Stellenbosch University Digital Elevation Model (SUDEM) 2013 Edition

Stellenbosch University Digital Elevation Model (SUDEM) 2013 Edition Adriaan van Niekerk Centre for Geographical Analysis Stellenbosch University 25 April 2014

CONTENTS CONTENTS... ii TABLES... iii FIGURES... iii 1 BACKGROUND... 1 3 Elevation data sources... 2 3.1 Contours and spot heights... 3 3.2 Global DEM... 3 3.3 WRC DEM... 3 3.4 Stereo aerial photography... 3 4 SUDEM PRODUCT LEVELS... 4 5 SUDEM LEVELS 1 & 2... 4 5.1 Data used for developing SUDEM 2013... 4 6 INPUT DATA VERIFICATION AND ERROR CORRECTION... 4 7 INTERPOLATION ALGORITHM... 5 8 DEM RESOLUTION... 5 9 ACCURACY ASSESSMENT... 6 9.1 LIMITATIONS & FUTURE IMPROVEMENTS... 9 9.1.1 Contour and spot height errors... 9 9.1.2 Interpolation artefacts... 10 9.1.3 SRTM errors... 10 9.1.4 Hydrological errors... 10 10 SUDEM LEVEL 3... 10 11 SUDEM PRICES... 11 12 REFERENCES... 12

TABLES iii Table 1 SUDEM processing levels and products... 4 Table 2 Vertical error of the Level1 and Level2 SUDEM products compared to the SRTM DEM... 7 Table 3 SUDEM prices (incl. VAT)... 12 FIGURES Figure 1 Areas for which 1:10000 and 1:50000 contour data are available... 2 Figure 2 Location (and number) of reference points used for accuracy assessment... 6 Figure 3 Hillshade of the (a) Level 1 and (b) Level 2 products for an area of moderate terrain indicating the higher detail of the Level 2 product... 8 Figure 4 Hillshades of the (a) SRTM DEM, (b) Level 1, and (c) Level 2 products.... 9 Figure 5 Comparison of SUDEM Level 2 (top) and Level 3 (bottom) products... 11

1 BACKGROUND 1 In an effort to address the need for a high-resolution digital elevation model (DEM) of the Western Cape, the Centre for Geographical Analysis (CGA) developed and released the Western Cape Digital Elevation Model (WCDEM) in 2001 (Van Niekerk 2001). The WCDEM was interpolated from 1:50000 scale (20m vertical interval) contours using the ANUDEM algorithm (Australian National University 2011). Notwithstanding subsequent releases of other DEM covering the Western Cape (e.g. SRTM and GDEM), the WCDEM still remains the DEM-of-choice for many applications. This is mainly due to its relatively high spatial resolution, consistency and accuracy 1. Since the release of the WCDEM, many users have expressed an interest in a similar DEM covering South Africa. The CGA also receives frequent requests for more detailed (i.e. larger-scaled) DEM and have produced many very high resolution (<20m) DEM from 1:10000 scaled contour data (where available) and stereographic aerial and satellite imagery. The idea of combining various sources of elevation data into a single national DEM was, consequently, a logical progression. Experiments showed, however, that incorporating various sources (and scales) of elevation data into existing interpolation algorithms often produce unsatisfactory results. Examples include visible seams at transition zones between differently-scaled sources, artefacts as a result of inconsistencies between data sources, and grid-like textures due to the fusion of data with different sampling resolutions. Another major obstacle in the development of a national DEM was the many attribute and spatial errors that occur in the available digitized contours and spot heights. Attribute errors can have a significant effect on the resulting DEM, as even a 5m offset can significantly alter the terrain characteristics of an area. In some cases these errors are large enough to identify using normal visualization methods (e.g. hillshade, slope, aspect, curvature and other focal operations), but in most cases the errors are in the order of one or two contour intervals and are not easily noticeable. An automated method was consequently developed to detect and correct even the slightest error. In addition, a methodology was developed to fuse elevation data from different sources and scales so that edge effects are minimised. The aim of this document is to give a short overview of the techniques developed for generating the Stellenbosch University DEM (SUDEM) and to provide a product description. 1 An accuracy assessment showed that the WCDEM is significantly more accurate than the SRTM DEM.

