Vertical Accuracy of Digital Elevation Models Based on Differential Global Positioning System

The Digital Elevation Model (DEM) has been known as a quantitative description of the surface of the Earth, which provides essential information about the terrain. DEMs are significant information sources for a number of practical applications that need surface elevation data. The open-source DEM datasets, such as the Advanced Space-borne Thermal Emission and Reflection Radiometer (ASTER), the Shuttle Radar Topography Mission (SRTM), and the Advanced Land Observing Satellite (ALOS) usually have approximately low accuracy and coarser resolution. The errors in many datasets of DEMs have already been generally examined for their importance, where their quality could be affected within different aspects, including the types of sensors, algorithms, terrain types, and other features. Ground control points (GCPs) used in this study were observed through the utilization of differential global positioning system (DGPS) with dual frequencies. Statistical indices were used to compare, evaluate, and validate the DEMs data against DGPS data. Statistical analysis of DEMs pointed out that SRTM accuracy was higher, with Root Mean Square Error (RMSE) of ±6.276m as compared to the other DEMs. ASTER showed the biggest residual error with an RMSE of ±10.241m. Nevertheless, ALOS was noticeably improved by having an RMSE of ±6.988m.


Introduction
The Digital Elevation Model (DEM) has been known as a quantitative description of the surface of the Earth, which provides essential information about the relief or terrain [1]. DEMs are significant information sources for a number of practical applications that need surface elevation data and they have been utilized as sources of elevation data in several geospatial applications and studies, including plant cover research, urban studies, tsunami assessments, geomorphology, glacier observations, topography, and archaeology [2]. They can be classified into two groups: Digital Surface Models (DSMs), which display the surface of the Earth and contain all natural and man-made objects, and Digital Terrain Models (DTMs), where there are no types of objects like trees and buildings [3]. The datasets of DEMs could be generated through the use of different mechanisms, such as techniques of air-borne (e.g. Light Detection and Ranging (LiDAR) and photogrammetric), conventional methods of surveying techniques (e.g. leveling and Global Positioning System (GPS)) and space-borne interferometry (e.g. satellite imagery and Radar altimetry) [4]. These data sources could be assessed in four various ways, namely the pre-processing, accuracy, resolution, and cost. Furthermore, every aforementioned mechanism includes both benefits and drawbacks [5]. Recently, research has always utilized the DEMs, which could be gained by techniques of remote sensing, instead of the mechanisms of direct measurement. This is due to the increasing quantity of satellite observations accompanied by increased spatial and temporal resolution with stereo abilities, homogeneous data quality, and wide coverage [6]. Also, the low production cost of modern DEMs is an additional advantage. DEMs can be seen in a format of raster data, that are arrays of square cells [7].
The open-source DEM datasets, that are freely accessible, such as the ASTER, the SRTM, and ALOS, usually have approximately low accuracy and coarse resolution [8]. The errors in many datasets of DEMs have already been generally examined for their importance, where their quality could be effected within different aspects, including the types of sensors, algorithms, terrain types, grid spacing, and other features [9]. As those of ASTER and SRTM, the data of ALOS are available to everyone. A plethora of studies have been concerned with examining the accuracy through the use of various mechanisms for validation. Nonetheless, the assessment of DEMs' accuracy necessitates a measurement of GPS data, including the regional ground-truth data, with an increased degree of precision, to validate the accuracy of DEM datasets [10].

