Different digital elevation model datasets agree well in altitude but significantly differ in their derivative parameters, which introduces significant errors in the estimation of surface-received solar radiation and data from river networks.
A digital elevation model (DEM) is a virtual representation of landforms, describing the surface elevation. As an important input parameter, DEM is widely used in hydrological (Liou et al. 2013), meteorological (Gu et al. 2012), and climatic modeling (Lee et al. 2011). In these applications, both the absolute values of surface elevations and some derivative parameters are used (Rayburg et al. 2009). For example, height and location of a mountain play important roles in airflow (Liang et al. 2005) and precipitation (Houze 2012). The slope and aspect of complex terrain impact both surface-received solar radiation (Li et al. 1999; Liou et al. 2007; Ryu et al. 2008; Yang et al. 2008) and surface emitted longwave radiation. Additionally, the slope and river network derived from DEM determine the distribution of precipitation, runoffs, evapotranspiration, and the recharge of groundwater (Walker and Willgoose 1999; Wechsler 2007).
DEM data can be derived from field surveys and airborne remote sensing techniques (Toutin 2001; Holland et al. 2006; Kornus et al. 2006). Currently, a number of global DEM datasets are widely used, such as the global 30 arc s elevation dataset (GTOPO30), the Global Land One-km Base Elevation Project (GLOBE), the Shuttle Radar Topography Mission (SRTM3), the Altimeter Corrected Elevations 2 (ACE2), the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM), and the Global Multi-Resolution Terrain Elevation Data 2010 (GMTED2010). These six datasets have been produced individually based on different data sources and have different spatial resolutions (see Table 1 for detailed information). However, impacts of DEM uncertainty on the applications remain unknown. More importantly, we demonstrate in this paper that the derivative parameters of DEM data, such as slope and aspect of the terrain, have important differences in spite of the fact that the absolute height values of the six DEM datasets agree very well. Using both the surface-received solar radiation and the extracted river network as examples, we demonstrate the levels of impacts of DEM uncertainty on the calculated surface-received solar radiation, as well as the extracted drainage river network, are introduced with different DEM datasets.
Basic information of the six global DEM data used in this study.
RESEARCH AREA.
It is difficult and unnecessary to compare the DEM datasets and their derivative parameters globally (Mukherjee et al. 2013). To highlight the differences among the datasets, we selected a research area with a size of 6° × 6° that includes both mountains and plains in Sichuan Province in southwest China (Fig. 1b). It has a complex topography contributing to radiation analysis, high in the west and lower in the east, with a standard deviation of approximately 1540 m in height. Compared with flat regions, the surface-received solar radiation over a complex terrain is effected by the slopes, aspects, and sky view factor. In most applications, these parameters can be calculated from DEM datasets.
(a) The location of the study areas; (b) the study area around Chuanxi Plateau and Chengdu Plain for calculating the surface-received solar radiation; and (c) the Pearl River basin used to extract the river networks. The landscape from high to low is colored from green to purple in (a) and the details of the topography for the two study areas are shown in (b),(c) (units in meters).
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
(a) The location of the study areas; (b) the study area around Chuanxi Plateau and Chengdu Plain for calculating the surface-received solar radiation; and (c) the Pearl River basin used to extract the river networks. The landscape from high to low is colored from green to purple in (a) and the details of the topography for the two study areas are shown in (b),(c) (units in meters).
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
(a) The location of the study areas; (b) the study area around Chuanxi Plateau and Chengdu Plain for calculating the surface-received solar radiation; and (c) the Pearl River basin used to extract the river networks. The landscape from high to low is colored from green to purple in (a) and the details of the topography for the two study areas are shown in (b),(c) (units in meters).
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
The Pearl River basin was selected to extract river networks, with a valley area of 440,000 km2 (Fig. 1c). The basin is located south of the Five Ridges, north of the South China Sea. In the basin, the terrain runs northwest to southeast, with mountains of altitudes between 1000 and 1500 m covering over 60% of the area, hills covering over 20%, and plains composing the remaining small occupying areas. The complex terrain is beneficial for river network comparisons between different DEM data. In addition, the study area is sufficiently typical, lest it should take too much operating time or be too inconvenient to analyze the differences of the extracted channels because of exceedingly small features.