3 ELEVATION DATA SOURCES 2 Contours and spot heights shown on the South African 1:50000 topographical maps series still remain a primary source of elevation data as it is the only official data set covering the country. These 20m (vertical interval) contours and spot heights were digitised by Chief Directorate National GeoSpatial Information (CDNGI) and are freely available to the public. Recently, CDNGI has also made available the 1:10000 contours (ranging from 5m to 20m vertical interval) and spot heights, which were digitised from the 1:10000 orthophoto map series. Unfortunately, this data set covers only 43% of South Africa (see Figure 1). In addition, the CDNGI has developed a 25m DEM (also known as the ORT-files ) covering only parts of South Africa. Other sources of elevation data covering South Africa include the 1km GTOPO30 DEM, the 90m Shuttle Radar Topography Mission (SRTM), the 30m Global DEM (GDEM) and various large-scale DEM that were developed for specific projects (usually restricted to very small areas). Figure 1 Areas for which 1:10000 and 1:50000 contour data are available

3 3.1 Contours and spot heights Contours are not ideal for interpolating DEM as their densities vary with slope gradient. Areas of low relief are particularly problematic as contours are often spaced far apart (horizontally) reducing the reliability of interpolations in such areas. Contour density is further reduced as the vertical interval of the contours increase (i.e. contours with a 5m vertical interval generally produce better DEM than contours with a 20m vertical interval) and as scale increases (i.e. contours with a 20m vertical interval captured at 1:10000 scale usually contain more detail than contours with the same vertical interval captured at 1:50000 scale). To alleviate the problem of low contour densities in areas of moderate terrain, additional spot heights are often shown at strategic locations on topographical and orthophoto maps. Although, the quality of a DEM can be improved by incorporating these elevation points in the interpolation process, the combined density of input points (i.e. contour vertices and spot heights) is often insufficient to represent subtle changes in terrain (e.g. floodplains and river banks), particularly in some areas where input points can be several kilometres apart. 3.2 Global DEM Global DEM (i.e. GTOPO30 DEM, SRTM DEM and GDEM) are generally considered to be unsuitable for some applications (e.g. flood modelling, geomorphometry, civil engineering) due to their relatively low resolutions (30m or less) and quality. For example, the ASTER-GDEM contains anomalies such as residual cloud patters and stripe effects (Hirt et al. 2010), while the SRTM DEM contains areas with no elevation information (i.e. voids). The vertical accuracy of the SRTM DEM is, however, relatively high (~6m), which makes it an attractive source for regional applications (Rodriguez et al. 2005). 3.3 WRC DEM The Water Research Commission (WRC) recently appointed Agricultural Research Council (ARC) to develop a hydrologically improved DEM of South Africa based on the SRTM DEM. This 30m resolution DEM was interpolated from the 90m SRTM DEM. The SRTM voids were filled with elevation values interpolated from 20m (1:50000 scale) vertical interval contours. The resulting DEM was hydrologically corrected by filling sinks and depressions. It should be noted that, in contrast to the SUDEM, contour data was not used to interpolate the WRC DEM outside of SRTM DEM voids. The SRTM data was consequently the main source of information for developing the WRC DEM and should consequently have a similar vertical accuracy and level of detail as the SRTM DEM (see Section 5.5). 3.4 Stereo aerial photography Stereo aerial photography and photogrammetry is the source of most contour and spot height data. The CDNGI has been acquiring very high resolution (0.5m) aerial photography since 2010. This data can be exploited to automatically extract elevation data at very high (<5m) resolutions.

