This booklet allows students to develop a macro landform classification procedure that has been used for mapping landforms around the world.
1. DEVELOPING LANDFORM MAPS USING ESRI’S ModelBuilder John M. Morgan, III, Ph.D., Professor of Geography and Director Ashley M. Lesh, GIS Specialist and Graduate Student Towson University Center for Geographic Information Sciences OVERVIEW AND OBJECTIVES Edward H. Hammond (1954, 1964a, 1964b) developed a macro landform classification procedure that has been used for mapping landforms around the world. Hammond’s classification is quantitative in nature with explicit definitions that can easily be applied by other researchers. Hammond’s procedure combines three important parameters—slope, relief, and profile type—to identify different landform, or terrain types. According to Hammond: Landform (terrain type) = Slope + Relief + Profile These landforms (terrain types) were subsequently grouped by Hammond into broader landform categories, such as nearly flat plains, rolling and irregular plains, plains with widely-spaced hills or mountains, partially dissected tablelands, hills, low mountains, and high mountains. Recently, Dikau (1989 and 1991) and True, et al. (n.d.) attempted to apply Hammond’s procedure using geographic information systems. These authors also modified Hammond’s three important parameters, and established their own groupings of broader landform categories. With this in mind, the objectives of this model are as follows: • Implement Hammond’s model using the U.S. Geological Survey’s 7.5 minute, 30-meter resolution National Elevation Dataset with ESRI’s ModelBuilder; • Generate both Dikau’s and True, et al.’s versions of Hammond’s landform maps; Completing these objectives required considerable testing of various neighborhood search radii for creation of various focal statistics used to create the slope, relief, and profile parameters. The model, per se, reflects this work in terms of the final radii used for all focal statistics calculations (i.e., 20 pixels). In addition to neighborhood radii search testing, we also spent a considerable amount of time checking all of the derived maps in this model to assure the accuracy of the model logic.
2. 2 In addition to our interest in landscape research, our purpose for developing this model is to provide an automated means for mapping landforms using NED data using ESRI’s ModelBuilder. Landform maps are useful for modeling erosion, characterizing watersheds, mapping “land units” for land management purposes, and for developing climate models (i.e., THE MODEL Initial (Pre-Model) Task • Reproject 7.5-Minute U.S. Geological Survey 30-meter resolution National Elevation Dataset (NED) data (1-arc second) to Maryland State Plane 83 meters (or whatever coordinate system is desired by the model user). Information regarding NED can be found at http://ned.usgs.gov/. NED data are available from a online seamless data distribution system or via CD. Initial Model Tasks • Create a map (Map1) with all non-zero elevation cells = 1 (reclassify NED so that all cells =1) • Run the sum (focal statistics) of Map1 within a 20 pixel radius circular window (Map2). The purpose of this step is to determine the number of cells within the 20 pixel radius circular window surrounding each pixel for use in later percentage calculations. The number of cells within the 20 pixel circular window surrounding each pixel is not constant due to the calculation of focal statistics along the border of NED data (or the clipped border of an irregular study area). • Create a floating point version of Map2 (Map3) via the MapAlgebra FLOAT operation. This map will be used later in the model for calculation of percentages within the 20 pixel radius circular window. Note that ArcGIS produces integer values in calculations involving integer maps. If one of the maps is floating point, ArcGIS will output floating point results from arithmetic calculations. Slope Sub-Model • Calculate slope from NED (Map4). • Reclassify slope (Map4) to the following categories: 0 Areas of greater than 8% slope 1 Areas of less than 8% slope
3. 3 This is the slope categories map (Map5). • Run the sum (focal statistics) of the slope categories map (Map5) within a 1.5 kilometer circular window (Map6). • Calculate the percent of near level land by dividing Map6 by Map3 (Map7). • Reclassify the percent of near level land (Map7) to the following categories: 400 0.00 - 0.2% 300 0.20 - 0.50% 200 0.50 - 0.80% 100 0.80 - 1.0% This is Hammond’s slope parameter map (Map8). Relief Sub-Model • Determine the maximum NED value within a 20 pixel circular window (Map9). • Determine the minimum NED value within a 20 pixel circular window (Map10). • Calculate the relief by subtracting the minimum NED value (Map10) from the maximum NED value (Map9) to create a relief map (Map11). • Reclassify the relief map (Map11) into the following categories: 10 0 - 30 meters 20 30 – 90 meters 30 90 – 150 meters 40 150 – 300 meters 50 300 – 900 meters 60 900 – 99999 meters This is Hammond’s relief parameter map (Map12). Profile Sub-Model • Calculate one-half of the maximum relief in the 20 pixel circular window by dividing the difference map (Map11) by 2 to create a local relief difference map (Map13).
