other DEMs

The statistical analysis of the entire process of height validation for the study area is summarized in Table 6. The statistical parameters used for this analysis included RMSEs, standard errors, standard deviations, medians, sample variances, minimums, and maximums. These parameters were selected, because the data was extracted using the random points method and because these parameters are helpful for the assessment of data obtained from different sources within a region. For example, standard error and standard deviation clarify the relative difference between the data used and the real ground data, where higher values indicate more spread-out data. A negative value indicates that the height level of the reference data is lower than that of the examined data, and a positive value indicates that the height level of the reference data is higher than that of the examined data. Statistical computation of the absolute vertical accuracy of the aerial photo, SRTM, GE, and ASTER GDEM 2 elevation data yielded values of ± 12.53 m, ± 40.41 m, ± 42.33 m, and ± 43.38 m, respectively (Table 6). The 10-m photo DEM featured a much greater absolute vertical accuracy than the other DEMs.

Table 6 Statistical differences between random points of the reference DEM and those of the other DEMs
Full size table

Assessment of relative and absolute accuracy

The accuracy assessment process was conducted by calculating indices such as standard deviation (SD) and root mean square error (RMSE), which represent absolute and relative accuracies, respectively (Table 4). To describe and compare the elevation distributions of each DEM, the elevations of 2000 random points were extracted from all the examined DEMs and compared with the reference DEM elevations to determine several descriptive statistical measures. All data points were organized along with their respective attributes in a spreadsheet table and analyzed using Excel software (Table 4).

The relative frequency distribution of the height differences between the reference DEM and the other examined DEMs is provided in Fig. 5. The overall range of the error distribution of the aerial photo DEM elevations was narrower than those of the other examined DEMs, indicating that the aerial photo DEM produced the best estimation of the spatial distribution of terrain elevations. The frequencies of positive errors were higher than those of negative errors for the ASTER GDEM 2, SRTM, and GE DEMs, indicating a small positive bias toward the reference elevations for these models.

Fig. 5
figure 5

Error distribution frequencies of a aerial photo, b ASTER GDEM 2, c SRTM, and d Google earth DEMs

Full size image

These differences between the reference DEM and the other DEMs used in this study could be traced back to the techniques used for DEM derivation. Stereo imagery and photogrammetric techniques were used to extract the aerial photo DEM with a resolution of up to 10 m and to extract the ASTER GDEM 2 with a resolution of up to 30 m. The ASTER GDEM 2 was reported to have a relative circular geolocation error (horizontal error) of less than 71 m. On the other hand, radar observations were used to construct the SRTM DEM through X-band and C-band InSAR with a vertical error of less than 10 and 16 m. The GE DEM used data collected by NASA’s SRTM, enabling a 3D view of the entire earth with a vertical error of less than ± 16 m.

Conclusion

Precise extraction and delineation of geomorphological features is one of the most important steps of many applications and studies. The determination of accurate features requires a DEM with a high level of accuracy. The comparison of certain statistical parameters with extracted geomorphological features (e.g., slope, aspect, hill shade, and some drainage parameters) and watershed network composition features (e.g., number of streams, stream lengths, number of watersheds, and watershed areas) revealed many differences between different DEM data sources, which would affect the extracted information if inaccurate DEM data were used. This paper showed that a DEM extracted from aerial photos had a high degree of accuracy and supported a high level of detail in the extraction of many geomorphological features related to many types of research and other applications.

An absolute and relative accuracy analysis revealed that the aerial photo DEM had the highest vertical accuracy,

followed by the SRTM, GE, and ASTER GDEM 2 DEMs, respectively, when 200 points were used as ground control points (GCPs). This means that the quality of the data extracted from aerial photos was similar to that of the reference DEM. However, the present study also showed that the aerial photo DEM was most accurate for drainage network studies. This analysis also indicated that the total stream length was observed at 40.9 km, 39 km, 39 km, and 34.4 km in the aerial photo, SRTM, ASTER GDEM, and GE DEMs, respectively. The total stream length observed in the reference DEM was 41.6 km.

This study used scientific methods to analyze geomorphological features by extracting data from different sources of DEMs. The accuracy of the extracted information was based on the quality of the DEM. This study showed that the aerial photo DEM was more accurate (relative to the reference DEM) than the ASTER GDEM 2, SRTM, and GE DEMs. However, DEM validation showed that some extracted geomorphological features of the examined DEMs, such as streams and watersheds, did not match the extracted data of the reference DEM.