4 4 SUDEM PRODUCT LEVELS The 2013 edition of the SUDEM includes three products that involve various levels of processing. A summary of these products is provided in Table 1. Table 1 SUDEM processing levels and products Level Description 1 Only contours and spot heights are used in the interpolation process. No hydrological corrections were carried out. 2 Contours and spot heights are supplemented with SRTM data in areas with low to moderate terrain. No hydrological corrections were carried out. 3 Elevations are extracted from 0.5m stereo aerial imagery. Level 1 & 2 is used only where extraction failures occurred (e.g. shadows). Details about the different products are provided in the next sections. 5 SUDEM LEVELS 1 & 2 5.1 Data used for developing SUDEM 2013 Three sets of input data were used to develop the 2013 edition of SUDEM. Preference was given to large scale (i.e. 1:10000) contours and spot heights, while smaller scale (i.e. 1:50000) data were only used in areas where large scale data were unavailable. Two products were developed from this data. The Level1 product was interpolated using only contours and elevation points, while the Level2 product combines contours and elevation points with the SRTM DEM. Our experiments showed that DEM quality in areas of moderate terrain can be significantly improved when the SRTM DEM ( research-grade version) is used in combination with contours and elevation points in relatively flat areas. The ASTER-GDEM was also considered for this purpose, but our accuracy assessments indicated significant deviations from our reference data. This observation is supported by others who have found that the SRTM DEM is superior to ASTER-GDEM. The 25m CDNGI DEM (also known as the CDSM ORT-files ) will likely be incorporated in future editions of the SUDEM, but only once verifications and corrections have been completed. 5.2 Input data verification and error correction A large proportion of the effort in developing the SUDEM was related to data verification and error correction. The main problems experienced with the input data were related to (1) attribute errors in the digitized contours and spot heights; (2) spatial errors such as gaps and mismatching contours at the edges of map sheets; and (3) voids in the SRTM DEM. This section provides an overview of the procedures followed to correct these errors.

5 Attribute errors refer to cases where the elevations stored in the HEIGHT field of contours and spot heights were incorrectly captured from the original maps. An algorithm was developed and implemented to identify and correct such errors. The algorithm examines vertical profiles (cross sections) created at regular intervals (determined by the extent of the smallest contour) within a specified area to find errors. Each profile is normalised (i.e. the horizontal distance between contours are unified) and tested against a set of topological rules. The algorithm not only identifies incorrect contours (or sequences of contours) but also corrects errors by examining each profile. These corrections are then verified by an operator. About 1% (2926 of 3479217) of the contours that were used as input to the SUDEM interpolation required attribute corrections. Hundreds of spatial edits were required. Spot heights that are likely incorrect were identified by comparing their heights to the height of its closest (corrected) contour. If the absolute height difference is more than twice the vertical interval of the contour, then the spot height was labelled as likely incorrect. These points were excluded from the interpolation process. Voids in the SRTM DEM were filled using elevation values interpolated from the corrected contours and spot heights. A similar procedure was used to remove elevation spikes in the SRTM DEM. 5.3 Interpolation algorithm The SUDEM was developed using a combination of algorithms. The ANUDEM algorithm was used for interpolating a DEM from contours and spot heights. This DEM, called the Level1 product, was employed to identify and correct the errors in the SRTM DEM (i.e. voids and spikes). Once corrected, the SRTM DEM was fused with the Level1 DEM using a newly-developed algorithm which ensures that the SRTM DEM is only applied in areas with low densities of contours and spot heights. Although it is recognised that the SRTM DEM is not a true DEM 2, the fusion procedure reduces the effect of surface objects. 5.4 Dem resolution Hengl (2006) suggests the use of Equation 1 to calculate the appropriate cell size when interpolating DEM from contours. When applying Equation 1 on contours of various intervals and scales, and in various types of terrains within South Africa, the optimal resolution varied between 5m and 50m. Consequently, it was decided to produce the DEM at a 5m resolution to ensure that no topographical variation is lost as a result of cell size. Producing the SUDEM at 5m resolution will also enable the incorporation of other DEM (e.g. those that were created using stereo images and LiDAR) in the future. 2 The SRTM DEM was developed using C-band radar technology. Objects on the ground (e.g. buildings) are consequently included in the signal, which results in a digital surface model (DSM) instead of a digital terrain model (DTM).

6 A p 2 l where p is the pixel size; A is the total size of the study area; and l represents contour length. Equation 1 5.5 Accuracy assessment To determine the accuracy of the SUDEM, elevation values were systematically compared to reference elevations. Highly accurate (centimetre) LiDAR points, obtained from Southern Mapping, were used as reference data. A total of 177 reference points were randomly selected from a range of LiDAR campaigns covering diverse terrain types and geographical areas in South Africa (see Figure 2). Figure 2 Location (and number) of reference points used for accuracy assessment To determine vertical accuracy, the mean absolute error (MAE) and root mean square error (RMSE) for the Level 1 and Level2 products were calculated using Equations 1 and 2 respectively (Bolstad & Smith). The results of the accuracy assessments are summarized in Table 2.