4. 4 • Calculate the average local relief by adding the minimum NED value (Map10) to the local relief difference map (Map13) to create a profile value map (Map 14). • Calculate the difference between the original NED value and the profile value map by subtracting NED from Map14 to determine an upland/lowland map (Map15). Note: pixel values of less than 0 in Map15 represent upland areas; pixel values greater than 0 in Map15 represent lowland areas. • Reclassify the upland/lowland map (Map15) into the following categories: 1 >0 (lowland) 2 <0 (upland) This is the profile type map (Map16). • Reclassify the profile type map (Map16) into the following categories: 1 1 (lowland) 0 2 (upland) This is the lowlands map (Map17). • Identify gentle slopes (i.e., slopes less than 8%) in lowlands by multiplying Map5 by Map17 to create a gentle slopes in lowlands map (Map18). • Determine the sum (focal statistics) of the gentle slopes in lowlands map (Map18) map within a 20 pixel circular window (Map19). • Create a floating point version of Map6 (Map20) via the MapAlgebra FLOAT operation. Note: as indicated above, this map was created for the percentage calculation in the following step. Map6 (and its floating point equivalent (Map20) represent the sum of the gentle slopes (i.e., slopes less than 8%) within a 20 pixel circular window. • Calculate the percentage of gentle slopes in lowlands by dividing Map19 by Map20 (Map21). • Mask any uplands pixels from the gentle slopes in lowlands map (Map21) by multiplying the lowlands map (Map17) by the gentle slopes in lowlands map (Map21) to create as masked gentle slopes in lowlands map (Map 22). Note: this step is needed to isolate gentle slopes in lowlands because all pixels in Map21 include a percentage calculation. Without this step, the next step produces incorrect results.
5. 5 • Reclassify the percentage of gentle slopes in lowlands (Map22) to the following categories: 0 0.00 % 2 0.50 - 0.75% 1 0.75 - 1.00% This is the gentle slopes in lowlands parameter map (Map23). • Reclassify the profile type map (Map16) into the following categories: 0 1 (lowland) 1 2 (upland) This is the uplands map (Map24) • Identify gentle slopes (i.e., slopes less than 8%) in uplands by multiplying Map5 by Map24 to create a gentle slopes in uplands map (Map25). • Determine the sum (focal statistics) of the gentle slopes in uplands map (Map25) map within a 20 pixel circular window (Map26). • Calculate the percentage of gentle slopes in uplands by dividing Map26 by Map20 (Map27). • Mask any uplands pixels from the gentle slopes in uplands map (Map27) by multiplying the uplands map (Map24) by the gentle slopes in uplands map (Map27) to create as masked gentle slopes in uplands map (Map 28). Note: this step is needed to isolate gentle slopes in uplands because all pixels in Map27 include a percentage calculation. Without this step, the next step produces incorrect results. • Reclassify the percentage of gentle slopes in uplands Map28 to the following categories: 0 0.00 % 3 0.50 - 0.75% 4 0.75 - 1.00% This is the gentle slopes in uplands parameter map (Map29). • Calculate Hammond’s profile parameter map by adding the gentle slopes in lowlands map Map23 to the gentle slopes in uplands map Map29 (Map30). • Reclassify Hammond’s profile parameter map (Map30) to the following categories:
6. 6 1 0 This is an adjusted profile parameter map (Map31). Note: this step is necessary to remove several isolated cells with the value of 0 in Map23 and Map29. The reclassification procedures that create these maps, for some reason, leave cells with isolated values of 0. We have spent a great deal of time trying to diagnose the way ArcGIS is calculating the percentages and reclassifying the percentages in the model. We anticipate removing this step in a future iteration of the model. Landform Classification Sub-Model • Calculate the Hammond terrain type code by adding Hammond’s slope parameter map (Map8) to Hammond’s relief parameter map (Map12) to create a temporary landform code map (Map32). • Calculate the Hammond terrain type code by adding the temporary landform code map (Map32) to Hammond’s adjusted profile parameter map (Map31) to create the final Hammond’s terrain type code map (Map33). Given the coding sequence used to prepare the slope parameter map (Map8), the relief parameter map (Map12), and the