DATA.
Six popular global DEM datasets with plausible vertical accuracies are used in this article: GTOPO30, GLOBE, SRTM3, ACE2, GDEM, and GMTED2010 (see Table 1). The spatial resolution of SRTM3 is 3 arc s (3") and of GDEM is 1". They are averaged into 30" to be consistent with the spatial resolution of the other DEM datasets, which is approximately 1 km on the ground. The World Geodetic System 1984 (WGS84) is the common horizontal coordinate system. The vertical coordinate system of GTOPO30 and GLOBE is the mean sea level (MSL), while the remaining four datasets use the Earth Gravitational Model 1996 (EGM96). It can be seen from Table 1 that the six global DEM data have different vertical accuracies and sources in our study area: GLOBE has nearly the same data forms composing the dataset as those of GTOPO30. Sources of SRTM3 and GMTED2010 are all SRTM data, albeit in different versions. The main sources of ACE2 are also SRTM data but warped with some radar data.
All six DEM datasets were used for assessing the impacts of data accuracy on the estimated surface-received solar radiation. Only the original SRTM3 (3") and GDEM (1") were utilized to extract the river networks, because the extracted channels could become much shorter when the spatial resolution of DEM data is too coarse (Chen et al. 2012).
METHODOLOGY: SOLAR RADIATION.
There have been many studies on the methods of calculating the surface total incoming solar radiation. The results of such studies indicate that the solar radiation is estimated by using DEM data as well as considering topographical factors and atmospheric conditions. Many methods have been developed to estimate the impact of the rugged surface on the surface-received solar radiation (Wang et al. 2005; Liou et al. 2007; Tian et al. 2007). In particular, Lee et al. (2011) developed a three-dimensional (3D) radiative transfer parameterization over mountain areas based on the Monte Carlo approach (Liou et al. 2007) for application to climate models, which has been implemented in a regional climate model in Gu et al. (2012), and further studies with longer-term simulations of hydrological model were reported in Liou et al. (2013). In this paper, we calculated the surface-received solar radiation over complex terrain following Wang et al. (2005). The method is based on conventional parameterization and slope, aspect, and sky view factor, which are calculated from DEM datasets.
A photo shows the impact of complex terrain on the surface-received solar radiation.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
A photo shows the impact of complex terrain on the surface-received solar radiation.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
A photo shows the impact of complex terrain on the surface-received solar radiation.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
METHODOLOGY: RIVER NETWORK.
The extracted river channels using DEM tremendously influence the outputs of surface runoffs and evapotranspiration. As a result, DEM play a decisive role to make the extracted river network realistic. Some geographic information system (GIS) software packages perform well in extracting drainage characteristics, such as the ArcView hydrology module and the U.S. Department of Agriculture (USDA) Topographic Parameterization Software (TOPAZ). In this study, Arc/Info Arc Hydro Tools were used.
RESULTS.
The absolute height values of the six DEM data are in excellent agreement with each other, with linear correlation coefficients surpassing 0.98 and without obvious bias (Fig. 3). The agreements among slopes derived from the six DEM datasets, as shown in Fig. 4, are much worse than those of the absolute height values. ACE2 has the best linear relationship with GTOPO30, far better than that with the others, whose correlation coefficients are less than 0.82 (Fig. 4a); Fig. 4e shows the analogous results of GTOPO30. Linear correlation coefficients between GMTED2010 and the other five datasets are all around 0.80 but have the closest relationship with GLOBE (Fig. 4b); Fig. 4d shows the analogous results of GLOBE. SRTM3 has a significantly close relationship with GDEM with a liner correlation coefficient of 0.99, far better than that of the others at approximately 0.80 (Fig. 4c); Fig. 4f shows the analogous results of GDEM.