7 MAE xi x j n where MAE is the mean absolute error; Equation 1 x i x j n is the DEM s elevation value; is the reference point s elevation value; and is the number of reference points. RMSE x x i n j 2 where RMSE is the root mean square error; x i x j n is the DEM s elevation value; is the reference point s elevation value; and is the number of reference points. Equation 2 Table 2 Vertical error of the Level1 and Level2 SUDEM products compared to the SRTM DEM PRODUCT MAE (m) RMSE (m) SRTM DEM 4.6 43.4 SUDEM Level1 2.1 10.1 SUDEM Level 2 2.2 10.2 It is clear from the results that the SUDEM products performed significantly better than the SRTM DEM in terms of MAE and RMSE. Although the Level 2 product has a slightly higher elevation error than the Level 1 product, it seems to be a good compromise between vertical accuracy and level of detail, particularly in flat terrain where contours are sparse. This was confirmed by a comprehensive qualitative assessment (visual interpretation) where the Level 2 product showed more detail than the Level 1 product in areas with moderate terrain. An example where this is apparent is provided in Figure 3.

8 (a) (b) Figure 3 Hillshade of the (a) Level 1 and (b) Level 2 products for an area of moderate terrain indicating the higher detail of the Level 2 product. The qualitative assessment also revealed that some of the artefacts (e.g. banding, tiger stripes, and wave effects), that are frequently present in contour-interpolated DEM (and visible in the Level 1 product), are reduced when the SRTM fusion is performed (see Figure 4). Additional methods to further reduce these artefacts are currently in development. From the quantitative and qualitative assessments it is clear that the SUDEM products are significantly more accurate than the SRTM DEM. The higher resolution of the SUDEM also enables the inclusion of more terrain detail. There is, however, much room for improvement and it is our intention to further enhance the SUDEM by incorporating other sources of information, including the latest version of the ASTER DEM, CD:NGI DEM, river lines, epipolar DEM (derived from high-resolution stereo aerial photographs and satellite imagery) and LiDAR data. These improved editions of the SUDEM are expected to be released annually.

9 Voids Artifacts Figure 4 Hillshades of the (a) SRTM DEM, (b) Level 1, and (c) Level 2 products. 5.5.1 Limitations & future improvements SUDEM 2011 has a number of limitations which users should take into consideration. This section provides an overview of these limitations and how they will be addressed in future editions. 5.5.1.1 Contour and spot height errors Although much time and effort was spent on identifying and correcting attribute errors in the contour and spot height data, some errors may have gone unnoticed or uncorrected. Hopefully all remaining attribute errors will be corrected in future versions. The spatial errors that occasionally occur in the contour data were described in Section 4. These errors can only be corrected manually and is a very time-consuming and laborious process. Many of these errors (e.g. gaps) do not have a significant impact on the quality of

10 the interpolated DEM (because interpolation is a gap-filling process). However, most of the major errors were corrected in preparation of SUDEM 2011. Minor errors will be corrected as they are identified. Users are encouraged to bring such errors to our attention 3. 5.5.1.2 Interpolation artefacts Contours are discrete representations of a continuous surface (terrain) and are consequently not ideal sources of terrain data. Interpolating from contours will always produce artefacts (e.g. banding, tiger stripes, and wave effects). Although several techniques were employed to limit these artefacts, some are still present in both the SUDEM products. Users may choose to employ filtering techniques to reduce the effect of these artefacts. However, this will reduce the integrity of the data. Research is currently being carried out to find alternative methods to reduce these artefacts. 5.5.1.3 SRTM errors Only major errors (e.g. voids and spikes) in the SRTM DEM were corrected in SUDEM 2011. Other errors, such as banding and striping, were not corrected. Methods to minimise such errors are currently being developed and will be incorporated in future releases. 5.5.1.4 Hydrological errors Although some interpolation algorithms (such as ANUDEM) can produce hydrologically-corrected (HC) DEM, they often produce artefacts such as artificially deep river channels and gorges, particularly in areas of relatively moderate terrain. Because these artefacts can have a significant impact on some applications (e.g. geometrical and terrain correction of imagery), the Level1 and Level2 products of the SUDEM were not hydrologically corrected. However, a HC version of each product was created by removing sinks and peaks from the original DEM. These products, signified by a processing level indicator B, are available on request, but should be regarded as an experimental version as work is currently in progress to develop a more robust correction methodology by using ancillary data (e.g. stream networks and water bodies). 6 SUDEM LEVEL 3 The SUDEM Level 3 product is a 2m resolution digital surface model (DSM) extracted from the CDNGI 0.5m aerial photography (see Section 3.4). The product is currently only available for selected areas, but can be produced on demand for anywhere in South Africa. The Level 3 product provides significantly more detail than the other SUDEM product levels (see Figure 1) 3 Kindly email the locations (i.e. latitude and longitude) of suspected errors to avn@sun.ac.za.