adjusted profile parameter map (Map31), the codes on the terrain type code map can potential range from 111 to 464. In other words, the slope parameter map codes (100, 200, 300, 400), plus the relief parameter map codes (10, 20, 30, 40, 50, 60), plus the adjusted profile parameter map codes (1, 2, 3, 4). • Reclassify the final Hammond’s terrain types code map (Map33) into the following categories: 11 411-414 12 421-424 13 311-312 14 321-324 21 433-434, 333-334 22 443-444, 343-344 23 453-454, 353-354 24 463-464, 363-364 31 431-432, 331-332 32 441-442, 341-342 33 451-452, 351-352 34 461-462, 361-362 41 211-214 42 221-224 43 231-234 44 241-244 45 251-254
7. 7 46 261-264 51 111-114 52 121-124 53 131-134 54 141-144 55 151-154 56 161-164 This is the Dikau landform code map (Map34) • Smooth the Dikau landform code map (Map34) using a majority filter operation with 8 neighbors (Map35). Note: the purpose of this step is to remove any “salt-and-pepper” pixels within landform units on the map. Mask the Dikau landform code map (Map35) using the non-zero elevation cells map (Map1). Note: by multiplying the two maps Map1 serves as a binary mask and clips any codes beyond the study area (i.e., codes created due to the focal statistics operations) for Map34. The output from this step is the final Dikau landform code map (Map36). The meaning for the codes on this map is as follows: 11 Flat or nearly flat plains 12 Smooth plains with some local relief 13 Irregular plains with low relief 14 Irregular plains with moderate relief 21 Tablelands with moderate relief 22 Tablelands with considerable relief 23 Tablelands with high relief 24 Tablelands with very high relief 31 Plains with hills 32 Plains with high hills 33 Plains with low mountains 34 Plains with high mountains 41 Open very low hills 42 Open low hills 43 Open moderate hills 44 Open high hills 45 Open low mountains 46 Open high mountains 51 Very low hills 52 Low hills 53 Moderate Hills 54 High hills 55 Low mountains 56 High mountains • Convert the final Dikau landform code map (Map36) from raster to vector (polygon) format (Map37.shp).
8. 8 In addition to Dikau landform code map, we also developed another version of Hammond’s landform map using a procedure suggested by the Missouri Resource Assessment Partnership (MORAP) (http://www.cerc.cr.usgs.gov/morap/projects.asp?project_id=17). MORAP is an interagency partnership of the University of Missouri. A sample of the map created by MORAP can be found on the ESRI Web site at The following are the steps used to create this very generalized version of Hammond’s landform map: • Reclassify the relief map (Map11) into the following categories: 1 0 - 15 meters 2 15-30 meters 3 30 – 90 meters 4 90 – 150 meters 5 150 – 300 meters 6 300 – 900 meters 7 900 – 99999 meters This is the MORAP relief parameter map (Map38). • Reclassify the percent of near level land (Map7) to the following categories: 10 0.50 - 1.00% 20 0.00 - 0.50% This is the MORAP slope parameter map (Map39). • Calculate the MORAP landform category by adding MORAP slope parameter map (Map39) to the MORAP relief parameter map (Map38) to create the MORAP landform map (Map40). • Smooth the MORAP landform map (Map40) using a majority filter operation with 8 neighbors (Map41). Note: the purpose of this step is to remove any “salt-and-pepper” pixels within landform units on the map. • Mask the MORAP landform map (Map41) using the non-zero elevation cells map (Map1). Note: by multiplying the two maps Map1 serves as a binary mask and clips any codes beyond the study area (i.e., codes created due to the focal statistics operations) for Map41. The output from this step is the final MORAP landform map (Map42). The meaning for the codes on this map is as follows:
9. 9 11 Flat plains 12 Smooth plains 13 Irregular plains 14 Plains with low hills 15 Plains with hills 16 Plains with low mountains 17 Plains with mountains 21 Rough plains 22 Rugged plains 23 Breaks 24 Low hills 25 Hills 26 Low mountains 27 Mountains • Convert the final MORAP landform map (Map42) from raster to vector (polygon) format (Map43.shp). • We also included an additional step in the model to generate an analytical hillshade for subsequent (non-model) mapping purposes (Hillshade). This procedure uses the defaults on options for this tool. The two maps shown in this document represent the final Dikau landform code map in vector format (Map37.shp) overlayed on the Hillshade map, and the final MORAP landform map in vector format (Map43.shp) overlayed on the Hillshade map. SCRIPTS/ALGORITHMS USED None. DATA USED Only one dataset is needed for use with this model, the 7.5-Minute U.S. Geological Survey 30-meter resolution National Elevation Dataset (NED) data (1- arc second) to Maryland State Plane 83 meters (or whatever coordinate system is desired by the model user). Information regarding NED can be found at http://ned.usgs.gov/. NED datasets are available for download from a online seamless data distribution system, or are available on CD from the Earth Resources Observations Systems (EROS) Data Center, Sioux Falls, South
10. 10 SUGGESTIONS FOR USING THE MODEL The following are several suggestions with regard to running our landform • We have run this model successfully using 30-meter resolution NED data for a single U.S. Geological Survey 7.5 minute quadrangle, for a single county, and for an entire state (Maryland). This model can be replicated nationwide given the availability of the 30-meter NED data. No attempt was made to test the model using either 10-meter resolution NED data, or smaller-scale DEM data, such as the U.S. Geological Survey’s 3-arc second DEM data. • We selected a search radius of 20 pixels for the focal statistics operations used in this model. This radius was based on suggestions in the literature, as well as considerable trial-and-error model testing. Furthermore, we used a circular “neighborhood” for the focal statistics operations. This number corresponds to recommendations by Hammond, MORAP, and others. The user can change this radius in order to see the effect of neighborhood size on landform classification. However, the user should keep in mind that increasing the size of the pixel search radius will correspondingly increase the time required to process the model. • Depending on the geographic area being mapped, as well as the local topography within the geographic area, not all Hammond, Dikau, or True, et al. (i.e., Missouri Resource Assessment Partnership, or MORAP) codes may be reflected in the final maps. Keep in mind that both landform classifications are designed for worldwide use. The geographic area you are mapping may have a limited range of topographic features due to limited elevation, slope, relief, or profile. • The model can be readily modified to incorporate changes to the slope, relief, and profile parameters, or to the grouping of the landform codes. • The model includes documentation of all of the steps described above. FUTURE Our work on this model is part of a continuing effort to characterize the landscape of the Mid-Atlantic region. We consider the model we submitted for the Best Practices in Science Modeling competition to be the first version of an ongoing landform modeling effort. We intend to continue our work to refine our model to better characterize landforms and incorporate the work of other researchers. Also, we will be developing a landform map of Maryland using the
11. 11 model that we will submit for the Map Gallery for the ESRI International User Conference in July. REFERENCES CITED Dikau, R. 1989. The application of a digital relief model to landform analysis. In Three Dimensional Applications in Geographical Information Systems, ed. J. F. Raper, 51-77. London: Taylor and Francis. Dikau, R., E. E. Brabb, and R. M Mark. 1991. Landform classification of New Mexico by Computer. U.S. Geological Survey Open File Report 91-634. Hammond, E. H. 1954. Small scale continental landform maps. Annals of the Association of American Geographers 44: 32-42. Hammond, E. H. 1964a. Analysis of properties in landform geography: An application to broadscale landform mapping. Annals of the Association of American Geographers 54(1):11-19. Hammond, E. H. 1964b. Classes of land surface form in the forty-eight states, U.S.A. Annals of the Association of American Geographers 54(1): map supplement True, C. D., T. Gordon, and D. Diamond. n.d. How the size of a sliding window impacts the generation of landforms. PowerPoint presentation on the Missouri Resource Assessment Partnership’s Web site at: http://www.cerc.cr.usgs.gov/morap/projects.asp?project_id=17&project_name =Landform%20Modeling&project_directory=landform_model.
+1-315-636-4485
+91 98151-74050 