Comparisons between each pair of the six DEM datasets in absolute height values; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in altitude.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Comparisons between each pair of the six DEM datasets in absolute height values; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in altitude.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Comparisons between each pair of the six DEM datasets in absolute height values; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in altitude.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Comparisons between each pair of the six DEM datasets in calculated slope values; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in slope.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Comparisons between each pair of the six DEM datasets in calculated slope values; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in slope.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Comparisons between each pair of the six DEM datasets in calculated slope values; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in slope.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
These substantial differences in slope introduce significant differences in the calculated relative solar radiation deviation using the six DEM datasets (Fig. 5). The agreements between the relative solar radiation deviations calculated from the six DEMs are even worse than those of the DEM slopes shown in Fig. 4. Among them, there are similar relationships as the slopes indicate. The correlation between GLOBE and GMTED2010 is the best while each of them has reasonable agreements with the other five DEMs. SRTM3 and GDEM agree very well with each other but have much worse correlations with the other four DEMs. It is also the case for ACE2 and GTOPO30.
Comparisons between each pair of the six DEM datasets in relative radiation deviation [Eq. (1)]; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in solar radiation.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Comparisons between each pair of the six DEM datasets in relative radiation deviation [Eq. (1)]; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in solar radiation.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Comparisons between each pair of the six DEM datasets in relative radiation deviation [Eq. (1)]; r denotes the linear correlation coefficients, with values closer to 1.0 indicating a stronger linear relationship in solar radiation.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
Overall, the extracted river networks of the Pearl River basin by GDEM and SRTM3 are nearly the same; the main channels are very close to the actual streamflow network obtained from the Chinese Ministry of Water Resources (Fig. 6a). The river length extracted by GDEM is approximately 1926 km while it is 1910 km for SRTM3 (the real Pearl River length is 2214 km (Wu et al., 2014). However, there are also obvious differences between the two extracted river networks in details. It can be concluded from the red rectangles in the Fig. 6 that the tributaries of SRTM3 extracted river networks are remarkably different from those of GDEM in some details, all shifting slightly to the actual channels. Besides, the DEM extraction results are all poor because of the extraction methods in some particular areas: for example, wide river channels (Fig. 6b) and loop waters (Fig. 6c).
The final extracted river networks by GDEM (in pink) and SRTM3 (in blue) in the Pearl River basin, with the actual streamflow shown in green. (a) Comparison of the entire drainage extraction results, with the river networks extracted by the two DEM data being nearly identical and very similar to the actual channels. However, it can be stated from the red rectangles that the tributaries of the SRTM3 extracted river networks are obviously different from those of GDEM in some details. (b) The significant differences in wide river channel extractions. The trends of the three colored lines are all distinct, which indicates that river networks extracted by DEM poorly represent the actual wide river channels, but channels extracted by SRTM3 are a little more accurate than those of GDEM. (c) Zoomed-in view to show the poor extraction results in loop waters.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
The final extracted river networks by GDEM (in pink) and SRTM3 (in blue) in the Pearl River basin, with the actual streamflow shown in green. (a) Comparison of the entire drainage extraction results, with the river networks extracted by the two DEM data being nearly identical and very similar to the actual channels. However, it can be stated from the red rectangles that the tributaries of the SRTM3 extracted river networks are obviously different from those of GDEM in some details. (b) The significant differences in wide river channel extractions. The trends of the three colored lines are all distinct, which indicates that river networks extracted by DEM poorly represent the actual wide river channels, but channels extracted by SRTM3 are a little more accurate than those of GDEM. (c) Zoomed-in view to show the poor extraction results in loop waters.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
The final extracted river networks by GDEM (in pink) and SRTM3 (in blue) in the Pearl River basin, with the actual streamflow shown in green. (a) Comparison of the entire drainage extraction results, with the river networks extracted by the two DEM data being nearly identical and very similar to the actual channels. However, it can be stated from the red rectangles that the tributaries of the SRTM3 extracted river networks are obviously different from those of GDEM in some details. (b) The significant differences in wide river channel extractions. The trends of the three colored lines are all distinct, which indicates that river networks extracted by DEM poorly represent the actual wide river channels, but channels extracted by SRTM3 are a little more accurate than those of GDEM. (c) Zoomed-in view to show the poor extraction results in loop waters.