Related Works
Recently, immense literature have been conducted on the generation of DEM data from images sent by satellite and the evaluation of their quality. Mukherjee et al., 2013, evaluated open-source DEMs (ASTER and SRTM) and their derived features, utilizing DEM high posting Cartosat and height information of Survey of India (SOI). Their results revealed that terrain characteristics' representation has been influenced by DEM coarse postings. The total vertical accuracy analysis revealed that RMS error values of ± 17.76 m and ± 12.62 m for SRTM and ASTER DEMs, respectively, compared with Cartosat DEM [1]. Khalid et al., 2016, made a comparison to evaluate the accuracy of SRTM, ASTER, and GMTED10 by using earth real data from the Global Positioning System (GPS). SRTM vertical accuracy displayed better findings against GMTED10 and ASTER, with an RMSE of ± 6.054 m [11]. Rabah et al., 2017, utilized the ellipsoidal observed heights of 601 points, compared to the ASTER and SRTM DEMs. The findings revealed that the most accurate one was the SRTM; it produced mean height differences and standard deviation values equal to ± 2.89 m and ± 8.65 m, respectively [12]. Elkhrachy, 2018, assessed the quality of DEMs gained by ASTER and SRTM. The levels of reference provided by topographic maps and elevations of GPS were utilized to evaluate the vertical accuracy of ASTAR and SRTM in Najran city, Saudi Arabia. The DGPS elevations showed RMSE values of ± 5.94 m and ± 5.07 m using SRTM and ASTER, respectively. Moreover, the gained accuracy values were ± 7.97 m and ± 6.87 m for ASTER and SRTM, respectively, through the use of elevation from the topographic map as a reference elevation [13]. Wessel et al., 2018, provided a unique assessment of the accuracy of the last TanDEM-X global DEM, utilising reference datasets of GPS point, toward the total characterization of the absolute height error. These comparisons proved a vertical Mean Error (ME) less than ± 0.20 m and an RMSE less than ± 1.4 m for TanDEM-X DEM [14].
This research aims to utilize notions of the Global Navigation Satellite System (GNSS) to validate the open DEMs accuracy. This is required to analyse remote sensing data and confirm that free available data provide a critical possibility for analysing spatial data and perform measurements for geomorphologic and topographic purposes. Study Area and Data A.
Study Area The research was conducted in the southeastern region of Iraq, as shown in Figure-1(A) , located between the geographic coordinates of 31˚ 59' 37.19" -32˚ 59' 31.64" N and 45˚ 05' 09.81" -45˚ 59' 08.22" E. The study area varied in terms of terrain. Geomorphologically, its eastern side was adjoining the mountain range, which represents the border of Iraq -Iran. Therefore, the northern part of the study area was higher than the southern part, which was a level territory with slight variety in heights. The Tigris River passes from the north to the south-east through the area of study.

B. Digital Elevation Model
The National Aeronautics and Space Administration (NASA) and the Japanese Ministry of Economy and Trade have developed the "The Advanced Space borne Thermal Emission and Reflection Radiometer" (ASTER). Its sensors have an along-track stereoscopic ability, through the use of near-infrared spectrum sensors. DEMs are generated with spatial resolution of 30m. The ASTER complete vertical accuracy is 20 m with a confidence level of 95% [ 51 ] . The National Imagery and Mapping Agency (NIMA) and NASA made the "Shuttle Radar Topography Mission" (SRTM) to gather the world elevation dataset. The SRTM data of elevation were taken from the Interferometric Synthetic Aperture Radar (InSAR). Data of SRTM are available at 90 m posting for the whole world and 30 m posting for USA. The data showed that the most complete vertical height accuracy was 16 m [11].
Japan Aerospace Exploration Agency (JAXA) launched the "Advanced Land Observing Satellite" (ALOS) which is a global digital elevation model that started functioning since 2006. ALOS is consisted of three panchromatic radiometers that procured along-track stereo images. The model has a 2.5 m spatial resolution in the nadir-looking radiometer and achieved coverage of the whole globe, producing a suitable potential candidate for exact global generation of digital elevation models [8].

C.
Ground Reference Data In the current research, DGPS survey data were utilized as ground truth data, which is capable of producing the highest accuracy of DEMs on land. As shown in Figure-1(B), the 219 (GCPs) used in this study were observed through the utilization of DGPS with dual frequencies, by using the mechanism of static observation, accompanied by vertical accuracy of 0.015 m+ 2.5 ppm and horizontal accuracy of 0.012 m + 2.5 ppm [12]. These data were gained through field surveys by the companies of oil and gas topographical survey functioning in Wasit district. Table-1  Data Pre-processing The World Geodetic System (WGS84) was used to project digital elevation models and make cuts to the extent of every testing area [16]. Then, the three digital elevation models were selected with ortho-metric heights regarding the computed geoid model, utilizing EGM2008. The elevation value extraction of the digital elevation model was carried out by the use of the Arc-Map Software's export functionality. Nevertheless, the observation data of DGPS provided the default vertical datum for computation purposes of relative heights. By the use of the software of Geodetic Datum Transformation Suite (GDTS), the digital elevation models were transformed into WGS84. After that, DEM elevation was changed to an ortho-metric height. Thus, the geometric height of DGPS was then changed to the orthometric height by subtracting the height of the geoid at the position of every point of DGPS. The fundamental formula for these transformations is [17], as shown in Figure-2: (1) where H is the vertical distance between geoid and topographic surface, h is the vertical distance between ellipsoid and topographic surface, and N is the vertical height of geoid from ellipsoid [17]. This change was made to the whole open-source DEMs which utilized a comparable reference (EGM2008) for vertical datum. The sample of 21 points with its coordinates and DEMs elevations is illustrated in Table-1.

