Creating a Map of Terrain Regions for using Digital Elevation Models

Craig Allison School of Geography & Earth Sciences, McMaster University, Hamilton, ON, Canada

1. Introduction 3. Results 3. Results (continued) The primary purpose of this project is to divide Italy into a Table 2: Comparison of DEMs by % of cells per terrain region type for Italy map of terrain regions using digital elevation models (DEMs) and a fully automated approach. Automation of % of Cells in Each Terrain Region Type traditional (manual) landform classification mapping is appealing because it is consistent, repeatable, updatable, DEM Plains w/ Hills and High Plains Tablelands hills or low and quantifiable. Terrain regions themselves are important Product mountains because they define boundary conditions for mountains mountains geomorphologic, hydrologic, ecologic, and pedologic ASTER 22.7 1.2 6.1 58.4 11.7 processes [1]. Further, they impact where human activities take place on the landscape. Italy has been chosen as the SRTM 24.6 1.6 8.9 53.3 11.6 study area because the most current effort at delineating (30m) terrain regions [2] can be significantly improved. SRTM 24.9 2.1 10.9 51.0 11.2 Specifically, these regions were derived using coarse (90m) resolution DEMs, highly correlated morphometric variables, and manual qualitative techniques that are not SRTM 30m and SRTM 90m DEM terrain region maps have repeatable. closer percentages of cells assigned to plains, plains with hills or mountains, and hills and low mountains (i.e. 3/5 This study has four objectives: terrain region types) (Table 2). Of all five terrain region 1. To create an updated and improved landform types, the percent values (Table 2) assigned to plains and classification map of Italy. high mountains are most similar between the three DEM 2. To create a tool using Esri technology to automate the products. Therefore, the type of DEM product used for a approach so that it can be applied to other project matters less when you are identifying plains and jurisdictions. high mountains. More generally, categories at the 3. To examine the impact that DEM products created extremes of elevation are less affected by resolution or from different satellite technologies have on the the DEM product used. Projects involving plains and high results. mountains should be less concerned about the specific 4. To explore the impact that resolution has on the DEM they are using. resulting classified map. Figure 2: Italian terrain regions derived from a 90m resolution SRTM Void Filled DEM Figure 4: Erroneous classification results when the ASTER DEM was used The at Italy’s northern border have been classified as “high mountains,” the Po Valley as “plains,” most of the as “hills and low mountains,” and 2. Data and Methods most of and Sardinia as hills and low mountains. Three DEM products – ASTER (a passive along-track scanning Note that prominent features like Corno Grande (tallest system with 30m resolution), SRTM (30m), and SRTM (90m) peak in the Apennines; located in central Mainland Italy) (an active scanning system, based on radar interferometry) – and (tall volcano in eastern Sicily) have been were retrieved using a combination of the U.S. Geological successfully identified as high mountains. Features like Survey’s EarthExplorer tool (earthexplorer.usgs.gov/) and the the Po Valley in the north, Campidano plain in southwest Consortium for Spatial Information (cgiar-csi.org/). Sardinia, and the Tavoliere in Puglia have been nicely extracted from the DEM as well.

The SRTM DEMs appear to classify terrain more accurately than the ASTER DEM (Figure 4). In Piemonte, for example, the ASTER DEM classification result indicates hills and low mountains where there are none (just north of the Monferrato Hills). The same is true for the southeastern most area of Apulia. There should also be no significant highland areas in the Po Plain (Figure 4 shows the Po Plain in Lombardia), but there are in the ASTER DEM classification result. SRTM DEM results did not contain these errors. Figure 1: ASTER, SRTM (30 m), and SRTM (90 m) DEMs for Italy The model for the landform classification was constructed based on the work of Hammond [3], Dikau [4], Morgan & Lesh [5], and Drescher & de Frey [6]. In the model, created in ModelBuilder, a circular window of 1.8km radius was used to 4. Conclusions calculate the percent of area occupied by gentle inclination (<8% slope gradient), the relief, and the percent of the gently A model for automatically dividing Italy into terrain sloping terrain occurring in the lower half of the local relief regions has been constructed using ModelBuilder. [7]. One step was added to the original model that solved a Results from the model show that the terrain region problem with the creation of NoData cells in the Alps. classification depends more on the type of DEM used Specifically, a value of 1 was added to the denominator map in and less on the resolution of the DEM product. one of the divide steps. Five main types of terrain regions are created based on merging 24 terrain types described by Morgan & Lesh (Table 1). In total the model required 38 steps to produce the results. 5. References Figure 3: Model results for three different DEM products for Italy Table 1: Assigning terrain types to terrain regions [1] Drăguţ, L., & Eisank, C. (2011). Automated classification of topography from SRTM data using object-based image analysis. Geomorphometry 2011, 7-9. Morgan & Lesh (2005) Corresponding Terrain Region [2] Guzzetti, F., & Reichenbach, P. (1994). Towards a definition of topographic Terrain Type The type of DEM (i.e. ASTER vs. SRTM) appears to have a divisions for Italy. Geomorphology, 11(1), 57-74. 11, 12, 13, 14 Plains [3] Hammond, E. H. (1964). Analysis of properties in land form geography: an greater influence on the classification result than does application to broad‐scale land form mapping. Annals of the Association of 21, 22, 23, 24 Tablelands resolution (i.e. 30m vs. 90m). More specifically, there is American Geographers, 54(1), 11-19. [4] Dikau, R. (1989). The application of a digital relief model to landform analysis visually more of a difference between the ASTER and 31, 32, 33, 34 Plains with hills or mountains in geomorphology. In: Raper, J. (ed.) Three dimensional applications in either SRTM map than there is between the SRTM (30m) geographical information systems, Taylor & Francis, London, UK 51-77. 41, 42, 43, 44, 45, 51, 52, 53, Hills and low mountains [5] Morgan, J. M., & Lesh, A. M. (2005). Developing landform maps using ESRI’S and SRTM (90m) maps. Model-Builder. In ESRI International User Conference. 54, 55 [6] Drescher, K., & de Frey, W. (2009). Landform classification using GIS. Position IT. 46, 56 High mountains [7] Gallant, A. L., Brown, D. D., & Hoffer, R. M. (2005). Automated mapping of Hammond's landforms. Geoscience and Remote Sensing Letters, IEEE, 2(4), 384-388.