Citation: Bulletin of the American Meteorological Society 96, 2; 10.1175/BAMS-D-13-00285.1
DISCUSSION AND CONCLUSIONS.
DEM is widely used in hydrological, meteorological, and climatic modeling. In these applications, both the absolute values of the surface elevations and some derivative parameters are investigated. Our results indicate that, although the DEM data themselves are nearly the same, impacts of the uncertainty on DEM derivative parameters could be significant.
The solar radiation over a complex terrain calculated from GDEM and SRTM3 are the most highly correlated, and they are nearly the same. GTOPO30 and ACE2 data show the second strongest linear relationship among the six datasets. It is amazing that, by only considering the data sources (see Table 1), the most similar group of GLOBE and GTOPO30 is less related compared with GTOPO30 and ACE2. SRTM data in different versions are the principal input of SRTM3, ACE2, and GMTED2010, but the radiation results of the three are not closer than any other DEMs.
The slope and aspect of the rugged surfaces are the key parameters impacting the surface-received solar radiation, which are determined by the spatial variation of the height rather than its absolute value. It is therefore very important to effectively use the spatial texture information to convert original data to DEM. This explains the reason that different data versions have significant different DEM derivative parameters and the calculated surface-received solar radiation.
The length of the river extracted by DEM grows obviously shorter when the DEM spatial resolution becomes coarser; however, the derived watershed area barely changes (Yi et al. 2007). Therefore, a large error exists in the drainage density extraction results between the 30" resolution DEM data and GDEM and SRTM3. For the highly correlated relationships (Figs. 3–5) and the accurate vertical resolution, the two DEM datasets result in similar river networks, which are in good agreement with the actual streamflows, but there are still remarkable distinctions in some details (Fig. 6).
In summary, although the DEM data themselves are nearly identical, impacts of the uncertainty of the DEM derivative parameters may be significant and thus cannot be ignored in the practical meteorology and climatology applications, such as radiation and river network extraction.
ACKNOWLEDGMENTS
This study is funded by the National Basic Research Program of China (2012CB955302) and the National Natural Science Foundation of China (41175126 and 91337111). We thank Dr. Qian Ma and M. D. Zhijun Lee for their insightful suggestions.
REFERENCES
Chen, Y. M., J. P. Wilson, Q. S. Zhu, and Q. M. Zhou, 2012: Comparison of drainage-constrained methods for DEM generalization. Comput. Geosci., 48, 41–49, doi:10.1016/j.cageo.2012.05.002.
Gu, Y., K. N. Liou, W. L. Lee, and L. R. Leung, 2012: Simulating 3-D radiative transfer effects over the Sierra Nevada Mountains using WRF. Atmos. Chem. Phys., 12, 9965–9976, doi:10.5194/acp-12-9965-2012.
Holland, D. A., D. S. Boyd, and P. Marshall, 2006: Updating topographic mapping in Great Britain using imagery from high-resolution satellite sensors. ISPRS J. Photogramm. Remote Sens., 60, 212–223, doi:10.1016/j.isprsjprs.2006.02.002.
Houze, R. A., 2012: Orographic effects on precipitating clouds. Rev. Geophys., 50, RG101, doi:10.1029/2011RG000365.
Kornus, W., R. Alamus, A. Ruiz, and J. Talaya, 2006: DEM generation from SPOT-5 3-fold along track stereoscopic imagery using autocalibration. ISPRS J. Photogramm. Remote Sens., 60, 147–159, doi:10.1016/j.isprsjprs.2005.12.004.
Lee, W.-L., K. N. Liou, and A. Hall, 2011: Parameterization of solar flux over mountain surfaces for application to climate models. J. Geophys. Res., 116, D01101, doi:10.1029/2010JD014722.