11 Figure 5 Comparison of SUDEM Level 2 (top) and Level 3 (bottom) products The relative accuracy of the Level 3 product is a few centimetres, while the absolute accuracy is estimated at less than 1m. 7 SUDEM PRICES Developing the SUDEM was very costly as a lot of manual inputs were required (as described in the previous sections). The intention is to do more research on developing even more accurate and higher resolution DEMs of South Africa (and other countries). The funds raised from selling the SUDEM will go towards this end. All products can be ordered per square km, with a minimum order of 200km 2 for Level 1 & 2 products and 300km for the Level 3 product. The price structure of the SUDEM is provided in Table 3. At the highest resolution (5m), the Level 1 and 2 products can also be ordered per 1:50 000 sheet starting at R12000 per sheet. The cost exponentially decreases as larger volumes are purchased (to as little as R3000 per sheet). The cost for the entire Level 1 and 2 data set is R1.5 million. Costs are significantly reduced

12 when supplied at lower resolutions. Students and researchers can apply for an academic discount of up to 30% on the Level 1 and 2 products. Table 3 SUDEM prices (incl. VAT) # of 1:50000 map sheets Cost @ 2m resolution (Level 3) Cost @ 5m resolution (Level 1/2) Cost @ 20m resolution (Level 1/2) Cost @ 40m resolution (Level 1/2) Cost @ 60m resolution (Level 1/2) Cost @ 80m resolution (Level 1/2) 1 NA R 12 000 NA NA NA NA 2-10 NA R 7 500 NA NA NA NA 11-20 NA R 5 000 NA NA NA NA 21-50 NA R 4 000 NA NA NA NA >50 NA R 3 000 NA NA NA NA Provincial NA R 250 000 R 125 000 R 62 500 R 31 250 R 15 625 National NA R 1 500 000 R 500 000 R 250 000 R 125 000 R 62 500 Cost per R 50* R 20** square km *Minimum order of 300km 2 rectangular area **Minimum order of 100km 2 rectangular area 8 REFERENCES Australian National University 2011. ANUDEM Version 5.2 [online]. Canberra: Fenner School of Environment and Society. Available from http://fennerschool.anu.edu.au/publications/software/anudem.php [Accessed 20 July 2011]. Bolstad PV & Smith JL 1995. Errors in GIS: Assessing spatial data accuracy. In Lyon JG & McCarthy J (eds) Wetland and environmental applications in GIS, 301-312. New York: Lewis. Hengl T 2006. Finding the right pixel size. Computers & Geosciences 32: 1283. Hirt C, Filmer MS & Featherstone WE 2010. Comparison and validation of the recent freely available ASTER-GDEM ver1, SRTM ver4.1 and GEODATA DEM-9S ver3 digital elevation models over Australia. Australian Journal of Earth Sciences 57: 337-347. Rodriguez E, Morris CS, Belz JE, Chaplin EC, Martin JM, Daffer W & Hensley S 2005. An assessment of the SRTM topographic products. Pasadena: Jet Propulsion Laboratory. Van Niekerk A 2001. Western Cape Digital Elevation Model: Product description. Stellenbosch: Centre for Geographical Analysis, Stellenbosch University.