B.
Validation of Data Statistical indices were used to compare, evaluate, and validate the DEMs data against DGPS data. Based on the current statistical research, DEM vertical accuracy relies on error propagation and trends amongst the DGPS and DEMs datasets [18]. The elevation error was estimated for each point as the difference between reference data and those of the model using equation 2 [9] -(2) where E model is DEMs' elevations , E ref is DGPS points' elevations, and E diff is the error of the elevations. After that, the values of RMSE and ME for each model were estimated by utilizing equations 3 and 4.
where n: number of points [9]. RMSE refers to surface quality measures and offers an understanding of the differences between two kinds of data (anticipated by the observed and model data) [19]. The agreement level of derived elevation values among SRTM, ASTER, and ALOS datasets, on one hand, and data of DGPS, on the other hand, was also assessed regarding correlation and linear regression through comparing the value of every DGPS validation point with the digital elevation model.

Results and Discussion
To evaluate the data obtained from open-source digital elevation model's data and ground truth data, 219 points were collected by DGPS and randomly distributed in the selected study area, which represents a flat land, coastal lines, and hilly land. Extracted DEMs elevation values were gained through the use of Arc Map software and validated using the DGPS points.
Nevertheless, regarding the validation of digital elevation models, the most accurate digital elevation model was demanded. Figure-3 illustrates the comparison of data of SRTM, ASTER, and ALOS against the ground-truth data around the study area. The correlation of the digital elevation models was calculated. By plotting the point between the heights of the digital elevation methods (SRTM, ALOS, and ASTER) and the ellipsoid height (DGPS), regression value findings are represented in Figure-  97 relationships among the ortho-metric (DGPS) and DEM heights could be identified by referring to the findings of the regression model. These findings indicate that the strongest and most suitable model was selected as the best technique for the digital elevation model.
The obtained results confirmed the vertical accuracy values for SRTM, ALOS, and ASTER DEMs, which have values of ± 6.276 m, ± 6.988 m, and ± 10.241 m, respectively, for the study area (see Table-2).

Conclusions
Three digital elevation models were compared in terms of vertical accuracy that is calculated through the utilization of DGPS reference data. Eventually, the differences in digital elevation models were discussed based on the assessment of statistical test results. After the removal of systematic and blunders errors, random errors were found to cause the distribution of the remaining errors. Regarding the accuracy of the evaluation data, results of statistical tests were presented and the characteristics of errors were investigated in the three sources of digital elevation models. The overall results showed the correlation of SRTM at 0.905, while ALOS and ASTER showed correlations of 0.764 and 0.427, respectively. Statistical analysis of the digital elevation model pointed out that the accuracy of SRTM was higher, with an RMSE value of ±6.276 m, as compared to the other digital elevation models (Table-2). ASTER showed the highest residual error value, with an RMSE of ±10.241 m. However, ALOS was noticeably improved with an RMSE of ±6.988 m.
Reviewing the study results is critical to comprehend the errors connected with the used digital elevation models. The spatial features of the digital elevation models error were also depicted for different slope, land cover, and terrain morphologies. The vertical accuracy of the digital elevation models is influenced by the features of terrain morphology, where the roughness of the terrain has affected the vertical accuracy in a negative way. Researchers recommend that DEMs are of crucial benefit in small scale regional level studies. Future research can be conducted to measure the vertical accuracy of open source digital elevation models. Moreover, it is recommended to investigate the variation of terrain attributes of the first and second order when computing with SRTM, ALOS, and ASTER DEMs.
Still, all the assessed DEMs are useful and confer a good substitute for those DEMs created based on topographic maps. Acquiring and preparing DEMs for the use at open sources takes less time than the vectorization of topographic maps and their further processing. The availability of high-resolution DEMs in open access and further improvement of their processing algorithms will promote more active use of DEMs in multidisciplinary research.