Li, X., G. D. Cheng, X. Z. Chen, and L. Lu, 1999: Modification of solar radiation model over rugged terrain. Chin. Sci. Bull., 44, 1345–1350, doi:10.1007/BF02885977.
Liang, X., Y. Liu, and G. Wu, 2005: The role of land-sea distribution in the formation of the Asian summer monsoon. Geophys. Res. Lett., 32, L03708, doi:10.1029/2004GL021587.
Liou, K. N., W.-L. Lee, and A. Hall, 2007: Radiative transfer in mountains: Application to the Tibetan Plateau. Geophys. Res. Lett., 34, L23809, doi:10.1029/2007GL031762.
Liou, K. N., Y. Gu, L. R. Leung, W. L. Lee, and R. G. Fovell, 2013: A WRF simulation of the impact of 3-D radiative transfer on surface hydrology over the Rocky Mountains and Sierra Nevada. Atmos. Chem. Phys., 13, 11 709–11 721, doi:10.5194/acp-13-11709-2013.
Mukherjee, S., P. K. Joshi, S. Mukherjee, A. Ghosh, R. D. Ghosh, and A. Mukhopadhyay, 2013: Evaluation of vertical accuracy of open source Digital Elevation Model (DEM). Int. J. Appl. Earth Obs., 21, 205–217, doi:10.1016/j.jag.2012.09.004.
Rayburg, S., M. Thoms, and M. Neave, 2009: A comparison of digital elevation models generated from different data sources. Geomorphology, 106, 261–270, doi:10.1016/j.geomorph.2008.11.007.
Ryu, Y., S. Kang, S. K. Moon, and J. Kim, 2008: Evaluation of land surface radiation balance derived from Moderate Resolution Imaging Spectroradiometer (MODIS) over complex terrain and heterogeneous landscape on clear sky days. Agric. For. Meteor., 148, 1538–1552, doi:10.1016/j.agrformet.2008.05.008.
Tian, H., J. Wen, Y. M. Ma, Z. B. Su, Z. G. Wei, J. M. Wang, T. T. Zhang, R. Liu, and X. Hu, 2007: Estimation of solar radiation over complex terrain of the Heihe River basin (in Chinese). Plateau Meteor., 26, 666–676.
Toutin, T., 2001: Elevation modeling from satellite visible and infrared (VIR) data. Int. J. Remote Sens., 22, 1097–1125, doi:10.1080/01431160117862.
Walker, J. P., and G. R. Willgoose, 1999: On the effect of digital elevation model accuracy on hydrology and geomorphology. Water Resour. Res., 35, 2259–2268, doi:10.1029/1999WR900034.
Wang, K. C., X. J. Zhou, J. M. Liu, and M. Sparrow, 2005: Estimating surface solar radiation over complex terrain using moderate-resolution satellite sensor data. Int. J. Remote Sens., 26, 47–58, doi:10.1080/01431160410001735111.
Wechsler, S. P., 2007: Uncertainties associated with digital elevation models for hydrologic applications: A review. Hydrol. Earth Syst. Sci., 11, 1481–1500, doi:10.5194/hess-11-1481-2007.
Wu, C., S. Yang, S. Huang, and S. Wang, 2014: Multi-scale variability of water discharge and sediment load in the Pearl River during 1954–2011. J. Geogr. Sci., 69, 422–432.
Yang, K., R. T. Pinker, Y. Ma, T. Koike, M. M. Wonsick, S. J. Cox, Y. Zhang, and P. Stackhouse, 2008: Evaluation of satellite estimates of downward shortwave radiation over the Tibetan Plateau. J. Geophys. Res., 113, D17204, doi:10.1029/2007JD009736.
Yi, W. H., J. M. Zhang, Y. S. Kuang, and L. N. Mi, 2007: Effect of horizontal resolution on the watershed features derived from DEM (in Chinese). Geogr. Geo-Inf. Sci., 23, 34–38.
